]>
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 | 18 | #include <linux/mempool.h> |
f26c5719 | 19 | #include <linux/dax.h> |
1da177e4 LT |
20 | #include <linux/slab.h> |
21 | #include <linux/idr.h> | |
3ac51e74 | 22 | #include <linux/hdreg.h> |
3f77316d | 23 | #include <linux/delay.h> |
ffcc3936 | 24 | #include <linux/wait.h> |
71cdb697 | 25 | #include <linux/pr.h> |
55782138 | 26 | |
72d94861 AK |
27 | #define DM_MSG_PREFIX "core" |
28 | ||
71a16736 NK |
29 | #ifdef CONFIG_PRINTK |
30 | /* | |
31 | * ratelimit state to be used in DMXXX_LIMIT(). | |
32 | */ | |
33 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
34 | DEFAULT_RATELIMIT_INTERVAL, | |
35 | DEFAULT_RATELIMIT_BURST); | |
36 | EXPORT_SYMBOL(dm_ratelimit_state); | |
37 | #endif | |
38 | ||
60935eb2 MB |
39 | /* |
40 | * Cookies are numeric values sent with CHANGE and REMOVE | |
41 | * uevents while resuming, removing or renaming the device. | |
42 | */ | |
43 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
44 | #define DM_COOKIE_LENGTH 24 | |
45 | ||
1da177e4 LT |
46 | static const char *_name = DM_NAME; |
47 | ||
48 | static unsigned int major = 0; | |
49 | static unsigned int _major = 0; | |
50 | ||
d15b774c AK |
51 | static DEFINE_IDR(_minor_idr); |
52 | ||
f32c10b0 | 53 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
54 | |
55 | static void do_deferred_remove(struct work_struct *w); | |
56 | ||
57 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
58 | ||
acfe0ad7 MP |
59 | static struct workqueue_struct *deferred_remove_workqueue; |
60 | ||
1da177e4 LT |
61 | /* |
62 | * One of these is allocated per bio. | |
63 | */ | |
64 | struct dm_io { | |
65 | struct mapped_device *md; | |
66 | int error; | |
1da177e4 | 67 | atomic_t io_count; |
6ae2fa67 | 68 | struct bio *bio; |
3eaf840e | 69 | unsigned long start_time; |
f88fb981 | 70 | spinlock_t endio_lock; |
fd2ed4d2 | 71 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
72 | }; |
73 | ||
ba61fdd1 JM |
74 | #define MINOR_ALLOCED ((void *)-1) |
75 | ||
1da177e4 LT |
76 | /* |
77 | * Bits for the md->flags field. | |
78 | */ | |
1eb787ec | 79 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 80 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 81 | #define DMF_FROZEN 2 |
fba9f90e | 82 | #define DMF_FREEING 3 |
5c6bd75d | 83 | #define DMF_DELETING 4 |
2e93ccc1 | 84 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
85 | #define DMF_DEFERRED_REMOVE 6 |
86 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 87 | |
115485e8 | 88 | #define DM_NUMA_NODE NUMA_NO_NODE |
115485e8 | 89 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 90 | |
e6ee8c0b KU |
91 | /* |
92 | * For mempools pre-allocation at the table loading time. | |
93 | */ | |
94 | struct dm_md_mempools { | |
95 | mempool_t *io_pool; | |
e6ee8c0b KU |
96 | struct bio_set *bs; |
97 | }; | |
98 | ||
86f1152b BM |
99 | struct table_device { |
100 | struct list_head list; | |
101 | atomic_t count; | |
102 | struct dm_dev dm_dev; | |
103 | }; | |
104 | ||
e18b890b | 105 | static struct kmem_cache *_io_cache; |
8fbf26ad | 106 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 107 | static struct kmem_cache *_rq_cache; |
94818742 | 108 | |
e8603136 MS |
109 | /* |
110 | * Bio-based DM's mempools' reserved IOs set by the user. | |
111 | */ | |
4cc96131 | 112 | #define RESERVED_BIO_BASED_IOS 16 |
e8603136 MS |
113 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
114 | ||
115485e8 MS |
115 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
116 | { | |
117 | int param = ACCESS_ONCE(*module_param); | |
118 | int modified_param = 0; | |
119 | bool modified = true; | |
120 | ||
121 | if (param < min) | |
122 | modified_param = min; | |
123 | else if (param > max) | |
124 | modified_param = max; | |
125 | else | |
126 | modified = false; | |
127 | ||
128 | if (modified) { | |
129 | (void)cmpxchg(module_param, param, modified_param); | |
130 | param = modified_param; | |
131 | } | |
132 | ||
133 | return param; | |
134 | } | |
135 | ||
4cc96131 MS |
136 | unsigned __dm_get_module_param(unsigned *module_param, |
137 | unsigned def, unsigned max) | |
f4790826 | 138 | { |
09c2d531 MS |
139 | unsigned param = ACCESS_ONCE(*module_param); |
140 | unsigned modified_param = 0; | |
f4790826 | 141 | |
09c2d531 MS |
142 | if (!param) |
143 | modified_param = def; | |
144 | else if (param > max) | |
145 | modified_param = max; | |
f4790826 | 146 | |
09c2d531 MS |
147 | if (modified_param) { |
148 | (void)cmpxchg(module_param, param, modified_param); | |
149 | param = modified_param; | |
f4790826 MS |
150 | } |
151 | ||
09c2d531 | 152 | return param; |
f4790826 MS |
153 | } |
154 | ||
e8603136 MS |
155 | unsigned dm_get_reserved_bio_based_ios(void) |
156 | { | |
09c2d531 | 157 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 158 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
159 | } |
160 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
161 | ||
115485e8 MS |
162 | static unsigned dm_get_numa_node(void) |
163 | { | |
164 | return __dm_get_module_param_int(&dm_numa_node, | |
165 | DM_NUMA_NODE, num_online_nodes() - 1); | |
166 | } | |
167 | ||
1da177e4 LT |
168 | static int __init local_init(void) |
169 | { | |
51157b4a | 170 | int r = -ENOMEM; |
1da177e4 | 171 | |
1da177e4 | 172 | /* allocate a slab for the dm_ios */ |
028867ac | 173 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 174 | if (!_io_cache) |
51157b4a | 175 | return r; |
1da177e4 | 176 | |
8fbf26ad KU |
177 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
178 | if (!_rq_tio_cache) | |
dba14160 | 179 | goto out_free_io_cache; |
8fbf26ad | 180 | |
eca7ee6d | 181 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
1ae49ea2 MS |
182 | __alignof__(struct request), 0, NULL); |
183 | if (!_rq_cache) | |
184 | goto out_free_rq_tio_cache; | |
185 | ||
51e5b2bd | 186 | r = dm_uevent_init(); |
51157b4a | 187 | if (r) |
1ae49ea2 | 188 | goto out_free_rq_cache; |
51e5b2bd | 189 | |
acfe0ad7 MP |
190 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
191 | if (!deferred_remove_workqueue) { | |
192 | r = -ENOMEM; | |
193 | goto out_uevent_exit; | |
194 | } | |
195 | ||
1da177e4 LT |
196 | _major = major; |
197 | r = register_blkdev(_major, _name); | |
51157b4a | 198 | if (r < 0) |
acfe0ad7 | 199 | goto out_free_workqueue; |
1da177e4 LT |
200 | |
201 | if (!_major) | |
202 | _major = r; | |
203 | ||
204 | return 0; | |
51157b4a | 205 | |
acfe0ad7 MP |
206 | out_free_workqueue: |
207 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
208 | out_uevent_exit: |
209 | dm_uevent_exit(); | |
1ae49ea2 MS |
210 | out_free_rq_cache: |
211 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
212 | out_free_rq_tio_cache: |
213 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
214 | out_free_io_cache: |
215 | kmem_cache_destroy(_io_cache); | |
216 | ||
217 | return r; | |
1da177e4 LT |
218 | } |
219 | ||
220 | static void local_exit(void) | |
221 | { | |
2c140a24 | 222 | flush_scheduled_work(); |
acfe0ad7 | 223 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 224 | |
1ae49ea2 | 225 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 226 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 227 | kmem_cache_destroy(_io_cache); |
00d59405 | 228 | unregister_blkdev(_major, _name); |
51e5b2bd | 229 | dm_uevent_exit(); |
1da177e4 LT |
230 | |
231 | _major = 0; | |
232 | ||
233 | DMINFO("cleaned up"); | |
234 | } | |
235 | ||
b9249e55 | 236 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
237 | local_init, |
238 | dm_target_init, | |
239 | dm_linear_init, | |
240 | dm_stripe_init, | |
952b3557 | 241 | dm_io_init, |
945fa4d2 | 242 | dm_kcopyd_init, |
1da177e4 | 243 | dm_interface_init, |
fd2ed4d2 | 244 | dm_statistics_init, |
1da177e4 LT |
245 | }; |
246 | ||
b9249e55 | 247 | static void (*_exits[])(void) = { |
1da177e4 LT |
248 | local_exit, |
249 | dm_target_exit, | |
250 | dm_linear_exit, | |
251 | dm_stripe_exit, | |
952b3557 | 252 | dm_io_exit, |
945fa4d2 | 253 | dm_kcopyd_exit, |
1da177e4 | 254 | dm_interface_exit, |
fd2ed4d2 | 255 | dm_statistics_exit, |
1da177e4 LT |
256 | }; |
257 | ||
258 | static int __init dm_init(void) | |
259 | { | |
260 | const int count = ARRAY_SIZE(_inits); | |
261 | ||
262 | int r, i; | |
263 | ||
264 | for (i = 0; i < count; i++) { | |
265 | r = _inits[i](); | |
266 | if (r) | |
267 | goto bad; | |
268 | } | |
269 | ||
270 | return 0; | |
271 | ||
272 | bad: | |
273 | while (i--) | |
274 | _exits[i](); | |
275 | ||
276 | return r; | |
277 | } | |
278 | ||
279 | static void __exit dm_exit(void) | |
280 | { | |
281 | int i = ARRAY_SIZE(_exits); | |
282 | ||
283 | while (i--) | |
284 | _exits[i](); | |
d15b774c AK |
285 | |
286 | /* | |
287 | * Should be empty by this point. | |
288 | */ | |
d15b774c | 289 | idr_destroy(&_minor_idr); |
1da177e4 LT |
290 | } |
291 | ||
292 | /* | |
293 | * Block device functions | |
294 | */ | |
432a212c MA |
295 | int dm_deleting_md(struct mapped_device *md) |
296 | { | |
297 | return test_bit(DMF_DELETING, &md->flags); | |
298 | } | |
299 | ||
fe5f9f2c | 300 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
301 | { |
302 | struct mapped_device *md; | |
303 | ||
fba9f90e JM |
304 | spin_lock(&_minor_lock); |
305 | ||
fe5f9f2c | 306 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
307 | if (!md) |
308 | goto out; | |
309 | ||
5c6bd75d | 310 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 311 | dm_deleting_md(md)) { |
fba9f90e JM |
312 | md = NULL; |
313 | goto out; | |
314 | } | |
315 | ||
1da177e4 | 316 | dm_get(md); |
5c6bd75d | 317 | atomic_inc(&md->open_count); |
fba9f90e JM |
318 | out: |
319 | spin_unlock(&_minor_lock); | |
320 | ||
321 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
322 | } |
323 | ||
db2a144b | 324 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 325 | { |
63a4f065 | 326 | struct mapped_device *md; |
6e9624b8 | 327 | |
4a1aeb98 MB |
328 | spin_lock(&_minor_lock); |
329 | ||
63a4f065 MS |
330 | md = disk->private_data; |
331 | if (WARN_ON(!md)) | |
332 | goto out; | |
333 | ||
2c140a24 MP |
334 | if (atomic_dec_and_test(&md->open_count) && |
335 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 336 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 337 | |
1da177e4 | 338 | dm_put(md); |
63a4f065 | 339 | out: |
4a1aeb98 | 340 | spin_unlock(&_minor_lock); |
1da177e4 LT |
341 | } |
342 | ||
5c6bd75d AK |
343 | int dm_open_count(struct mapped_device *md) |
344 | { | |
345 | return atomic_read(&md->open_count); | |
346 | } | |
347 | ||
348 | /* | |
349 | * Guarantees nothing is using the device before it's deleted. | |
350 | */ | |
2c140a24 | 351 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
352 | { |
353 | int r = 0; | |
354 | ||
355 | spin_lock(&_minor_lock); | |
356 | ||
2c140a24 | 357 | if (dm_open_count(md)) { |
5c6bd75d | 358 | r = -EBUSY; |
2c140a24 MP |
359 | if (mark_deferred) |
