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> |
174cd4b1 | 15 | #include <linux/sched/signal.h> |
1da177e4 LT |
16 | #include <linux/blkpg.h> |
17 | #include <linux/bio.h> | |
1da177e4 LT |
18 | #include <linux/mempool.h> |
19 | #include <linux/slab.h> | |
20 | #include <linux/idr.h> | |
3ac51e74 | 21 | #include <linux/hdreg.h> |
3f77316d | 22 | #include <linux/delay.h> |
ffcc3936 | 23 | #include <linux/wait.h> |
71cdb697 | 24 | #include <linux/pr.h> |
55782138 | 25 | |
72d94861 AK |
26 | #define DM_MSG_PREFIX "core" |
27 | ||
71a16736 NK |
28 | #ifdef CONFIG_PRINTK |
29 | /* | |
30 | * ratelimit state to be used in DMXXX_LIMIT(). | |
31 | */ | |
32 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
33 | DEFAULT_RATELIMIT_INTERVAL, | |
34 | DEFAULT_RATELIMIT_BURST); | |
35 | EXPORT_SYMBOL(dm_ratelimit_state); | |
36 | #endif | |
37 | ||
60935eb2 MB |
38 | /* |
39 | * Cookies are numeric values sent with CHANGE and REMOVE | |
40 | * uevents while resuming, removing or renaming the device. | |
41 | */ | |
42 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
43 | #define DM_COOKIE_LENGTH 24 | |
44 | ||
1da177e4 LT |
45 | static const char *_name = DM_NAME; |
46 | ||
47 | static unsigned int major = 0; | |
48 | static unsigned int _major = 0; | |
49 | ||
d15b774c AK |
50 | static DEFINE_IDR(_minor_idr); |
51 | ||
f32c10b0 | 52 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
53 | |
54 | static void do_deferred_remove(struct work_struct *w); | |
55 | ||
56 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
57 | ||
acfe0ad7 MP |
58 | static struct workqueue_struct *deferred_remove_workqueue; |
59 | ||
1da177e4 LT |
60 | /* |
61 | * One of these is allocated per bio. | |
62 | */ | |
63 | struct dm_io { | |
64 | struct mapped_device *md; | |
65 | int error; | |
1da177e4 | 66 | atomic_t io_count; |
6ae2fa67 | 67 | struct bio *bio; |
3eaf840e | 68 | unsigned long start_time; |
f88fb981 | 69 | spinlock_t endio_lock; |
fd2ed4d2 | 70 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
71 | }; |
72 | ||
ba61fdd1 JM |
73 | #define MINOR_ALLOCED ((void *)-1) |
74 | ||
1da177e4 LT |
75 | /* |
76 | * Bits for the md->flags field. | |
77 | */ | |
1eb787ec | 78 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 79 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 80 | #define DMF_FROZEN 2 |
fba9f90e | 81 | #define DMF_FREEING 3 |
5c6bd75d | 82 | #define DMF_DELETING 4 |
2e93ccc1 | 83 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
84 | #define DMF_DEFERRED_REMOVE 6 |
85 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 86 | |
115485e8 | 87 | #define DM_NUMA_NODE NUMA_NO_NODE |
115485e8 | 88 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 89 | |
e6ee8c0b KU |
90 | /* |
91 | * For mempools pre-allocation at the table loading time. | |
92 | */ | |
93 | struct dm_md_mempools { | |
94 | mempool_t *io_pool; | |
e6ee8c0b KU |
95 | struct bio_set *bs; |
96 | }; | |
97 | ||
86f1152b BM |
98 | struct table_device { |
99 | struct list_head list; | |
100 | atomic_t count; | |
101 | struct dm_dev dm_dev; | |
102 | }; | |
103 | ||
e18b890b | 104 | static struct kmem_cache *_io_cache; |
8fbf26ad | 105 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 106 | static struct kmem_cache *_rq_cache; |
94818742 | 107 | |
e8603136 MS |
108 | /* |
109 | * Bio-based DM's mempools' reserved IOs set by the user. | |
110 | */ | |
4cc96131 | 111 | #define RESERVED_BIO_BASED_IOS 16 |
e8603136 MS |
112 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
113 | ||
115485e8 MS |
114 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
115 | { | |
116 | int param = ACCESS_ONCE(*module_param); | |
117 | int modified_param = 0; | |
118 | bool modified = true; | |
119 | ||
120 | if (param < min) | |
121 | modified_param = min; | |
122 | else if (param > max) | |
123 | modified_param = max; | |
124 | else | |
125 | modified = false; | |
126 | ||
127 | if (modified) { | |
128 | (void)cmpxchg(module_param, param, modified_param); | |
129 | param = modified_param; | |
130 | } | |
131 | ||
132 | return param; | |
133 | } | |
134 | ||
4cc96131 MS |
135 | unsigned __dm_get_module_param(unsigned *module_param, |
136 | unsigned def, unsigned max) | |
f4790826 | 137 | { |
09c2d531 MS |
138 | unsigned param = ACCESS_ONCE(*module_param); |
139 | unsigned modified_param = 0; | |
f4790826 | 140 | |
09c2d531 MS |
141 | if (!param) |
142 | modified_param = def; | |
143 | else if (param > max) | |
144 | modified_param = max; | |
f4790826 | 145 | |
09c2d531 MS |
146 | if (modified_param) { |
147 | (void)cmpxchg(module_param, param, modified_param); | |
148 | param = modified_param; | |
f4790826 MS |
149 | } |
150 | ||
09c2d531 | 151 | return param; |
f4790826 MS |
152 | } |
153 | ||
e8603136 MS |
154 | unsigned dm_get_reserved_bio_based_ios(void) |
155 | { | |
09c2d531 | 156 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 157 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
158 | } |
159 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
160 | ||
115485e8 MS |
161 | static unsigned dm_get_numa_node(void) |
162 | { | |
163 | return __dm_get_module_param_int(&dm_numa_node, | |
164 | DM_NUMA_NODE, num_online_nodes() - 1); | |
165 | } | |
166 | ||
1da177e4 LT |
167 | static int __init local_init(void) |
168 | { | |
51157b4a | 169 | int r = -ENOMEM; |
1da177e4 | 170 | |
1da177e4 | 171 | /* allocate a slab for the dm_ios */ |
028867ac | 172 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 173 | if (!_io_cache) |
51157b4a | 174 | return r; |
1da177e4 | 175 | |
8fbf26ad KU |
176 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
177 | if (!_rq_tio_cache) | |
dba14160 | 178 | goto out_free_io_cache; |
8fbf26ad | 179 | |
eca7ee6d | 180 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
1ae49ea2 MS |
181 | __alignof__(struct request), 0, NULL); |
182 | if (!_rq_cache) | |
183 | goto out_free_rq_tio_cache; | |
184 | ||
51e5b2bd | 185 | r = dm_uevent_init(); |
51157b4a | 186 | if (r) |
1ae49ea2 | 187 | goto out_free_rq_cache; |
51e5b2bd | 188 | |
acfe0ad7 MP |
189 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
190 | if (!deferred_remove_workqueue) { | |
191 | r = -ENOMEM; | |
192 | goto out_uevent_exit; | |
193 | } | |
194 | ||
1da177e4 LT |
195 | _major = major; |
196 | r = register_blkdev(_major, _name); | |
51157b4a | 197 | if (r < 0) |
acfe0ad7 | 198 | goto out_free_workqueue; |
1da177e4 LT |
199 | |
200 | if (!_major) | |
201 | _major = r; | |
202 | ||
203 | return 0; | |
51157b4a | 204 | |
acfe0ad7 MP |
205 | out_free_workqueue: |
206 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
207 | out_uevent_exit: |
208 | dm_uevent_exit(); | |
1ae49ea2 MS |
209 | out_free_rq_cache: |
210 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
211 | out_free_rq_tio_cache: |
212 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
213 | out_free_io_cache: |
214 | kmem_cache_destroy(_io_cache); | |
215 | ||
216 | return r; | |
1da177e4 LT |
217 | } |
218 | ||
219 | static void local_exit(void) | |
220 | { | |
2c140a24 | 221 | flush_scheduled_work(); |
acfe0ad7 | 222 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 223 | |
1ae49ea2 | 224 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 225 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 226 | kmem_cache_destroy(_io_cache); |
00d59405 | 227 | unregister_blkdev(_major, _name); |
51e5b2bd | 228 | dm_uevent_exit(); |
1da177e4 LT |
229 | |
230 | _major = 0; | |
231 | ||
232 | DMINFO("cleaned up"); | |
233 | } | |
234 | ||
b9249e55 | 235 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
236 | local_init, |
237 | dm_target_init, | |
238 | dm_linear_init, | |
239 | dm_stripe_init, | |
952b3557 | 240 | dm_io_init, |
945fa4d2 | 241 | dm_kcopyd_init, |
1da177e4 | 242 | dm_interface_init, |
fd2ed4d2 | 243 | dm_statistics_init, |
1da177e4 LT |
244 | }; |
245 | ||
b9249e55 | 246 | static void (*_exits[])(void) = { |
1da177e4 LT |
247 | local_exit, |
248 | dm_target_exit, | |
249 | dm_linear_exit, | |
250 | dm_stripe_exit, | |
952b3557 | 251 | dm_io_exit, |
945fa4d2 | 252 | dm_kcopyd_exit, |
1da177e4 | 253 | dm_interface_exit, |
fd2ed4d2 | 254 | dm_statistics_exit, |
1da177e4 LT |
255 | }; |
256 | ||
257 | static int __init dm_init(void) | |
258 | { | |
259 | const int count = ARRAY_SIZE(_inits); | |
260 | ||
261 | int r, i; | |
262 | ||
263 | for (i = 0; i < count; i++) { | |
264 | r = _inits[i](); | |
265 | if (r) | |
266 | goto bad; | |
267 | } | |
268 | ||
269 | return 0; | |
270 | ||
271 | bad: | |
272 | while (i--) | |
273 | _exits[i](); | |
274 | ||
275 | return r; | |
276 | } | |
277 | ||
278 | static void __exit dm_exit(void) | |
279 | { | |
280 | int i = ARRAY_SIZE(_exits); | |
281 | ||
282 | while (i--) | |
283 | _exits[i](); | |
d15b774c AK |
284 | |
285 | /* | |
286 | * Should be empty by this point. | |
287 | */ | |
d15b774c | 288 | idr_destroy(&_minor_idr); |
1da177e4 LT |
289 | } |
290 | ||
291 | /* | |
292 | * Block device functions | |
293 | */ | |
432a212c MA |
294 | int dm_deleting_md(struct mapped_device *md) |
295 | { | |
296 | return test_bit(DMF_DELETING, &md->flags); | |
297 | } | |
298 | ||
fe5f9f2c | 299 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
300 | { |
301 | struct mapped_device *md; | |
302 | ||
fba9f90e JM |
303 | spin_lock(&_minor_lock); |
304 | ||
fe5f9f2c | 305 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
306 | if (!md) |
307 | goto out; | |
308 | ||
5c6bd75d | 309 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 310 | dm_deleting_md(md)) { |
fba9f90e JM |
311 | md = NULL; |
312 | goto out; | |
313 | } | |
314 | ||
1da177e4 | 315 | dm_get(md); |
5c6bd75d | 316 | atomic_inc(&md->open_count); |
fba9f90e JM |
317 | out: |
318 | spin_unlock(&_minor_lock); | |
319 | ||
320 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
321 | } |
322 | ||
db2a144b | 323 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 324 | { |
63a4f065 | 325 | struct mapped_device *md; |
6e9624b8 | 326 | |
4a1aeb98 MB |
327 | spin_lock(&_minor_lock); |
328 | ||
63a4f065 MS |
329 | md = disk->private_data; |
330 | if (WARN_ON(!md)) | |
331 | goto out; | |
332 | ||
2c140a24 MP |
333 | if (atomic_dec_and_test(&md->open_count) && |
334 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 335 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 336 | |
1da177e4 | 337 | dm_put(md); |
63a4f065 | 338 | out: |
4a1aeb98 | 339 | spin_unlock(&_minor_lock); |
1da177e4 LT |
340 | } |
341 | ||
5c6bd75d AK |
342 | int dm_open_count(struct mapped_device *md) |
343 | { | |
344 | return atomic_read(&md->open_count); | |
345 | } | |
346 | ||
347 | /* | |
348 | * Guarantees nothing is using the device before it's deleted. | |
349 | */ | |
2c140a24 | 350 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
351 | { |
352 | int r = 0; | |
353 | ||
354 | spin_lock(&_minor_lock); | |
355 | ||
2c140a24 | 356 | if (dm_open_count(md)) { |
5c6bd75d | 357 | r = -EBUSY; |
2c140a24 MP |
358 | if (mark_deferred) |
359 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
360 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
361 | r = -EEXIST; | |
5c6bd75d AK |
362 | else |
363 | set_bit(DMF_DELETING, &md->flags); | |
364 | ||
365 | spin_unlock(&_minor_lock); | |
366 | ||
367 | return r; | |
368 | } | |
369 | ||
2c140a24 MP |
370 | int dm_cancel_deferred_remove(struct mapped_device *md) |
371 | { | |
372 | int r = 0; | |
373 | ||
374 | spin_lock(&_minor_lock); | |
375 | ||
376 | if (test_bit(DMF_DELETING, &md->flags)) | |
377 | r = -EBUSY; | |
378 | else | |
379 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
380 | ||
381 | spin_unlock(&_minor_lock); | |
382 | ||
383 | return r; | |
384 | } | |
385 | ||
386 | static void do_deferred_remove(struct work_struct *w) | |
387 | { | |
388 | dm_deferred_remove(); | |
389 | } | |
390 | ||
fd2ed4d2 MP |
391 | sector_t dm_get_size(struct mapped_device *md) |
392 | { | |
393 | return get_capacity(md->disk); | |
394 | } | |
395 | ||
9974fa2c MS |
396 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
397 | { | |
398 | return md->queue; | |
399 | } | |
400 | ||
fd2ed4d2 MP |
401 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
