]>
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 | ||
4cc96131 MS |
8 | #include "dm-core.h" |
9 | #include "dm-rq.h" | |
51e5b2bd | 10 | #include "dm-uevent.h" |
1da177e4 LT |
11 | |
12 | #include <linux/init.h> | |
13 | #include <linux/module.h> | |
48c9c27b | 14 | #include <linux/mutex.h> |
1da177e4 LT |
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> |
ffcc3936 | 22 | #include <linux/wait.h> |
71cdb697 | 23 | #include <linux/pr.h> |
55782138 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
71a16736 NK |
27 | #ifdef CONFIG_PRINTK |
28 | /* | |
29 | * ratelimit state to be used in DMXXX_LIMIT(). | |
30 | */ | |
31 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
32 | DEFAULT_RATELIMIT_INTERVAL, | |
33 | DEFAULT_RATELIMIT_BURST); | |
34 | EXPORT_SYMBOL(dm_ratelimit_state); | |
35 | #endif | |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
1da177e4 LT |
44 | static const char *_name = DM_NAME; |
45 | ||
46 | static unsigned int major = 0; | |
47 | static unsigned int _major = 0; | |
48 | ||
d15b774c AK |
49 | static DEFINE_IDR(_minor_idr); |
50 | ||
f32c10b0 | 51 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
52 | |
53 | static void do_deferred_remove(struct work_struct *w); | |
54 | ||
55 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
56 | ||
acfe0ad7 MP |
57 | static struct workqueue_struct *deferred_remove_workqueue; |
58 | ||
1da177e4 LT |
59 | /* |
60 | * One of these is allocated per bio. | |
61 | */ | |
62 | struct dm_io { | |
63 | struct mapped_device *md; | |
64 | int error; | |
1da177e4 | 65 | atomic_t io_count; |
6ae2fa67 | 66 | struct bio *bio; |
3eaf840e | 67 | unsigned long start_time; |
f88fb981 | 68 | spinlock_t endio_lock; |
fd2ed4d2 | 69 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
70 | }; |
71 | ||
ba61fdd1 JM |
72 | #define MINOR_ALLOCED ((void *)-1) |
73 | ||
1da177e4 LT |
74 | /* |
75 | * Bits for the md->flags field. | |
76 | */ | |
1eb787ec | 77 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 78 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 79 | #define DMF_FROZEN 2 |
fba9f90e | 80 | #define DMF_FREEING 3 |
5c6bd75d | 81 | #define DMF_DELETING 4 |
2e93ccc1 | 82 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
83 | #define DMF_DEFERRED_REMOVE 6 |
84 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 85 | |
115485e8 | 86 | #define DM_NUMA_NODE NUMA_NO_NODE |
115485e8 | 87 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 88 | |
e6ee8c0b KU |
89 | /* |
90 | * For mempools pre-allocation at the table loading time. | |
91 | */ | |
92 | struct dm_md_mempools { | |
93 | mempool_t *io_pool; | |
e6ee8c0b KU |
94 | struct bio_set *bs; |
95 | }; | |
96 | ||
86f1152b BM |
97 | struct table_device { |
98 | struct list_head list; | |
99 | atomic_t count; | |
100 | struct dm_dev dm_dev; | |
101 | }; | |
102 | ||
e18b890b | 103 | static struct kmem_cache *_io_cache; |
8fbf26ad | 104 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 105 | static struct kmem_cache *_rq_cache; |
94818742 | 106 | |
e8603136 MS |
107 | /* |
108 | * Bio-based DM's mempools' reserved IOs set by the user. | |
109 | */ | |
4cc96131 | 110 | #define RESERVED_BIO_BASED_IOS 16 |
e8603136 MS |
111 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
112 | ||
115485e8 MS |
113 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
114 | { | |
115 | int param = ACCESS_ONCE(*module_param); | |
116 | int modified_param = 0; | |
117 | bool modified = true; | |
118 | ||
119 | if (param < min) | |
120 | modified_param = min; | |
121 | else if (param > max) | |
122 | modified_param = max; | |
123 | else | |
124 | modified = false; | |
125 | ||
126 | if (modified) { | |
127 | (void)cmpxchg(module_param, param, modified_param); | |
128 | param = modified_param; | |
129 | } | |
130 | ||
131 | return param; | |
132 | } | |
133 | ||
4cc96131 MS |
134 | unsigned __dm_get_module_param(unsigned *module_param, |
135 | unsigned def, unsigned max) | |
f4790826 | 136 | { |
09c2d531 MS |
137 | unsigned param = ACCESS_ONCE(*module_param); |
138 | unsigned modified_param = 0; | |
f4790826 | 139 | |
09c2d531 MS |
140 | if (!param) |
141 | modified_param = def; | |
142 | else if (param > max) | |
143 | modified_param = max; | |
f4790826 | 144 | |
09c2d531 MS |
145 | if (modified_param) { |
146 | (void)cmpxchg(module_param, param, modified_param); | |
147 | param = modified_param; | |
f4790826 MS |
148 | } |
149 | ||
09c2d531 | 150 | return param; |
f4790826 MS |
151 | } |
152 | ||
e8603136 MS |
153 | unsigned dm_get_reserved_bio_based_ios(void) |
154 | { | |
09c2d531 | 155 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 156 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
157 | } |
158 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
159 | ||
115485e8 MS |
160 | static unsigned dm_get_numa_node(void) |
161 | { | |
162 | return __dm_get_module_param_int(&dm_numa_node, | |
163 | DM_NUMA_NODE, num_online_nodes() - 1); | |
164 | } | |
165 | ||
1da177e4 LT |
166 | static int __init local_init(void) |
167 | { | |
51157b4a | 168 | int r = -ENOMEM; |
1da177e4 | 169 | |
1da177e4 | 170 | /* allocate a slab for the dm_ios */ |
028867ac | 171 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 172 | if (!_io_cache) |
51157b4a | 173 | return r; |
1da177e4 | 174 | |
8fbf26ad KU |
175 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
176 | if (!_rq_tio_cache) | |
dba14160 | 177 | goto out_free_io_cache; |
8fbf26ad | 178 | |
eca7ee6d | 179 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
1ae49ea2 MS |
180 | __alignof__(struct request), 0, NULL); |
181 | if (!_rq_cache) | |
182 | goto out_free_rq_tio_cache; | |
183 | ||
51e5b2bd | 184 | r = dm_uevent_init(); |
51157b4a | 185 | if (r) |
1ae49ea2 | 186 | goto out_free_rq_cache; |
51e5b2bd | 187 | |
acfe0ad7 MP |
188 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
189 | if (!deferred_remove_workqueue) { | |
190 | r = -ENOMEM; | |
191 | goto out_uevent_exit; | |
192 | } | |
193 | ||
1da177e4 LT |
194 | _major = major; |
195 | r = register_blkdev(_major, _name); | |
51157b4a | 196 | if (r < 0) |
acfe0ad7 | 197 | goto out_free_workqueue; |
1da177e4 LT |
198 | |
199 | if (!_major) | |
200 | _major = r; | |
201 | ||
202 | return 0; | |
51157b4a | 203 | |
acfe0ad7 MP |
204 | out_free_workqueue: |
205 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
206 | out_uevent_exit: |
207 | dm_uevent_exit(); | |
1ae49ea2 MS |
208 | out_free_rq_cache: |
209 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
210 | out_free_rq_tio_cache: |
211 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
212 | out_free_io_cache: |
213 | kmem_cache_destroy(_io_cache); | |
214 | ||
215 | return r; | |
1da177e4 LT |
216 | } |
217 | ||
218 | static void local_exit(void) | |
219 | { | |
2c140a24 | 220 | flush_scheduled_work(); |
acfe0ad7 | 221 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 222 | |
1ae49ea2 | 223 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 224 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 225 | kmem_cache_destroy(_io_cache); |
00d59405 | 226 | unregister_blkdev(_major, _name); |
51e5b2bd | 227 | dm_uevent_exit(); |
1da177e4 LT |
228 | |
229 | _major = 0; | |
230 | ||
231 | DMINFO("cleaned up"); | |
232 | } | |
233 | ||
b9249e55 | 234 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
235 | local_init, |
236 | dm_target_init, | |
237 | dm_linear_init, | |
238 | dm_stripe_init, | |
952b3557 | 239 | dm_io_init, |
945fa4d2 | 240 | dm_kcopyd_init, |
1da177e4 | 241 | dm_interface_init, |
fd2ed4d2 | 242 | dm_statistics_init, |
1da177e4 LT |
243 | }; |
244 | ||
b9249e55 | 245 | static void (*_exits[])(void) = { |
1da177e4 LT |
246 | local_exit, |
247 | dm_target_exit, | |
248 | dm_linear_exit, | |
249 | dm_stripe_exit, | |
952b3557 | 250 | dm_io_exit, |
945fa4d2 | 251 | dm_kcopyd_exit, |
1da177e4 | 252 | dm_interface_exit, |
fd2ed4d2 | 253 | dm_statistics_exit, |
1da177e4 LT |
254 | }; |
255 | ||
256 | static int __init dm_init(void) | |
257 | { | |
258 | const int count = ARRAY_SIZE(_inits); | |
259 | ||
260 | int r, i; | |
261 | ||
262 | for (i = 0; i < count; i++) { | |
263 | r = _inits[i](); | |
264 | if (r) | |
265 | goto bad; | |
266 | } | |
267 | ||
268 | return 0; | |
269 | ||
270 | bad: | |
271 | while (i--) | |
272 | _exits[i](); | |
273 | ||
274 | return r; | |
275 | } | |
276 | ||
277 | static void __exit dm_exit(void) | |
278 | { | |
279 | int i = ARRAY_SIZE(_exits); | |
280 | ||
281 | while (i--) | |
282 | _exits[i](); | |
d15b774c AK |
283 | |
284 | /* | |
285 | * Should be empty by this point. | |
286 | */ | |
d15b774c | 287 | idr_destroy(&_minor_idr); |
1da177e4 LT |
288 | } |
289 | ||
290 | /* | |
291 | * Block device functions | |
292 | */ | |
432a212c MA |
293 | int dm_deleting_md(struct mapped_device *md) |
294 | { | |
295 | return test_bit(DMF_DELETING, &md->flags); | |
296 | } | |
297 | ||
fe5f9f2c | 298 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
299 | { |
300 | struct mapped_device *md; | |
301 | ||
fba9f90e JM |
302 | spin_lock(&_minor_lock); |
303 | ||
fe5f9f2c | 304 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
305 | if (!md) |
306 | goto out; | |
307 | ||
5c6bd75d | 308 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 309 | dm_deleting_md(md)) { |
fba9f90e JM |
310 | md = NULL; |
311 | goto out; | |
312 | } | |
313 | ||
1da177e4 | 314 | dm_get(md); |
5c6bd75d | 315 | atomic_inc(&md->open_count); |
fba9f90e JM |
316 | out: |
317 | spin_unlock(&_minor_lock); | |
318 | ||
319 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
320 | } |
321 | ||
db2a144b | 322 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 323 | { |
63a4f065 | 324 | struct mapped_device *md; |
6e9624b8 | 325 | |
4a1aeb98 MB |
326 | spin_lock(&_minor_lock); |
327 | ||
63a4f065 MS |
328 | md = disk->private_data; |
329 | if (WARN_ON(!md)) | |
330 | goto out; | |
331 | ||
2c140a24 MP |
332 | if (atomic_dec_and_test(&md->open_count) && |
333 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 334 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 335 | |
1da177e4 | 336 | dm_put(md); |
63a4f065 | 337 | out: |
4a1aeb98 | 338 | spin_unlock(&_minor_lock); |
1da177e4 LT |
339 | } |
340 | ||
5c6bd75d AK |
341 | int dm_open_count(struct mapped_device *md) |
342 | { | |
343 | return atomic_read(&md->open_count); | |
344 | } | |
345 | ||
346 | /* | |
347 | * Guarantees nothing is using the device before it's deleted. | |
348 | */ | |
2c140a24 | 349 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
350 | { |
351 | int r = 0; | |
352 | ||
353 | spin_lock(&_minor_lock); | |
354 | ||
2c140a24 | 355 | if (dm_open_count(md)) { |
5c6bd75d | 356 | r = -EBUSY; |
2c140a24 MP |
357 | if (mark_deferred) |
358 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
359 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
360 | r = -EEXIST; | |
5c6bd75d AK |
361 | else |
362 | set_bit(DMF_DELETING, &md->flags); | |
363 | ||
364 | spin_unlock(&_minor_lock); | |
365 | ||
366 | return r; | |
367 | } | |
368 | ||
2c140a24 MP |
369 | int dm_cancel_deferred_remove(struct mapped_device *md) |
370 | { | |
371 | int r = 0; | |
372 | ||
373 | spin_lock(&_minor_lock); | |
374 | ||
375 | if (test_bit(DMF_DELETING, &md->flags)) | |
376 | r = -EBUSY; | |
377 | else | |
378 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
379 | ||
380 | spin_unlock(&_minor_lock); | |
381 | ||
382 | return r; | |
383 | } | |
384 | ||
385 | static void do_deferred_remove(struct work_struct *w) | |
386 | { | |
387 | dm_deferred_remove(); | |
388 | } | |
389 | ||
fd2ed4d2 MP |
390 | sector_t dm_get_size(struct mapped_device *md) |
391 | { | |
392 | return get_capacity(md->disk); | |
393 | } | |
394 | ||
9974fa2c MS |
395 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
396 | { | |
397 | return md->queue; | |
398 | } | |
399 | ||
fd2ed4d2 MP |
400 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