360 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
361 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
362 | r = -EEXIST; | |
5c6bd75d AK |
363 | else |
364 | set_bit(DMF_DELETING, &md->flags); | |
365 | ||
366 | spin_unlock(&_minor_lock); | |
367 | ||
368 | return r; | |
369 | } | |
370 | ||
2c140a24 MP |
371 | int dm_cancel_deferred_remove(struct mapped_device *md) |
372 | { | |
373 | int r = 0; | |
374 | ||
375 | spin_lock(&_minor_lock); | |
376 | ||
377 | if (test_bit(DMF_DELETING, &md->flags)) | |
378 | r = -EBUSY; | |
379 | else | |
380 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
381 | ||
382 | spin_unlock(&_minor_lock); | |
383 | ||
384 | return r; | |
385 | } | |
386 | ||
387 | static void do_deferred_remove(struct work_struct *w) | |
388 | { | |
389 | dm_deferred_remove(); | |
390 | } | |
391 | ||
fd2ed4d2 MP |
392 | sector_t dm_get_size(struct mapped_device *md) |
393 | { | |
394 | return get_capacity(md->disk); | |
395 | } | |
396 | ||
9974fa2c MS |
397 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
398 | { | |
399 | return md->queue; | |
400 | } | |
401 | ||
fd2ed4d2 MP |
402 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
403 | { | |
404 | return &md->stats; | |
405 | } | |
406 | ||
3ac51e74 DW |
407 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
408 | { | |
409 | struct mapped_device *md = bdev->bd_disk->private_data; | |
410 | ||
411 | return dm_get_geometry(md, geo); | |
412 | } | |
413 | ||
956a4025 MS |
414 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
415 | struct block_device **bdev, | |
416 | fmode_t *mode) | |
aa129a22 | 417 | { |
66482026 | 418 | struct dm_target *tgt; |
6c182cd8 | 419 | struct dm_table *map; |
956a4025 | 420 | int srcu_idx, r; |
aa129a22 | 421 | |
6c182cd8 | 422 | retry: |
e56f81e0 | 423 | r = -ENOTTY; |
956a4025 | 424 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
425 | if (!map || !dm_table_get_size(map)) |
426 | goto out; | |
427 | ||
428 | /* We only support devices that have a single target */ | |
429 | if (dm_table_get_num_targets(map) != 1) | |
430 | goto out; | |
431 | ||
66482026 MS |
432 | tgt = dm_table_get_target(map, 0); |
433 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 434 | goto out; |
aa129a22 | 435 | |
4f186f8b | 436 | if (dm_suspended_md(md)) { |
aa129a22 MB |
437 | r = -EAGAIN; |
438 | goto out; | |
439 | } | |
440 | ||
66482026 | 441 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
442 | if (r < 0) |
443 | goto out; | |
aa129a22 | 444 | |
956a4025 MS |
445 | bdgrab(*bdev); |
446 | dm_put_live_table(md, srcu_idx); | |
e56f81e0 | 447 | return r; |
aa129a22 | 448 | |
aa129a22 | 449 | out: |
956a4025 | 450 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 451 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
452 | msleep(10); |
453 | goto retry; | |
454 | } | |
e56f81e0 CH |
455 | return r; |
456 | } | |
457 | ||
458 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
459 | unsigned int cmd, unsigned long arg) | |
460 | { | |
461 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 462 | int r; |
e56f81e0 | 463 | |
956a4025 | 464 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
465 | if (r < 0) |
466 | return r; | |
6c182cd8 | 467 | |
e56f81e0 CH |
468 | if (r > 0) { |
469 | /* | |
e980f623 CH |
470 | * Target determined this ioctl is being issued against a |
471 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 472 | */ |
e980f623 CH |
473 | if (!capable(CAP_SYS_RAWIO)) { |
474 | DMWARN_LIMIT( | |
475 | "%s: sending ioctl %x to DM device without required privilege.", | |
476 | current->comm, cmd); | |
477 | r = -ENOIOCTLCMD; | |
e56f81e0 | 478 | goto out; |
e980f623 | 479 | } |
e56f81e0 | 480 | } |
6c182cd8 | 481 | |
66482026 | 482 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 483 | out: |
956a4025 | 484 | bdput(bdev); |
aa129a22 MB |
485 | return r; |
486 | } | |
487 | ||
028867ac | 488 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
489 | { |
490 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
491 | } | |
492 | ||
028867ac | 493 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
494 | { |
495 | mempool_free(io, md->io_pool); | |
496 | } | |
497 | ||
cfae7529 | 498 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 499 | { |
dba14160 | 500 | bio_put(&tio->clone); |
1da177e4 LT |
501 | } |
502 | ||
4cc96131 | 503 | int md_in_flight(struct mapped_device *md) |
90abb8c4 KU |
504 | { |
505 | return atomic_read(&md->pending[READ]) + | |
506 | atomic_read(&md->pending[WRITE]); | |
507 | } | |
508 | ||
3eaf840e JNN |
509 | static void start_io_acct(struct dm_io *io) |
510 | { | |
511 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 512 | struct bio *bio = io->bio; |
c9959059 | 513 | int cpu; |
fd2ed4d2 | 514 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
515 | |
516 | io->start_time = jiffies; | |
517 | ||
074a7aca TH |
518 | cpu = part_stat_lock(); |
519 | part_round_stats(cpu, &dm_disk(md)->part0); | |
520 | part_stat_unlock(); | |
1e9bb880 SL |
521 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
522 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
523 | |
524 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
525 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
526 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
527 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
528 | } |
529 | ||
d221d2e7 | 530 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
531 | { |
532 | struct mapped_device *md = io->md; | |
533 | struct bio *bio = io->bio; | |
534 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 535 | int pending; |
3eaf840e JNN |
536 | int rw = bio_data_dir(bio); |
537 | ||
18c0b223 | 538 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 539 | |
fd2ed4d2 | 540 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
541 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
542 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
543 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 544 | |
af7e466a MP |
545 | /* |
546 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 547 | * a flush. |
af7e466a | 548 | */ |
1e9bb880 SL |
549 | pending = atomic_dec_return(&md->pending[rw]); |
550 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 551 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 552 | |
d221d2e7 MP |
553 | /* nudge anyone waiting on suspend queue */ |
554 | if (!pending) | |
555 | wake_up(&md->wait); | |
3eaf840e JNN |
556 | } |
557 | ||
1da177e4 LT |
558 | /* |
559 | * Add the bio to the list of deferred io. | |
560 | */ | |
92c63902 | 561 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 562 | { |
05447420 | 563 | unsigned long flags; |
1da177e4 | 564 | |
05447420 | 565 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 566 | bio_list_add(&md->deferred, bio); |
05447420 | 567 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 568 | queue_work(md->wq, &md->work); |
1da177e4 LT |
569 | } |
570 | ||
571 | /* | |
572 | * Everyone (including functions in this file), should use this | |
573 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 574 | * dm_put_live_table() when finished. |
1da177e4 | 575 | */ |
83d5e5b0 | 576 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 577 | { |
83d5e5b0 MP |
578 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
579 | ||
580 | return srcu_dereference(md->map, &md->io_barrier); | |
581 | } | |
1da177e4 | 582 | |
83d5e5b0 MP |
583 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
584 | { | |
585 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
586 | } | |
587 | ||
588 | void dm_sync_table(struct mapped_device *md) | |
589 | { | |
590 | synchronize_srcu(&md->io_barrier); | |
591 | synchronize_rcu_expedited(); | |
592 | } | |
593 | ||
594 | /* | |
595 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
596 | * The caller must not block between these two functions. | |
597 | */ | |
598 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
599 | { | |
600 | rcu_read_lock(); | |
601 | return rcu_dereference(md->map); | |
602 | } | |
1da177e4 | 603 | |
83d5e5b0 MP |
604 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
605 | { | |
606 | rcu_read_unlock(); | |
1da177e4 LT |
607 | } |
608 | ||
86f1152b BM |
609 | /* |
610 | * Open a table device so we can use it as a map destination. | |
611 | */ | |
612 | static int open_table_device(struct table_device *td, dev_t dev, | |
613 | struct mapped_device *md) | |
614 | { | |
615 | static char *_claim_ptr = "I belong to device-mapper"; | |
616 | struct block_device *bdev; | |
617 | ||
618 | int r; | |
619 | ||
620 | BUG_ON(td->dm_dev.bdev); | |
621 | ||
622 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
623 | if (IS_ERR(bdev)) | |
624 | return PTR_ERR(bdev); | |
625 | ||
626 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
627 | if (r) { | |
628 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
629 | return r; | |
630 | } | |
631 | ||
632 | td->dm_dev.bdev = bdev; | |
817bf402 | 633 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
634 | return 0; |
635 | } | |
636 | ||
637 | /* | |
638 | * Close a table device that we've been using. | |
639 | */ | |
640 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
641 | { | |
642 | if (!td->dm_dev.bdev) | |
643 | return; | |
644 | ||
645 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
646 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 647 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 648 | td->dm_dev.bdev = NULL; |
817bf402 | 649 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
650 | } |
651 | ||
652 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
653 | fmode_t mode) { | |
654 | struct table_device *td; | |
655 | ||
656 | list_for_each_entry(td, l, list) | |
657 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
658 | return td; | |
659 | ||
660 | return NULL; | |
661 | } | |
662 | ||
663 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
664 | struct dm_dev **result) { | |
665 | int r; | |
666 | struct table_device *td; | |
667 | ||
668 | mutex_lock(&md->table_devices_lock); | |
669 | td = find_table_device(&md->table_devices, dev, mode); | |
670 | if (!td) { | |
115485e8 | 671 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
672 | if (!td) { |
673 | mutex_unlock(&md->table_devices_lock); | |
674 | return -ENOMEM; | |
675 | } | |
676 | ||
677 | td->dm_dev.mode = mode; | |
678 | td->dm_dev.bdev = NULL; | |
679 | ||
680 | if ((r = open_table_device(td, dev, md))) { | |
681 | mutex_unlock(&md->table_devices_lock); | |
682 | kfree(td); | |
683 | return r; | |
684 | } | |
685 | ||
686 | format_dev_t(td->dm_dev.name, dev); | |
687 | ||
688 | atomic_set(&td->count, 0); | |
689 | list_add(&td->list, &md->table_devices); | |
690 | } | |
691 | atomic_inc(&td->count); | |
692 | mutex_unlock(&md->table_devices_lock); | |
693 | ||
694 | *result = &td->dm_dev; | |
695 | return 0; | |
696 | } | |