402 | { | |
403 | return &md->stats; | |
404 | } | |
405 | ||
3ac51e74 DW |
406 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
407 | { | |
408 | struct mapped_device *md = bdev->bd_disk->private_data; | |
409 | ||
410 | return dm_get_geometry(md, geo); | |
411 | } | |
412 | ||
956a4025 MS |
413 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
414 | struct block_device **bdev, | |
415 | fmode_t *mode) | |
aa129a22 | 416 | { |
66482026 | 417 | struct dm_target *tgt; |
6c182cd8 | 418 | struct dm_table *map; |
956a4025 | 419 | int srcu_idx, r; |
aa129a22 | 420 | |
6c182cd8 | 421 | retry: |
e56f81e0 | 422 | r = -ENOTTY; |
956a4025 | 423 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
424 | if (!map || !dm_table_get_size(map)) |
425 | goto out; | |
426 | ||
427 | /* We only support devices that have a single target */ | |
428 | if (dm_table_get_num_targets(map) != 1) | |
429 | goto out; | |
430 | ||
66482026 MS |
431 | tgt = dm_table_get_target(map, 0); |
432 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 433 | goto out; |
aa129a22 | 434 | |
4f186f8b | 435 | if (dm_suspended_md(md)) { |
aa129a22 MB |
436 | r = -EAGAIN; |
437 | goto out; | |
438 | } | |
439 | ||
66482026 | 440 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
441 | if (r < 0) |
442 | goto out; | |
aa129a22 | 443 | |
956a4025 MS |
444 | bdgrab(*bdev); |
445 | dm_put_live_table(md, srcu_idx); | |
e56f81e0 | 446 | return r; |
aa129a22 | 447 | |
aa129a22 | 448 | out: |
956a4025 | 449 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 450 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
451 | msleep(10); |
452 | goto retry; | |
453 | } | |
e56f81e0 CH |
454 | return r; |
455 | } | |
456 | ||
457 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
458 | unsigned int cmd, unsigned long arg) | |
459 | { | |
460 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 461 | int r; |
e56f81e0 | 462 | |
956a4025 | 463 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
464 | if (r < 0) |
465 | return r; | |
6c182cd8 | 466 | |
e56f81e0 CH |
467 | if (r > 0) { |
468 | /* | |
e980f623 CH |
469 | * Target determined this ioctl is being issued against a |
470 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 471 | */ |
e980f623 CH |
472 | if (!capable(CAP_SYS_RAWIO)) { |
473 | DMWARN_LIMIT( | |
474 | "%s: sending ioctl %x to DM device without required privilege.", | |
475 | current->comm, cmd); | |
476 | r = -ENOIOCTLCMD; | |
e56f81e0 | 477 | goto out; |
e980f623 | 478 | } |
e56f81e0 | 479 | } |
6c182cd8 | 480 | |
66482026 | 481 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 482 | out: |
956a4025 | 483 | bdput(bdev); |
aa129a22 MB |
484 | return r; |
485 | } | |
486 | ||
028867ac | 487 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
488 | { |
489 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
490 | } | |
491 | ||
028867ac | 492 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
493 | { |
494 | mempool_free(io, md->io_pool); | |
495 | } | |
496 | ||
cfae7529 | 497 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 498 | { |
dba14160 | 499 | bio_put(&tio->clone); |
1da177e4 LT |
500 | } |
501 | ||
4cc96131 | 502 | int md_in_flight(struct mapped_device *md) |
90abb8c4 KU |
503 | { |
504 | return atomic_read(&md->pending[READ]) + | |
505 | atomic_read(&md->pending[WRITE]); | |
506 | } | |
507 | ||
3eaf840e JNN |
508 | static void start_io_acct(struct dm_io *io) |
509 | { | |
510 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 511 | struct bio *bio = io->bio; |
c9959059 | 512 | int cpu; |
fd2ed4d2 | 513 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
514 | |
515 | io->start_time = jiffies; | |
516 | ||
074a7aca TH |
517 | cpu = part_stat_lock(); |
518 | part_round_stats(cpu, &dm_disk(md)->part0); | |
519 | part_stat_unlock(); | |
1e9bb880 SL |
520 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
521 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
522 | |
523 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
524 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
525 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
526 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
527 | } |
528 | ||
d221d2e7 | 529 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
530 | { |
531 | struct mapped_device *md = io->md; | |
532 | struct bio *bio = io->bio; | |
533 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 534 | int pending; |
3eaf840e JNN |
535 | int rw = bio_data_dir(bio); |
536 | ||
18c0b223 | 537 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 538 | |
fd2ed4d2 | 539 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
540 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
541 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
542 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 543 | |
af7e466a MP |
544 | /* |
545 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 546 | * a flush. |
af7e466a | 547 | */ |
1e9bb880 SL |
548 | pending = atomic_dec_return(&md->pending[rw]); |
549 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 550 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 551 | |
d221d2e7 MP |
552 | /* nudge anyone waiting on suspend queue */ |
553 | if (!pending) | |
554 | wake_up(&md->wait); | |
3eaf840e JNN |
555 | } |
556 | ||
1da177e4 LT |
557 | /* |
558 | * Add the bio to the list of deferred io. | |
559 | */ | |
92c63902 | 560 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 561 | { |
05447420 | 562 | unsigned long flags; |
1da177e4 | 563 | |
05447420 | 564 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 565 | bio_list_add(&md->deferred, bio); |
05447420 | 566 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 567 | queue_work(md->wq, &md->work); |
1da177e4 LT |
568 | } |
569 | ||
570 | /* | |
571 | * Everyone (including functions in this file), should use this | |
572 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 573 | * dm_put_live_table() when finished. |
1da177e4 | 574 | */ |
83d5e5b0 | 575 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 576 | { |
83d5e5b0 MP |
577 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
578 | ||
579 | return srcu_dereference(md->map, &md->io_barrier); | |
580 | } | |
1da177e4 | 581 | |
83d5e5b0 MP |
582 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
583 | { | |
584 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
585 | } | |
586 | ||
587 | void dm_sync_table(struct mapped_device *md) | |
588 | { | |
589 | synchronize_srcu(&md->io_barrier); | |
590 | synchronize_rcu_expedited(); | |
591 | } | |
592 | ||
593 | /* | |
594 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
595 | * The caller must not block between these two functions. | |
596 | */ | |
597 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
598 | { | |
599 | rcu_read_lock(); | |
600 | return rcu_dereference(md->map); | |
601 | } | |
1da177e4 | 602 | |
83d5e5b0 MP |
603 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
604 | { | |
605 | rcu_read_unlock(); | |
1da177e4 LT |
606 | } |
607 | ||
86f1152b BM |
608 | /* |
609 | * Open a table device so we can use it as a map destination. | |
610 | */ | |
611 | static int open_table_device(struct table_device *td, dev_t dev, | |
612 | struct mapped_device *md) | |
613 | { | |
614 | static char *_claim_ptr = "I belong to device-mapper"; | |
615 | struct block_device *bdev; | |
616 | ||
617 | int r; | |
618 | ||
619 | BUG_ON(td->dm_dev.bdev); | |
620 | ||
621 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
622 | if (IS_ERR(bdev)) | |
623 | return PTR_ERR(bdev); | |
624 | ||
625 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
626 | if (r) { | |
627 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
628 | return r; | |
629 | } | |
630 | ||
631 | td->dm_dev.bdev = bdev; | |
632 | return 0; | |
633 | } | |
634 | ||
635 | /* | |
636 | * Close a table device that we've been using. | |
637 | */ | |
638 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
639 | { | |
640 | if (!td->dm_dev.bdev) | |
641 | return; | |
642 | ||
643 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
644 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
645 | td->dm_dev.bdev = NULL; | |
646 | } | |
647 | ||
648 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
649 | fmode_t mode) { | |
650 | struct table_device *td; | |
651 | ||
652 | list_for_each_entry(td, l, list) | |
653 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
654 | return td; | |
655 | ||
656 | return NULL; | |
657 | } | |
658 | ||
659 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
660 | struct dm_dev **result) { | |
661 | int r; | |
662 | struct table_device *td; | |
663 | ||
664 | mutex_lock(&md->table_devices_lock); | |
665 | td = find_table_device(&md->table_devices, dev, mode); | |
666 | if (!td) { | |
115485e8 | 667 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
668 | if (!td) { |
669 | mutex_unlock(&md->table_devices_lock); | |
670 | return -ENOMEM; | |
671 | } | |
672 | ||
673 | td->dm_dev.mode = mode; | |
674 | td->dm_dev.bdev = NULL; | |
675 | ||
676 | if ((r = open_table_device(td, dev, md))) { | |
677 | mutex_unlock(&md->table_devices_lock); | |
678 | kfree(td); | |
679 | return r; | |
680 | } | |
681 | ||
682 | format_dev_t(td->dm_dev.name, dev); | |
683 | ||
684 | atomic_set(&td->count, 0); | |
685 | list_add(&td->list, &md->table_devices); | |
686 | } | |
687 | atomic_inc(&td->count); | |
688 | mutex_unlock(&md->table_devices_lock); | |
689 | ||
690 | *result = &td->dm_dev; | |
691 | return 0; | |
692 | } | |
693 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
694 | ||
695 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
696 | { | |
697 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
698 | ||
699 | mutex_lock(&md->table_devices_lock); | |
700 | if (atomic_dec_and_test(&td->count)) { | |
701 | close_table_device(td, md); | |
702 | list_del(&td->list); | |
703 | kfree(td); | |
704 | } | |
705 | mutex_unlock(&md->table_devices_lock); | |
706 | } | |
707 | EXPORT_SYMBOL(dm_put_table_device); | |
708 | ||
709 | static void free_table_devices(struct list_head *devices) | |
710 | { | |
711 | struct list_head *tmp, *next; | |
712 | ||
713 | list_for_each_safe(tmp, next, devices) { | |
714 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
715 | ||
716 | DMWARN("dm_destroy: %s still exists with %d references", | |
717 | td->dm_dev.name, atomic_read(&td->count)); | |
718 | kfree(td); | |
719 | } | |
720 | } | |
721 | ||
3ac51e74 DW |
722 | /* |
723 | * Get the geometry associated with a dm device | |
724 | */ | |
725 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
726 | { | |
727 | *geo = md->geometry; | |
728 | ||
729 | return 0; | |
730 | } | |
731 | ||
732 | /* | |
733 | * Set the geometry of a device. | |
734 | */ | |
735 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
736 | { | |
737 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
738 | ||
739 | if (geo->start > sz) { | |
740 | DMWARN("Start sector is beyond the geometry limits."); | |
741 | return -EINVAL; | |
742 | } | |
743 | ||
744 | md->geometry = *geo; | |
745 | ||
746 | return 0; | |
747 | } | |
748 | ||
1da177e4 LT |
749 | /*----------------------------------------------------------------- |
750 | * CRUD START: | |
751 | * A more elegant soln is in the works that uses the queue | |
752 | * merge fn, unfortunately there are a couple of changes to | |
753 | * the block layer that I want to make for this. So in the | |
754 | * interests of getting something for people to use I give | |
755 | * you this clearly demarcated crap. | |
756 | *---------------------------------------------------------------*/ | |
757 | ||
2e93ccc1 KU |
758 | static int __noflush_suspending(struct mapped_device *md) |
759 | { | |
760 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
761 | } | |
762 | ||
1da177e4 LT |
763 | /* |
764 | * Decrements the number of outstanding ios that a bio has been | |
765 | * cloned into, completing the original io if necc. | |
766 | */ | |
858119e1 | 767 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 768 | { |
2e93ccc1 | 769 | unsigned long flags; |
b35f8caa MB |
770 | int io_error; |