401 | { | |
402 | return &md->stats; | |
403 | } | |
404 | ||
3ac51e74 DW |
405 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
406 | { | |
407 | struct mapped_device *md = bdev->bd_disk->private_data; | |
408 | ||
409 | return dm_get_geometry(md, geo); | |
410 | } | |
411 | ||
956a4025 MS |
412 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
413 | struct block_device **bdev, | |
414 | fmode_t *mode) | |
aa129a22 | 415 | { |
66482026 | 416 | struct dm_target *tgt; |
6c182cd8 | 417 | struct dm_table *map; |
956a4025 | 418 | int srcu_idx, r; |
aa129a22 | 419 | |
6c182cd8 | 420 | retry: |
e56f81e0 | 421 | r = -ENOTTY; |
956a4025 | 422 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
423 | if (!map || !dm_table_get_size(map)) |
424 | goto out; | |
425 | ||
426 | /* We only support devices that have a single target */ | |
427 | if (dm_table_get_num_targets(map) != 1) | |
428 | goto out; | |
429 | ||
66482026 MS |
430 | tgt = dm_table_get_target(map, 0); |
431 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 432 | goto out; |
aa129a22 | 433 | |
4f186f8b | 434 | if (dm_suspended_md(md)) { |
aa129a22 MB |
435 | r = -EAGAIN; |
436 | goto out; | |
437 | } | |
438 | ||
66482026 | 439 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
440 | if (r < 0) |
441 | goto out; | |
aa129a22 | 442 | |
956a4025 MS |
443 | bdgrab(*bdev); |
444 | dm_put_live_table(md, srcu_idx); | |
e56f81e0 | 445 | return r; |
aa129a22 | 446 | |
aa129a22 | 447 | out: |
956a4025 | 448 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 449 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
450 | msleep(10); |
451 | goto retry; | |
452 | } | |
e56f81e0 CH |
453 | return r; |
454 | } | |
455 | ||
456 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
457 | unsigned int cmd, unsigned long arg) | |
458 | { | |
459 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 460 | int r; |
e56f81e0 | 461 | |
956a4025 | 462 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
463 | if (r < 0) |
464 | return r; | |
6c182cd8 | 465 | |
e56f81e0 CH |
466 | if (r > 0) { |
467 | /* | |
468 | * Target determined this ioctl is being issued against | |
469 | * a logical partition of the parent bdev; so extra | |
470 | * validation is needed. | |
471 | */ | |
472 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
473 | if (r) | |
474 | goto out; | |
475 | } | |
6c182cd8 | 476 | |
66482026 | 477 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 478 | out: |
956a4025 | 479 | bdput(bdev); |
aa129a22 MB |
480 | return r; |
481 | } | |
482 | ||
028867ac | 483 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
484 | { |
485 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
486 | } | |
487 | ||
028867ac | 488 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
489 | { |
490 | mempool_free(io, md->io_pool); | |
491 | } | |
492 | ||
cfae7529 | 493 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 494 | { |
dba14160 | 495 | bio_put(&tio->clone); |
1da177e4 LT |
496 | } |
497 | ||
4cc96131 | 498 | int md_in_flight(struct mapped_device *md) |
90abb8c4 KU |
499 | { |
500 | return atomic_read(&md->pending[READ]) + | |
501 | atomic_read(&md->pending[WRITE]); | |
502 | } | |
503 | ||
3eaf840e JNN |
504 | static void start_io_acct(struct dm_io *io) |
505 | { | |
506 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 507 | struct bio *bio = io->bio; |
c9959059 | 508 | int cpu; |
fd2ed4d2 | 509 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
510 | |
511 | io->start_time = jiffies; | |
512 | ||
074a7aca TH |
513 | cpu = part_stat_lock(); |
514 | part_round_stats(cpu, &dm_disk(md)->part0); | |
515 | part_stat_unlock(); | |
1e9bb880 SL |
516 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
517 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
518 | |
519 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
520 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
521 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
522 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
523 | } |
524 | ||
d221d2e7 | 525 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
526 | { |
527 | struct mapped_device *md = io->md; | |
528 | struct bio *bio = io->bio; | |
529 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 530 | int pending; |
3eaf840e JNN |
531 | int rw = bio_data_dir(bio); |
532 | ||
18c0b223 | 533 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 534 | |
fd2ed4d2 | 535 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
536 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
537 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
538 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 539 | |
af7e466a MP |
540 | /* |
541 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 542 | * a flush. |
af7e466a | 543 | */ |
1e9bb880 SL |
544 | pending = atomic_dec_return(&md->pending[rw]); |
545 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 546 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 547 | |
d221d2e7 MP |
548 | /* nudge anyone waiting on suspend queue */ |
549 | if (!pending) | |
550 | wake_up(&md->wait); | |
3eaf840e JNN |
551 | } |
552 | ||
1da177e4 LT |
553 | /* |
554 | * Add the bio to the list of deferred io. | |
555 | */ | |
92c63902 | 556 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 557 | { |
05447420 | 558 | unsigned long flags; |
1da177e4 | 559 | |
05447420 | 560 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 561 | bio_list_add(&md->deferred, bio); |
05447420 | 562 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 563 | queue_work(md->wq, &md->work); |
1da177e4 LT |
564 | } |
565 | ||
566 | /* | |
567 | * Everyone (including functions in this file), should use this | |
568 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 569 | * dm_put_live_table() when finished. |
1da177e4 | 570 | */ |
83d5e5b0 | 571 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 572 | { |
83d5e5b0 MP |
573 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
574 | ||
575 | return srcu_dereference(md->map, &md->io_barrier); | |
576 | } | |
1da177e4 | 577 | |
83d5e5b0 MP |
578 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
579 | { | |
580 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
581 | } | |
582 | ||
583 | void dm_sync_table(struct mapped_device *md) | |
584 | { | |
585 | synchronize_srcu(&md->io_barrier); | |
586 | synchronize_rcu_expedited(); | |
587 | } | |
588 | ||
589 | /* | |
590 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
591 | * The caller must not block between these two functions. | |
592 | */ | |
593 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
594 | { | |
595 | rcu_read_lock(); | |
596 | return rcu_dereference(md->map); | |
597 | } | |
1da177e4 | 598 | |
83d5e5b0 MP |
599 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
600 | { | |
601 | rcu_read_unlock(); | |
1da177e4 LT |
602 | } |
603 | ||
86f1152b BM |
604 | /* |
605 | * Open a table device so we can use it as a map destination. | |
606 | */ | |
607 | static int open_table_device(struct table_device *td, dev_t dev, | |
608 | struct mapped_device *md) | |
609 | { | |
610 | static char *_claim_ptr = "I belong to device-mapper"; | |
611 | struct block_device *bdev; | |
612 | ||
613 | int r; | |
614 | ||
615 | BUG_ON(td->dm_dev.bdev); | |
616 | ||
617 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
618 | if (IS_ERR(bdev)) | |
619 | return PTR_ERR(bdev); | |
620 | ||
621 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
622 | if (r) { | |
623 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
624 | return r; | |
625 | } | |
626 | ||
627 | td->dm_dev.bdev = bdev; | |
628 | return 0; | |
629 | } | |
630 | ||
631 | /* | |
632 | * Close a table device that we've been using. | |
633 | */ | |
634 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
635 | { | |
636 | if (!td->dm_dev.bdev) | |
637 | return; | |
638 | ||
639 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
640 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
641 | td->dm_dev.bdev = NULL; | |
642 | } | |
643 | ||
644 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
645 | fmode_t mode) { | |
646 | struct table_device *td; | |
647 | ||
648 | list_for_each_entry(td, l, list) | |
649 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
650 | return td; | |
651 | ||
652 | return NULL; | |
653 | } | |
654 | ||
655 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
656 | struct dm_dev **result) { | |
657 | int r; | |
658 | struct table_device *td; | |
659 | ||
660 | mutex_lock(&md->table_devices_lock); | |
661 | td = find_table_device(&md->table_devices, dev, mode); | |
662 | if (!td) { | |
115485e8 | 663 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
664 | if (!td) { |
665 | mutex_unlock(&md->table_devices_lock); | |
666 | return -ENOMEM; | |
667 | } | |
668 | ||
669 | td->dm_dev.mode = mode; | |
670 | td->dm_dev.bdev = NULL; | |
671 | ||
672 | if ((r = open_table_device(td, dev, md))) { | |
673 | mutex_unlock(&md->table_devices_lock); | |
674 | kfree(td); | |
675 | return r; | |
676 | } | |
677 | ||
678 | format_dev_t(td->dm_dev.name, dev); | |
679 | ||
680 | atomic_set(&td->count, 0); | |
681 | list_add(&td->list, &md->table_devices); | |
682 | } | |
683 | atomic_inc(&td->count); | |
684 | mutex_unlock(&md->table_devices_lock); | |
685 | ||
686 | *result = &td->dm_dev; | |
687 | return 0; | |
688 | } | |
689 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
690 | ||
691 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
692 | { | |
693 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
694 | ||
695 | mutex_lock(&md->table_devices_lock); | |
696 | if (atomic_dec_and_test(&td->count)) { | |
697 | close_table_device(td, md); | |
698 | list_del(&td->list); | |
699 | kfree(td); | |
700 | } | |
701 | mutex_unlock(&md->table_devices_lock); | |
702 | } | |
703 | EXPORT_SYMBOL(dm_put_table_device); | |
704 | ||
705 | static void free_table_devices(struct list_head *devices) | |
706 | { | |
707 | struct list_head *tmp, *next; | |
708 | ||
709 | list_for_each_safe(tmp, next, devices) { | |
710 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
711 | ||
712 | DMWARN("dm_destroy: %s still exists with %d references", | |
713 | td->dm_dev.name, atomic_read(&td->count)); | |
714 | kfree(td); | |
715 | } | |
716 | } | |
717 | ||
3ac51e74 DW |
718 | /* |
719 | * Get the geometry associated with a dm device | |
720 | */ | |
721 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
722 | { | |
723 | *geo = md->geometry; | |
724 | ||
725 | return 0; | |
726 | } | |
727 | ||
728 | /* | |
729 | * Set the geometry of a device. | |
730 | */ | |
731 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
732 | { | |
733 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
734 | ||
735 | if (geo->start > sz) { | |
736 | DMWARN("Start sector is beyond the geometry limits."); | |
737 | return -EINVAL; | |
738 | } | |
739 | ||
740 | md->geometry = *geo; | |
741 | ||
742 | return 0; | |
743 | } | |
744 | ||
1da177e4 LT |
745 | /*----------------------------------------------------------------- |
746 | * CRUD START: | |
747 | * A more elegant soln is in the works that uses the queue | |
748 | * merge fn, unfortunately there are a couple of changes to | |
749 | * the block layer that I want to make for this. So in the | |
750 | * interests of getting something for people to use I give | |
751 | * you this clearly demarcated crap. | |
752 | *---------------------------------------------------------------*/ | |
753 | ||
2e93ccc1 KU |
754 | static int __noflush_suspending(struct mapped_device *md) |
755 | { | |
756 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
757 | } | |
758 | ||
1da177e4 LT |
759 | /* |
760 | * Decrements the number of outstanding ios that a bio has been | |
761 | * cloned into, completing the original io if necc. | |