697 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
698 | ||
699 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
700 | { | |
701 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
702 | ||
703 | mutex_lock(&md->table_devices_lock); | |
704 | if (atomic_dec_and_test(&td->count)) { | |
705 | close_table_device(td, md); | |
706 | list_del(&td->list); | |
707 | kfree(td); | |
708 | } | |
709 | mutex_unlock(&md->table_devices_lock); | |
710 | } | |
711 | EXPORT_SYMBOL(dm_put_table_device); | |
712 | ||
713 | static void free_table_devices(struct list_head *devices) | |
714 | { | |
715 | struct list_head *tmp, *next; | |
716 | ||
717 | list_for_each_safe(tmp, next, devices) { | |
718 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
719 | ||
720 | DMWARN("dm_destroy: %s still exists with %d references", | |
721 | td->dm_dev.name, atomic_read(&td->count)); | |
722 | kfree(td); | |
723 | } | |
724 | } | |
725 | ||
3ac51e74 DW |
726 | /* |
727 | * Get the geometry associated with a dm device | |
728 | */ | |
729 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
730 | { | |
731 | *geo = md->geometry; | |
732 | ||
733 | return 0; | |
734 | } | |
735 | ||
736 | /* | |
737 | * Set the geometry of a device. | |
738 | */ | |
739 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
740 | { | |
741 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
742 | ||
743 | if (geo->start > sz) { | |
744 | DMWARN("Start sector is beyond the geometry limits."); | |
745 | return -EINVAL; | |
746 | } | |
747 | ||
748 | md->geometry = *geo; | |
749 | ||
750 | return 0; | |
751 | } | |
752 | ||
1da177e4 LT |
753 | /*----------------------------------------------------------------- |
754 | * CRUD START: | |
755 | * A more elegant soln is in the works that uses the queue | |
756 | * merge fn, unfortunately there are a couple of changes to | |
757 | * the block layer that I want to make for this. So in the | |
758 | * interests of getting something for people to use I give | |
759 | * you this clearly demarcated crap. | |
760 | *---------------------------------------------------------------*/ | |
761 | ||
2e93ccc1 KU |
762 | static int __noflush_suspending(struct mapped_device *md) |
763 | { | |
764 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
765 | } | |
766 | ||
1da177e4 LT |
767 | /* |
768 | * Decrements the number of outstanding ios that a bio has been | |
769 | * cloned into, completing the original io if necc. | |
770 | */ | |
858119e1 | 771 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 772 | { |
2e93ccc1 | 773 | unsigned long flags; |
b35f8caa MB |
774 | int io_error; |
775 | struct bio *bio; | |
776 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
777 | |
778 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
779 | if (unlikely(error)) { |
780 | spin_lock_irqsave(&io->endio_lock, flags); | |
781 | if (!(io->error > 0 && __noflush_suspending(md))) | |
782 | io->error = error; | |
783 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
784 | } | |
1da177e4 LT |
785 | |
786 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
787 | if (io->error == DM_ENDIO_REQUEUE) { |
788 | /* | |
789 | * Target requested pushing back the I/O. | |
2e93ccc1 | 790 | */ |
022c2611 | 791 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
792 | if (__noflush_suspending(md)) |
793 | bio_list_add_head(&md->deferred, io->bio); | |
794 | else | |
2e93ccc1 KU |
795 | /* noflush suspend was interrupted. */ |
796 | io->error = -EIO; | |
022c2611 | 797 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
798 | } |
799 | ||
b35f8caa MB |
800 | io_error = io->error; |
801 | bio = io->bio; | |
6a8736d1 TH |
802 | end_io_acct(io); |
803 | free_io(md, io); | |
804 | ||
805 | if (io_error == DM_ENDIO_REQUEUE) | |
806 | return; | |
2e93ccc1 | 807 | |
1eff9d32 | 808 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 809 | /* |
6a8736d1 | 810 | * Preflush done for flush with data, reissue |
28a8f0d3 | 811 | * without REQ_PREFLUSH. |
af7e466a | 812 | */ |
1eff9d32 | 813 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 814 | queue_io(md, bio); |
af7e466a | 815 | } else { |
b372d360 | 816 | /* done with normal IO or empty flush */ |
4246a0b6 CH |
817 | bio->bi_error = io_error; |
818 | bio_endio(bio); | |
b35f8caa | 819 | } |
1da177e4 LT |
820 | } |
821 | } | |
822 | ||
4cc96131 | 823 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
824 | { |
825 | struct queue_limits *limits = dm_get_queue_limits(md); | |
826 | ||
827 | /* device doesn't really support WRITE SAME, disable it */ | |
828 | limits->max_write_same_sectors = 0; | |
829 | } | |
830 | ||
ac62d620 CH |
831 | void disable_write_zeroes(struct mapped_device *md) |
832 | { | |
833 | struct queue_limits *limits = dm_get_queue_limits(md); | |
834 | ||
835 | /* device doesn't really support WRITE ZEROES, disable it */ | |
836 | limits->max_write_zeroes_sectors = 0; | |
837 | } | |
838 | ||
4246a0b6 | 839 | static void clone_endio(struct bio *bio) |
1da177e4 | 840 | { |
4246a0b6 | 841 | int error = bio->bi_error; |
5164bece | 842 | int r = error; |
bfc6d41c | 843 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 844 | struct dm_io *io = tio->io; |
9faf400f | 845 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
846 | dm_endio_fn endio = tio->ti->type->end_io; |
847 | ||
1da177e4 | 848 | if (endio) { |
7de3ee57 | 849 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
850 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
851 | /* | |
852 | * error and requeue request are handled | |
853 | * in dec_pending(). | |
854 | */ | |
1da177e4 | 855 | error = r; |
45cbcd79 KU |
856 | else if (r == DM_ENDIO_INCOMPLETE) |
857 | /* The target will handle the io */ | |
6712ecf8 | 858 | return; |
45cbcd79 KU |
859 | else if (r) { |
860 | DMWARN("unimplemented target endio return value: %d", r); | |
861 | BUG(); | |
862 | } | |
1da177e4 LT |
863 | } |
864 | ||
ac62d620 CH |
865 | if (unlikely(r == -EREMOTEIO)) { |
866 | if (bio_op(bio) == REQ_OP_WRITE_SAME && | |
867 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors) | |
868 | disable_write_same(md); | |
869 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && | |
870 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_zeroes_sectors) | |
871 | disable_write_zeroes(md); | |
872 | } | |
7eee4ae2 | 873 | |
cfae7529 | 874 | free_tio(tio); |
b35f8caa | 875 | dec_pending(io, error); |
1da177e4 LT |
876 | } |
877 | ||
56a67df7 MS |
878 | /* |
879 | * Return maximum size of I/O possible at the supplied sector up to the current | |
880 | * target boundary. | |
881 | */ | |
882 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
883 | { | |
884 | sector_t target_offset = dm_target_offset(ti, sector); | |
885 | ||
886 | return ti->len - target_offset; | |
887 | } | |
888 | ||
889 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 890 | { |
56a67df7 | 891 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 892 | sector_t offset, max_len; |
1da177e4 LT |
893 | |
894 | /* | |
542f9038 | 895 | * Does the target need to split even further? |
1da177e4 | 896 | */ |
542f9038 MS |
897 | if (ti->max_io_len) { |
898 | offset = dm_target_offset(ti, sector); | |
899 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
900 | max_len = sector_div(offset, ti->max_io_len); | |
901 | else | |
902 | max_len = offset & (ti->max_io_len - 1); | |
903 | max_len = ti->max_io_len - max_len; | |
904 | ||
905 | if (len > max_len) | |
906 | len = max_len; | |
1da177e4 LT |
907 | } |
908 | ||
909 | return len; | |
910 | } | |
911 | ||
542f9038 MS |
912 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
913 | { | |
914 | if (len > UINT_MAX) { | |
915 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
916 | (unsigned long long)len, UINT_MAX); | |
917 | ti->error = "Maximum size of target IO is too large"; | |
918 | return -EINVAL; | |
919 | } | |
920 | ||
921 | ti->max_io_len = (uint32_t) len; | |
922 | ||
923 | return 0; | |
924 | } | |
925 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
926 | ||
f26c5719 DW |
927 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
928 | sector_t sector, int *srcu_idx) | |
545ed20e | 929 | { |
545ed20e TK |
930 | struct dm_table *map; |
931 | struct dm_target *ti; | |
545ed20e | 932 | |
f26c5719 | 933 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 934 | if (!map) |
f26c5719 | 935 | return NULL; |
545ed20e TK |
936 | |
937 | ti = dm_table_find_target(map, sector); | |
938 | if (!dm_target_is_valid(ti)) | |
f26c5719 | 939 | return NULL; |
545ed20e | 940 | |
f26c5719 DW |
941 | return ti; |
942 | } | |
545ed20e | 943 | |
f26c5719 DW |
944 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
945 | long nr_pages, void **kaddr, pfn_t *pfn) | |
946 | { | |
947 | struct mapped_device *md = dax_get_private(dax_dev); | |
948 | sector_t sector = pgoff * PAGE_SECTORS; | |
949 | struct dm_target *ti; | |
950 | long len, ret = -EIO; | |
951 | int srcu_idx; | |
545ed20e | 952 | |
f26c5719 | 953 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 954 | |
f26c5719 DW |
955 | if (!ti) |
956 | goto out; | |
957 | if (!ti->type->direct_access) | |
958 | goto out; | |
959 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
960 | if (len < 1) | |
961 | goto out; | |
962 | nr_pages = min(len, nr_pages); | |
545ed20e | 963 | if (ti->type->direct_access) |
817bf402 DW |
964 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
965 | ||
f26c5719 | 966 | out: |
545ed20e | 967 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
968 | |
969 | return ret; | |
545ed20e TK |
970 | } |
971 | ||
1dd40c3e MP |
972 | /* |
973 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
28a8f0d3 | 974 | * allowed for all bio types except REQ_PREFLUSH. |
1dd40c3e MP |
975 | * |
976 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
977 | * additional n_sectors sectors of the bio and the rest of the data should be | |
978 | * sent in a next bio. | |
979 | * | |
980 | * A diagram that explains the arithmetics: | |
981 | * +--------------------+---------------+-------+ | |
982 | * | 1 | 2 | 3 | | |
983 | * +--------------------+---------------+-------+ | |
984 | * | |
985 | * <-------------- *tio->len_ptr ---------------> | |
986 | * <------- bi_size -------> | |
987 | * <-- n_sectors --> | |
988 | * | |
989 | * Region 1 was already iterated over with bio_advance or similar function. | |
990 | * (it may be empty if the target doesn't use bio_advance) | |
991 | * Region 2 is the remaining bio size that the target wants to process. | |
992 | * (it may be empty if region 1 is non-empty, although there is no reason | |
993 | * to make it empty) | |
994 | * The target requires that region 3 is to be sent in the next bio. | |
995 | * | |
996 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
997 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
998 | * copies of the bio. | |
999 | */ | |
1000 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1001 | { | |