771 | struct bio *bio; | |
772 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
773 | |
774 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
775 | if (unlikely(error)) { |
776 | spin_lock_irqsave(&io->endio_lock, flags); | |
777 | if (!(io->error > 0 && __noflush_suspending(md))) | |
778 | io->error = error; | |
779 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
780 | } | |
1da177e4 LT |
781 | |
782 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
783 | if (io->error == DM_ENDIO_REQUEUE) { |
784 | /* | |
785 | * Target requested pushing back the I/O. | |
2e93ccc1 | 786 | */ |
022c2611 | 787 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
788 | if (__noflush_suspending(md)) |
789 | bio_list_add_head(&md->deferred, io->bio); | |
790 | else | |
2e93ccc1 KU |
791 | /* noflush suspend was interrupted. */ |
792 | io->error = -EIO; | |
022c2611 | 793 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
794 | } |
795 | ||
b35f8caa MB |
796 | io_error = io->error; |
797 | bio = io->bio; | |
6a8736d1 TH |
798 | end_io_acct(io); |
799 | free_io(md, io); | |
800 | ||
801 | if (io_error == DM_ENDIO_REQUEUE) | |
802 | return; | |
2e93ccc1 | 803 | |
1eff9d32 | 804 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 805 | /* |
6a8736d1 | 806 | * Preflush done for flush with data, reissue |
28a8f0d3 | 807 | * without REQ_PREFLUSH. |
af7e466a | 808 | */ |
1eff9d32 | 809 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 810 | queue_io(md, bio); |
af7e466a | 811 | } else { |
b372d360 | 812 | /* done with normal IO or empty flush */ |
4246a0b6 CH |
813 | bio->bi_error = io_error; |
814 | bio_endio(bio); | |
b35f8caa | 815 | } |
1da177e4 LT |
816 | } |
817 | } | |
818 | ||
4cc96131 | 819 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
820 | { |
821 | struct queue_limits *limits = dm_get_queue_limits(md); | |
822 | ||
823 | /* device doesn't really support WRITE SAME, disable it */ | |
824 | limits->max_write_same_sectors = 0; | |
825 | } | |
826 | ||
4246a0b6 | 827 | static void clone_endio(struct bio *bio) |
1da177e4 | 828 | { |
4246a0b6 | 829 | int error = bio->bi_error; |
5164bece | 830 | int r = error; |
bfc6d41c | 831 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 832 | struct dm_io *io = tio->io; |
9faf400f | 833 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
834 | dm_endio_fn endio = tio->ti->type->end_io; |
835 | ||
1da177e4 | 836 | if (endio) { |
7de3ee57 | 837 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
838 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
839 | /* | |
840 | * error and requeue request are handled | |
841 | * in dec_pending(). | |
842 | */ | |
1da177e4 | 843 | error = r; |
45cbcd79 KU |
844 | else if (r == DM_ENDIO_INCOMPLETE) |
845 | /* The target will handle the io */ | |
6712ecf8 | 846 | return; |
45cbcd79 KU |
847 | else if (r) { |
848 | DMWARN("unimplemented target endio return value: %d", r); | |
849 | BUG(); | |
850 | } | |
1da177e4 LT |
851 | } |
852 | ||
e6047149 | 853 | if (unlikely(r == -EREMOTEIO && (bio_op(bio) == REQ_OP_WRITE_SAME) && |
7eee4ae2 MS |
854 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) |
855 | disable_write_same(md); | |
856 | ||
cfae7529 | 857 | free_tio(tio); |
b35f8caa | 858 | dec_pending(io, error); |
1da177e4 LT |
859 | } |
860 | ||
56a67df7 MS |
861 | /* |
862 | * Return maximum size of I/O possible at the supplied sector up to the current | |
863 | * target boundary. | |
864 | */ | |
865 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
866 | { | |
867 | sector_t target_offset = dm_target_offset(ti, sector); | |
868 | ||
869 | return ti->len - target_offset; | |
870 | } | |
871 | ||
872 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 873 | { |
56a67df7 | 874 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 875 | sector_t offset, max_len; |
1da177e4 LT |
876 | |
877 | /* | |
542f9038 | 878 | * Does the target need to split even further? |
1da177e4 | 879 | */ |
542f9038 MS |
880 | if (ti->max_io_len) { |
881 | offset = dm_target_offset(ti, sector); | |
882 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
883 | max_len = sector_div(offset, ti->max_io_len); | |
884 | else | |
885 | max_len = offset & (ti->max_io_len - 1); | |
886 | max_len = ti->max_io_len - max_len; | |
887 | ||
888 | if (len > max_len) | |
889 | len = max_len; | |
1da177e4 LT |
890 | } |
891 | ||
892 | return len; | |
893 | } | |
894 | ||
542f9038 MS |
895 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
896 | { | |
897 | if (len > UINT_MAX) { | |
898 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
899 | (unsigned long long)len, UINT_MAX); | |
900 | ti->error = "Maximum size of target IO is too large"; | |
901 | return -EINVAL; | |
902 | } | |
903 | ||
904 | ti->max_io_len = (uint32_t) len; | |
905 | ||
906 | return 0; | |
907 | } | |
908 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
909 | ||
545ed20e | 910 | static long dm_blk_direct_access(struct block_device *bdev, sector_t sector, |
f0c98ebc | 911 | void **kaddr, pfn_t *pfn, long size) |
545ed20e TK |
912 | { |
913 | struct mapped_device *md = bdev->bd_disk->private_data; | |
914 | struct dm_table *map; | |
915 | struct dm_target *ti; | |
916 | int srcu_idx; | |
917 | long len, ret = -EIO; | |
918 | ||
919 | map = dm_get_live_table(md, &srcu_idx); | |
920 | if (!map) | |
921 | goto out; | |
922 | ||
923 | ti = dm_table_find_target(map, sector); | |
924 | if (!dm_target_is_valid(ti)) | |
925 | goto out; | |
926 | ||
927 | len = max_io_len(sector, ti) << SECTOR_SHIFT; | |
928 | size = min(len, size); | |
929 | ||
930 | if (ti->type->direct_access) | |
931 | ret = ti->type->direct_access(ti, sector, kaddr, pfn, size); | |
932 | out: | |
933 | dm_put_live_table(md, srcu_idx); | |
934 | return min(ret, size); | |
935 | } | |
936 | ||
1dd40c3e MP |
937 | /* |
938 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
28a8f0d3 | 939 | * allowed for all bio types except REQ_PREFLUSH. |
1dd40c3e MP |
940 | * |
941 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
942 | * additional n_sectors sectors of the bio and the rest of the data should be | |
943 | * sent in a next bio. | |
944 | * | |
945 | * A diagram that explains the arithmetics: | |
946 | * +--------------------+---------------+-------+ | |
947 | * | 1 | 2 | 3 | | |
948 | * +--------------------+---------------+-------+ | |
949 | * | |
950 | * <-------------- *tio->len_ptr ---------------> | |
951 | * <------- bi_size -------> | |
952 | * <-- n_sectors --> | |
953 | * | |
954 | * Region 1 was already iterated over with bio_advance or similar function. | |
955 | * (it may be empty if the target doesn't use bio_advance) | |
956 | * Region 2 is the remaining bio size that the target wants to process. | |
957 | * (it may be empty if region 1 is non-empty, although there is no reason | |
958 | * to make it empty) | |
959 | * The target requires that region 3 is to be sent in the next bio. | |
960 | * | |
961 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
962 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
963 | * copies of the bio. | |
964 | */ | |
965 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
966 | { | |
967 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
968 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 969 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
970 | BUG_ON(bi_size > *tio->len_ptr); |
971 | BUG_ON(n_sectors > bi_size); | |
972 | *tio->len_ptr -= bi_size - n_sectors; | |
973 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
974 | } | |
975 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
976 | ||
d67a5f4b MP |
977 | /* |
978 | * Flush current->bio_list when the target map method blocks. | |
979 | * This fixes deadlocks in snapshot and possibly in other targets. | |
980 | */ | |
981 | struct dm_offload { | |
982 | struct blk_plug plug; | |
983 | struct blk_plug_cb cb; | |
984 | }; | |
985 | ||
986 | static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule) | |
987 | { | |
988 | struct dm_offload *o = container_of(cb, struct dm_offload, cb); | |
989 | struct bio_list list; | |
990 | struct bio *bio; | |
f5fe1b51 | 991 | int i; |
d67a5f4b MP |
992 | |
993 | INIT_LIST_HEAD(&o->cb.list); | |
994 | ||
995 | if (unlikely(!current->bio_list)) | |
996 | return; | |
997 | ||
f5fe1b51 N |
998 | for (i = 0; i < 2; i++) { |
999 | list = current->bio_list[i]; | |
1000 | bio_list_init(¤t->bio_list[i]); | |
1001 | ||
1002 | while ((bio = bio_list_pop(&list))) { | |
1003 | struct bio_set *bs = bio->bi_pool; | |
1004 | if (unlikely(!bs) || bs == fs_bio_set) { | |
1005 | bio_list_add(¤t->bio_list[i], bio); | |
1006 | continue; | |
1007 | } | |
1008 | ||
1009 | spin_lock(&bs->rescue_lock); | |
1010 | bio_list_add(&bs->rescue_list, bio); | |
1011 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
1012 | spin_unlock(&bs->rescue_lock); | |
d67a5f4b | 1013 | } |
d67a5f4b MP |
1014 | } |
1015 | } | |
1016 | ||
1017 | static void dm_offload_start(struct dm_offload *o) | |
1018 | { | |
1019 | blk_start_plug(&o->plug); | |
1020 | o->cb.callback = flush_current_bio_list; | |
1021 | list_add(&o->cb.list, ¤t->plug->cb_list); | |
1022 | } | |
1023 | ||
1024 | static void dm_offload_end(struct dm_offload *o) | |
1025 | { | |
1026 | list_del(&o->cb.list); | |
1027 | blk_finish_plug(&o->plug); | |
1028 | } | |
1029 | ||
bd2a49b8 | 1030 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1031 | { |
1032 | int r; | |
2056a782 | 1033 | sector_t sector; |
d67a5f4b | 1034 | struct dm_offload o; |
dba14160 | 1035 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1036 | struct dm_target *ti = tio->ti; |
1da177e4 | 1037 | |
1da177e4 | 1038 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1039 | |
1040 | /* | |
1041 | * Map the clone. If r == 0 we don't need to do | |
1042 | * anything, the target has assumed ownership of | |
1043 | * this io. | |
1044 | */ | |
1045 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1046 | sector = clone->bi_iter.bi_sector; |
d67a5f4b MP |
1047 | |
1048 | dm_offload_start(&o); | |
7de3ee57 | 1049 | r = ti->type->map(ti, clone); |
d67a5f4b MP |
1050 | dm_offload_end(&o); |
1051 | ||
45cbcd79 | 1052 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1053 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1054 | |
d07335e5 MS |
1055 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1056 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1057 | |
1da177e4 | 1058 | generic_make_request(clone); |
2e93ccc1 KU |
1059 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1060 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f | 1061 | dec_pending(tio->io, r); |
cfae7529 | 1062 | free_tio(tio); |
ab37844d | 1063 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1064 | DMWARN("unimplemented target map return value: %d", r); |
1065 | BUG(); | |
1da177e4 LT |
1066 | } |
1067 | } | |
1068 | ||
1069 | struct clone_info { | |
1070 | struct mapped_device *md; | |
1071 | struct dm_table *map; | |
1072 | struct bio *bio; | |
1073 | struct dm_io *io; | |
1074 | sector_t sector; | |
e0d6609a | 1075 | unsigned sector_count; |
1da177e4 LT |
1076 | }; |
1077 | ||
e0d6609a | 1078 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1079 | { |
4f024f37 KO |
1080 | bio->bi_iter.bi_sector = sector; |
1081 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1082 | } |
1083 | ||
1084 | /* | |
1085 | * Creates a bio that consists of range of complete bvecs. | |
1086 | */ | |
c80914e8 MS |
1087 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1088 | sector_t sector, unsigned len) | |
1da177e4 | 1089 | { |
dba14160 | 1090 | struct bio *clone = &tio->clone; |
1da177e4 | 1091 | |
1c3b13e6 KO |
1092 | __bio_clone_fast(clone, bio); |
1093 | ||
c80914e8 MS |
1094 | if (bio_integrity(bio)) { |
1095 | int r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
1096 | if (r < 0) | |
1097 | return r; | |
1098 | } | |
bd2a49b8 | 1099 | |
1c3b13e6 KO |
1100 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1101 | clone->bi_iter.bi_size = to_bytes(len); | |
1102 | ||
1103 | if (bio_integrity(bio)) | |
1104 | bio_integrity_trim(clone, 0, len); | |
c80914e8 MS |