762 | */ | |
858119e1 | 763 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 764 | { |
2e93ccc1 | 765 | unsigned long flags; |
b35f8caa MB |
766 | int io_error; |
767 | struct bio *bio; | |
768 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
769 | |
770 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
771 | if (unlikely(error)) { |
772 | spin_lock_irqsave(&io->endio_lock, flags); | |
773 | if (!(io->error > 0 && __noflush_suspending(md))) | |
774 | io->error = error; | |
775 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
776 | } | |
1da177e4 LT |
777 | |
778 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
779 | if (io->error == DM_ENDIO_REQUEUE) { |
780 | /* | |
781 | * Target requested pushing back the I/O. | |
2e93ccc1 | 782 | */ |
022c2611 | 783 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
784 | if (__noflush_suspending(md)) |
785 | bio_list_add_head(&md->deferred, io->bio); | |
786 | else | |
2e93ccc1 KU |
787 | /* noflush suspend was interrupted. */ |
788 | io->error = -EIO; | |
022c2611 | 789 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
790 | } |
791 | ||
b35f8caa MB |
792 | io_error = io->error; |
793 | bio = io->bio; | |
6a8736d1 TH |
794 | end_io_acct(io); |
795 | free_io(md, io); | |
796 | ||
797 | if (io_error == DM_ENDIO_REQUEUE) | |
798 | return; | |
2e93ccc1 | 799 | |
1eff9d32 | 800 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 801 | /* |
6a8736d1 | 802 | * Preflush done for flush with data, reissue |
28a8f0d3 | 803 | * without REQ_PREFLUSH. |
af7e466a | 804 | */ |
1eff9d32 | 805 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 806 | queue_io(md, bio); |
af7e466a | 807 | } else { |
b372d360 | 808 | /* done with normal IO or empty flush */ |
0a82a8d1 | 809 | trace_block_bio_complete(md->queue, bio, io_error); |
4246a0b6 CH |
810 | bio->bi_error = io_error; |
811 | bio_endio(bio); | |
b35f8caa | 812 | } |
1da177e4 LT |
813 | } |
814 | } | |
815 | ||
4cc96131 | 816 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
817 | { |
818 | struct queue_limits *limits = dm_get_queue_limits(md); | |
819 | ||
820 | /* device doesn't really support WRITE SAME, disable it */ | |
821 | limits->max_write_same_sectors = 0; | |
822 | } | |
823 | ||
4246a0b6 | 824 | static void clone_endio(struct bio *bio) |
1da177e4 | 825 | { |
4246a0b6 | 826 | int error = bio->bi_error; |
5164bece | 827 | int r = error; |
bfc6d41c | 828 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 829 | struct dm_io *io = tio->io; |
9faf400f | 830 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
831 | dm_endio_fn endio = tio->ti->type->end_io; |
832 | ||
1da177e4 | 833 | if (endio) { |
7de3ee57 | 834 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
835 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
836 | /* | |
837 | * error and requeue request are handled | |
838 | * in dec_pending(). | |
839 | */ | |
1da177e4 | 840 | error = r; |
45cbcd79 KU |
841 | else if (r == DM_ENDIO_INCOMPLETE) |
842 | /* The target will handle the io */ | |
6712ecf8 | 843 | return; |
45cbcd79 KU |
844 | else if (r) { |
845 | DMWARN("unimplemented target endio return value: %d", r); | |
846 | BUG(); | |
847 | } | |
1da177e4 LT |
848 | } |
849 | ||
e6047149 | 850 | if (unlikely(r == -EREMOTEIO && (bio_op(bio) == REQ_OP_WRITE_SAME) && |
7eee4ae2 MS |
851 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) |
852 | disable_write_same(md); | |
853 | ||
cfae7529 | 854 | free_tio(tio); |
b35f8caa | 855 | dec_pending(io, error); |
1da177e4 LT |
856 | } |
857 | ||
56a67df7 MS |
858 | /* |
859 | * Return maximum size of I/O possible at the supplied sector up to the current | |
860 | * target boundary. | |
861 | */ | |
862 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
863 | { | |
864 | sector_t target_offset = dm_target_offset(ti, sector); | |
865 | ||
866 | return ti->len - target_offset; | |
867 | } | |
868 | ||
869 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 870 | { |
56a67df7 | 871 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 872 | sector_t offset, max_len; |
1da177e4 LT |
873 | |
874 | /* | |
542f9038 | 875 | * Does the target need to split even further? |
1da177e4 | 876 | */ |
542f9038 MS |
877 | if (ti->max_io_len) { |
878 | offset = dm_target_offset(ti, sector); | |
879 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
880 | max_len = sector_div(offset, ti->max_io_len); | |
881 | else | |
882 | max_len = offset & (ti->max_io_len - 1); | |
883 | max_len = ti->max_io_len - max_len; | |
884 | ||
885 | if (len > max_len) | |
886 | len = max_len; | |
1da177e4 LT |
887 | } |
888 | ||
889 | return len; | |
890 | } | |
891 | ||
542f9038 MS |
892 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
893 | { | |
894 | if (len > UINT_MAX) { | |
895 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
896 | (unsigned long long)len, UINT_MAX); | |
897 | ti->error = "Maximum size of target IO is too large"; | |
898 | return -EINVAL; | |
899 | } | |
900 | ||
901 | ti->max_io_len = (uint32_t) len; | |
902 | ||
903 | return 0; | |
904 | } | |
905 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
906 | ||
545ed20e | 907 | static long dm_blk_direct_access(struct block_device *bdev, sector_t sector, |
f0c98ebc | 908 | void **kaddr, pfn_t *pfn, long size) |
545ed20e TK |
909 | { |
910 | struct mapped_device *md = bdev->bd_disk->private_data; | |
911 | struct dm_table *map; | |
912 | struct dm_target *ti; | |
913 | int srcu_idx; | |
914 | long len, ret = -EIO; | |
915 | ||
916 | map = dm_get_live_table(md, &srcu_idx); | |
917 | if (!map) | |
918 | goto out; | |
919 | ||
920 | ti = dm_table_find_target(map, sector); | |
921 | if (!dm_target_is_valid(ti)) | |
922 | goto out; | |
923 | ||
924 | len = max_io_len(sector, ti) << SECTOR_SHIFT; | |
925 | size = min(len, size); | |
926 | ||
927 | if (ti->type->direct_access) | |
928 | ret = ti->type->direct_access(ti, sector, kaddr, pfn, size); | |
929 | out: | |
930 | dm_put_live_table(md, srcu_idx); | |
931 | return min(ret, size); | |
932 | } | |
933 | ||
1dd40c3e MP |
934 | /* |
935 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
28a8f0d3 | 936 | * allowed for all bio types except REQ_PREFLUSH. |
1dd40c3e MP |
937 | * |
938 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
939 | * additional n_sectors sectors of the bio and the rest of the data should be | |
940 | * sent in a next bio. | |
941 | * | |
942 | * A diagram that explains the arithmetics: | |
943 | * +--------------------+---------------+-------+ | |
944 | * | 1 | 2 | 3 | | |
945 | * +--------------------+---------------+-------+ | |
946 | * | |
947 | * <-------------- *tio->len_ptr ---------------> | |
948 | * <------- bi_size -------> | |
949 | * <-- n_sectors --> | |
950 | * | |
951 | * Region 1 was already iterated over with bio_advance or similar function. | |
952 | * (it may be empty if the target doesn't use bio_advance) | |
953 | * Region 2 is the remaining bio size that the target wants to process. | |
954 | * (it may be empty if region 1 is non-empty, although there is no reason | |
955 | * to make it empty) | |
956 | * The target requires that region 3 is to be sent in the next bio. | |
957 | * | |
958 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
959 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
960 | * copies of the bio. | |
961 | */ | |
962 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
963 | { | |
964 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
965 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 966 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
967 | BUG_ON(bi_size > *tio->len_ptr); |
968 | BUG_ON(n_sectors > bi_size); | |
969 | *tio->len_ptr -= bi_size - n_sectors; | |
970 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
971 | } | |
972 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
973 | ||
bd2a49b8 | 974 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
975 | { |
976 | int r; | |
2056a782 | 977 | sector_t sector; |
dba14160 | 978 | struct bio *clone = &tio->clone; |
bd2a49b8 | 979 | struct dm_target *ti = tio->ti; |
1da177e4 | 980 | |
1da177e4 | 981 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
982 | |
983 | /* | |
984 | * Map the clone. If r == 0 we don't need to do | |
985 | * anything, the target has assumed ownership of | |
986 | * this io. | |
987 | */ | |
988 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 989 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 990 | r = ti->type->map(ti, clone); |
45cbcd79 | 991 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 992 | /* the bio has been remapped so dispatch it */ |
2056a782 | 993 | |
d07335e5 MS |
994 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
995 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 996 | |
1da177e4 | 997 | generic_make_request(clone); |
2e93ccc1 KU |
998 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
999 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f | 1000 | dec_pending(tio->io, r); |
cfae7529 | 1001 | free_tio(tio); |
ab37844d | 1002 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1003 | DMWARN("unimplemented target map return value: %d", r); |
1004 | BUG(); | |
1da177e4 LT |
1005 | } |
1006 | } | |
1007 | ||
1008 | struct clone_info { | |
1009 | struct mapped_device *md; | |
1010 | struct dm_table *map; | |
1011 | struct bio *bio; | |
1012 | struct dm_io *io; | |
1013 | sector_t sector; | |
e0d6609a | 1014 | unsigned sector_count; |
1da177e4 LT |
1015 | }; |
1016 | ||
e0d6609a | 1017 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1018 | { |
4f024f37 KO |
1019 | bio->bi_iter.bi_sector = sector; |
1020 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1021 | } |
1022 | ||
1023 | /* | |
1024 | * Creates a bio that consists of range of complete bvecs. | |
1025 | */ | |
c80914e8 MS |
1026 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1027 | sector_t sector, unsigned len) | |
1da177e4 | 1028 | { |
dba14160 | 1029 | struct bio *clone = &tio->clone; |
1da177e4 | 1030 | |
1c3b13e6 KO |
1031 | __bio_clone_fast(clone, bio); |
1032 | ||
c80914e8 MS |
1033 | if (bio_integrity(bio)) { |
1034 | int r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
1035 | if (r < 0) | |
1036 | return r; | |
1037 | } | |
bd2a49b8 | 1038 | |
1c3b13e6 KO |
1039 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1040 | clone->bi_iter.bi_size = to_bytes(len); | |
1041 | ||
1042 | if (bio_integrity(bio)) | |
1043 | bio_integrity_trim(clone, 0, len); | |
c80914e8 MS |
1044 | |
1045 | return 0; | |
1da177e4 LT |
1046 | } |
1047 | ||
9015df24 | 1048 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1049 | struct dm_target *ti, |
55a62eef | 1050 | unsigned target_bio_nr) |
f9ab94ce | 1051 | { |
dba14160 MP |
1052 | struct dm_target_io *tio; |
1053 | struct bio *clone; | |
1054 | ||
99778273 | 1055 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1056 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1057 | |
1058 | tio->io = ci->io; | |
1059 | tio->ti = ti; | |
55a62eef | 1060 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1061 | |
1062 | return tio; | |
1063 | } | |
1064 | ||
14fe594d AK |
1065 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1066 | struct dm_target *ti, | |
1dd40c3e | 1067 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1068 | { |
99778273 | 1069 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1070 | struct bio *clone = &tio->clone; |
9015df24 | 1071 | |
1dd40c3e MP |
1072 | tio->len_ptr = len; |
1073 | ||
99778273 | 1074 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1075 | if (len) |
1dd40c3e | 1076 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1077 | |
bd2a49b8 | 1078 | __map_bio(tio); |
f9ab94ce MP |
1079 | } |
1080 | ||
14fe594d | 1081 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1082 | unsigned num_bios, unsigned *len) |
06a426ce | 1083 | { |