1002 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1003 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1004 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1005 | BUG_ON(bi_size > *tio->len_ptr); |
1006 | BUG_ON(n_sectors > bi_size); | |
1007 | *tio->len_ptr -= bi_size - n_sectors; | |
1008 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1009 | } | |
1010 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1011 | ||
d67a5f4b MP |
1012 | /* |
1013 | * Flush current->bio_list when the target map method blocks. | |
1014 | * This fixes deadlocks in snapshot and possibly in other targets. | |
1015 | */ | |
1016 | struct dm_offload { | |
1017 | struct blk_plug plug; | |
1018 | struct blk_plug_cb cb; | |
1019 | }; | |
1020 | ||
1021 | static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule) | |
1022 | { | |
1023 | struct dm_offload *o = container_of(cb, struct dm_offload, cb); | |
1024 | struct bio_list list; | |
1025 | struct bio *bio; | |
f5fe1b51 | 1026 | int i; |
d67a5f4b MP |
1027 | |
1028 | INIT_LIST_HEAD(&o->cb.list); | |
1029 | ||
1030 | if (unlikely(!current->bio_list)) | |
1031 | return; | |
1032 | ||
f5fe1b51 N |
1033 | for (i = 0; i < 2; i++) { |
1034 | list = current->bio_list[i]; | |
1035 | bio_list_init(¤t->bio_list[i]); | |
1036 | ||
1037 | while ((bio = bio_list_pop(&list))) { | |
1038 | struct bio_set *bs = bio->bi_pool; | |
1039 | if (unlikely(!bs) || bs == fs_bio_set) { | |
1040 | bio_list_add(¤t->bio_list[i], bio); | |
1041 | continue; | |
1042 | } | |
1043 | ||
1044 | spin_lock(&bs->rescue_lock); | |
1045 | bio_list_add(&bs->rescue_list, bio); | |
1046 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
1047 | spin_unlock(&bs->rescue_lock); | |
d67a5f4b | 1048 | } |
d67a5f4b MP |
1049 | } |
1050 | } | |
1051 | ||
1052 | static void dm_offload_start(struct dm_offload *o) | |
1053 | { | |
1054 | blk_start_plug(&o->plug); | |
1055 | o->cb.callback = flush_current_bio_list; | |
1056 | list_add(&o->cb.list, ¤t->plug->cb_list); | |
1057 | } | |
1058 | ||
1059 | static void dm_offload_end(struct dm_offload *o) | |
1060 | { | |
1061 | list_del(&o->cb.list); | |
1062 | blk_finish_plug(&o->plug); | |
1063 | } | |
1064 | ||
bd2a49b8 | 1065 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1066 | { |
1067 | int r; | |
2056a782 | 1068 | sector_t sector; |
d67a5f4b | 1069 | struct dm_offload o; |
dba14160 | 1070 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1071 | struct dm_target *ti = tio->ti; |
1da177e4 | 1072 | |
1da177e4 | 1073 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1074 | |
1075 | /* | |
1076 | * Map the clone. If r == 0 we don't need to do | |
1077 | * anything, the target has assumed ownership of | |
1078 | * this io. | |
1079 | */ | |
1080 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1081 | sector = clone->bi_iter.bi_sector; |
d67a5f4b MP |
1082 | |
1083 | dm_offload_start(&o); | |
7de3ee57 | 1084 | r = ti->type->map(ti, clone); |
d67a5f4b MP |
1085 | dm_offload_end(&o); |
1086 | ||
45cbcd79 | 1087 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1088 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1089 | |
d07335e5 MS |
1090 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1091 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1092 | |
1da177e4 | 1093 | generic_make_request(clone); |
2e93ccc1 KU |
1094 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1095 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f | 1096 | dec_pending(tio->io, r); |
cfae7529 | 1097 | free_tio(tio); |
ab37844d | 1098 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1099 | DMWARN("unimplemented target map return value: %d", r); |
1100 | BUG(); | |
1da177e4 LT |
1101 | } |
1102 | } | |
1103 | ||
1104 | struct clone_info { | |
1105 | struct mapped_device *md; | |
1106 | struct dm_table *map; | |
1107 | struct bio *bio; | |
1108 | struct dm_io *io; | |
1109 | sector_t sector; | |
e0d6609a | 1110 | unsigned sector_count; |
1da177e4 LT |
1111 | }; |
1112 | ||
e0d6609a | 1113 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1114 | { |
4f024f37 KO |
1115 | bio->bi_iter.bi_sector = sector; |
1116 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1117 | } |
1118 | ||
1119 | /* | |
1120 | * Creates a bio that consists of range of complete bvecs. | |
1121 | */ | |
c80914e8 MS |
1122 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1123 | sector_t sector, unsigned len) | |
1da177e4 | 1124 | { |
dba14160 | 1125 | struct bio *clone = &tio->clone; |
1da177e4 | 1126 | |
1c3b13e6 KO |
1127 | __bio_clone_fast(clone, bio); |
1128 | ||
e2460f2a MP |
1129 | if (unlikely(bio_integrity(bio) != NULL)) { |
1130 | int r; | |
1131 | ||
1132 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1133 | !dm_target_passes_integrity(tio->ti->type))) { | |
1134 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1135 | dm_device_name(tio->io->md), | |
1136 | tio->ti->type->name); | |
1137 | return -EIO; | |
1138 | } | |
1139 | ||
1140 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1141 | if (r < 0) |
1142 | return r; | |
1143 | } | |
bd2a49b8 | 1144 | |
1c3b13e6 KO |
1145 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1146 | clone->bi_iter.bi_size = to_bytes(len); | |
1147 | ||
e2460f2a | 1148 | if (unlikely(bio_integrity(bio) != NULL)) |
1c3b13e6 | 1149 | bio_integrity_trim(clone, 0, len); |
c80914e8 MS |
1150 | |
1151 | return 0; | |
1da177e4 LT |
1152 | } |
1153 | ||
9015df24 | 1154 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1155 | struct dm_target *ti, |
55a62eef | 1156 | unsigned target_bio_nr) |
f9ab94ce | 1157 | { |
dba14160 MP |
1158 | struct dm_target_io *tio; |
1159 | struct bio *clone; | |
1160 | ||
99778273 | 1161 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1162 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1163 | |
1164 | tio->io = ci->io; | |
1165 | tio->ti = ti; | |
55a62eef | 1166 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1167 | |
1168 | return tio; | |
1169 | } | |
1170 | ||
14fe594d AK |
1171 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1172 | struct dm_target *ti, | |
1dd40c3e | 1173 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1174 | { |
99778273 | 1175 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1176 | struct bio *clone = &tio->clone; |
9015df24 | 1177 | |
1dd40c3e MP |
1178 | tio->len_ptr = len; |
1179 | ||
99778273 | 1180 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1181 | if (len) |
1dd40c3e | 1182 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1183 | |
bd2a49b8 | 1184 | __map_bio(tio); |
f9ab94ce MP |
1185 | } |
1186 | ||
14fe594d | 1187 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1188 | unsigned num_bios, unsigned *len) |
06a426ce | 1189 | { |
55a62eef | 1190 | unsigned target_bio_nr; |
06a426ce | 1191 | |
55a62eef | 1192 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1193 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1194 | } |
1195 | ||
14fe594d | 1196 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1197 | { |
06a426ce | 1198 | unsigned target_nr = 0; |
f9ab94ce MP |
1199 | struct dm_target *ti; |
1200 | ||
b372d360 | 1201 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1202 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1203 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1204 | |
f9ab94ce MP |
1205 | return 0; |
1206 | } | |
1207 | ||
c80914e8 | 1208 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1209 | sector_t sector, unsigned *len) |
5ae89a87 | 1210 | { |
dba14160 | 1211 | struct bio *bio = ci->bio; |
5ae89a87 | 1212 | struct dm_target_io *tio; |
b0d8ed4d AK |
1213 | unsigned target_bio_nr; |
1214 | unsigned num_target_bios = 1; | |
c80914e8 | 1215 | int r = 0; |
5ae89a87 | 1216 | |
b0d8ed4d AK |
1217 | /* |
1218 | * Does the target want to receive duplicate copies of the bio? | |
1219 | */ | |
1220 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1221 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1222 | |
b0d8ed4d | 1223 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1224 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e | 1225 | tio->len_ptr = len; |
c80914e8 | 1226 | r = clone_bio(tio, bio, sector, *len); |
072623de | 1227 | if (r < 0) { |
cfae7529 | 1228 | free_tio(tio); |
c80914e8 | 1229 | break; |
072623de | 1230 | } |
b0d8ed4d AK |
1231 | __map_bio(tio); |
1232 | } | |
c80914e8 MS |
1233 | |
1234 | return r; | |
5ae89a87 MS |
1235 | } |
1236 | ||
55a62eef | 1237 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1238 | |
55a62eef | 1239 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1240 | { |
55a62eef | 1241 | return ti->num_discard_bios; |
23508a96 MS |
1242 | } |
1243 | ||
55a62eef | 1244 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1245 | { |
55a62eef | 1246 | return ti->num_write_same_bios; |
23508a96 MS |
1247 | } |
1248 | ||
ac62d620 CH |
1249 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1250 | { | |
1251 | return ti->num_write_zeroes_bios; | |
1252 | } | |
1253 | ||
23508a96 | 1254 | typedef bool (*is_split_required_fn)(struct dm_target *ti); |
9eef87da | 1255 | |
23508a96 MS |
1256 | static bool is_split_required_for_discard(struct dm_target *ti) |
1257 | { | |
55a62eef | 1258 | return ti->split_discard_bios; |
cec47e3d KU |
1259 | } |
1260 | ||
14fe594d AK |
1261 | static int __send_changing_extent_only(struct clone_info *ci, |
1262 | get_num_bios_fn get_num_bios, | |
1263 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1264 | { |
5ae89a87 | 1265 | struct dm_target *ti; |
e0d6609a | 1266 | unsigned len; |
55a62eef | 1267 | unsigned num_bios; |
ba1cbad9 | 1268 | |
a79245b3 MS |
1269 | do { |
1270 | ti = dm_table_find_target(ci->map, ci->sector); | |
1271 | if (!dm_target_is_valid(ti)) | |
1272 | return -EIO; | |
2eb6e1e3 | 1273 | |
5ae89a87 | 1274 | /* |
23508a96 MS |
1275 | * Even though the device advertised support for this type of |
1276 | * request, that does not mean every target supports it, and | |
936688d7 | 1277 | * reconfiguration might also have changed that since the |
a79245b3 | 1278 | * check was performed. |
5ae89a87 | 1279 | */ |
55a62eef AK |
1280 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1281 | if (!num_bios) | |
a79245b3 | 1282 | return -EOPNOTSUPP; |
ba1cbad9 | 1283 | |
23508a96 | 1284 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1285 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1286 | else |
e0d6609a | 1287 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
de3ec86d | 1288 | |
1dd40c3e | 1289 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1290 | |
a79245b3 MS |
1291 | ci->sector += len; |
1292 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1293 | |
1294 | return 0; | |
ba1cbad9 MS |
1295 | } |
1296 | ||
14fe594d | 1297 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1298 | { |
14fe594d AK |