1105 | |
1106 | return 0; | |
1da177e4 LT |
1107 | } |
1108 | ||
9015df24 | 1109 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1110 | struct dm_target *ti, |
55a62eef | 1111 | unsigned target_bio_nr) |
f9ab94ce | 1112 | { |
dba14160 MP |
1113 | struct dm_target_io *tio; |
1114 | struct bio *clone; | |
1115 | ||
99778273 | 1116 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1117 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1118 | |
1119 | tio->io = ci->io; | |
1120 | tio->ti = ti; | |
55a62eef | 1121 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1122 | |
1123 | return tio; | |
1124 | } | |
1125 | ||
14fe594d AK |
1126 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1127 | struct dm_target *ti, | |
1dd40c3e | 1128 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1129 | { |
99778273 | 1130 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1131 | struct bio *clone = &tio->clone; |
9015df24 | 1132 | |
1dd40c3e MP |
1133 | tio->len_ptr = len; |
1134 | ||
99778273 | 1135 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1136 | if (len) |
1dd40c3e | 1137 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1138 | |
bd2a49b8 | 1139 | __map_bio(tio); |
f9ab94ce MP |
1140 | } |
1141 | ||
14fe594d | 1142 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1143 | unsigned num_bios, unsigned *len) |
06a426ce | 1144 | { |
55a62eef | 1145 | unsigned target_bio_nr; |
06a426ce | 1146 | |
55a62eef | 1147 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1148 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1149 | } |
1150 | ||
14fe594d | 1151 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1152 | { |
06a426ce | 1153 | unsigned target_nr = 0; |
f9ab94ce MP |
1154 | struct dm_target *ti; |
1155 | ||
b372d360 | 1156 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1157 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1158 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1159 | |
f9ab94ce MP |
1160 | return 0; |
1161 | } | |
1162 | ||
c80914e8 | 1163 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1164 | sector_t sector, unsigned *len) |
5ae89a87 | 1165 | { |
dba14160 | 1166 | struct bio *bio = ci->bio; |
5ae89a87 | 1167 | struct dm_target_io *tio; |
b0d8ed4d AK |
1168 | unsigned target_bio_nr; |
1169 | unsigned num_target_bios = 1; | |
c80914e8 | 1170 | int r = 0; |
5ae89a87 | 1171 | |
b0d8ed4d AK |
1172 | /* |
1173 | * Does the target want to receive duplicate copies of the bio? | |
1174 | */ | |
1175 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1176 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1177 | |
b0d8ed4d | 1178 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1179 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e | 1180 | tio->len_ptr = len; |
c80914e8 | 1181 | r = clone_bio(tio, bio, sector, *len); |
072623de | 1182 | if (r < 0) { |
cfae7529 | 1183 | free_tio(tio); |
c80914e8 | 1184 | break; |
072623de | 1185 | } |
b0d8ed4d AK |
1186 | __map_bio(tio); |
1187 | } | |
c80914e8 MS |
1188 | |
1189 | return r; | |
5ae89a87 MS |
1190 | } |
1191 | ||
55a62eef | 1192 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1193 | |
55a62eef | 1194 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1195 | { |
55a62eef | 1196 | return ti->num_discard_bios; |
23508a96 MS |
1197 | } |
1198 | ||
55a62eef | 1199 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1200 | { |
55a62eef | 1201 | return ti->num_write_same_bios; |
23508a96 MS |
1202 | } |
1203 | ||
1204 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
9eef87da | 1205 | |
23508a96 MS |
1206 | static bool is_split_required_for_discard(struct dm_target *ti) |
1207 | { | |
55a62eef | 1208 | return ti->split_discard_bios; |
cec47e3d KU |
1209 | } |
1210 | ||
14fe594d AK |
1211 | static int __send_changing_extent_only(struct clone_info *ci, |
1212 | get_num_bios_fn get_num_bios, | |
1213 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1214 | { |
5ae89a87 | 1215 | struct dm_target *ti; |
e0d6609a | 1216 | unsigned len; |
55a62eef | 1217 | unsigned num_bios; |
ba1cbad9 | 1218 | |
a79245b3 MS |
1219 | do { |
1220 | ti = dm_table_find_target(ci->map, ci->sector); | |
1221 | if (!dm_target_is_valid(ti)) | |
1222 | return -EIO; | |
2eb6e1e3 | 1223 | |
5ae89a87 | 1224 | /* |
23508a96 MS |
1225 | * Even though the device advertised support for this type of |
1226 | * request, that does not mean every target supports it, and | |
936688d7 | 1227 | * reconfiguration might also have changed that since the |
a79245b3 | 1228 | * check was performed. |
5ae89a87 | 1229 | */ |
55a62eef AK |
1230 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1231 | if (!num_bios) | |
a79245b3 | 1232 | return -EOPNOTSUPP; |
ba1cbad9 | 1233 | |
23508a96 | 1234 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1235 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1236 | else |
e0d6609a | 1237 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
de3ec86d | 1238 | |
1dd40c3e | 1239 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1240 | |
a79245b3 MS |
1241 | ci->sector += len; |
1242 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1243 | |
1244 | return 0; | |
ba1cbad9 MS |
1245 | } |
1246 | ||
14fe594d | 1247 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1248 | { |
14fe594d AK |
1249 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1250 | is_split_required_for_discard); | |
23508a96 | 1251 | } |
0ce65797 | 1252 | |
14fe594d | 1253 | static int __send_write_same(struct clone_info *ci) |
0ce65797 | 1254 | { |
14fe594d | 1255 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
0ce65797 MS |
1256 | } |
1257 | ||
e4c93811 AK |
1258 | /* |
1259 | * Select the correct strategy for processing a non-flush bio. | |
1260 | */ | |
14fe594d | 1261 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1262 | { |
dba14160 | 1263 | struct bio *bio = ci->bio; |
512875bd | 1264 | struct dm_target *ti; |
1c3b13e6 | 1265 | unsigned len; |
c80914e8 | 1266 | int r; |
0ce65797 | 1267 | |
e6047149 | 1268 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
14fe594d | 1269 | return __send_discard(ci); |
e6047149 | 1270 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
14fe594d | 1271 | return __send_write_same(ci); |
0ce65797 | 1272 | |
512875bd JN |
1273 | ti = dm_table_find_target(ci->map, ci->sector); |
1274 | if (!dm_target_is_valid(ti)) | |
1275 | return -EIO; | |
1276 | ||
1c3b13e6 | 1277 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
0ce65797 | 1278 | |
c80914e8 MS |
1279 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1280 | if (r < 0) | |
1281 | return r; | |
0ce65797 | 1282 | |
1c3b13e6 KO |
1283 | ci->sector += len; |
1284 | ci->sector_count -= len; | |
0ce65797 | 1285 | |
1c3b13e6 | 1286 | return 0; |
0ce65797 MS |
1287 | } |
1288 | ||
1da177e4 | 1289 | /* |
14fe594d | 1290 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1291 | */ |
83d5e5b0 MP |
1292 | static void __split_and_process_bio(struct mapped_device *md, |
1293 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1294 | { |
1da177e4 | 1295 | struct clone_info ci; |
512875bd | 1296 | int error = 0; |
1da177e4 | 1297 | |
83d5e5b0 | 1298 | if (unlikely(!map)) { |
6a8736d1 | 1299 | bio_io_error(bio); |
f0b9a450 MP |
1300 | return; |
1301 | } | |
692d0eb9 | 1302 | |
83d5e5b0 | 1303 | ci.map = map; |
1da177e4 | 1304 | ci.md = md; |
1da177e4 LT |
1305 | ci.io = alloc_io(md); |
1306 | ci.io->error = 0; | |
1307 | atomic_set(&ci.io->io_count, 1); | |
1308 | ci.io->bio = bio; | |
1309 | ci.io->md = md; | |
f88fb981 | 1310 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1311 | ci.sector = bio->bi_iter.bi_sector; |
0ce65797 | 1312 | |
3eaf840e | 1313 | start_io_acct(ci.io); |
0ce65797 | 1314 | |
1eff9d32 | 1315 | if (bio->bi_opf & REQ_PREFLUSH) { |
b372d360 MS |
1316 | ci.bio = &ci.md->flush_bio; |
1317 | ci.sector_count = 0; | |
14fe594d | 1318 | error = __send_empty_flush(&ci); |
b372d360 MS |
1319 | /* dec_pending submits any data associated with flush */ |
1320 | } else { | |
6a8736d1 | 1321 | ci.bio = bio; |
d87f4c14 | 1322 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1323 | while (ci.sector_count && !error) |
14fe594d | 1324 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1325 | } |
0ce65797 | 1326 | |
1da177e4 | 1327 | /* drop the extra reference count */ |
512875bd | 1328 | dec_pending(ci.io, error); |
0ce65797 | 1329 | } |
1da177e4 LT |
1330 | /*----------------------------------------------------------------- |
1331 | * CRUD END | |
1332 | *---------------------------------------------------------------*/ | |
0ce65797 | 1333 | |
cec47e3d | 1334 | /* |
1da177e4 LT |
1335 | * The request function that just remaps the bio built up by |
1336 | * dm_merge_bvec. | |
cec47e3d | 1337 | */ |
dece1635 | 1338 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
cec47e3d | 1339 | { |
12f03a49 | 1340 | int rw = bio_data_dir(bio); |
cec47e3d | 1341 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1342 | int srcu_idx; |
1343 | struct dm_table *map; | |
cec47e3d | 1344 | |
83d5e5b0 | 1345 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1346 | |
18c0b223 | 1347 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
d0bcb878 | 1348 | |
6a8736d1 TH |
1349 | /* if we're suspended, we have to queue this io for later */ |
1350 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1351 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1352 | |
1eff9d32 | 1353 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1354 | queue_io(md, bio); |
1355 | else | |
54d9a1b4 | 1356 | bio_io_error(bio); |
dece1635 | 1357 | return BLK_QC_T_NONE; |
cec47e3d | 1358 | } |
1da177e4 | 1359 | |
83d5e5b0 MP |
1360 | __split_and_process_bio(md, map, bio); |
1361 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1362 | return BLK_QC_T_NONE; |
cec47e3d KU |
1363 | } |
1364 | ||
1da177e4 LT |
1365 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1366 | { | |
8a57dfc6 CS |
1367 | int r = bdi_bits; |
1368 | struct mapped_device *md = congested_data; | |
1369 | struct dm_table *map; | |
1da177e4 | 1370 | |
1eb787ec | 1371 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1372 | if (dm_request_based(md)) { |
cec47e3d | 1373 | /* |
e522c039 MS |
1374 | * With request-based DM we only need to check the |
1375 | * top-level queue for congestion. | |
cec47e3d | 1376 | */ |
dc3b17cc | 1377 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1378 | } else { |
1379 | map = dm_get_live_table_fast(md); | |
1380 | if (map) | |
cec47e3d | 1381 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1382 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1383 | } |
1384 | } | |
1385 | ||
1da177e4 LT |
1386 | return r; |
1387 | } | |
1388 | ||
1389 | /*----------------------------------------------------------------- | |
1390 | * An IDR is used to keep track of allocated minor numbers. | |
1391 | *---------------------------------------------------------------*/ | |
2b06cfff | 1392 | static void free_minor(int minor) |
1da177e4 | 1393 | { |
f32c10b0 | 1394 | spin_lock(&_minor_lock); |
1da177e4 | 1395 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1396 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1397 | } |
1398 | ||
1399 | /* | |
1400 | * See if the device with a specific minor # is free. | |
1401 | */ | |
cf13ab8e | 1402 | static int specific_minor(int minor) |
1da177e4 | 1403 | { |
c9d76be6 | 1404 | int r; |
1da177e4 LT |
1405 | |
1406 | if (minor >= (1 << MINORBITS)) | |
1407 | return -EINVAL; | |
1408 | ||
c9d76be6 | 1409 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1410 | spin_lock(&_minor_lock); |
1da177e4 | 1411 | |
c9d76be6 | 1412 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1413 | |
f32c10b0 | 1414 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1415 | idr_preload_end(); |
1416 | if (r < 0) | |
1417 | return r == -ENOSPC ? -EBUSY : r; | |
1418 | return 0; | |
1da177e4 LT |
1419 | } |
1420 | ||
cf13ab8e | 1421 | static int next_free_minor(int *minor) |
1da177e4 | 1422 | { |
c9d76be6 | 1423 | int r; |
62f75c2f | 1424 | |
c9d76be6 | 1425 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1426 | spin_lock(&_minor_lock); |