55a62eef | 1084 | unsigned target_bio_nr; |
06a426ce | 1085 | |
55a62eef | 1086 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1087 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1088 | } |
1089 | ||
14fe594d | 1090 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1091 | { |
06a426ce | 1092 | unsigned target_nr = 0; |
f9ab94ce MP |
1093 | struct dm_target *ti; |
1094 | ||
b372d360 | 1095 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1096 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1097 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1098 | |
f9ab94ce MP |
1099 | return 0; |
1100 | } | |
1101 | ||
c80914e8 | 1102 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1103 | sector_t sector, unsigned *len) |
5ae89a87 | 1104 | { |
dba14160 | 1105 | struct bio *bio = ci->bio; |
5ae89a87 | 1106 | struct dm_target_io *tio; |
b0d8ed4d AK |
1107 | unsigned target_bio_nr; |
1108 | unsigned num_target_bios = 1; | |
c80914e8 | 1109 | int r = 0; |
5ae89a87 | 1110 | |
b0d8ed4d AK |
1111 | /* |
1112 | * Does the target want to receive duplicate copies of the bio? | |
1113 | */ | |
1114 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1115 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1116 | |
b0d8ed4d | 1117 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1118 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e | 1119 | tio->len_ptr = len; |
c80914e8 | 1120 | r = clone_bio(tio, bio, sector, *len); |
072623de | 1121 | if (r < 0) { |
cfae7529 | 1122 | free_tio(tio); |
c80914e8 | 1123 | break; |
072623de | 1124 | } |
b0d8ed4d AK |
1125 | __map_bio(tio); |
1126 | } | |
c80914e8 MS |
1127 | |
1128 | return r; | |
5ae89a87 MS |
1129 | } |
1130 | ||
55a62eef | 1131 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1132 | |
55a62eef | 1133 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1134 | { |
55a62eef | 1135 | return ti->num_discard_bios; |
23508a96 MS |
1136 | } |
1137 | ||
55a62eef | 1138 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1139 | { |
55a62eef | 1140 | return ti->num_write_same_bios; |
23508a96 MS |
1141 | } |
1142 | ||
1143 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
9eef87da | 1144 | |
23508a96 MS |
1145 | static bool is_split_required_for_discard(struct dm_target *ti) |
1146 | { | |
55a62eef | 1147 | return ti->split_discard_bios; |
cec47e3d KU |
1148 | } |
1149 | ||
14fe594d AK |
1150 | static int __send_changing_extent_only(struct clone_info *ci, |
1151 | get_num_bios_fn get_num_bios, | |
1152 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1153 | { |
5ae89a87 | 1154 | struct dm_target *ti; |
e0d6609a | 1155 | unsigned len; |
55a62eef | 1156 | unsigned num_bios; |
ba1cbad9 | 1157 | |
a79245b3 MS |
1158 | do { |
1159 | ti = dm_table_find_target(ci->map, ci->sector); | |
1160 | if (!dm_target_is_valid(ti)) | |
1161 | return -EIO; | |
2eb6e1e3 | 1162 | |
5ae89a87 | 1163 | /* |
23508a96 MS |
1164 | * Even though the device advertised support for this type of |
1165 | * request, that does not mean every target supports it, and | |
936688d7 | 1166 | * reconfiguration might also have changed that since the |
a79245b3 | 1167 | * check was performed. |
5ae89a87 | 1168 | */ |
55a62eef AK |
1169 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1170 | if (!num_bios) | |
a79245b3 | 1171 | return -EOPNOTSUPP; |
ba1cbad9 | 1172 | |
23508a96 | 1173 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1174 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1175 | else |
e0d6609a | 1176 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
de3ec86d | 1177 | |
1dd40c3e | 1178 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1179 | |
a79245b3 MS |
1180 | ci->sector += len; |
1181 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1182 | |
1183 | return 0; | |
ba1cbad9 MS |
1184 | } |
1185 | ||
14fe594d | 1186 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1187 | { |
14fe594d AK |
1188 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1189 | is_split_required_for_discard); | |
23508a96 | 1190 | } |
0ce65797 | 1191 | |
14fe594d | 1192 | static int __send_write_same(struct clone_info *ci) |
0ce65797 | 1193 | { |
14fe594d | 1194 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
0ce65797 MS |
1195 | } |
1196 | ||
e4c93811 AK |
1197 | /* |
1198 | * Select the correct strategy for processing a non-flush bio. | |
1199 | */ | |
14fe594d | 1200 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1201 | { |
dba14160 | 1202 | struct bio *bio = ci->bio; |
512875bd | 1203 | struct dm_target *ti; |
1c3b13e6 | 1204 | unsigned len; |
c80914e8 | 1205 | int r; |
0ce65797 | 1206 | |
e6047149 | 1207 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
14fe594d | 1208 | return __send_discard(ci); |
e6047149 | 1209 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
14fe594d | 1210 | return __send_write_same(ci); |
0ce65797 | 1211 | |
512875bd JN |
1212 | ti = dm_table_find_target(ci->map, ci->sector); |
1213 | if (!dm_target_is_valid(ti)) | |
1214 | return -EIO; | |
1215 | ||
1c3b13e6 | 1216 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
0ce65797 | 1217 | |
c80914e8 MS |
1218 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1219 | if (r < 0) | |
1220 | return r; | |
0ce65797 | 1221 | |
1c3b13e6 KO |
1222 | ci->sector += len; |
1223 | ci->sector_count -= len; | |
0ce65797 | 1224 | |
1c3b13e6 | 1225 | return 0; |
0ce65797 MS |
1226 | } |
1227 | ||
1da177e4 | 1228 | /* |
14fe594d | 1229 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1230 | */ |
83d5e5b0 MP |
1231 | static void __split_and_process_bio(struct mapped_device *md, |
1232 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1233 | { |
1da177e4 | 1234 | struct clone_info ci; |
512875bd | 1235 | int error = 0; |
1da177e4 | 1236 | |
83d5e5b0 | 1237 | if (unlikely(!map)) { |
6a8736d1 | 1238 | bio_io_error(bio); |
f0b9a450 MP |
1239 | return; |
1240 | } | |
692d0eb9 | 1241 | |
83d5e5b0 | 1242 | ci.map = map; |
1da177e4 | 1243 | ci.md = md; |
1da177e4 LT |
1244 | ci.io = alloc_io(md); |
1245 | ci.io->error = 0; | |
1246 | atomic_set(&ci.io->io_count, 1); | |
1247 | ci.io->bio = bio; | |
1248 | ci.io->md = md; | |
f88fb981 | 1249 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1250 | ci.sector = bio->bi_iter.bi_sector; |
0ce65797 | 1251 | |
3eaf840e | 1252 | start_io_acct(ci.io); |
0ce65797 | 1253 | |
1eff9d32 | 1254 | if (bio->bi_opf & REQ_PREFLUSH) { |
b372d360 MS |
1255 | ci.bio = &ci.md->flush_bio; |
1256 | ci.sector_count = 0; | |
14fe594d | 1257 | error = __send_empty_flush(&ci); |
b372d360 MS |
1258 | /* dec_pending submits any data associated with flush */ |
1259 | } else { | |
6a8736d1 | 1260 | ci.bio = bio; |
d87f4c14 | 1261 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1262 | while (ci.sector_count && !error) |
14fe594d | 1263 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1264 | } |
0ce65797 | 1265 | |
1da177e4 | 1266 | /* drop the extra reference count */ |
512875bd | 1267 | dec_pending(ci.io, error); |
0ce65797 | 1268 | } |
1da177e4 LT |
1269 | /*----------------------------------------------------------------- |
1270 | * CRUD END | |
1271 | *---------------------------------------------------------------*/ | |
0ce65797 | 1272 | |
cec47e3d | 1273 | /* |
1da177e4 LT |
1274 | * The request function that just remaps the bio built up by |
1275 | * dm_merge_bvec. | |
cec47e3d | 1276 | */ |
dece1635 | 1277 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
cec47e3d | 1278 | { |
12f03a49 | 1279 | int rw = bio_data_dir(bio); |
cec47e3d | 1280 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1281 | int srcu_idx; |
1282 | struct dm_table *map; | |
cec47e3d | 1283 | |
83d5e5b0 | 1284 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1285 | |
18c0b223 | 1286 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
d0bcb878 | 1287 | |
6a8736d1 TH |
1288 | /* if we're suspended, we have to queue this io for later */ |
1289 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1290 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1291 | |
1eff9d32 | 1292 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1293 | queue_io(md, bio); |
1294 | else | |
54d9a1b4 | 1295 | bio_io_error(bio); |
dece1635 | 1296 | return BLK_QC_T_NONE; |
cec47e3d | 1297 | } |
1da177e4 | 1298 | |
83d5e5b0 MP |
1299 | __split_and_process_bio(md, map, bio); |
1300 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1301 | return BLK_QC_T_NONE; |
cec47e3d KU |
1302 | } |
1303 | ||
1da177e4 LT |
1304 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1305 | { | |
8a57dfc6 CS |
1306 | int r = bdi_bits; |
1307 | struct mapped_device *md = congested_data; | |
1308 | struct dm_table *map; | |
1da177e4 | 1309 | |
1eb787ec | 1310 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1311 | if (dm_request_based(md)) { |
cec47e3d | 1312 | /* |
e522c039 MS |
1313 | * With request-based DM we only need to check the |
1314 | * top-level queue for congestion. | |
cec47e3d | 1315 | */ |
dc3b17cc | 1316 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1317 | } else { |
1318 | map = dm_get_live_table_fast(md); | |
1319 | if (map) | |
cec47e3d | 1320 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1321 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1322 | } |
1323 | } | |
1324 | ||
1da177e4 LT |
1325 | return r; |
1326 | } | |
1327 | ||
1328 | /*----------------------------------------------------------------- | |
1329 | * An IDR is used to keep track of allocated minor numbers. | |
1330 | *---------------------------------------------------------------*/ | |
2b06cfff | 1331 | static void free_minor(int minor) |
1da177e4 | 1332 | { |
f32c10b0 | 1333 | spin_lock(&_minor_lock); |
1da177e4 | 1334 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1335 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1336 | } |
1337 | ||
1338 | /* | |
1339 | * See if the device with a specific minor # is free. | |
1340 | */ | |
cf13ab8e | 1341 | static int specific_minor(int minor) |
1da177e4 | 1342 | { |
c9d76be6 | 1343 | int r; |
1da177e4 LT |
1344 | |
1345 | if (minor >= (1 << MINORBITS)) | |
1346 | return -EINVAL; | |
1347 | ||
c9d76be6 | 1348 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1349 | spin_lock(&_minor_lock); |
1da177e4 | 1350 | |
c9d76be6 | 1351 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1352 | |
f32c10b0 | 1353 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1354 | idr_preload_end(); |
1355 | if (r < 0) | |
1356 | return r == -ENOSPC ? -EBUSY : r; | |
1357 | return 0; | |
1da177e4 LT |
1358 | } |
1359 | ||
cf13ab8e | 1360 | static int next_free_minor(int *minor) |
1da177e4 | 1361 | { |
c9d76be6 | 1362 | int r; |
62f75c2f | 1363 | |
c9d76be6 | 1364 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1365 | spin_lock(&_minor_lock); |
1da177e4 | 1366 | |
c9d76be6 | 1367 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1368 | |
f32c10b0 | 1369 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1370 | idr_preload_end(); |
1371 | if (r < 0) | |
1372 | return r; | |
1373 | *minor = r; | |
1374 | return 0; | |
1da177e4 LT |
1375 | } |
1376 | ||
83d5cde4 | 1377 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1378 | |
53d5914f MP |
1379 | static void dm_wq_work(struct work_struct *work); |
1380 | ||
4cc96131 | 1381 | void dm_init_md_queue(struct mapped_device *md) |
4a0b4ddf MS |
1382 | { |
1383 | /* | |
1384 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 1385 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
1386 | * The type is decided at the first table loading time. |
1387 | * To prevent problematic device stacking, clear the queue flag | |
1388 | * for request stacking support until then. | |
1389 | * | |
1390 | * This queue is new, so no concurrency on the queue_flags. | |
1391 | */ | |
1392 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
1393 | |
1394 | /* | |