1299 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1300 | is_split_required_for_discard); | |
23508a96 | 1301 | } |
0ce65797 | 1302 | |
14fe594d | 1303 | static int __send_write_same(struct clone_info *ci) |
0ce65797 | 1304 | { |
14fe594d | 1305 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
0ce65797 MS |
1306 | } |
1307 | ||
ac62d620 CH |
1308 | static int __send_write_zeroes(struct clone_info *ci) |
1309 | { | |
1310 | return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL); | |
1311 | } | |
1312 | ||
e4c93811 AK |
1313 | /* |
1314 | * Select the correct strategy for processing a non-flush bio. | |
1315 | */ | |
14fe594d | 1316 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1317 | { |
dba14160 | 1318 | struct bio *bio = ci->bio; |
512875bd | 1319 | struct dm_target *ti; |
1c3b13e6 | 1320 | unsigned len; |
c80914e8 | 1321 | int r; |
0ce65797 | 1322 | |
e6047149 | 1323 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
14fe594d | 1324 | return __send_discard(ci); |
e6047149 | 1325 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
14fe594d | 1326 | return __send_write_same(ci); |
ac62d620 CH |
1327 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES)) |
1328 | return __send_write_zeroes(ci); | |
0ce65797 | 1329 | |
512875bd JN |
1330 | ti = dm_table_find_target(ci->map, ci->sector); |
1331 | if (!dm_target_is_valid(ti)) | |
1332 | return -EIO; | |
1333 | ||
1c3b13e6 | 1334 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
0ce65797 | 1335 | |
c80914e8 MS |
1336 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1337 | if (r < 0) | |
1338 | return r; | |
0ce65797 | 1339 | |
1c3b13e6 KO |
1340 | ci->sector += len; |
1341 | ci->sector_count -= len; | |
0ce65797 | 1342 | |
1c3b13e6 | 1343 | return 0; |
0ce65797 MS |
1344 | } |
1345 | ||
1da177e4 | 1346 | /* |
14fe594d | 1347 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1348 | */ |
83d5e5b0 MP |
1349 | static void __split_and_process_bio(struct mapped_device *md, |
1350 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1351 | { |
1da177e4 | 1352 | struct clone_info ci; |
512875bd | 1353 | int error = 0; |
1da177e4 | 1354 | |
83d5e5b0 | 1355 | if (unlikely(!map)) { |
6a8736d1 | 1356 | bio_io_error(bio); |
f0b9a450 MP |
1357 | return; |
1358 | } | |
692d0eb9 | 1359 | |
83d5e5b0 | 1360 | ci.map = map; |
1da177e4 | 1361 | ci.md = md; |
1da177e4 LT |
1362 | ci.io = alloc_io(md); |
1363 | ci.io->error = 0; | |
1364 | atomic_set(&ci.io->io_count, 1); | |
1365 | ci.io->bio = bio; | |
1366 | ci.io->md = md; | |
f88fb981 | 1367 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1368 | ci.sector = bio->bi_iter.bi_sector; |
0ce65797 | 1369 | |
3eaf840e | 1370 | start_io_acct(ci.io); |
0ce65797 | 1371 | |
1eff9d32 | 1372 | if (bio->bi_opf & REQ_PREFLUSH) { |
b372d360 MS |
1373 | ci.bio = &ci.md->flush_bio; |
1374 | ci.sector_count = 0; | |
14fe594d | 1375 | error = __send_empty_flush(&ci); |
b372d360 MS |
1376 | /* dec_pending submits any data associated with flush */ |
1377 | } else { | |
6a8736d1 | 1378 | ci.bio = bio; |
d87f4c14 | 1379 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1380 | while (ci.sector_count && !error) |
14fe594d | 1381 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1382 | } |
0ce65797 | 1383 | |
1da177e4 | 1384 | /* drop the extra reference count */ |
512875bd | 1385 | dec_pending(ci.io, error); |
0ce65797 | 1386 | } |
1da177e4 LT |
1387 | /*----------------------------------------------------------------- |
1388 | * CRUD END | |
1389 | *---------------------------------------------------------------*/ | |
0ce65797 | 1390 | |
cec47e3d | 1391 | /* |
1da177e4 LT |
1392 | * The request function that just remaps the bio built up by |
1393 | * dm_merge_bvec. | |
cec47e3d | 1394 | */ |
dece1635 | 1395 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
cec47e3d | 1396 | { |
12f03a49 | 1397 | int rw = bio_data_dir(bio); |
cec47e3d | 1398 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1399 | int srcu_idx; |
1400 | struct dm_table *map; | |
cec47e3d | 1401 | |
83d5e5b0 | 1402 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1403 | |
18c0b223 | 1404 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
d0bcb878 | 1405 | |
6a8736d1 TH |
1406 | /* if we're suspended, we have to queue this io for later */ |
1407 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1408 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1409 | |
1eff9d32 | 1410 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1411 | queue_io(md, bio); |
1412 | else | |
54d9a1b4 | 1413 | bio_io_error(bio); |
dece1635 | 1414 | return BLK_QC_T_NONE; |
cec47e3d | 1415 | } |
1da177e4 | 1416 | |
83d5e5b0 MP |
1417 | __split_and_process_bio(md, map, bio); |
1418 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1419 | return BLK_QC_T_NONE; |
cec47e3d KU |
1420 | } |
1421 | ||
1da177e4 LT |
1422 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1423 | { | |
8a57dfc6 CS |
1424 | int r = bdi_bits; |
1425 | struct mapped_device *md = congested_data; | |
1426 | struct dm_table *map; | |
1da177e4 | 1427 | |
1eb787ec | 1428 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1429 | if (dm_request_based(md)) { |
cec47e3d | 1430 | /* |
e522c039 MS |
1431 | * With request-based DM we only need to check the |
1432 | * top-level queue for congestion. | |
cec47e3d | 1433 | */ |
dc3b17cc | 1434 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1435 | } else { |
1436 | map = dm_get_live_table_fast(md); | |
1437 | if (map) | |
cec47e3d | 1438 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1439 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1440 | } |
1441 | } | |
1442 | ||
1da177e4 LT |
1443 | return r; |
1444 | } | |
1445 | ||
1446 | /*----------------------------------------------------------------- | |
1447 | * An IDR is used to keep track of allocated minor numbers. | |
1448 | *---------------------------------------------------------------*/ | |
2b06cfff | 1449 | static void free_minor(int minor) |
1da177e4 | 1450 | { |
f32c10b0 | 1451 | spin_lock(&_minor_lock); |
1da177e4 | 1452 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1453 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1454 | } |
1455 | ||
1456 | /* | |
1457 | * See if the device with a specific minor # is free. | |
1458 | */ | |
cf13ab8e | 1459 | static int specific_minor(int minor) |
1da177e4 | 1460 | { |
c9d76be6 | 1461 | int r; |
1da177e4 LT |
1462 | |
1463 | if (minor >= (1 << MINORBITS)) | |
1464 | return -EINVAL; | |
1465 | ||
c9d76be6 | 1466 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1467 | spin_lock(&_minor_lock); |
1da177e4 | 1468 | |
c9d76be6 | 1469 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1470 | |
f32c10b0 | 1471 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1472 | idr_preload_end(); |
1473 | if (r < 0) | |
1474 | return r == -ENOSPC ? -EBUSY : r; | |
1475 | return 0; | |
1da177e4 LT |
1476 | } |
1477 | ||
cf13ab8e | 1478 | static int next_free_minor(int *minor) |
1da177e4 | 1479 | { |
c9d76be6 | 1480 | int r; |
62f75c2f | 1481 | |
c9d76be6 | 1482 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1483 | spin_lock(&_minor_lock); |
1da177e4 | 1484 | |
c9d76be6 | 1485 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1486 | |
f32c10b0 | 1487 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1488 | idr_preload_end(); |
1489 | if (r < 0) | |
1490 | return r; | |
1491 | *minor = r; | |
1492 | return 0; | |
1da177e4 LT |
1493 | } |
1494 | ||
83d5cde4 | 1495 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1496 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1497 | |
53d5914f MP |
1498 | static void dm_wq_work(struct work_struct *work); |
1499 | ||
4cc96131 | 1500 | void dm_init_md_queue(struct mapped_device *md) |
4a0b4ddf MS |
1501 | { |
1502 | /* | |
1503 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 1504 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
1505 | * The type is decided at the first table loading time. |
1506 | * To prevent problematic device stacking, clear the queue flag | |
1507 | * for request stacking support until then. | |
1508 | * | |
1509 | * This queue is new, so no concurrency on the queue_flags. | |
1510 | */ | |
1511 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
1512 | |
1513 | /* | |
1514 | * Initialize data that will only be used by a non-blk-mq DM queue | |
1515 | * - must do so here (in alloc_dev callchain) before queue is used | |
1516 | */ | |
1517 | md->queue->queuedata = md; | |
dc3b17cc | 1518 | md->queue->backing_dev_info->congested_data = md; |
bfebd1cd | 1519 | } |
4a0b4ddf | 1520 | |
4cc96131 | 1521 | void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1522 | { |
17e149b8 | 1523 | md->use_blk_mq = false; |
bfebd1cd MS |
1524 | dm_init_md_queue(md); |
1525 | ||
1526 | /* | |
1527 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1528 | */ | |
dc3b17cc | 1529 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf | 1530 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
1531 | } |
1532 | ||
0f20972f MS |
1533 | static void cleanup_mapped_device(struct mapped_device *md) |
1534 | { | |
0f20972f MS |
1535 | if (md->wq) |
1536 | destroy_workqueue(md->wq); | |
1537 | if (md->kworker_task) | |
1538 | kthread_stop(md->kworker_task); | |
6f65985e | 1539 | mempool_destroy(md->io_pool); |
0f20972f MS |
1540 | if (md->bs) |
1541 | bioset_free(md->bs); | |
1542 | ||
f26c5719 DW |
1543 | if (md->dax_dev) { |
1544 | kill_dax(md->dax_dev); | |
1545 | put_dax(md->dax_dev); | |
1546 | md->dax_dev = NULL; | |
1547 | } | |
1548 | ||
0f20972f MS |
1549 | if (md->disk) { |
1550 | spin_lock(&_minor_lock); | |
1551 | md->disk->private_data = NULL; | |
1552 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1553 | del_gendisk(md->disk); |
1554 | put_disk(md->disk); | |
1555 | } | |
1556 | ||
1557 | if (md->queue) | |
1558 | blk_cleanup_queue(md->queue); | |
1559 | ||
d09960b0 TE |
1560 | cleanup_srcu_struct(&md->io_barrier); |
1561 | ||
0f20972f MS |
1562 | if (md->bdev) { |
1563 | bdput(md->bdev); | |
1564 | md->bdev = NULL; | |
1565 | } | |
4cc96131 MS |
1566 | |
1567 | dm_mq_cleanup_mapped_device(md); | |
0f20972f MS |
1568 | } |
1569 | ||
1da177e4 LT |
1570 | /* |
1571 | * Allocate and initialise a blank device with a given minor. | |
1572 | */ | |
2b06cfff | 1573 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1574 | { |
115485e8 | 1575 | int r, numa_node_id = dm_get_numa_node(); |
f26c5719 | 1576 | struct dax_device *dax_dev; |
115485e8 | 1577 | struct mapped_device *md; |
ba61fdd1 | 1578 | void *old_md; |
1da177e4 | 1579 | |
115485e8 | 1580 | md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1581 | if (!md) { |
1582 | DMWARN("unable to allocate device, out of memory."); | |
1583 | return NULL; | |
1584 | } | |
1585 | ||
10da4f79 | 1586 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1587 | goto bad_module_get; |
10da4f79 | 1588 | |
1da177e4 | 1589 | /* get a minor number for the dev */ |