1da177e4 | 1427 | |
c9d76be6 | 1428 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1429 | |
f32c10b0 | 1430 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1431 | idr_preload_end(); |
1432 | if (r < 0) | |
1433 | return r; | |
1434 | *minor = r; | |
1435 | return 0; | |
1da177e4 LT |
1436 | } |
1437 | ||
83d5cde4 | 1438 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1439 | |
53d5914f MP |
1440 | static void dm_wq_work(struct work_struct *work); |
1441 | ||
4cc96131 | 1442 | void dm_init_md_queue(struct mapped_device *md) |
4a0b4ddf MS |
1443 | { |
1444 | /* | |
1445 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 1446 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
1447 | * The type is decided at the first table loading time. |
1448 | * To prevent problematic device stacking, clear the queue flag | |
1449 | * for request stacking support until then. | |
1450 | * | |
1451 | * This queue is new, so no concurrency on the queue_flags. | |
1452 | */ | |
1453 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
1454 | |
1455 | /* | |
1456 | * Initialize data that will only be used by a non-blk-mq DM queue | |
1457 | * - must do so here (in alloc_dev callchain) before queue is used | |
1458 | */ | |
1459 | md->queue->queuedata = md; | |
dc3b17cc | 1460 | md->queue->backing_dev_info->congested_data = md; |
bfebd1cd | 1461 | } |
4a0b4ddf | 1462 | |
4cc96131 | 1463 | void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1464 | { |
17e149b8 | 1465 | md->use_blk_mq = false; |
bfebd1cd MS |
1466 | dm_init_md_queue(md); |
1467 | ||
1468 | /* | |
1469 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1470 | */ | |
dc3b17cc | 1471 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf | 1472 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
1473 | } |
1474 | ||
0f20972f MS |
1475 | static void cleanup_mapped_device(struct mapped_device *md) |
1476 | { | |
0f20972f MS |
1477 | if (md->wq) |
1478 | destroy_workqueue(md->wq); | |
1479 | if (md->kworker_task) | |
1480 | kthread_stop(md->kworker_task); | |
6f65985e | 1481 | mempool_destroy(md->io_pool); |
0f20972f MS |
1482 | if (md->bs) |
1483 | bioset_free(md->bs); | |
1484 | ||
1485 | if (md->disk) { | |
1486 | spin_lock(&_minor_lock); | |
1487 | md->disk->private_data = NULL; | |
1488 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1489 | del_gendisk(md->disk); |
1490 | put_disk(md->disk); | |
1491 | } | |
1492 | ||
1493 | if (md->queue) | |
1494 | blk_cleanup_queue(md->queue); | |
1495 | ||
d09960b0 TE |
1496 | cleanup_srcu_struct(&md->io_barrier); |
1497 | ||
0f20972f MS |
1498 | if (md->bdev) { |
1499 | bdput(md->bdev); | |
1500 | md->bdev = NULL; | |
1501 | } | |
4cc96131 MS |
1502 | |
1503 | dm_mq_cleanup_mapped_device(md); | |
0f20972f MS |
1504 | } |
1505 | ||
1da177e4 LT |
1506 | /* |
1507 | * Allocate and initialise a blank device with a given minor. | |
1508 | */ | |
2b06cfff | 1509 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1510 | { |
115485e8 MS |
1511 | int r, numa_node_id = dm_get_numa_node(); |
1512 | struct mapped_device *md; | |
ba61fdd1 | 1513 | void *old_md; |
1da177e4 | 1514 | |
115485e8 | 1515 | md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1516 | if (!md) { |
1517 | DMWARN("unable to allocate device, out of memory."); | |
1518 | return NULL; | |
1519 | } | |
1520 | ||
10da4f79 | 1521 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1522 | goto bad_module_get; |
10da4f79 | 1523 | |
1da177e4 | 1524 | /* get a minor number for the dev */ |
2b06cfff | 1525 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1526 | r = next_free_minor(&minor); |
2b06cfff | 1527 | else |
cf13ab8e | 1528 | r = specific_minor(minor); |
1da177e4 | 1529 | if (r < 0) |
6ed7ade8 | 1530 | goto bad_minor; |
1da177e4 | 1531 | |
83d5e5b0 MP |
1532 | r = init_srcu_struct(&md->io_barrier); |
1533 | if (r < 0) | |
1534 | goto bad_io_barrier; | |
1535 | ||
115485e8 | 1536 | md->numa_node_id = numa_node_id; |
4cc96131 | 1537 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1538 | md->init_tio_pdu = false; |
a5664dad | 1539 | md->type = DM_TYPE_NONE; |
e61290a4 | 1540 | mutex_init(&md->suspend_lock); |
a5664dad | 1541 | mutex_init(&md->type_lock); |
86f1152b | 1542 | mutex_init(&md->table_devices_lock); |
022c2611 | 1543 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1544 | atomic_set(&md->holders, 1); |
5c6bd75d | 1545 | atomic_set(&md->open_count, 0); |
1da177e4 | 1546 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1547 | atomic_set(&md->uevent_seq, 0); |
1548 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1549 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1550 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1551 | |
115485e8 | 1552 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1553 | if (!md->queue) |
0f20972f | 1554 | goto bad; |
1da177e4 | 1555 | |
4a0b4ddf | 1556 | dm_init_md_queue(md); |
9faf400f | 1557 | |
115485e8 | 1558 | md->disk = alloc_disk_node(1, numa_node_id); |
1da177e4 | 1559 | if (!md->disk) |
0f20972f | 1560 | goto bad; |
1da177e4 | 1561 | |
316d315b NK |
1562 | atomic_set(&md->pending[0], 0); |
1563 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1564 | init_waitqueue_head(&md->wait); |
53d5914f | 1565 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1566 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1567 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1568 | md->kworker_task = NULL; |
f0b04115 | 1569 | |
1da177e4 LT |
1570 | md->disk->major = _major; |
1571 | md->disk->first_minor = minor; | |
1572 | md->disk->fops = &dm_blk_dops; | |
1573 | md->disk->queue = md->queue; | |
1574 | md->disk->private_data = md; | |
1575 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1576 | add_disk(md->disk); | |
7e51f257 | 1577 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1578 | |
670368a8 | 1579 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1580 | if (!md->wq) |
0f20972f | 1581 | goto bad; |
304f3f6a | 1582 | |
32a926da MP |
1583 | md->bdev = bdget_disk(md->disk, 0); |
1584 | if (!md->bdev) | |
0f20972f | 1585 | goto bad; |
32a926da | 1586 | |
3a83f467 | 1587 | bio_init(&md->flush_bio, NULL, 0); |
6a8736d1 | 1588 | md->flush_bio.bi_bdev = md->bdev; |
70fd7614 | 1589 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
6a8736d1 | 1590 | |
fd2ed4d2 MP |
1591 | dm_stats_init(&md->stats); |
1592 | ||
ba61fdd1 | 1593 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1594 | spin_lock(&_minor_lock); |
ba61fdd1 | 1595 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1596 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1597 | |
1598 | BUG_ON(old_md != MINOR_ALLOCED); | |
1599 | ||
1da177e4 LT |
1600 | return md; |
1601 | ||
0f20972f MS |
1602 | bad: |
1603 | cleanup_mapped_device(md); | |
83d5e5b0 | 1604 | bad_io_barrier: |
1da177e4 | 1605 | free_minor(minor); |
6ed7ade8 | 1606 | bad_minor: |
10da4f79 | 1607 | module_put(THIS_MODULE); |
6ed7ade8 | 1608 | bad_module_get: |
1da177e4 LT |
1609 | kfree(md); |
1610 | return NULL; | |
1611 | } | |
1612 | ||
ae9da83f JN |
1613 | static void unlock_fs(struct mapped_device *md); |
1614 | ||
1da177e4 LT |
1615 | static void free_dev(struct mapped_device *md) |
1616 | { | |
f331c029 | 1617 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1618 | |
32a926da | 1619 | unlock_fs(md); |
2eb6e1e3 | 1620 | |
0f20972f | 1621 | cleanup_mapped_device(md); |
63a4f065 | 1622 | |
86f1152b | 1623 | free_table_devices(&md->table_devices); |
63a4f065 | 1624 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1625 | free_minor(minor); |
1626 | ||
10da4f79 | 1627 | module_put(THIS_MODULE); |
1da177e4 LT |
1628 | kfree(md); |
1629 | } | |
1630 | ||
e6ee8c0b KU |
1631 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1632 | { | |
c0820cf5 | 1633 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1634 | |
4e6e36c3 MS |
1635 | if (md->bs) { |
1636 | /* The md already has necessary mempools. */ | |
545ed20e | 1637 | if (dm_table_bio_based(t)) { |
16245bdc JN |
1638 | /* |
1639 | * Reload bioset because front_pad may have changed | |
1640 | * because a different table was loaded. | |
1641 | */ | |
1642 | bioset_free(md->bs); | |
1643 | md->bs = p->bs; | |
1644 | p->bs = NULL; | |
16245bdc | 1645 | } |
4e6e36c3 MS |
1646 | /* |
1647 | * There's no need to reload with request-based dm | |
1648 | * because the size of front_pad doesn't change. | |
1649 | * Note for future: If you are to reload bioset, | |
1650 | * prep-ed requests in the queue may refer | |
1651 | * to bio from the old bioset, so you must walk | |
1652 | * through the queue to unprep. | |
1653 | */ | |
1654 | goto out; | |
c0820cf5 | 1655 | } |
e6ee8c0b | 1656 | |
eb8db831 | 1657 | BUG_ON(!p || md->io_pool || md->bs); |
cbc4e3c1 | 1658 | |
e6ee8c0b KU |
1659 | md->io_pool = p->io_pool; |
1660 | p->io_pool = NULL; | |
e6ee8c0b KU |
1661 | md->bs = p->bs; |
1662 | p->bs = NULL; | |
4e6e36c3 | 1663 | |
e6ee8c0b | 1664 | out: |
02233342 | 1665 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1666 | dm_table_free_md_mempools(t); |
1667 | } | |
1668 | ||
1da177e4 LT |
1669 | /* |
1670 | * Bind a table to the device. | |
1671 | */ | |
1672 | static void event_callback(void *context) | |
1673 | { | |
7a8c3d3b MA |
1674 | unsigned long flags; |
1675 | LIST_HEAD(uevents); | |
1da177e4 LT |
1676 | struct mapped_device *md = (struct mapped_device *) context; |
1677 | ||
7a8c3d3b MA |
1678 | spin_lock_irqsave(&md->uevent_lock, flags); |
1679 | list_splice_init(&md->uevent_list, &uevents); | |
1680 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1681 | ||
ed9e1982 | 1682 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1683 | |
1da177e4 LT |
1684 | atomic_inc(&md->event_nr); |
1685 | wake_up(&md->eventq); | |
1686 | } | |
1687 | ||
c217649b MS |
1688 | /* |
1689 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1690 | */ | |
4e90188b | 1691 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1692 | { |
4e90188b | 1693 | set_capacity(md->disk, size); |
1da177e4 | 1694 | |
db8fef4f | 1695 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
1696 | } |
1697 | ||
042d2a9b AK |
1698 | /* |
1699 | * Returns old map, which caller must destroy. | |
1700 | */ | |
1701 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
1702 | struct queue_limits *limits) | |
1da177e4 | 1703 | { |
042d2a9b | 1704 | struct dm_table *old_map; |
165125e1 | 1705 | struct request_queue *q = md->queue; |
1da177e4 LT |
1706 | sector_t size; |
1707 | ||
5a8f1f80 BVA |
1708 | lockdep_assert_held(&md->suspend_lock); |
1709 | ||
1da177e4 | 1710 | size = dm_table_get_size(t); |
3ac51e74 DW |
1711 | |
1712 | /* | |
1713 | * Wipe any geometry if the size of the table changed. | |
1714 | */ | |
fd2ed4d2 | 1715 | if (size != dm_get_size(md)) |
3ac51e74 DW |
1716 | memset(&md->geometry, 0, sizeof(md->geometry)); |
1717 | ||
32a926da | 1718 | __set_size(md, size); |
d5816876 | 1719 | |
2ca3310e AK |
1720 | dm_table_event_callback(t, event_callback, md); |
1721 | ||
e6ee8c0b KU |
1722 | /* |
1723 | * The queue hasn't been stopped yet, if the old table type wasn't | |
1724 | * for request-based during suspension. So stop it to prevent | |
1725 | * I/O mapping before resume. | |
1726 | * This must be done before setting the queue restrictions, | |
1727 | * because request-based dm may be run just after the setting. | |
1728 | */ | |
16f12266 | 1729 | if (dm_table_request_based(t)) { |
eca7ee6d | 1730 | dm_stop_queue(q); |
16f12266 MS |
1731 | /* |
1732 | * Leverage the fact that request-based DM targets are | |
1733 | * immutable singletons and establish md->immutable_target | |
1734 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
1735 | */ | |
1736 | md->immutable_target = dm_table_get_immutable_target(t); | |
1737 | } | |
e6ee8c0b KU |
1738 | |
1739 | __bind_mempools(md, t); | |
1740 | ||
a12f5d48 | 1741 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 1742 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
1743 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
1744 | ||
754c5fc7 | 1745 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
1746 | if (old_map) |
1747 | dm_sync_table(md); | |