1395 | * Initialize data that will only be used by a non-blk-mq DM queue | |
1396 | * - must do so here (in alloc_dev callchain) before queue is used | |
1397 | */ | |
1398 | md->queue->queuedata = md; | |
dc3b17cc | 1399 | md->queue->backing_dev_info->congested_data = md; |
bfebd1cd | 1400 | } |
4a0b4ddf | 1401 | |
4cc96131 | 1402 | void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1403 | { |
17e149b8 | 1404 | md->use_blk_mq = false; |
bfebd1cd MS |
1405 | dm_init_md_queue(md); |
1406 | ||
1407 | /* | |
1408 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1409 | */ | |
dc3b17cc | 1410 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf | 1411 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
1412 | } |
1413 | ||
0f20972f MS |
1414 | static void cleanup_mapped_device(struct mapped_device *md) |
1415 | { | |
0f20972f MS |
1416 | if (md->wq) |
1417 | destroy_workqueue(md->wq); | |
1418 | if (md->kworker_task) | |
1419 | kthread_stop(md->kworker_task); | |
6f65985e | 1420 | mempool_destroy(md->io_pool); |
0f20972f MS |
1421 | if (md->bs) |
1422 | bioset_free(md->bs); | |
1423 | ||
1424 | if (md->disk) { | |
1425 | spin_lock(&_minor_lock); | |
1426 | md->disk->private_data = NULL; | |
1427 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1428 | del_gendisk(md->disk); |
1429 | put_disk(md->disk); | |
1430 | } | |
1431 | ||
1432 | if (md->queue) | |
1433 | blk_cleanup_queue(md->queue); | |
1434 | ||
d09960b0 TE |
1435 | cleanup_srcu_struct(&md->io_barrier); |
1436 | ||
0f20972f MS |
1437 | if (md->bdev) { |
1438 | bdput(md->bdev); | |
1439 | md->bdev = NULL; | |
1440 | } | |
4cc96131 MS |
1441 | |
1442 | dm_mq_cleanup_mapped_device(md); | |
0f20972f MS |
1443 | } |
1444 | ||
1da177e4 LT |
1445 | /* |
1446 | * Allocate and initialise a blank device with a given minor. | |
1447 | */ | |
2b06cfff | 1448 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1449 | { |
115485e8 MS |
1450 | int r, numa_node_id = dm_get_numa_node(); |
1451 | struct mapped_device *md; | |
ba61fdd1 | 1452 | void *old_md; |
1da177e4 | 1453 | |
115485e8 | 1454 | md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1455 | if (!md) { |
1456 | DMWARN("unable to allocate device, out of memory."); | |
1457 | return NULL; | |
1458 | } | |
1459 | ||
10da4f79 | 1460 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1461 | goto bad_module_get; |
10da4f79 | 1462 | |
1da177e4 | 1463 | /* get a minor number for the dev */ |
2b06cfff | 1464 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1465 | r = next_free_minor(&minor); |
2b06cfff | 1466 | else |
cf13ab8e | 1467 | r = specific_minor(minor); |
1da177e4 | 1468 | if (r < 0) |
6ed7ade8 | 1469 | goto bad_minor; |
1da177e4 | 1470 | |
83d5e5b0 MP |
1471 | r = init_srcu_struct(&md->io_barrier); |
1472 | if (r < 0) | |
1473 | goto bad_io_barrier; | |
1474 | ||
115485e8 | 1475 | md->numa_node_id = numa_node_id; |
4cc96131 | 1476 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1477 | md->init_tio_pdu = false; |
a5664dad | 1478 | md->type = DM_TYPE_NONE; |
e61290a4 | 1479 | mutex_init(&md->suspend_lock); |
a5664dad | 1480 | mutex_init(&md->type_lock); |
86f1152b | 1481 | mutex_init(&md->table_devices_lock); |
022c2611 | 1482 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1483 | atomic_set(&md->holders, 1); |
5c6bd75d | 1484 | atomic_set(&md->open_count, 0); |
1da177e4 | 1485 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1486 | atomic_set(&md->uevent_seq, 0); |
1487 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1488 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1489 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1490 | |
115485e8 | 1491 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1492 | if (!md->queue) |
0f20972f | 1493 | goto bad; |
1da177e4 | 1494 | |
4a0b4ddf | 1495 | dm_init_md_queue(md); |
9faf400f | 1496 | |
115485e8 | 1497 | md->disk = alloc_disk_node(1, numa_node_id); |
1da177e4 | 1498 | if (!md->disk) |
0f20972f | 1499 | goto bad; |
1da177e4 | 1500 | |
316d315b NK |
1501 | atomic_set(&md->pending[0], 0); |
1502 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1503 | init_waitqueue_head(&md->wait); |
53d5914f | 1504 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1505 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1506 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1507 | md->kworker_task = NULL; |
f0b04115 | 1508 | |
1da177e4 LT |
1509 | md->disk->major = _major; |
1510 | md->disk->first_minor = minor; | |
1511 | md->disk->fops = &dm_blk_dops; | |
1512 | md->disk->queue = md->queue; | |
1513 | md->disk->private_data = md; | |
1514 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1515 | add_disk(md->disk); | |
7e51f257 | 1516 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1517 | |
670368a8 | 1518 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1519 | if (!md->wq) |
0f20972f | 1520 | goto bad; |
304f3f6a | 1521 | |
32a926da MP |
1522 | md->bdev = bdget_disk(md->disk, 0); |
1523 | if (!md->bdev) | |
0f20972f | 1524 | goto bad; |
32a926da | 1525 | |
3a83f467 | 1526 | bio_init(&md->flush_bio, NULL, 0); |
6a8736d1 | 1527 | md->flush_bio.bi_bdev = md->bdev; |
70fd7614 | 1528 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
6a8736d1 | 1529 | |
fd2ed4d2 MP |
1530 | dm_stats_init(&md->stats); |
1531 | ||
ba61fdd1 | 1532 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1533 | spin_lock(&_minor_lock); |
ba61fdd1 | 1534 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1535 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1536 | |
1537 | BUG_ON(old_md != MINOR_ALLOCED); | |
1538 | ||
1da177e4 LT |
1539 | return md; |
1540 | ||
0f20972f MS |
1541 | bad: |
1542 | cleanup_mapped_device(md); | |
83d5e5b0 | 1543 | bad_io_barrier: |
1da177e4 | 1544 | free_minor(minor); |
6ed7ade8 | 1545 | bad_minor: |
10da4f79 | 1546 | module_put(THIS_MODULE); |
6ed7ade8 | 1547 | bad_module_get: |
1da177e4 LT |
1548 | kfree(md); |
1549 | return NULL; | |
1550 | } | |
1551 | ||
ae9da83f JN |
1552 | static void unlock_fs(struct mapped_device *md); |
1553 | ||
1da177e4 LT |
1554 | static void free_dev(struct mapped_device *md) |
1555 | { | |
f331c029 | 1556 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1557 | |
32a926da | 1558 | unlock_fs(md); |
2eb6e1e3 | 1559 | |
0f20972f | 1560 | cleanup_mapped_device(md); |
63a4f065 | 1561 | |
86f1152b | 1562 | free_table_devices(&md->table_devices); |
63a4f065 | 1563 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1564 | free_minor(minor); |
1565 | ||
10da4f79 | 1566 | module_put(THIS_MODULE); |
1da177e4 LT |
1567 | kfree(md); |
1568 | } | |
1569 | ||
e6ee8c0b KU |
1570 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1571 | { | |
c0820cf5 | 1572 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1573 | |
4e6e36c3 MS |
1574 | if (md->bs) { |
1575 | /* The md already has necessary mempools. */ | |
545ed20e | 1576 | if (dm_table_bio_based(t)) { |
16245bdc JN |
1577 | /* |
1578 | * Reload bioset because front_pad may have changed | |
1579 | * because a different table was loaded. | |
1580 | */ | |
1581 | bioset_free(md->bs); | |
1582 | md->bs = p->bs; | |
1583 | p->bs = NULL; | |
16245bdc | 1584 | } |
4e6e36c3 MS |
1585 | /* |
1586 | * There's no need to reload with request-based dm | |
1587 | * because the size of front_pad doesn't change. | |
1588 | * Note for future: If you are to reload bioset, | |
1589 | * prep-ed requests in the queue may refer | |
1590 | * to bio from the old bioset, so you must walk | |
1591 | * through the queue to unprep. | |
1592 | */ | |
1593 | goto out; | |
c0820cf5 | 1594 | } |
e6ee8c0b | 1595 | |
eb8db831 | 1596 | BUG_ON(!p || md->io_pool || md->bs); |
cbc4e3c1 | 1597 | |
e6ee8c0b KU |
1598 | md->io_pool = p->io_pool; |
1599 | p->io_pool = NULL; | |
e6ee8c0b KU |
1600 | md->bs = p->bs; |
1601 | p->bs = NULL; | |
4e6e36c3 | 1602 | |
e6ee8c0b | 1603 | out: |
02233342 | 1604 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1605 | dm_table_free_md_mempools(t); |
1606 | } | |
1607 | ||
1da177e4 LT |
1608 | /* |
1609 | * Bind a table to the device. | |
1610 | */ | |
1611 | static void event_callback(void *context) | |
1612 | { | |
7a8c3d3b MA |
1613 | unsigned long flags; |
1614 | LIST_HEAD(uevents); | |
1da177e4 LT |
1615 | struct mapped_device *md = (struct mapped_device *) context; |
1616 | ||
7a8c3d3b MA |
1617 | spin_lock_irqsave(&md->uevent_lock, flags); |
1618 | list_splice_init(&md->uevent_list, &uevents); | |
1619 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1620 | ||
ed9e1982 | 1621 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1622 | |
1da177e4 LT |
1623 | atomic_inc(&md->event_nr); |
1624 | wake_up(&md->eventq); | |
1625 | } | |
1626 | ||
c217649b MS |
1627 | /* |
1628 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1629 | */ | |
4e90188b | 1630 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1631 | { |
4e90188b | 1632 | set_capacity(md->disk, size); |
1da177e4 | 1633 | |
db8fef4f | 1634 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
1635 | } |
1636 | ||
042d2a9b AK |
1637 | /* |
1638 | * Returns old map, which caller must destroy. | |
1639 | */ | |
1640 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
1641 | struct queue_limits *limits) | |
1da177e4 | 1642 | { |
042d2a9b | 1643 | struct dm_table *old_map; |
165125e1 | 1644 | struct request_queue *q = md->queue; |
1da177e4 LT |
1645 | sector_t size; |
1646 | ||
5a8f1f80 BVA |
1647 | lockdep_assert_held(&md->suspend_lock); |
1648 | ||
1da177e4 | 1649 | size = dm_table_get_size(t); |
3ac51e74 DW |
1650 | |
1651 | /* | |
1652 | * Wipe any geometry if the size of the table changed. | |
1653 | */ | |
fd2ed4d2 | 1654 | if (size != dm_get_size(md)) |
3ac51e74 DW |
1655 | memset(&md->geometry, 0, sizeof(md->geometry)); |
1656 | ||
32a926da | 1657 | __set_size(md, size); |
d5816876 | 1658 | |
2ca3310e AK |
1659 | dm_table_event_callback(t, event_callback, md); |
1660 | ||
e6ee8c0b KU |
1661 | /* |
1662 | * The queue hasn't been stopped yet, if the old table type wasn't | |
1663 | * for request-based during suspension. So stop it to prevent | |
1664 | * I/O mapping before resume. | |
1665 | * This must be done before setting the queue restrictions, | |
1666 | * because request-based dm may be run just after the setting. | |
1667 | */ | |
16f12266 | 1668 | if (dm_table_request_based(t)) { |
eca7ee6d | 1669 | dm_stop_queue(q); |
16f12266 MS |
1670 | /* |
1671 | * Leverage the fact that request-based DM targets are | |
1672 | * immutable singletons and establish md->immutable_target | |
1673 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
1674 | */ | |
1675 | md->immutable_target = dm_table_get_immutable_target(t); | |
1676 | } | |
e6ee8c0b KU |
1677 | |
1678 | __bind_mempools(md, t); | |
1679 | ||
a12f5d48 | 1680 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 1681 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
1682 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
1683 | ||
754c5fc7 | 1684 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
1685 | if (old_map) |
1686 | dm_sync_table(md); | |
1da177e4 | 1687 | |
042d2a9b | 1688 | return old_map; |
1da177e4 LT |
1689 | } |
1690 | ||
a7940155 AK |
1691 | /* |
1692 | * Returns unbound table for the caller to free. | |
1693 | */ | |
1694 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 1695 | { |
a12f5d48 | 1696 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
1697 | |
1698 | if (!map) | |
a7940155 | 1699 | return NULL; |
1da177e4 LT |
1700 | |
1701 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 1702 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 1703 | dm_sync_table(md); |
a7940155 AK |
1704 | |
1705 | return map; | |
1da177e4 LT |
1706 | } |
1707 | ||
1708 | /* | |
1709 | * Constructor for a new device. | |
1710 | */ | |
2b06cfff | 1711 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1712 | { |
1713 | struct mapped_device *md; | |
1714 | ||