2b06cfff | 1590 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1591 | r = next_free_minor(&minor); |
2b06cfff | 1592 | else |
cf13ab8e | 1593 | r = specific_minor(minor); |
1da177e4 | 1594 | if (r < 0) |
6ed7ade8 | 1595 | goto bad_minor; |
1da177e4 | 1596 | |
83d5e5b0 MP |
1597 | r = init_srcu_struct(&md->io_barrier); |
1598 | if (r < 0) | |
1599 | goto bad_io_barrier; | |
1600 | ||
115485e8 | 1601 | md->numa_node_id = numa_node_id; |
4cc96131 | 1602 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1603 | md->init_tio_pdu = false; |
a5664dad | 1604 | md->type = DM_TYPE_NONE; |
e61290a4 | 1605 | mutex_init(&md->suspend_lock); |
a5664dad | 1606 | mutex_init(&md->type_lock); |
86f1152b | 1607 | mutex_init(&md->table_devices_lock); |
022c2611 | 1608 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1609 | atomic_set(&md->holders, 1); |
5c6bd75d | 1610 | atomic_set(&md->open_count, 0); |
1da177e4 | 1611 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1612 | atomic_set(&md->uevent_seq, 0); |
1613 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1614 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1615 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1616 | |
115485e8 | 1617 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1618 | if (!md->queue) |
0f20972f | 1619 | goto bad; |
1da177e4 | 1620 | |
4a0b4ddf | 1621 | dm_init_md_queue(md); |
9faf400f | 1622 | |
115485e8 | 1623 | md->disk = alloc_disk_node(1, numa_node_id); |
1da177e4 | 1624 | if (!md->disk) |
0f20972f | 1625 | goto bad; |
1da177e4 | 1626 | |
316d315b NK |
1627 | atomic_set(&md->pending[0], 0); |
1628 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1629 | init_waitqueue_head(&md->wait); |
53d5914f | 1630 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1631 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1632 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1633 | md->kworker_task = NULL; |
f0b04115 | 1634 | |
1da177e4 LT |
1635 | md->disk->major = _major; |
1636 | md->disk->first_minor = minor; | |
1637 | md->disk->fops = &dm_blk_dops; | |
1638 | md->disk->queue = md->queue; | |
1639 | md->disk->private_data = md; | |
1640 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 DW |
1641 | |
1642 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); | |
1643 | if (!dax_dev) | |
1644 | goto bad; | |
1645 | md->dax_dev = dax_dev; | |
1646 | ||
1da177e4 | 1647 | add_disk(md->disk); |
7e51f257 | 1648 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1649 | |
670368a8 | 1650 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1651 | if (!md->wq) |
0f20972f | 1652 | goto bad; |
304f3f6a | 1653 | |
32a926da MP |
1654 | md->bdev = bdget_disk(md->disk, 0); |
1655 | if (!md->bdev) | |
0f20972f | 1656 | goto bad; |
32a926da | 1657 | |
3a83f467 | 1658 | bio_init(&md->flush_bio, NULL, 0); |
6a8736d1 | 1659 | md->flush_bio.bi_bdev = md->bdev; |
70fd7614 | 1660 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
6a8736d1 | 1661 | |
fd2ed4d2 MP |
1662 | dm_stats_init(&md->stats); |
1663 | ||
ba61fdd1 | 1664 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1665 | spin_lock(&_minor_lock); |
ba61fdd1 | 1666 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1667 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1668 | |
1669 | BUG_ON(old_md != MINOR_ALLOCED); | |
1670 | ||
1da177e4 LT |
1671 | return md; |
1672 | ||
0f20972f MS |
1673 | bad: |
1674 | cleanup_mapped_device(md); | |
83d5e5b0 | 1675 | bad_io_barrier: |
1da177e4 | 1676 | free_minor(minor); |
6ed7ade8 | 1677 | bad_minor: |
10da4f79 | 1678 | module_put(THIS_MODULE); |
6ed7ade8 | 1679 | bad_module_get: |
1da177e4 LT |
1680 | kfree(md); |
1681 | return NULL; | |
1682 | } | |
1683 | ||
ae9da83f JN |
1684 | static void unlock_fs(struct mapped_device *md); |
1685 | ||
1da177e4 LT |
1686 | static void free_dev(struct mapped_device *md) |
1687 | { | |
f331c029 | 1688 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1689 | |
32a926da | 1690 | unlock_fs(md); |
2eb6e1e3 | 1691 | |
0f20972f | 1692 | cleanup_mapped_device(md); |
63a4f065 | 1693 | |
86f1152b | 1694 | free_table_devices(&md->table_devices); |
63a4f065 | 1695 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1696 | free_minor(minor); |
1697 | ||
10da4f79 | 1698 | module_put(THIS_MODULE); |
1da177e4 LT |
1699 | kfree(md); |
1700 | } | |
1701 | ||
e6ee8c0b KU |
1702 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1703 | { | |
c0820cf5 | 1704 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1705 | |
4e6e36c3 MS |
1706 | if (md->bs) { |
1707 | /* The md already has necessary mempools. */ | |
545ed20e | 1708 | if (dm_table_bio_based(t)) { |
16245bdc JN |
1709 | /* |
1710 | * Reload bioset because front_pad may have changed | |
1711 | * because a different table was loaded. | |
1712 | */ | |
1713 | bioset_free(md->bs); | |
1714 | md->bs = p->bs; | |
1715 | p->bs = NULL; | |
16245bdc | 1716 | } |
4e6e36c3 MS |
1717 | /* |
1718 | * There's no need to reload with request-based dm | |
1719 | * because the size of front_pad doesn't change. | |
1720 | * Note for future: If you are to reload bioset, | |
1721 | * prep-ed requests in the queue may refer | |
1722 | * to bio from the old bioset, so you must walk | |
1723 | * through the queue to unprep. | |
1724 | */ | |
1725 | goto out; | |
c0820cf5 | 1726 | } |
e6ee8c0b | 1727 | |
eb8db831 | 1728 | BUG_ON(!p || md->io_pool || md->bs); |
cbc4e3c1 | 1729 | |
e6ee8c0b KU |
1730 | md->io_pool = p->io_pool; |
1731 | p->io_pool = NULL; | |
e6ee8c0b KU |
1732 | md->bs = p->bs; |
1733 | p->bs = NULL; | |
4e6e36c3 | 1734 | |
e6ee8c0b | 1735 | out: |
02233342 | 1736 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1737 | dm_table_free_md_mempools(t); |
1738 | } | |
1739 | ||
1da177e4 LT |
1740 | /* |
1741 | * Bind a table to the device. | |
1742 | */ | |
1743 | static void event_callback(void *context) | |
1744 | { | |
7a8c3d3b MA |
1745 | unsigned long flags; |
1746 | LIST_HEAD(uevents); | |
1da177e4 LT |
1747 | struct mapped_device *md = (struct mapped_device *) context; |
1748 | ||
7a8c3d3b MA |
1749 | spin_lock_irqsave(&md->uevent_lock, flags); |
1750 | list_splice_init(&md->uevent_list, &uevents); | |
1751 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1752 | ||
ed9e1982 | 1753 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1754 | |
1da177e4 LT |
1755 | atomic_inc(&md->event_nr); |
1756 | wake_up(&md->eventq); | |
1757 | } | |
1758 | ||
c217649b MS |
1759 | /* |
1760 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1761 | */ | |
4e90188b | 1762 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1763 | { |
1ea0654e BVA |
1764 | lockdep_assert_held(&md->suspend_lock); |
1765 | ||
4e90188b | 1766 | set_capacity(md->disk, size); |
1da177e4 | 1767 | |
db8fef4f | 1768 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
1769 | } |
1770 | ||
042d2a9b AK |
1771 | /* |
1772 | * Returns old map, which caller must destroy. | |
1773 | */ | |
1774 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
1775 | struct queue_limits *limits) | |
1da177e4 | 1776 | { |
042d2a9b | 1777 | struct dm_table *old_map; |
165125e1 | 1778 | struct request_queue *q = md->queue; |
1da177e4 LT |
1779 | sector_t size; |
1780 | ||
5a8f1f80 BVA |
1781 | lockdep_assert_held(&md->suspend_lock); |
1782 | ||
1da177e4 | 1783 | size = dm_table_get_size(t); |
3ac51e74 DW |
1784 | |
1785 | /* | |
1786 | * Wipe any geometry if the size of the table changed. | |
1787 | */ | |
fd2ed4d2 | 1788 | if (size != dm_get_size(md)) |
3ac51e74 DW |
1789 | memset(&md->geometry, 0, sizeof(md->geometry)); |
1790 | ||
32a926da | 1791 | __set_size(md, size); |
d5816876 | 1792 | |
2ca3310e AK |
1793 | dm_table_event_callback(t, event_callback, md); |
1794 | ||
e6ee8c0b KU |
1795 | /* |
1796 | * The queue hasn't been stopped yet, if the old table type wasn't | |
1797 | * for request-based during suspension. So stop it to prevent | |
1798 | * I/O mapping before resume. | |
1799 | * This must be done before setting the queue restrictions, | |
1800 | * because request-based dm may be run just after the setting. | |
1801 | */ | |
16f12266 | 1802 | if (dm_table_request_based(t)) { |
eca7ee6d | 1803 | dm_stop_queue(q); |
16f12266 MS |
1804 | /* |
1805 | * Leverage the fact that request-based DM targets are | |
1806 | * immutable singletons and establish md->immutable_target | |
1807 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
1808 | */ | |
1809 | md->immutable_target = dm_table_get_immutable_target(t); | |
1810 | } | |
e6ee8c0b KU |
1811 | |
1812 | __bind_mempools(md, t); | |
1813 | ||
a12f5d48 | 1814 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 1815 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
1816 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
1817 | ||
754c5fc7 | 1818 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
1819 | if (old_map) |
1820 | dm_sync_table(md); | |
1da177e4 | 1821 | |
042d2a9b | 1822 | return old_map; |
1da177e4 LT |
1823 | } |
1824 | ||
a7940155 AK |
1825 | /* |
1826 | * Returns unbound table for the caller to free. | |
1827 | */ | |
1828 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 1829 | { |
a12f5d48 | 1830 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
1831 | |
1832 | if (!map) | |
a7940155 | 1833 | return NULL; |
1da177e4 LT |
1834 | |
1835 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 1836 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 1837 | dm_sync_table(md); |
a7940155 AK |
1838 | |
1839 | return map; | |
1da177e4 LT |
1840 | } |
1841 | ||
1842 | /* | |
1843 | * Constructor for a new device. | |
1844 | */ | |
2b06cfff | 1845 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1846 | { |
1847 | struct mapped_device *md; | |
1848 | ||
2b06cfff | 1849 | md = alloc_dev(minor); |
1da177e4 LT |
1850 | if (!md) |
1851 | return -ENXIO; | |
1852 | ||
784aae73 MB |
1853 | dm_sysfs_init(md); |
1854 | ||
1da177e4 LT |
1855 | *result = md; |
1856 | return 0; | |
1857 | } | |
1858 | ||
a5664dad MS |
1859 | /* |
1860 | * Functions to manage md->type. | |
1861 | * All are required to hold md->type_lock. | |
1862 | */ | |
1863 | void dm_lock_md_type(struct mapped_device *md) | |
1864 | { | |
1865 | mutex_lock(&md->type_lock); | |
1866 | } | |
1867 | ||
1868 | void dm_unlock_md_type(struct mapped_device *md) | |
1869 | { | |
1870 | mutex_unlock(&md->type_lock); | |
1871 | } | |
1872 | ||
7e0d574f | 1873 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 1874 | { |
00c4fc3b | 1875 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
1876 | md->type = type; |
1877 | } | |
1878 | ||
7e0d574f | 1879 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
1880 | { |
1881 | return md->type; | |
1882 | } | |
1883 | ||
36a0456f AK |
1884 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
1885 | { | |
1886 | return md->immutable_target_type; | |
1887 | } | |
1888 | ||
f84cb8a4 MS |