1da177e4 | 1748 | |
042d2a9b | 1749 | return old_map; |
1da177e4 LT |
1750 | } |
1751 | ||
a7940155 AK |
1752 | /* |
1753 | * Returns unbound table for the caller to free. | |
1754 | */ | |
1755 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 1756 | { |
a12f5d48 | 1757 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
1758 | |
1759 | if (!map) | |
a7940155 | 1760 | return NULL; |
1da177e4 LT |
1761 | |
1762 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 1763 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 1764 | dm_sync_table(md); |
a7940155 AK |
1765 | |
1766 | return map; | |
1da177e4 LT |
1767 | } |
1768 | ||
1769 | /* | |
1770 | * Constructor for a new device. | |
1771 | */ | |
2b06cfff | 1772 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1773 | { |
1774 | struct mapped_device *md; | |
1775 | ||
2b06cfff | 1776 | md = alloc_dev(minor); |
1da177e4 LT |
1777 | if (!md) |
1778 | return -ENXIO; | |
1779 | ||
784aae73 MB |
1780 | dm_sysfs_init(md); |
1781 | ||
1da177e4 LT |
1782 | *result = md; |
1783 | return 0; | |
1784 | } | |
1785 | ||
a5664dad MS |
1786 | /* |
1787 | * Functions to manage md->type. | |
1788 | * All are required to hold md->type_lock. | |
1789 | */ | |
1790 | void dm_lock_md_type(struct mapped_device *md) | |
1791 | { | |
1792 | mutex_lock(&md->type_lock); | |
1793 | } | |
1794 | ||
1795 | void dm_unlock_md_type(struct mapped_device *md) | |
1796 | { | |
1797 | mutex_unlock(&md->type_lock); | |
1798 | } | |
1799 | ||
1800 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
1801 | { | |
00c4fc3b | 1802 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
1803 | md->type = type; |
1804 | } | |
1805 | ||
1806 | unsigned dm_get_md_type(struct mapped_device *md) | |
1807 | { | |
1808 | return md->type; | |
1809 | } | |
1810 | ||
36a0456f AK |
1811 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
1812 | { | |
1813 | return md->immutable_target_type; | |
1814 | } | |
1815 | ||
f84cb8a4 MS |
1816 | /* |
1817 | * The queue_limits are only valid as long as you have a reference | |
1818 | * count on 'md'. | |
1819 | */ | |
1820 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
1821 | { | |
1822 | BUG_ON(!atomic_read(&md->holders)); | |
1823 | return &md->queue->limits; | |
1824 | } | |
1825 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
1826 | ||
4a0b4ddf MS |
1827 | /* |
1828 | * Setup the DM device's queue based on md's type | |
1829 | */ | |
591ddcfc | 1830 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 1831 | { |
bfebd1cd | 1832 | int r; |
545ed20e | 1833 | unsigned type = dm_get_md_type(md); |
bfebd1cd | 1834 | |
545ed20e | 1835 | switch (type) { |
bfebd1cd | 1836 | case DM_TYPE_REQUEST_BASED: |
eb8db831 | 1837 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 1838 | if (r) { |
eca7ee6d | 1839 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 1840 | return r; |
ff36ab34 | 1841 | } |
bfebd1cd MS |
1842 | break; |
1843 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 1844 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 1845 | if (r) { |
eca7ee6d | 1846 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
1847 | return r; |
1848 | } | |
1849 | break; | |
1850 | case DM_TYPE_BIO_BASED: | |
545ed20e | 1851 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 1852 | dm_init_normal_md_queue(md); |
ff36ab34 | 1853 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
1854 | /* |
1855 | * DM handles splitting bios as needed. Free the bio_split bioset | |
1856 | * since it won't be used (saves 1 process per bio-based DM device). | |
1857 | */ | |
1858 | bioset_free(md->queue->bio_split); | |
1859 | md->queue->bio_split = NULL; | |
545ed20e TK |
1860 | |
1861 | if (type == DM_TYPE_DAX_BIO_BASED) | |
1862 | queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue); | |
bfebd1cd | 1863 | break; |
4a0b4ddf MS |
1864 | } |
1865 | ||
1866 | return 0; | |
1867 | } | |
1868 | ||
2bec1f4a | 1869 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
1870 | { |
1871 | struct mapped_device *md; | |
1da177e4 LT |
1872 | unsigned minor = MINOR(dev); |
1873 | ||
1874 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1875 | return NULL; | |
1876 | ||
f32c10b0 | 1877 | spin_lock(&_minor_lock); |
1da177e4 LT |
1878 | |
1879 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
1880 | if (md) { |
1881 | if ((md == MINOR_ALLOCED || | |
1882 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
1883 | dm_deleting_md(md) || | |
1884 | test_bit(DMF_FREEING, &md->flags))) { | |
1885 | md = NULL; | |
1886 | goto out; | |
1887 | } | |
1888 | dm_get(md); | |
fba9f90e | 1889 | } |
1da177e4 | 1890 | |
fba9f90e | 1891 | out: |
f32c10b0 | 1892 | spin_unlock(&_minor_lock); |
1da177e4 | 1893 | |
637842cf DT |
1894 | return md; |
1895 | } | |
3cf2e4ba | 1896 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 1897 | |
9ade92a9 | 1898 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1899 | { |
9ade92a9 | 1900 | return md->interface_ptr; |
1da177e4 LT |
1901 | } |
1902 | ||
1903 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1904 | { | |
1905 | md->interface_ptr = ptr; | |
1906 | } | |
1907 | ||
1908 | void dm_get(struct mapped_device *md) | |
1909 | { | |
1910 | atomic_inc(&md->holders); | |
3f77316d | 1911 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
1912 | } |
1913 | ||
09ee96b2 MP |
1914 | int dm_hold(struct mapped_device *md) |
1915 | { | |
1916 | spin_lock(&_minor_lock); | |
1917 | if (test_bit(DMF_FREEING, &md->flags)) { | |
1918 | spin_unlock(&_minor_lock); | |
1919 | return -EBUSY; | |
1920 | } | |
1921 | dm_get(md); | |
1922 | spin_unlock(&_minor_lock); | |
1923 | return 0; | |
1924 | } | |
1925 | EXPORT_SYMBOL_GPL(dm_hold); | |
1926 | ||
72d94861 AK |
1927 | const char *dm_device_name(struct mapped_device *md) |
1928 | { | |
1929 | return md->name; | |
1930 | } | |
1931 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1932 | ||
3f77316d | 1933 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 1934 | { |
3b785fbc | 1935 | struct request_queue *q = dm_get_md_queue(md); |
1134e5ae | 1936 | struct dm_table *map; |
83d5e5b0 | 1937 | int srcu_idx; |
1da177e4 | 1938 | |
3f77316d | 1939 | might_sleep(); |
fba9f90e | 1940 | |
63a4f065 | 1941 | spin_lock(&_minor_lock); |
3f77316d KU |
1942 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
1943 | set_bit(DMF_FREEING, &md->flags); | |
1944 | spin_unlock(&_minor_lock); | |
3b785fbc | 1945 | |
2e91c369 | 1946 | blk_set_queue_dying(q); |
3f77316d | 1947 | |
02233342 | 1948 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 1949 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 1950 | |
ab7c7bb6 MP |
1951 | /* |
1952 | * Take suspend_lock so that presuspend and postsuspend methods | |
1953 | * do not race with internal suspend. | |
1954 | */ | |
1955 | mutex_lock(&md->suspend_lock); | |
2a708cff | 1956 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
1957 | if (!dm_suspended_md(md)) { |
1958 | dm_table_presuspend_targets(map); | |
1959 | dm_table_postsuspend_targets(map); | |
1da177e4 | 1960 | } |
83d5e5b0 MP |
1961 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
1962 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 1963 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 1964 | |
3f77316d KU |
1965 | /* |
1966 | * Rare, but there may be I/O requests still going to complete, | |
1967 | * for example. Wait for all references to disappear. | |
1968 | * No one should increment the reference count of the mapped_device, | |
1969 | * after the mapped_device state becomes DMF_FREEING. | |
1970 | */ | |
1971 | if (wait) | |
1972 | while (atomic_read(&md->holders)) | |
1973 | msleep(1); | |
1974 | else if (atomic_read(&md->holders)) | |
1975 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
1976 | dm_device_name(md), atomic_read(&md->holders)); | |
1977 | ||
1978 | dm_sysfs_exit(md); | |
3f77316d KU |
1979 | dm_table_destroy(__unbind(md)); |
1980 | free_dev(md); | |
1981 | } | |
1982 | ||
1983 | void dm_destroy(struct mapped_device *md) | |
1984 | { | |
1985 | __dm_destroy(md, true); | |
1986 | } | |
1987 | ||
1988 | void dm_destroy_immediate(struct mapped_device *md) | |
1989 | { | |
1990 | __dm_destroy(md, false); | |
1991 | } | |
1992 | ||
1993 | void dm_put(struct mapped_device *md) | |
1994 | { | |
1995 | atomic_dec(&md->holders); | |
1da177e4 | 1996 | } |
79eb885c | 1997 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 1998 | |
b48633f8 | 1999 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2000 | { |
2001 | int r = 0; | |
9f4c3f87 | 2002 | DEFINE_WAIT(wait); |
46125c1c MB |
2003 | |
2004 | while (1) { | |
9f4c3f87 | 2005 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2006 | |
b4324fee | 2007 | if (!md_in_flight(md)) |
46125c1c MB |
2008 | break; |
2009 | ||
e3fabdfd | 2010 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2011 | r = -EINTR; |
2012 | break; | |
2013 | } | |
2014 | ||
2015 | io_schedule(); | |
2016 | } | |
9f4c3f87 | 2017 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2018 | |
46125c1c MB |
2019 | return r; |
2020 | } | |
2021 | ||
1da177e4 LT |
2022 | /* |
2023 | * Process the deferred bios | |
2024 | */ | |
ef208587 | 2025 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2026 | { |
ef208587 MP |
2027 | struct mapped_device *md = container_of(work, struct mapped_device, |
2028 | work); | |
6d6f10df | 2029 | struct bio *c; |
83d5e5b0 MP |
2030 | int srcu_idx; |
2031 | struct dm_table *map; | |
1da177e4 | 2032 | |
83d5e5b0 | 2033 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2034 | |
3b00b203 | 2035 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2036 | spin_lock_irq(&md->deferred_lock); |
2037 | c = bio_list_pop(&md->deferred); | |
2038 | spin_unlock_irq(&md->deferred_lock); | |
2039 | ||
6a8736d1 | 2040 | if (!c) |
df12ee99 | 2041 | break; |
022c2611 | 2042 | |
e6ee8c0b KU |
2043 | if (dm_request_based(md)) |
2044 | generic_make_request(c); | |
6a8736d1 | 2045 | else |
83d5e5b0 | 2046 | __split_and_process_bio(md, map, c); |
022c2611 | 2047 | } |
73d410c0 | 2048 | |
83d5e5b0 | 2049 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2050 | } |
2051 | ||
9a1fb464 | 2052 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2053 | { |
3b00b203 | 2054 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2055 | smp_mb__after_atomic(); |
53d5914f | 2056 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2057 | } |
2058 | ||
1da177e4 | 2059 | /* |
042d2a9b | 2060 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2061 | */ |
042d2a9b | 2062 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2063 | { |
87eb5b21 | 2064 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2065 | struct queue_limits limits; |
042d2a9b | 2066 | int r; |
1da177e4 | 2067 | |
e61290a4 | 2068 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2069 | |
2070 | /* device must be suspended */ | |
4f186f8b | 2071 | if (!dm_suspended_md(md)) |
93c534ae | 2072 | goto out; |
1da177e4 | 2073 | |
3ae70656 MS |
2074 | /* |
2075 | * If the new table has no data devices, retain the existing limits. | |
2076 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2077 | * then new I/O is queued based on these limits, and then some paths | |
2078 | * reappear. | |
2079 | */ | |
2080 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2081 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2082 | if (live_map) |
2083 | limits = md->queue->limits; | |
83d5e5b0 | 2084 | dm_put_live_table_fast(md); |
3ae70656 MS |
2085 | } |
2086 | ||
87eb5b21 MC |
2087 | if (!live_map) { |
2088 | r = dm_calculate_queue_limits(table, &limits); | |
2089 | if (r) { | |
2090 | map = ERR_PTR(r); | |
2091 | goto out; | |
2092 | } | |
042d2a9b | 2093 | } |
754c5fc7 | 2094 | |
042d2a9b | 2095 | map = __bind(md, table, &limits); |
1da177e4 | 2096 | |
93c534ae | 2097 | out: |
e61290a4 | 2098 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2099 | return map; |
1da177e4 LT |
2100 | } |
2101 | ||
2102 | /* | |
2103 | * Functions to lock and unlock any filesystem running on the | |
2104 | * device. | |
2105 | */ | |
2ca3310e | 2106 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2107 | { |