2b06cfff | 1715 | md = alloc_dev(minor); |
1da177e4 LT |
1716 | if (!md) |
1717 | return -ENXIO; | |
1718 | ||
784aae73 MB |
1719 | dm_sysfs_init(md); |
1720 | ||
1da177e4 LT |
1721 | *result = md; |
1722 | return 0; | |
1723 | } | |
1724 | ||
a5664dad MS |
1725 | /* |
1726 | * Functions to manage md->type. | |
1727 | * All are required to hold md->type_lock. | |
1728 | */ | |
1729 | void dm_lock_md_type(struct mapped_device *md) | |
1730 | { | |
1731 | mutex_lock(&md->type_lock); | |
1732 | } | |
1733 | ||
1734 | void dm_unlock_md_type(struct mapped_device *md) | |
1735 | { | |
1736 | mutex_unlock(&md->type_lock); | |
1737 | } | |
1738 | ||
1739 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
1740 | { | |
00c4fc3b | 1741 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
1742 | md->type = type; |
1743 | } | |
1744 | ||
1745 | unsigned dm_get_md_type(struct mapped_device *md) | |
1746 | { | |
1747 | return md->type; | |
1748 | } | |
1749 | ||
36a0456f AK |
1750 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
1751 | { | |
1752 | return md->immutable_target_type; | |
1753 | } | |
1754 | ||
f84cb8a4 MS |
1755 | /* |
1756 | * The queue_limits are only valid as long as you have a reference | |
1757 | * count on 'md'. | |
1758 | */ | |
1759 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
1760 | { | |
1761 | BUG_ON(!atomic_read(&md->holders)); | |
1762 | return &md->queue->limits; | |
1763 | } | |
1764 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
1765 | ||
4a0b4ddf MS |
1766 | /* |
1767 | * Setup the DM device's queue based on md's type | |
1768 | */ | |
591ddcfc | 1769 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 1770 | { |
bfebd1cd | 1771 | int r; |
545ed20e | 1772 | unsigned type = dm_get_md_type(md); |
bfebd1cd | 1773 | |
545ed20e | 1774 | switch (type) { |
bfebd1cd | 1775 | case DM_TYPE_REQUEST_BASED: |
eb8db831 | 1776 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 1777 | if (r) { |
eca7ee6d | 1778 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 1779 | return r; |
ff36ab34 | 1780 | } |
bfebd1cd MS |
1781 | break; |
1782 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 1783 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 1784 | if (r) { |
eca7ee6d | 1785 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
1786 | return r; |
1787 | } | |
1788 | break; | |
1789 | case DM_TYPE_BIO_BASED: | |
545ed20e | 1790 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 1791 | dm_init_normal_md_queue(md); |
ff36ab34 | 1792 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
1793 | /* |
1794 | * DM handles splitting bios as needed. Free the bio_split bioset | |
1795 | * since it won't be used (saves 1 process per bio-based DM device). | |
1796 | */ | |
1797 | bioset_free(md->queue->bio_split); | |
1798 | md->queue->bio_split = NULL; | |
545ed20e TK |
1799 | |
1800 | if (type == DM_TYPE_DAX_BIO_BASED) | |
1801 | queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue); | |
bfebd1cd | 1802 | break; |
4a0b4ddf MS |
1803 | } |
1804 | ||
1805 | return 0; | |
1806 | } | |
1807 | ||
2bec1f4a | 1808 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
1809 | { |
1810 | struct mapped_device *md; | |
1da177e4 LT |
1811 | unsigned minor = MINOR(dev); |
1812 | ||
1813 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1814 | return NULL; | |
1815 | ||
f32c10b0 | 1816 | spin_lock(&_minor_lock); |
1da177e4 LT |
1817 | |
1818 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
1819 | if (md) { |
1820 | if ((md == MINOR_ALLOCED || | |
1821 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
1822 | dm_deleting_md(md) || | |
1823 | test_bit(DMF_FREEING, &md->flags))) { | |
1824 | md = NULL; | |
1825 | goto out; | |
1826 | } | |
1827 | dm_get(md); | |
fba9f90e | 1828 | } |
1da177e4 | 1829 | |
fba9f90e | 1830 | out: |
f32c10b0 | 1831 | spin_unlock(&_minor_lock); |
1da177e4 | 1832 | |
637842cf DT |
1833 | return md; |
1834 | } | |
3cf2e4ba | 1835 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 1836 | |
9ade92a9 | 1837 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1838 | { |
9ade92a9 | 1839 | return md->interface_ptr; |
1da177e4 LT |
1840 | } |
1841 | ||
1842 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1843 | { | |
1844 | md->interface_ptr = ptr; | |
1845 | } | |
1846 | ||
1847 | void dm_get(struct mapped_device *md) | |
1848 | { | |
1849 | atomic_inc(&md->holders); | |
3f77316d | 1850 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
1851 | } |
1852 | ||
09ee96b2 MP |
1853 | int dm_hold(struct mapped_device *md) |
1854 | { | |
1855 | spin_lock(&_minor_lock); | |
1856 | if (test_bit(DMF_FREEING, &md->flags)) { | |
1857 | spin_unlock(&_minor_lock); | |
1858 | return -EBUSY; | |
1859 | } | |
1860 | dm_get(md); | |
1861 | spin_unlock(&_minor_lock); | |
1862 | return 0; | |
1863 | } | |
1864 | EXPORT_SYMBOL_GPL(dm_hold); | |
1865 | ||
72d94861 AK |
1866 | const char *dm_device_name(struct mapped_device *md) |
1867 | { | |
1868 | return md->name; | |
1869 | } | |
1870 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1871 | ||
3f77316d | 1872 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 1873 | { |
3b785fbc | 1874 | struct request_queue *q = dm_get_md_queue(md); |
1134e5ae | 1875 | struct dm_table *map; |
83d5e5b0 | 1876 | int srcu_idx; |
1da177e4 | 1877 | |
3f77316d | 1878 | might_sleep(); |
fba9f90e | 1879 | |
63a4f065 | 1880 | spin_lock(&_minor_lock); |
3f77316d KU |
1881 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
1882 | set_bit(DMF_FREEING, &md->flags); | |
1883 | spin_unlock(&_minor_lock); | |
3b785fbc | 1884 | |
2e91c369 | 1885 | blk_set_queue_dying(q); |
3f77316d | 1886 | |
02233342 | 1887 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 1888 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 1889 | |
ab7c7bb6 MP |
1890 | /* |
1891 | * Take suspend_lock so that presuspend and postsuspend methods | |
1892 | * do not race with internal suspend. | |
1893 | */ | |
1894 | mutex_lock(&md->suspend_lock); | |
2a708cff | 1895 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
1896 | if (!dm_suspended_md(md)) { |
1897 | dm_table_presuspend_targets(map); | |
1898 | dm_table_postsuspend_targets(map); | |
1da177e4 | 1899 | } |
83d5e5b0 MP |
1900 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
1901 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 1902 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 1903 | |
3f77316d KU |
1904 | /* |
1905 | * Rare, but there may be I/O requests still going to complete, | |
1906 | * for example. Wait for all references to disappear. | |
1907 | * No one should increment the reference count of the mapped_device, | |
1908 | * after the mapped_device state becomes DMF_FREEING. | |
1909 | */ | |
1910 | if (wait) | |
1911 | while (atomic_read(&md->holders)) | |
1912 | msleep(1); | |
1913 | else if (atomic_read(&md->holders)) | |
1914 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
1915 | dm_device_name(md), atomic_read(&md->holders)); | |
1916 | ||
1917 | dm_sysfs_exit(md); | |
3f77316d KU |
1918 | dm_table_destroy(__unbind(md)); |
1919 | free_dev(md); | |
1920 | } | |
1921 | ||
1922 | void dm_destroy(struct mapped_device *md) | |
1923 | { | |
1924 | __dm_destroy(md, true); | |
1925 | } | |
1926 | ||
1927 | void dm_destroy_immediate(struct mapped_device *md) | |
1928 | { | |
1929 | __dm_destroy(md, false); | |
1930 | } | |
1931 | ||
1932 | void dm_put(struct mapped_device *md) | |
1933 | { | |
1934 | atomic_dec(&md->holders); | |
1da177e4 | 1935 | } |
79eb885c | 1936 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 1937 | |
b48633f8 | 1938 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
1939 | { |
1940 | int r = 0; | |
9f4c3f87 | 1941 | DEFINE_WAIT(wait); |
46125c1c MB |
1942 | |
1943 | while (1) { | |
9f4c3f87 | 1944 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 1945 | |
b4324fee | 1946 | if (!md_in_flight(md)) |
46125c1c MB |
1947 | break; |
1948 | ||
e3fabdfd | 1949 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
1950 | r = -EINTR; |
1951 | break; | |
1952 | } | |
1953 | ||
1954 | io_schedule(); | |
1955 | } | |
9f4c3f87 | 1956 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 1957 | |
46125c1c MB |
1958 | return r; |
1959 | } | |
1960 | ||
1da177e4 LT |
1961 | /* |
1962 | * Process the deferred bios | |
1963 | */ | |
ef208587 | 1964 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 1965 | { |
ef208587 MP |
1966 | struct mapped_device *md = container_of(work, struct mapped_device, |
1967 | work); | |
6d6f10df | 1968 | struct bio *c; |
83d5e5b0 MP |
1969 | int srcu_idx; |
1970 | struct dm_table *map; | |
1da177e4 | 1971 | |
83d5e5b0 | 1972 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 1973 | |
3b00b203 | 1974 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
1975 | spin_lock_irq(&md->deferred_lock); |
1976 | c = bio_list_pop(&md->deferred); | |
1977 | spin_unlock_irq(&md->deferred_lock); | |
1978 | ||
6a8736d1 | 1979 | if (!c) |
df12ee99 | 1980 | break; |
022c2611 | 1981 | |
e6ee8c0b KU |
1982 | if (dm_request_based(md)) |
1983 | generic_make_request(c); | |
6a8736d1 | 1984 | else |
83d5e5b0 | 1985 | __split_and_process_bio(md, map, c); |
022c2611 | 1986 | } |
73d410c0 | 1987 | |
83d5e5b0 | 1988 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
1989 | } |
1990 | ||
9a1fb464 | 1991 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 1992 | { |
3b00b203 | 1993 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 1994 | smp_mb__after_atomic(); |
53d5914f | 1995 | queue_work(md->wq, &md->work); |
304f3f6a MB |
1996 | } |
1997 | ||
1da177e4 | 1998 | /* |
042d2a9b | 1999 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2000 | */ |
042d2a9b | 2001 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2002 | { |
87eb5b21 | 2003 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2004 | struct queue_limits limits; |
042d2a9b | 2005 | int r; |
1da177e4 | 2006 | |
e61290a4 | 2007 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2008 | |
2009 | /* device must be suspended */ | |
4f186f8b | 2010 | if (!dm_suspended_md(md)) |
93c534ae | 2011 | goto out; |
1da177e4 | 2012 | |
3ae70656 MS |
2013 | /* |
2014 | * If the new table has no data devices, retain the existing limits. | |
2015 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2016 | * then new I/O is queued based on these limits, and then some paths | |
2017 | * reappear. | |
2018 | */ | |
2019 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2020 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2021 | if (live_map) |
2022 | limits = md->queue->limits; | |
83d5e5b0 | 2023 | dm_put_live_table_fast(md); |
3ae70656 MS |
2024 | } |
2025 | ||
87eb5b21 MC |
2026 | if (!live_map) { |
2027 | r = dm_calculate_queue_limits(table, &limits); | |
2028 | if (r) { | |
2029 | map = ERR_PTR(r); | |
2030 | goto out; | |
2031 | } | |
042d2a9b | 2032 | } |
754c5fc7 | 2033 | |
042d2a9b | 2034 | map = __bind(md, table, &limits); |
1da177e4 | 2035 | |
93c534ae | 2036 | out: |
e61290a4 | 2037 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2038 | return map; |
1da177e4 LT |
2039 | } |
2040 | ||
2041 | /* | |
2042 | * Functions to lock and unlock any filesystem running on the | |
2043 | * device. | |
2044 | */ | |
2ca3310e | 2045 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2046 | { |
e39e2e95 | 2047 | int r; |
1da177e4 LT |
2048 | |
2049 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2050 | |
db8fef4f | 2051 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2052 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2053 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2054 | md->frozen_sb = NULL; |
2055 | return r; | |
dfbe03f6 AK |
2056 | } |
2057 | ||
aa8d7c2f AK |
2058 | set_bit(DMF_FROZEN, &md->flags); |
2059 | ||
1da177e4 LT |
2060 | return 0; |
2061 | } | |
2062 | ||
2ca3310e | 2063 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2064 | { |