1889 | /* |
1890 | * The queue_limits are only valid as long as you have a reference | |
1891 | * count on 'md'. | |
1892 | */ | |
1893 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
1894 | { | |
1895 | BUG_ON(!atomic_read(&md->holders)); | |
1896 | return &md->queue->limits; | |
1897 | } | |
1898 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
1899 | ||
4a0b4ddf MS |
1900 | /* |
1901 | * Setup the DM device's queue based on md's type | |
1902 | */ | |
591ddcfc | 1903 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 1904 | { |
bfebd1cd | 1905 | int r; |
7e0d574f | 1906 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 1907 | |
545ed20e | 1908 | switch (type) { |
bfebd1cd | 1909 | case DM_TYPE_REQUEST_BASED: |
eb8db831 | 1910 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 1911 | if (r) { |
eca7ee6d | 1912 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 1913 | return r; |
ff36ab34 | 1914 | } |
bfebd1cd MS |
1915 | break; |
1916 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 1917 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 1918 | if (r) { |
eca7ee6d | 1919 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
1920 | return r; |
1921 | } | |
1922 | break; | |
1923 | case DM_TYPE_BIO_BASED: | |
545ed20e | 1924 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 1925 | dm_init_normal_md_queue(md); |
ff36ab34 | 1926 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
1927 | /* |
1928 | * DM handles splitting bios as needed. Free the bio_split bioset | |
1929 | * since it won't be used (saves 1 process per bio-based DM device). | |
1930 | */ | |
1931 | bioset_free(md->queue->bio_split); | |
1932 | md->queue->bio_split = NULL; | |
545ed20e TK |
1933 | |
1934 | if (type == DM_TYPE_DAX_BIO_BASED) | |
1935 | queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue); | |
bfebd1cd | 1936 | break; |
7e0d574f BVA |
1937 | case DM_TYPE_NONE: |
1938 | WARN_ON_ONCE(true); | |
1939 | break; | |
4a0b4ddf MS |
1940 | } |
1941 | ||
1942 | return 0; | |
1943 | } | |
1944 | ||
2bec1f4a | 1945 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
1946 | { |
1947 | struct mapped_device *md; | |
1da177e4 LT |
1948 | unsigned minor = MINOR(dev); |
1949 | ||
1950 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1951 | return NULL; | |
1952 | ||
f32c10b0 | 1953 | spin_lock(&_minor_lock); |
1da177e4 LT |
1954 | |
1955 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
1956 | if (md) { |
1957 | if ((md == MINOR_ALLOCED || | |
1958 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
1959 | dm_deleting_md(md) || | |
1960 | test_bit(DMF_FREEING, &md->flags))) { | |
1961 | md = NULL; | |
1962 | goto out; | |
1963 | } | |
1964 | dm_get(md); | |
fba9f90e | 1965 | } |
1da177e4 | 1966 | |
fba9f90e | 1967 | out: |
f32c10b0 | 1968 | spin_unlock(&_minor_lock); |
1da177e4 | 1969 | |
637842cf DT |
1970 | return md; |
1971 | } | |
3cf2e4ba | 1972 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 1973 | |
9ade92a9 | 1974 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1975 | { |
9ade92a9 | 1976 | return md->interface_ptr; |
1da177e4 LT |
1977 | } |
1978 | ||
1979 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1980 | { | |
1981 | md->interface_ptr = ptr; | |
1982 | } | |
1983 | ||
1984 | void dm_get(struct mapped_device *md) | |
1985 | { | |
1986 | atomic_inc(&md->holders); | |
3f77316d | 1987 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
1988 | } |
1989 | ||
09ee96b2 MP |
1990 | int dm_hold(struct mapped_device *md) |
1991 | { | |
1992 | spin_lock(&_minor_lock); | |
1993 | if (test_bit(DMF_FREEING, &md->flags)) { | |
1994 | spin_unlock(&_minor_lock); | |
1995 | return -EBUSY; | |
1996 | } | |
1997 | dm_get(md); | |
1998 | spin_unlock(&_minor_lock); | |
1999 | return 0; | |
2000 | } | |
2001 | EXPORT_SYMBOL_GPL(dm_hold); | |
2002 | ||
72d94861 AK |
2003 | const char *dm_device_name(struct mapped_device *md) |
2004 | { | |
2005 | return md->name; | |
2006 | } | |
2007 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2008 | ||
3f77316d | 2009 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2010 | { |
3b785fbc | 2011 | struct request_queue *q = dm_get_md_queue(md); |
1134e5ae | 2012 | struct dm_table *map; |
83d5e5b0 | 2013 | int srcu_idx; |
1da177e4 | 2014 | |
3f77316d | 2015 | might_sleep(); |
fba9f90e | 2016 | |
63a4f065 | 2017 | spin_lock(&_minor_lock); |
3f77316d KU |
2018 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2019 | set_bit(DMF_FREEING, &md->flags); | |
2020 | spin_unlock(&_minor_lock); | |
3b785fbc | 2021 | |
2e91c369 | 2022 | blk_set_queue_dying(q); |
3f77316d | 2023 | |
02233342 | 2024 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 2025 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2026 | |
ab7c7bb6 MP |
2027 | /* |
2028 | * Take suspend_lock so that presuspend and postsuspend methods | |
2029 | * do not race with internal suspend. | |
2030 | */ | |
2031 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2032 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2033 | if (!dm_suspended_md(md)) { |
2034 | dm_table_presuspend_targets(map); | |
2035 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2036 | } |
83d5e5b0 MP |
2037 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2038 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2039 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2040 | |
3f77316d KU |
2041 | /* |
2042 | * Rare, but there may be I/O requests still going to complete, | |
2043 | * for example. Wait for all references to disappear. | |
2044 | * No one should increment the reference count of the mapped_device, | |
2045 | * after the mapped_device state becomes DMF_FREEING. | |
2046 | */ | |
2047 | if (wait) | |
2048 | while (atomic_read(&md->holders)) | |
2049 | msleep(1); | |
2050 | else if (atomic_read(&md->holders)) | |
2051 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2052 | dm_device_name(md), atomic_read(&md->holders)); | |
2053 | ||
2054 | dm_sysfs_exit(md); | |
3f77316d KU |
2055 | dm_table_destroy(__unbind(md)); |
2056 | free_dev(md); | |
2057 | } | |
2058 | ||
2059 | void dm_destroy(struct mapped_device *md) | |
2060 | { | |
2061 | __dm_destroy(md, true); | |
2062 | } | |
2063 | ||
2064 | void dm_destroy_immediate(struct mapped_device *md) | |
2065 | { | |
2066 | __dm_destroy(md, false); | |
2067 | } | |
2068 | ||
2069 | void dm_put(struct mapped_device *md) | |
2070 | { | |
2071 | atomic_dec(&md->holders); | |
1da177e4 | 2072 | } |
79eb885c | 2073 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2074 | |
b48633f8 | 2075 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2076 | { |
2077 | int r = 0; | |
9f4c3f87 | 2078 | DEFINE_WAIT(wait); |
46125c1c MB |
2079 | |
2080 | while (1) { | |
9f4c3f87 | 2081 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2082 | |
b4324fee | 2083 | if (!md_in_flight(md)) |
46125c1c MB |
2084 | break; |
2085 | ||
e3fabdfd | 2086 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2087 | r = -EINTR; |
2088 | break; | |
2089 | } | |
2090 | ||
2091 | io_schedule(); | |
2092 | } | |
9f4c3f87 | 2093 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2094 | |
46125c1c MB |
2095 | return r; |
2096 | } | |
2097 | ||
1da177e4 LT |
2098 | /* |
2099 | * Process the deferred bios | |
2100 | */ | |
ef208587 | 2101 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2102 | { |
ef208587 MP |
2103 | struct mapped_device *md = container_of(work, struct mapped_device, |
2104 | work); | |
6d6f10df | 2105 | struct bio *c; |
83d5e5b0 MP |
2106 | int srcu_idx; |
2107 | struct dm_table *map; | |
1da177e4 | 2108 | |
83d5e5b0 | 2109 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2110 | |
3b00b203 | 2111 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2112 | spin_lock_irq(&md->deferred_lock); |
2113 | c = bio_list_pop(&md->deferred); | |
2114 | spin_unlock_irq(&md->deferred_lock); | |
2115 | ||
6a8736d1 | 2116 | if (!c) |
df12ee99 | 2117 | break; |
022c2611 | 2118 | |
e6ee8c0b KU |
2119 | if (dm_request_based(md)) |
2120 | generic_make_request(c); | |
6a8736d1 | 2121 | else |
83d5e5b0 | 2122 | __split_and_process_bio(md, map, c); |
022c2611 | 2123 | } |
73d410c0 | 2124 | |
83d5e5b0 | 2125 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2126 | } |
2127 | ||
9a1fb464 | 2128 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2129 | { |
3b00b203 | 2130 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2131 | smp_mb__after_atomic(); |
53d5914f | 2132 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2133 | } |
2134 | ||
1da177e4 | 2135 | /* |
042d2a9b | 2136 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2137 | */ |
042d2a9b | 2138 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2139 | { |
87eb5b21 | 2140 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2141 | struct queue_limits limits; |
042d2a9b | 2142 | int r; |
1da177e4 | 2143 | |
e61290a4 | 2144 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2145 | |
2146 | /* device must be suspended */ | |
4f186f8b | 2147 | if (!dm_suspended_md(md)) |
93c534ae | 2148 | goto out; |
1da177e4 | 2149 | |
3ae70656 MS |
2150 | /* |
2151 | * If the new table has no data devices, retain the existing limits. | |
2152 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2153 | * then new I/O is queued based on these limits, and then some paths | |
2154 | * reappear. | |
2155 | */ | |
2156 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2157 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2158 | if (live_map) |
2159 | limits = md->queue->limits; | |
83d5e5b0 | 2160 | dm_put_live_table_fast(md); |
3ae70656 MS |
2161 | } |
2162 | ||
87eb5b21 MC |
2163 | if (!live_map) { |
2164 | r = dm_calculate_queue_limits(table, &limits); | |
2165 | if (r) { | |
2166 | map = ERR_PTR(r); | |
2167 | goto out; | |
2168 | } | |
042d2a9b | 2169 | } |
754c5fc7 | 2170 | |
042d2a9b | 2171 | map = __bind(md, table, &limits); |
1da177e4 | 2172 | |
93c534ae | 2173 | out: |
e61290a4 | 2174 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2175 | return map; |
1da177e4 LT |
2176 | } |
2177 | ||
2178 | /* | |
2179 | * Functions to lock and unlock any filesystem running on the | |
2180 | * device. | |
2181 | */ | |
2ca3310e | 2182 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2183 | { |
e39e2e95 | 2184 | int r; |
1da177e4 LT |
2185 | |
2186 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2187 | |
db8fef4f | 2188 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2189 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2190 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2191 | md->frozen_sb = NULL; |
2192 | return r; | |
dfbe03f6 AK |
2193 | } |
2194 | ||
aa8d7c2f AK |
2195 | set_bit(DMF_FROZEN, &md->flags); |
2196 | ||
1da177e4 LT |
2197 | return 0; |
2198 | } | |
2199 | ||
2ca3310e | 2200 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2201 | { |
aa8d7c2f AK |