e39e2e95 | 2108 | int r; |
1da177e4 LT |
2109 | |
2110 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2111 | |
db8fef4f | 2112 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2113 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2114 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2115 | md->frozen_sb = NULL; |
2116 | return r; | |
dfbe03f6 AK |
2117 | } |
2118 | ||
aa8d7c2f AK |
2119 | set_bit(DMF_FROZEN, &md->flags); |
2120 | ||
1da177e4 LT |
2121 | return 0; |
2122 | } | |
2123 | ||
2ca3310e | 2124 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2125 | { |
aa8d7c2f AK |
2126 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2127 | return; | |
2128 | ||
db8fef4f | 2129 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2130 | md->frozen_sb = NULL; |
aa8d7c2f | 2131 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2132 | } |
2133 | ||
2134 | /* | |
b48633f8 BVA |
2135 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2136 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2137 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2138 | * | |
ffcc3936 MS |
2139 | * If __dm_suspend returns 0, the device is completely quiescent |
2140 | * now. There is no request-processing activity. All new requests | |
2141 | * are being added to md->deferred list. | |
cec47e3d | 2142 | * |
ffcc3936 | 2143 | * Caller must hold md->suspend_lock |
cec47e3d | 2144 | */ |
ffcc3936 | 2145 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2146 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2147 | int dmf_suspended_flag) |
1da177e4 | 2148 | { |
ffcc3936 MS |
2149 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2150 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2151 | int r; | |
1da177e4 | 2152 | |
5a8f1f80 BVA |
2153 | lockdep_assert_held(&md->suspend_lock); |
2154 | ||
2e93ccc1 KU |
2155 | /* |
2156 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2157 | * This flag is cleared before dm_suspend returns. | |
2158 | */ | |
2159 | if (noflush) | |
2160 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2161 | ||
d67ee213 MS |
2162 | /* |
2163 | * This gets reverted if there's an error later and the targets | |
2164 | * provide the .presuspend_undo hook. | |
2165 | */ | |
cf222b37 AK |
2166 | dm_table_presuspend_targets(map); |
2167 | ||
32a926da | 2168 | /* |
9f518b27 KU |
2169 | * Flush I/O to the device. |
2170 | * Any I/O submitted after lock_fs() may not be flushed. | |
2171 | * noflush takes precedence over do_lockfs. | |
2172 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2173 | */ |
2174 | if (!noflush && do_lockfs) { | |
2175 | r = lock_fs(md); | |
d67ee213 MS |
2176 | if (r) { |
2177 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2178 | return r; |
d67ee213 | 2179 | } |
aa8d7c2f | 2180 | } |
1da177e4 LT |
2181 | |
2182 | /* | |
3b00b203 MP |
2183 | * Here we must make sure that no processes are submitting requests |
2184 | * to target drivers i.e. no one may be executing | |
2185 | * __split_and_process_bio. This is called from dm_request and | |
2186 | * dm_wq_work. | |
2187 | * | |
2188 | * To get all processes out of __split_and_process_bio in dm_request, | |
2189 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2190 | * __split_and_process_bio from dm_request and quiesce the thread |
2191 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2192 | * flush_workqueue(md->wq). | |
1da177e4 | 2193 | */ |
1eb787ec | 2194 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2195 | if (map) |
2196 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2197 | |
d0bcb878 | 2198 | /* |
29e4013d TH |
2199 | * Stop md->queue before flushing md->wq in case request-based |
2200 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2201 | */ |
2eb6e1e3 | 2202 | if (dm_request_based(md)) { |
eca7ee6d | 2203 | dm_stop_queue(md->queue); |
02233342 | 2204 | if (md->kworker_task) |
3989144f | 2205 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2206 | } |
cec47e3d | 2207 | |
d0bcb878 KU |
2208 | flush_workqueue(md->wq); |
2209 | ||
1da177e4 | 2210 | /* |
3b00b203 MP |
2211 | * At this point no more requests are entering target request routines. |
2212 | * We call dm_wait_for_completion to wait for all existing requests | |
2213 | * to finish. | |
1da177e4 | 2214 | */ |
b48633f8 | 2215 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2216 | if (!r) |
2217 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2218 | |
6d6f10df | 2219 | if (noflush) |
022c2611 | 2220 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2221 | if (map) |
2222 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2223 | |
1da177e4 | 2224 | /* were we interrupted ? */ |
46125c1c | 2225 | if (r < 0) { |
9a1fb464 | 2226 | dm_queue_flush(md); |
73d410c0 | 2227 | |
cec47e3d | 2228 | if (dm_request_based(md)) |
eca7ee6d | 2229 | dm_start_queue(md->queue); |
cec47e3d | 2230 | |
2ca3310e | 2231 | unlock_fs(md); |
d67ee213 | 2232 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2233 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2234 | } |
1da177e4 | 2235 | |
ffcc3936 MS |
2236 | return r; |
2237 | } | |
2238 | ||
2239 | /* | |
2240 | * We need to be able to change a mapping table under a mounted | |
2241 | * filesystem. For example we might want to move some data in | |
2242 | * the background. Before the table can be swapped with | |
2243 | * dm_bind_table, dm_suspend must be called to flush any in | |
2244 | * flight bios and ensure that any further io gets deferred. | |
2245 | */ | |
2246 | /* | |
2247 | * Suspend mechanism in request-based dm. | |
2248 | * | |
2249 | * 1. Flush all I/Os by lock_fs() if needed. | |
2250 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2251 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2252 | * | |
2253 | * To abort suspend, start the request_queue. | |
2254 | */ | |
2255 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2256 | { | |
2257 | struct dm_table *map = NULL; | |
2258 | int r = 0; | |
2259 | ||
2260 | retry: | |
2261 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2262 | ||
2263 | if (dm_suspended_md(md)) { | |
2264 | r = -EINVAL; | |
2265 | goto out_unlock; | |
2266 | } | |
2267 | ||
2268 | if (dm_suspended_internally_md(md)) { | |
2269 | /* already internally suspended, wait for internal resume */ | |
2270 | mutex_unlock(&md->suspend_lock); | |
2271 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2272 | if (r) | |
2273 | return r; | |
2274 | goto retry; | |
2275 | } | |
2276 | ||
a12f5d48 | 2277 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2278 | |
eaf9a736 | 2279 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2280 | if (r) |
2281 | goto out_unlock; | |
3b00b203 | 2282 | |
4d4471cb KU |
2283 | dm_table_postsuspend_targets(map); |
2284 | ||
d287483d | 2285 | out_unlock: |
e61290a4 | 2286 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2287 | return r; |
1da177e4 LT |
2288 | } |
2289 | ||
ffcc3936 MS |
2290 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2291 | { | |
2292 | if (map) { | |
2293 | int r = dm_table_resume_targets(map); | |
2294 | if (r) | |
2295 | return r; | |
2296 | } | |
2297 | ||
2298 | dm_queue_flush(md); | |
2299 | ||
2300 | /* | |
2301 | * Flushing deferred I/Os must be done after targets are resumed | |
2302 | * so that mapping of targets can work correctly. | |
2303 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2304 | */ | |
2305 | if (dm_request_based(md)) | |
eca7ee6d | 2306 | dm_start_queue(md->queue); |
ffcc3936 MS |
2307 | |
2308 | unlock_fs(md); | |
2309 | ||
2310 | return 0; | |
2311 | } | |
2312 | ||
1da177e4 LT |
2313 | int dm_resume(struct mapped_device *md) |
2314 | { | |
8dc23658 | 2315 | int r; |
cf222b37 | 2316 | struct dm_table *map = NULL; |
1da177e4 | 2317 | |
ffcc3936 | 2318 | retry: |
8dc23658 | 2319 | r = -EINVAL; |
ffcc3936 MS |
2320 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2321 | ||
4f186f8b | 2322 | if (!dm_suspended_md(md)) |
cf222b37 | 2323 | goto out; |
cf222b37 | 2324 | |
ffcc3936 MS |
2325 | if (dm_suspended_internally_md(md)) { |
2326 | /* already internally suspended, wait for internal resume */ | |
2327 | mutex_unlock(&md->suspend_lock); | |
2328 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2329 | if (r) | |
2330 | return r; | |
2331 | goto retry; | |
2332 | } | |
2333 | ||
a12f5d48 | 2334 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2335 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2336 | goto out; |
1da177e4 | 2337 | |
ffcc3936 | 2338 | r = __dm_resume(md, map); |
8757b776 MB |
2339 | if (r) |
2340 | goto out; | |
2ca3310e | 2341 | |
2ca3310e | 2342 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2343 | out: |
e61290a4 | 2344 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2345 | |
cf222b37 | 2346 | return r; |
1da177e4 LT |
2347 | } |
2348 | ||
fd2ed4d2 MP |
2349 | /* |
2350 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2351 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2352 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2353 | */ |
2354 | ||
ffcc3936 | 2355 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2356 | { |
ffcc3936 MS |
2357 | struct dm_table *map = NULL; |
2358 | ||
96b26c8c | 2359 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2360 | return; /* nested internal suspend */ |
2361 | ||
2362 | if (dm_suspended_md(md)) { | |
2363 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2364 | return; /* nest suspend */ | |
2365 | } | |
2366 | ||
a12f5d48 | 2367 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2368 | |
2369 | /* | |
2370 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2371 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2372 | * would require changing .presuspend to return an error -- avoid this | |
2373 | * until there is a need for more elaborate variants of internal suspend. | |
2374 | */ | |
eaf9a736 MS |
2375 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2376 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2377 | |
2378 | dm_table_postsuspend_targets(map); | |
2379 | } | |
2380 | ||
2381 | static void __dm_internal_resume(struct mapped_device *md) | |
2382 | { | |
96b26c8c MP |
2383 | BUG_ON(!md->internal_suspend_count); |
2384 | ||
2385 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2386 | return; /* resume from nested internal suspend */ |
2387 | ||
fd2ed4d2 | 2388 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2389 | goto done; /* resume from nested suspend */ |
2390 | ||
2391 | /* | |
2392 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2393 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2394 | */ | |
2395 | (void) __dm_resume(md, NULL); | |
2396 | ||
2397 | done: | |
2398 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2399 | smp_mb__after_atomic(); | |
2400 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2401 | } | |
2402 | ||
2403 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2404 | { | |
2405 | mutex_lock(&md->suspend_lock); | |
2406 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2407 | mutex_unlock(&md->suspend_lock); | |
2408 | } | |
2409 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2410 | ||
2411 | void dm_internal_resume(struct mapped_device *md) | |
2412 | { | |
2413 | mutex_lock(&md->suspend_lock); | |
2414 | __dm_internal_resume(md); | |
2415 | mutex_unlock(&md->suspend_lock); | |
2416 | } | |
2417 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2418 | ||
2419 | /* | |
2420 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2421 | * which prevents interaction with userspace-driven suspend. | |
2422 | */ | |
2423 | ||
2424 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2425 | { | |
2426 | mutex_lock(&md->suspend_lock); | |
2427 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2428 | return; |
2429 | ||
2430 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2431 | synchronize_srcu(&md->io_barrier); | |
2432 | flush_workqueue(md->wq); | |
2433 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2434 | } | |
b735fede | 2435 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2436 | |
ffcc3936 | 2437 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2438 | { |