aa8d7c2f AK |
2065 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2066 | return; | |
2067 | ||
db8fef4f | 2068 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2069 | md->frozen_sb = NULL; |
aa8d7c2f | 2070 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2071 | } |
2072 | ||
2073 | /* | |
b48633f8 BVA |
2074 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2075 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2076 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2077 | * | |
ffcc3936 MS |
2078 | * If __dm_suspend returns 0, the device is completely quiescent |
2079 | * now. There is no request-processing activity. All new requests | |
2080 | * are being added to md->deferred list. | |
cec47e3d | 2081 | * |
ffcc3936 | 2082 | * Caller must hold md->suspend_lock |
cec47e3d | 2083 | */ |
ffcc3936 | 2084 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2085 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2086 | int dmf_suspended_flag) |
1da177e4 | 2087 | { |
ffcc3936 MS |
2088 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2089 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2090 | int r; | |
1da177e4 | 2091 | |
5a8f1f80 BVA |
2092 | lockdep_assert_held(&md->suspend_lock); |
2093 | ||
2e93ccc1 KU |
2094 | /* |
2095 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2096 | * This flag is cleared before dm_suspend returns. | |
2097 | */ | |
2098 | if (noflush) | |
2099 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2100 | ||
d67ee213 MS |
2101 | /* |
2102 | * This gets reverted if there's an error later and the targets | |
2103 | * provide the .presuspend_undo hook. | |
2104 | */ | |
cf222b37 AK |
2105 | dm_table_presuspend_targets(map); |
2106 | ||
32a926da | 2107 | /* |
9f518b27 KU |
2108 | * Flush I/O to the device. |
2109 | * Any I/O submitted after lock_fs() may not be flushed. | |
2110 | * noflush takes precedence over do_lockfs. | |
2111 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2112 | */ |
2113 | if (!noflush && do_lockfs) { | |
2114 | r = lock_fs(md); | |
d67ee213 MS |
2115 | if (r) { |
2116 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2117 | return r; |
d67ee213 | 2118 | } |
aa8d7c2f | 2119 | } |
1da177e4 LT |
2120 | |
2121 | /* | |
3b00b203 MP |
2122 | * Here we must make sure that no processes are submitting requests |
2123 | * to target drivers i.e. no one may be executing | |
2124 | * __split_and_process_bio. This is called from dm_request and | |
2125 | * dm_wq_work. | |
2126 | * | |
2127 | * To get all processes out of __split_and_process_bio in dm_request, | |
2128 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2129 | * __split_and_process_bio from dm_request and quiesce the thread |
2130 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2131 | * flush_workqueue(md->wq). | |
1da177e4 | 2132 | */ |
1eb787ec | 2133 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2134 | if (map) |
2135 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2136 | |
d0bcb878 | 2137 | /* |
29e4013d TH |
2138 | * Stop md->queue before flushing md->wq in case request-based |
2139 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2140 | */ |
2eb6e1e3 | 2141 | if (dm_request_based(md)) { |
eca7ee6d | 2142 | dm_stop_queue(md->queue); |
02233342 | 2143 | if (md->kworker_task) |
3989144f | 2144 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2145 | } |
cec47e3d | 2146 | |
d0bcb878 KU |
2147 | flush_workqueue(md->wq); |
2148 | ||
1da177e4 | 2149 | /* |
3b00b203 MP |
2150 | * At this point no more requests are entering target request routines. |
2151 | * We call dm_wait_for_completion to wait for all existing requests | |
2152 | * to finish. | |
1da177e4 | 2153 | */ |
b48633f8 | 2154 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2155 | if (!r) |
2156 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2157 | |
6d6f10df | 2158 | if (noflush) |
022c2611 | 2159 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2160 | if (map) |
2161 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2162 | |
1da177e4 | 2163 | /* were we interrupted ? */ |
46125c1c | 2164 | if (r < 0) { |
9a1fb464 | 2165 | dm_queue_flush(md); |
73d410c0 | 2166 | |
cec47e3d | 2167 | if (dm_request_based(md)) |
eca7ee6d | 2168 | dm_start_queue(md->queue); |
cec47e3d | 2169 | |
2ca3310e | 2170 | unlock_fs(md); |
d67ee213 | 2171 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2172 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2173 | } |
1da177e4 | 2174 | |
ffcc3936 MS |
2175 | return r; |
2176 | } | |
2177 | ||
2178 | /* | |
2179 | * We need to be able to change a mapping table under a mounted | |
2180 | * filesystem. For example we might want to move some data in | |
2181 | * the background. Before the table can be swapped with | |
2182 | * dm_bind_table, dm_suspend must be called to flush any in | |
2183 | * flight bios and ensure that any further io gets deferred. | |
2184 | */ | |
2185 | /* | |
2186 | * Suspend mechanism in request-based dm. | |
2187 | * | |
2188 | * 1. Flush all I/Os by lock_fs() if needed. | |
2189 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2190 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2191 | * | |
2192 | * To abort suspend, start the request_queue. | |
2193 | */ | |
2194 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2195 | { | |
2196 | struct dm_table *map = NULL; | |
2197 | int r = 0; | |
2198 | ||
2199 | retry: | |
2200 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2201 | ||
2202 | if (dm_suspended_md(md)) { | |
2203 | r = -EINVAL; | |
2204 | goto out_unlock; | |
2205 | } | |
2206 | ||
2207 | if (dm_suspended_internally_md(md)) { | |
2208 | /* already internally suspended, wait for internal resume */ | |
2209 | mutex_unlock(&md->suspend_lock); | |
2210 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2211 | if (r) | |
2212 | return r; | |
2213 | goto retry; | |
2214 | } | |
2215 | ||
a12f5d48 | 2216 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2217 | |
eaf9a736 | 2218 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2219 | if (r) |
2220 | goto out_unlock; | |
3b00b203 | 2221 | |
4d4471cb KU |
2222 | dm_table_postsuspend_targets(map); |
2223 | ||
d287483d | 2224 | out_unlock: |
e61290a4 | 2225 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2226 | return r; |
1da177e4 LT |
2227 | } |
2228 | ||
ffcc3936 MS |
2229 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2230 | { | |
2231 | if (map) { | |
2232 | int r = dm_table_resume_targets(map); | |
2233 | if (r) | |
2234 | return r; | |
2235 | } | |
2236 | ||
2237 | dm_queue_flush(md); | |
2238 | ||
2239 | /* | |
2240 | * Flushing deferred I/Os must be done after targets are resumed | |
2241 | * so that mapping of targets can work correctly. | |
2242 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2243 | */ | |
2244 | if (dm_request_based(md)) | |
eca7ee6d | 2245 | dm_start_queue(md->queue); |
ffcc3936 MS |
2246 | |
2247 | unlock_fs(md); | |
2248 | ||
2249 | return 0; | |
2250 | } | |
2251 | ||
1da177e4 LT |
2252 | int dm_resume(struct mapped_device *md) |
2253 | { | |
8dc23658 | 2254 | int r; |
cf222b37 | 2255 | struct dm_table *map = NULL; |
1da177e4 | 2256 | |
ffcc3936 | 2257 | retry: |
8dc23658 | 2258 | r = -EINVAL; |
ffcc3936 MS |
2259 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2260 | ||
4f186f8b | 2261 | if (!dm_suspended_md(md)) |
cf222b37 | 2262 | goto out; |
cf222b37 | 2263 | |
ffcc3936 MS |
2264 | if (dm_suspended_internally_md(md)) { |
2265 | /* already internally suspended, wait for internal resume */ | |
2266 | mutex_unlock(&md->suspend_lock); | |
2267 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2268 | if (r) | |
2269 | return r; | |
2270 | goto retry; | |
2271 | } | |
2272 | ||
a12f5d48 | 2273 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2274 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2275 | goto out; |
1da177e4 | 2276 | |
ffcc3936 | 2277 | r = __dm_resume(md, map); |
8757b776 MB |
2278 | if (r) |
2279 | goto out; | |
2ca3310e | 2280 | |
2ca3310e | 2281 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2282 | out: |
e61290a4 | 2283 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2284 | |
cf222b37 | 2285 | return r; |
1da177e4 LT |
2286 | } |
2287 | ||
fd2ed4d2 MP |
2288 | /* |
2289 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2290 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2291 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2292 | */ |
2293 | ||
ffcc3936 | 2294 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2295 | { |
ffcc3936 MS |
2296 | struct dm_table *map = NULL; |
2297 | ||
96b26c8c | 2298 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2299 | return; /* nested internal suspend */ |
2300 | ||
2301 | if (dm_suspended_md(md)) { | |
2302 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2303 | return; /* nest suspend */ | |
2304 | } | |
2305 | ||
a12f5d48 | 2306 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2307 | |
2308 | /* | |
2309 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2310 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2311 | * would require changing .presuspend to return an error -- avoid this | |
2312 | * until there is a need for more elaborate variants of internal suspend. | |
2313 | */ | |
eaf9a736 MS |
2314 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2315 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2316 | |
2317 | dm_table_postsuspend_targets(map); | |
2318 | } | |
2319 | ||
2320 | static void __dm_internal_resume(struct mapped_device *md) | |
2321 | { | |
96b26c8c MP |
2322 | BUG_ON(!md->internal_suspend_count); |
2323 | ||
2324 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2325 | return; /* resume from nested internal suspend */ |
2326 | ||
fd2ed4d2 | 2327 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2328 | goto done; /* resume from nested suspend */ |
2329 | ||
2330 | /* | |
2331 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2332 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2333 | */ | |
2334 | (void) __dm_resume(md, NULL); | |
2335 | ||
2336 | done: | |
2337 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2338 | smp_mb__after_atomic(); | |
2339 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2340 | } | |
2341 | ||
2342 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2343 | { | |
2344 | mutex_lock(&md->suspend_lock); | |
2345 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2346 | mutex_unlock(&md->suspend_lock); | |
2347 | } | |
2348 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2349 | ||
2350 | void dm_internal_resume(struct mapped_device *md) | |
2351 | { | |
2352 | mutex_lock(&md->suspend_lock); | |
2353 | __dm_internal_resume(md); | |
2354 | mutex_unlock(&md->suspend_lock); | |
2355 | } | |
2356 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2357 | ||
2358 | /* | |
2359 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2360 | * which prevents interaction with userspace-driven suspend. | |
2361 | */ | |
2362 | ||
2363 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2364 | { | |
2365 | mutex_lock(&md->suspend_lock); | |
2366 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2367 | return; |
2368 | ||
2369 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2370 | synchronize_srcu(&md->io_barrier); | |
2371 | flush_workqueue(md->wq); | |
2372 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2373 | } | |
b735fede | 2374 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2375 | |
ffcc3936 | 2376 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2377 | { |
ffcc3936 | 2378 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2379 | goto done; |
2380 | ||
2381 | dm_queue_flush(md); | |
2382 | ||
2383 | done: | |
2384 | mutex_unlock(&md->suspend_lock); | |
2385 | } | |
b735fede | 2386 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2387 | |
1da177e4 LT |
2388 | /*----------------------------------------------------------------- |
2389 | * Event notification. | |
2390 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2391 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2392 | unsigned cookie) |
69267a30 | 2393 | { |
60935eb2 MB |
2394 | char udev_cookie[DM_COOKIE_LENGTH]; |
2395 | char *envp[] = { udev_cookie, NULL }; | |
2396 | ||
2397 | if (!cookie) | |
3abf85b5 | 2398 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2399 | else { |
2400 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2401 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2402 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2403 | action, envp); | |
60935eb2 | 2404 | } |
69267a30 AK |
2405 | } |
2406 | ||
7a8c3d3b MA |
2407 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2408 | { | |
2409 | return atomic_add_return(1, &md->uevent_seq); | |
2410 | } | |
2411 | ||
1da177e4 LT |
2412 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2413 | { | |
2414 | return atomic_read(&md->event_nr); | |
2415 | } | |
2416 | ||
2417 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2418 | { | |
2419 | return wait_event_interruptible(md->eventq, | |
2420 | (event_nr != atomic_read(&md->event_nr))); | |
2421 | } | |
2422 | ||
7a8c3d3b MA |
2423 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2424 | { | |
2425 | unsigned long flags; | |
2426 | ||
2427 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2428 | list_add(elist, &md->uevent_list); | |
2429 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2430 | } | |
2431 | ||
1da177e4 LT |
2432 | /* |
2433 | * The gendisk is only valid as long as you have a reference | |
2434 | * count on 'md'. | |
2435 | */ | |
2436 | struct gendisk *dm_disk(struct mapped_device *md) | |
2437 | { | |
2438 | return md->disk; | |
2439 | } | |
65ff5b7d | 2440 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2441 | |
784aae73 MB |
2442 | struct kobject *dm_kobject(struct mapped_device *md) |
2443 | { | |
2995fa78 | 2444 | return &md->kobj_holder.kobj; |
784aae73 MB |
2445 | } |
2446 | ||
784aae73 MB |
2447 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2448 | { | |
2449 | struct mapped_device *md; | |
2450 | ||
2995fa78 | 2451 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2452 | |
4d89b7b4 | 2453 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2454 | dm_deleting_md(md)) |
4d89b7b4 MB |
2455 | return NULL; |
2456 | ||
784aae73 MB |
2457 | dm_get(md); |
2458 | return md; | |
2459 | } | |
2460 | ||
4f186f8b | 2461 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2462 | { |
2463 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2464 | } | |
2465 | ||
ffcc3936 MS |
2466 | int dm_suspended_internally_md(struct mapped_device *md) |
2467 | { | |
2468 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2469 | } | |
2470 | ||
2c140a24 MP |
2471 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2472 | { | |
2473 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2474 | } | |
2475 | ||
64dbce58 KU |
2476 | int dm_suspended(struct dm_target *ti) |
2477 | { | |
ecdb2e25 | 2478 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2479 | } |
2480 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2481 | ||
2e93ccc1 KU |
2482 | int dm_noflush_suspending(struct dm_target *ti) |
2483 | { | |
ecdb2e25 | 2484 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2485 | } |
2486 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2487 | ||
78d8e58a | 2488 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
30187e1d | 2489 | unsigned integrity, unsigned per_io_data_size) |
e6ee8c0b | 2490 | { |
115485e8 | 2491 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2492 | unsigned int pool_size = 0; |
5f015204 | 2493 | unsigned int front_pad; |
e6ee8c0b KU |
2494 | |
2495 | if (!pools) | |
4e6e36c3 | 2496 | return NULL; |
e6ee8c0b | 2497 | |
78d8e58a MS |
2498 | switch (type) { |
2499 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2500 | case DM_TYPE_DAX_BIO_BASED: |
78d8e58a | 2501 | pool_size = dm_get_reserved_bio_based_ios(); |
30187e1d | 2502 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
eb8db831 CH |
2503 | |
2504 | pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); | |
2505 | if (!pools->io_pool) | |
2506 | goto out; | |
78d8e58a MS |
2507 | break; |
2508 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2509 | case DM_TYPE_MQ_REQUEST_BASED: |
eb8db831 | 2510 | pool_size = dm_get_reserved_rq_based_ios(); |
78d8e58a | 2511 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2512 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2513 | break; |
2514 | default: | |
2515 | BUG(); | |
2516 | } | |
2517 | ||
3d8aab2d | 2518 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 2519 | if (!pools->bs) |
5f015204 | 2520 | goto out; |
e6ee8c0b | 2521 | |
a91a2785 | 2522 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2523 | goto out; |
a91a2785 | 2524 | |
e6ee8c0b | 2525 | return pools; |
5f1b670d | 2526 | |
5f1b670d CH |
2527 | out: |
2528 | dm_free_md_mempools(pools); | |
78d8e58a | 2529 | |
4e6e36c3 | 2530 | return NULL; |
e6ee8c0b KU |
2531 | } |
2532 | ||
2533 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2534 | { | |
2535 | if (!pools) | |
2536 | return; | |
2537 | ||
6f65985e | 2538 | mempool_destroy(pools->io_pool); |
1ae49ea2 | 2539 | |
e6ee8c0b KU |
2540 | if (pools->bs) |
2541 | bioset_free(pools->bs); | |
2542 | ||
2543 | kfree(pools); | |
2544 | } | |
2545 | ||
9c72bad1 CH |
2546 | struct dm_pr { |
2547 | u64 old_key; | |
2548 | u64 new_key; | |
2549 | u32 flags; | |
2550 | bool fail_early; | |
2551 | }; | |
2552 | ||
2553 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2554 | void *data) | |
71cdb697 CH |
2555 | { |
2556 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2557 | struct dm_table *table; |
2558 | struct dm_target *ti; | |
2559 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2560 | |
9c72bad1 CH |
2561 | table = dm_get_live_table(md, &srcu_idx); |
2562 | if (!table || !dm_table_get_size(table)) | |
2563 | goto out; | |
71cdb697 | 2564 | |
9c72bad1 CH |
2565 | /* We only support devices that have a single target */ |
2566 | if (dm_table_get_num_targets(table) != 1) | |
2567 | goto out; | |
2568 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2569 | |
9c72bad1 CH |
2570 | ret = -EINVAL; |
2571 | if (!ti->type->iterate_devices) | |
2572 | goto out; | |
2573 | ||
2574 | ret = ti->type->iterate_devices(ti, fn, data); | |
2575 | out: | |
2576 | dm_put_live_table(md, srcu_idx); | |
2577 | return ret; | |
2578 | } | |
2579 | ||
2580 | /* | |
2581 | * For register / unregister we need to manually call out to every path. | |
2582 | */ | |
2583 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
2584 | sector_t start, sector_t len, void *data) | |
2585 | { | |
2586 | struct dm_pr *pr = data; | |
2587 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
2588 | ||
2589 | if (!ops || !ops->pr_register) | |
2590 | return -EOPNOTSUPP; | |
2591 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
2592 | } | |
2593 | ||
2594 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
2595 | u32 flags) | |
2596 | { | |
2597 | struct dm_pr pr = { | |
2598 | .old_key = old_key, | |
2599 | .new_key = new_key, | |
2600 | .flags = flags, | |
2601 | .fail_early = true, | |
2602 | }; | |
2603 | int ret; | |
2604 | ||
2605 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
2606 | if (ret && new_key) { | |
2607 | /* unregister all paths if we failed to register any path */ | |
2608 | pr.old_key = new_key; | |
2609 | pr.new_key = 0; | |
2610 | pr.flags = 0; | |
2611 | pr.fail_early = false; | |
2612 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
2613 | } | |
2614 | ||
2615 | return ret; | |
71cdb697 CH |
2616 | } |
2617 | ||
2618 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 2619 | u32 flags) |
71cdb697 CH |
2620 | { |
2621 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2622 | const struct pr_ops *ops; | |
71cdb697 | 2623 | fmode_t mode; |
956a4025 | 2624 | int r; |
71cdb697 | 2625 | |
956a4025 | 2626 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2627 | if (r < 0) |
2628 | return r; | |
2629 | ||
2630 | ops = bdev->bd_disk->fops->pr_ops; | |
2631 | if (ops && ops->pr_reserve) | |
2632 | r = ops->pr_reserve(bdev, key, type, flags); | |
2633 | else | |
2634 | r = -EOPNOTSUPP; | |
2635 | ||
956a4025 | 2636 | bdput(bdev); |
71cdb697 CH |
2637 | return r; |
2638 | } | |
2639 | ||
2640 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
2641 | { | |
2642 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2643 | const struct pr_ops *ops; | |
71cdb697 | 2644 | fmode_t mode; |
956a4025 | 2645 | int r; |
71cdb697 | 2646 | |
956a4025 | 2647 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2648 | if (r < 0) |
2649 | return r; | |
2650 | ||
2651 | ops = bdev->bd_disk->fops->pr_ops; | |
2652 | if (ops && ops->pr_release) | |
2653 | r = ops->pr_release(bdev, key, type); | |
2654 | else | |
2655 | r = -EOPNOTSUPP; | |
2656 | ||
956a4025 | 2657 | bdput(bdev); |
71cdb697 CH |
2658 | return r; |
2659 | } | |
2660 | ||
2661 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 2662 | enum pr_type type, bool abort) |
71cdb697 CH |
2663 | { |
2664 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2665 | const struct pr_ops *ops; | |
71cdb697 | 2666 | fmode_t mode; |
956a4025 | 2667 | int r; |
71cdb697 | 2668 | |
956a4025 | 2669 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2670 | if (r < 0) |
2671 | return r; | |
2672 | ||
2673 | ops = bdev->bd_disk->fops->pr_ops; | |
2674 | if (ops && ops->pr_preempt) | |
2675 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
2676 | else | |
2677 | r = -EOPNOTSUPP; | |
2678 | ||
956a4025 | 2679 | bdput(bdev); |
71cdb697 CH |
2680 | return r; |
2681 | } | |
2682 | ||
2683 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
2684 | { | |
2685 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2686 | const struct pr_ops *ops; | |
71cdb697 | 2687 | fmode_t mode; |
956a4025 | 2688 | int r; |
71cdb697 | 2689 | |
956a4025 | 2690 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2691 | if (r < 0) |
2692 | return r; | |
2693 | ||
2694 | ops = bdev->bd_disk->fops->pr_ops; | |
2695 | if (ops && ops->pr_clear) | |
2696 | r = ops->pr_clear(bdev, key); | |
2697 | else | |
2698 | r = -EOPNOTSUPP; | |
2699 | ||
956a4025 | 2700 | bdput(bdev); |
71cdb697 CH |
2701 | return r; |
2702 | } | |
2703 | ||
2704 | static const struct pr_ops dm_pr_ops = { | |
2705 | .pr_register = dm_pr_register, | |
2706 | .pr_reserve = dm_pr_reserve, | |
2707 | .pr_release = dm_pr_release, | |
2708 | .pr_preempt = dm_pr_preempt, | |
2709 | .pr_clear = dm_pr_clear, | |
2710 | }; | |
2711 | ||
83d5cde4 | 2712 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2713 | .open = dm_blk_open, |
2714 | .release = dm_blk_close, | |
aa129a22 | 2715 | .ioctl = dm_blk_ioctl, |
545ed20e | 2716 | .direct_access = dm_blk_direct_access, |
3ac51e74 | 2717 | .getgeo = dm_blk_getgeo, |
71cdb697 | 2718 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
2719 | .owner = THIS_MODULE |
2720 | }; | |
2721 | ||
1da177e4 LT |
2722 | /* |
2723 | * module hooks | |
2724 | */ | |
2725 | module_init(dm_init); | |
2726 | module_exit(dm_exit); | |
2727 | ||
2728 | module_param(major, uint, 0); | |
2729 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 2730 | |
e8603136 MS |
2731 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
2732 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
2733 | ||
115485e8 MS |
2734 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
2735 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
2736 | ||
1da177e4 LT |
2737 | MODULE_DESCRIPTION(DM_NAME " driver"); |
2738 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2739 | MODULE_LICENSE("GPL"); |