ffcc3936 | 2439 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2440 | goto done; |
2441 | ||
2442 | dm_queue_flush(md); | |
2443 | ||
2444 | done: | |
2445 | mutex_unlock(&md->suspend_lock); | |
2446 | } | |
b735fede | 2447 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2448 | |
1da177e4 LT |
2449 | /*----------------------------------------------------------------- |
2450 | * Event notification. | |
2451 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2452 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2453 | unsigned cookie) |
69267a30 | 2454 | { |
60935eb2 MB |
2455 | char udev_cookie[DM_COOKIE_LENGTH]; |
2456 | char *envp[] = { udev_cookie, NULL }; | |
2457 | ||
2458 | if (!cookie) | |
3abf85b5 | 2459 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2460 | else { |
2461 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2462 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2463 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2464 | action, envp); | |
60935eb2 | 2465 | } |
69267a30 AK |
2466 | } |
2467 | ||
7a8c3d3b MA |
2468 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2469 | { | |
2470 | return atomic_add_return(1, &md->uevent_seq); | |
2471 | } | |
2472 | ||
1da177e4 LT |
2473 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2474 | { | |
2475 | return atomic_read(&md->event_nr); | |
2476 | } | |
2477 | ||
2478 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2479 | { | |
2480 | return wait_event_interruptible(md->eventq, | |
2481 | (event_nr != atomic_read(&md->event_nr))); | |
2482 | } | |
2483 | ||
7a8c3d3b MA |
2484 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2485 | { | |
2486 | unsigned long flags; | |
2487 | ||
2488 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2489 | list_add(elist, &md->uevent_list); | |
2490 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2491 | } | |
2492 | ||
1da177e4 LT |
2493 | /* |
2494 | * The gendisk is only valid as long as you have a reference | |
2495 | * count on 'md'. | |
2496 | */ | |
2497 | struct gendisk *dm_disk(struct mapped_device *md) | |
2498 | { | |
2499 | return md->disk; | |
2500 | } | |
65ff5b7d | 2501 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2502 | |
784aae73 MB |
2503 | struct kobject *dm_kobject(struct mapped_device *md) |
2504 | { | |
2995fa78 | 2505 | return &md->kobj_holder.kobj; |
784aae73 MB |
2506 | } |
2507 | ||
784aae73 MB |
2508 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2509 | { | |
2510 | struct mapped_device *md; | |
2511 | ||
2995fa78 | 2512 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2513 | |
4d89b7b4 | 2514 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2515 | dm_deleting_md(md)) |
4d89b7b4 MB |
2516 | return NULL; |
2517 | ||
784aae73 MB |
2518 | dm_get(md); |
2519 | return md; | |
2520 | } | |
2521 | ||
4f186f8b | 2522 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2523 | { |
2524 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2525 | } | |
2526 | ||
ffcc3936 MS |
2527 | int dm_suspended_internally_md(struct mapped_device *md) |
2528 | { | |
2529 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2530 | } | |
2531 | ||
2c140a24 MP |
2532 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2533 | { | |
2534 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2535 | } | |
2536 | ||
64dbce58 KU |
2537 | int dm_suspended(struct dm_target *ti) |
2538 | { | |
ecdb2e25 | 2539 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2540 | } |
2541 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2542 | ||
2e93ccc1 KU |
2543 | int dm_noflush_suspending(struct dm_target *ti) |
2544 | { | |
ecdb2e25 | 2545 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2546 | } |
2547 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2548 | ||
78d8e58a | 2549 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
30187e1d | 2550 | unsigned integrity, unsigned per_io_data_size) |
e6ee8c0b | 2551 | { |
115485e8 | 2552 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2553 | unsigned int pool_size = 0; |
5f015204 | 2554 | unsigned int front_pad; |
e6ee8c0b KU |
2555 | |
2556 | if (!pools) | |
4e6e36c3 | 2557 | return NULL; |
e6ee8c0b | 2558 | |
78d8e58a MS |
2559 | switch (type) { |
2560 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2561 | case DM_TYPE_DAX_BIO_BASED: |
78d8e58a | 2562 | pool_size = dm_get_reserved_bio_based_ios(); |
30187e1d | 2563 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
eb8db831 CH |
2564 | |
2565 | pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); | |
2566 | if (!pools->io_pool) | |
2567 | goto out; | |
78d8e58a MS |
2568 | break; |
2569 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2570 | case DM_TYPE_MQ_REQUEST_BASED: |
eb8db831 | 2571 | pool_size = dm_get_reserved_rq_based_ios(); |
78d8e58a | 2572 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2573 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2574 | break; |
2575 | default: | |
2576 | BUG(); | |
2577 | } | |
2578 | ||
3d8aab2d | 2579 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 2580 | if (!pools->bs) |
5f015204 | 2581 | goto out; |
e6ee8c0b | 2582 | |
a91a2785 | 2583 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2584 | goto out; |
a91a2785 | 2585 | |
e6ee8c0b | 2586 | return pools; |
5f1b670d | 2587 | |
5f1b670d CH |
2588 | out: |
2589 | dm_free_md_mempools(pools); | |
78d8e58a | 2590 | |
4e6e36c3 | 2591 | return NULL; |
e6ee8c0b KU |
2592 | } |
2593 | ||
2594 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2595 | { | |
2596 | if (!pools) | |
2597 | return; | |
2598 | ||
6f65985e | 2599 | mempool_destroy(pools->io_pool); |
1ae49ea2 | 2600 | |
e6ee8c0b KU |
2601 | if (pools->bs) |
2602 | bioset_free(pools->bs); | |
2603 | ||
2604 | kfree(pools); | |
2605 | } | |
2606 | ||
9c72bad1 CH |
2607 | struct dm_pr { |
2608 | u64 old_key; | |
2609 | u64 new_key; | |
2610 | u32 flags; | |
2611 | bool fail_early; | |
2612 | }; | |
2613 | ||
2614 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2615 | void *data) | |
71cdb697 CH |
2616 | { |
2617 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2618 | struct dm_table *table; |
2619 | struct dm_target *ti; | |
2620 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2621 | |
9c72bad1 CH |
2622 | table = dm_get_live_table(md, &srcu_idx); |
2623 | if (!table || !dm_table_get_size(table)) | |
2624 | goto out; | |
71cdb697 | 2625 | |
9c72bad1 CH |
2626 | /* We only support devices that have a single target */ |
2627 | if (dm_table_get_num_targets(table) != 1) | |
2628 | goto out; | |
2629 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2630 | |
9c72bad1 CH |
2631 | ret = -EINVAL; |
2632 | if (!ti->type->iterate_devices) | |
2633 | goto out; | |
2634 | ||
2635 | ret = ti->type->iterate_devices(ti, fn, data); | |
2636 | out: | |
2637 | dm_put_live_table(md, srcu_idx); | |
2638 | return ret; | |
2639 | } | |
2640 | ||
2641 | /* | |
2642 | * For register / unregister we need to manually call out to every path. | |
2643 | */ | |
2644 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
2645 | sector_t start, sector_t len, void *data) | |
2646 | { | |
2647 | struct dm_pr *pr = data; | |
2648 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
2649 | ||
2650 | if (!ops || !ops->pr_register) | |
2651 | return -EOPNOTSUPP; | |
2652 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
2653 | } | |
2654 | ||
2655 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
2656 | u32 flags) | |
2657 | { | |
2658 | struct dm_pr pr = { | |
2659 | .old_key = old_key, | |
2660 | .new_key = new_key, | |
2661 | .flags = flags, | |
2662 | .fail_early = true, | |
2663 | }; | |
2664 | int ret; | |
2665 | ||
2666 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
2667 | if (ret && new_key) { | |
2668 | /* unregister all paths if we failed to register any path */ | |
2669 | pr.old_key = new_key; | |
2670 | pr.new_key = 0; | |
2671 | pr.flags = 0; | |
2672 | pr.fail_early = false; | |
2673 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
2674 | } | |
2675 | ||
2676 | return ret; | |
71cdb697 CH |
2677 | } |
2678 | ||
2679 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 2680 | u32 flags) |
71cdb697 CH |
2681 | { |
2682 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2683 | const struct pr_ops *ops; | |
71cdb697 | 2684 | fmode_t mode; |
956a4025 | 2685 | int r; |
71cdb697 | 2686 | |
956a4025 | 2687 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2688 | if (r < 0) |
2689 | return r; | |
2690 | ||
2691 | ops = bdev->bd_disk->fops->pr_ops; | |
2692 | if (ops && ops->pr_reserve) | |
2693 | r = ops->pr_reserve(bdev, key, type, flags); | |
2694 | else | |
2695 | r = -EOPNOTSUPP; | |
2696 | ||
956a4025 | 2697 | bdput(bdev); |
71cdb697 CH |
2698 | return r; |
2699 | } | |
2700 | ||
2701 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
2702 | { | |
2703 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2704 | const struct pr_ops *ops; | |
71cdb697 | 2705 | fmode_t mode; |
956a4025 | 2706 | int r; |
71cdb697 | 2707 | |
956a4025 | 2708 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2709 | if (r < 0) |
2710 | return r; | |
2711 | ||
2712 | ops = bdev->bd_disk->fops->pr_ops; | |
2713 | if (ops && ops->pr_release) | |
2714 | r = ops->pr_release(bdev, key, type); | |
2715 | else | |
2716 | r = -EOPNOTSUPP; | |
2717 | ||
956a4025 | 2718 | bdput(bdev); |
71cdb697 CH |
2719 | return r; |
2720 | } | |
2721 | ||
2722 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 2723 | enum pr_type type, bool abort) |
71cdb697 CH |
2724 | { |
2725 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2726 | const struct pr_ops *ops; | |
71cdb697 | 2727 | fmode_t mode; |
956a4025 | 2728 | int r; |
71cdb697 | 2729 | |
956a4025 | 2730 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2731 | if (r < 0) |
2732 | return r; | |
2733 | ||
2734 | ops = bdev->bd_disk->fops->pr_ops; | |
2735 | if (ops && ops->pr_preempt) | |
2736 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
2737 | else | |
2738 | r = -EOPNOTSUPP; | |
2739 | ||
956a4025 | 2740 | bdput(bdev); |
71cdb697 CH |
2741 | return r; |
2742 | } | |
2743 | ||
2744 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
2745 | { | |
2746 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2747 | const struct pr_ops *ops; | |
71cdb697 | 2748 | fmode_t mode; |
956a4025 | 2749 | int r; |
71cdb697 | 2750 | |
956a4025 | 2751 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2752 | if (r < 0) |
2753 | return r; | |
2754 | ||
2755 | ops = bdev->bd_disk->fops->pr_ops; | |
2756 | if (ops && ops->pr_clear) | |
2757 | r = ops->pr_clear(bdev, key); | |
2758 | else | |
2759 | r = -EOPNOTSUPP; | |
2760 | ||
956a4025 | 2761 | bdput(bdev); |
71cdb697 CH |
2762 | return r; |
2763 | } | |
2764 | ||
2765 | static const struct pr_ops dm_pr_ops = { | |
2766 | .pr_register = dm_pr_register, | |
2767 | .pr_reserve = dm_pr_reserve, | |
2768 | .pr_release = dm_pr_release, | |
2769 | .pr_preempt = dm_pr_preempt, | |
2770 | .pr_clear = dm_pr_clear, | |
2771 | }; | |
2772 | ||
83d5cde4 | 2773 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2774 | .open = dm_blk_open, |
2775 | .release = dm_blk_close, | |
aa129a22 | 2776 | .ioctl = dm_blk_ioctl, |
545ed20e | 2777 | .direct_access = dm_blk_direct_access, |
3ac51e74 | 2778 | .getgeo = dm_blk_getgeo, |
71cdb697 | 2779 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
2780 | .owner = THIS_MODULE |
2781 | }; | |
2782 | ||
1da177e4 LT |
2783 | /* |
2784 | * module hooks | |
2785 | */ | |
2786 | module_init(dm_init); | |
2787 | module_exit(dm_exit); | |
2788 | ||
2789 | module_param(major, uint, 0); | |
2790 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 2791 | |
e8603136 MS |
2792 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
2793 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
2794 | ||
115485e8 MS |
2795 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
2796 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
2797 | ||
1da177e4 LT |
2798 | MODULE_DESCRIPTION(DM_NAME " driver"); |
2799 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2800 | MODULE_LICENSE("GPL"); |