2202 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2203 | return; | |
2204 | ||
db8fef4f | 2205 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2206 | md->frozen_sb = NULL; |
aa8d7c2f | 2207 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2208 | } |
2209 | ||
2210 | /* | |
b48633f8 BVA |
2211 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2212 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2213 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2214 | * | |
ffcc3936 MS |
2215 | * If __dm_suspend returns 0, the device is completely quiescent |
2216 | * now. There is no request-processing activity. All new requests | |
2217 | * are being added to md->deferred list. | |
cec47e3d | 2218 | */ |
ffcc3936 | 2219 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2220 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2221 | int dmf_suspended_flag) |
1da177e4 | 2222 | { |
ffcc3936 MS |
2223 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2224 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2225 | int r; | |
1da177e4 | 2226 | |
5a8f1f80 BVA |
2227 | lockdep_assert_held(&md->suspend_lock); |
2228 | ||
2e93ccc1 KU |
2229 | /* |
2230 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2231 | * This flag is cleared before dm_suspend returns. | |
2232 | */ | |
2233 | if (noflush) | |
2234 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 BVA |
2235 | else |
2236 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); | |
2e93ccc1 | 2237 | |
d67ee213 MS |
2238 | /* |
2239 | * This gets reverted if there's an error later and the targets | |
2240 | * provide the .presuspend_undo hook. | |
2241 | */ | |
cf222b37 AK |
2242 | dm_table_presuspend_targets(map); |
2243 | ||
32a926da | 2244 | /* |
9f518b27 KU |
2245 | * Flush I/O to the device. |
2246 | * Any I/O submitted after lock_fs() may not be flushed. | |
2247 | * noflush takes precedence over do_lockfs. | |
2248 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2249 | */ |
2250 | if (!noflush && do_lockfs) { | |
2251 | r = lock_fs(md); | |
d67ee213 MS |
2252 | if (r) { |
2253 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2254 | return r; |
d67ee213 | 2255 | } |
aa8d7c2f | 2256 | } |
1da177e4 LT |
2257 | |
2258 | /* | |
3b00b203 MP |
2259 | * Here we must make sure that no processes are submitting requests |
2260 | * to target drivers i.e. no one may be executing | |
2261 | * __split_and_process_bio. This is called from dm_request and | |
2262 | * dm_wq_work. | |
2263 | * | |
2264 | * To get all processes out of __split_and_process_bio in dm_request, | |
2265 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2266 | * __split_and_process_bio from dm_request and quiesce the thread |
2267 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2268 | * flush_workqueue(md->wq). | |
1da177e4 | 2269 | */ |
1eb787ec | 2270 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2271 | if (map) |
2272 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2273 | |
d0bcb878 | 2274 | /* |
29e4013d TH |
2275 | * Stop md->queue before flushing md->wq in case request-based |
2276 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2277 | */ |
2eb6e1e3 | 2278 | if (dm_request_based(md)) { |
eca7ee6d | 2279 | dm_stop_queue(md->queue); |
02233342 | 2280 | if (md->kworker_task) |
3989144f | 2281 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2282 | } |
cec47e3d | 2283 | |
d0bcb878 KU |
2284 | flush_workqueue(md->wq); |
2285 | ||
1da177e4 | 2286 | /* |
3b00b203 MP |
2287 | * At this point no more requests are entering target request routines. |
2288 | * We call dm_wait_for_completion to wait for all existing requests | |
2289 | * to finish. | |
1da177e4 | 2290 | */ |
b48633f8 | 2291 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2292 | if (!r) |
2293 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2294 | |
6d6f10df | 2295 | if (noflush) |
022c2611 | 2296 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2297 | if (map) |
2298 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2299 | |
1da177e4 | 2300 | /* were we interrupted ? */ |
46125c1c | 2301 | if (r < 0) { |
9a1fb464 | 2302 | dm_queue_flush(md); |
73d410c0 | 2303 | |
cec47e3d | 2304 | if (dm_request_based(md)) |
eca7ee6d | 2305 | dm_start_queue(md->queue); |
cec47e3d | 2306 | |
2ca3310e | 2307 | unlock_fs(md); |
d67ee213 | 2308 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2309 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2310 | } |
1da177e4 | 2311 | |
ffcc3936 MS |
2312 | return r; |
2313 | } | |
2314 | ||
2315 | /* | |
2316 | * We need to be able to change a mapping table under a mounted | |
2317 | * filesystem. For example we might want to move some data in | |
2318 | * the background. Before the table can be swapped with | |
2319 | * dm_bind_table, dm_suspend must be called to flush any in | |
2320 | * flight bios and ensure that any further io gets deferred. | |
2321 | */ | |
2322 | /* | |
2323 | * Suspend mechanism in request-based dm. | |
2324 | * | |
2325 | * 1. Flush all I/Os by lock_fs() if needed. | |
2326 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2327 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2328 | * | |
2329 | * To abort suspend, start the request_queue. | |
2330 | */ | |
2331 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2332 | { | |
2333 | struct dm_table *map = NULL; | |
2334 | int r = 0; | |
2335 | ||
2336 | retry: | |
2337 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2338 | ||
2339 | if (dm_suspended_md(md)) { | |
2340 | r = -EINVAL; | |
2341 | goto out_unlock; | |
2342 | } | |
2343 | ||
2344 | if (dm_suspended_internally_md(md)) { | |
2345 | /* already internally suspended, wait for internal resume */ | |
2346 | mutex_unlock(&md->suspend_lock); | |
2347 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2348 | if (r) | |
2349 | return r; | |
2350 | goto retry; | |
2351 | } | |
2352 | ||
a12f5d48 | 2353 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2354 | |
eaf9a736 | 2355 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2356 | if (r) |
2357 | goto out_unlock; | |
3b00b203 | 2358 | |
4d4471cb KU |
2359 | dm_table_postsuspend_targets(map); |
2360 | ||
d287483d | 2361 | out_unlock: |
e61290a4 | 2362 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2363 | return r; |
1da177e4 LT |
2364 | } |
2365 | ||
ffcc3936 MS |
2366 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2367 | { | |
2368 | if (map) { | |
2369 | int r = dm_table_resume_targets(map); | |
2370 | if (r) | |
2371 | return r; | |
2372 | } | |
2373 | ||
2374 | dm_queue_flush(md); | |
2375 | ||
2376 | /* | |
2377 | * Flushing deferred I/Os must be done after targets are resumed | |
2378 | * so that mapping of targets can work correctly. | |
2379 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2380 | */ | |
2381 | if (dm_request_based(md)) | |
eca7ee6d | 2382 | dm_start_queue(md->queue); |
ffcc3936 MS |
2383 | |
2384 | unlock_fs(md); | |
2385 | ||
2386 | return 0; | |
2387 | } | |
2388 | ||
1da177e4 LT |
2389 | int dm_resume(struct mapped_device *md) |
2390 | { | |
8dc23658 | 2391 | int r; |
cf222b37 | 2392 | struct dm_table *map = NULL; |
1da177e4 | 2393 | |
ffcc3936 | 2394 | retry: |
8dc23658 | 2395 | r = -EINVAL; |
ffcc3936 MS |
2396 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2397 | ||
4f186f8b | 2398 | if (!dm_suspended_md(md)) |
cf222b37 | 2399 | goto out; |
cf222b37 | 2400 | |
ffcc3936 MS |
2401 | if (dm_suspended_internally_md(md)) { |
2402 | /* already internally suspended, wait for internal resume */ | |
2403 | mutex_unlock(&md->suspend_lock); | |
2404 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2405 | if (r) | |
2406 | return r; | |
2407 | goto retry; | |
2408 | } | |
2409 | ||
a12f5d48 | 2410 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2411 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2412 | goto out; |
1da177e4 | 2413 | |
ffcc3936 | 2414 | r = __dm_resume(md, map); |
8757b776 MB |
2415 | if (r) |
2416 | goto out; | |
2ca3310e | 2417 | |
2ca3310e | 2418 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2419 | out: |
e61290a4 | 2420 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2421 | |
cf222b37 | 2422 | return r; |
1da177e4 LT |
2423 | } |
2424 | ||
fd2ed4d2 MP |
2425 | /* |
2426 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2427 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2428 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2429 | */ |
2430 | ||
ffcc3936 | 2431 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2432 | { |
ffcc3936 MS |
2433 | struct dm_table *map = NULL; |
2434 | ||
1ea0654e BVA |
2435 | lockdep_assert_held(&md->suspend_lock); |
2436 | ||
96b26c8c | 2437 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2438 | return; /* nested internal suspend */ |
2439 | ||
2440 | if (dm_suspended_md(md)) { | |
2441 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2442 | return; /* nest suspend */ | |
2443 | } | |
2444 | ||
a12f5d48 | 2445 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2446 | |
2447 | /* | |
2448 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2449 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2450 | * would require changing .presuspend to return an error -- avoid this | |
2451 | * until there is a need for more elaborate variants of internal suspend. | |
2452 | */ | |
eaf9a736 MS |
2453 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2454 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2455 | |
2456 | dm_table_postsuspend_targets(map); | |
2457 | } | |
2458 | ||
2459 | static void __dm_internal_resume(struct mapped_device *md) | |
2460 | { | |
96b26c8c MP |
2461 | BUG_ON(!md->internal_suspend_count); |
2462 | ||
2463 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2464 | return; /* resume from nested internal suspend */ |
2465 | ||
fd2ed4d2 | 2466 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2467 | goto done; /* resume from nested suspend */ |
2468 | ||
2469 | /* | |
2470 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2471 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2472 | */ | |
2473 | (void) __dm_resume(md, NULL); | |
2474 | ||
2475 | done: | |
2476 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2477 | smp_mb__after_atomic(); | |
2478 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2479 | } | |
2480 | ||
2481 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2482 | { | |
2483 | mutex_lock(&md->suspend_lock); | |
2484 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2485 | mutex_unlock(&md->suspend_lock); | |
2486 | } | |
2487 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2488 | ||
2489 | void dm_internal_resume(struct mapped_device *md) | |
2490 | { | |
2491 | mutex_lock(&md->suspend_lock); | |
2492 | __dm_internal_resume(md); | |
2493 | mutex_unlock(&md->suspend_lock); | |
2494 | } | |
2495 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2496 | ||
2497 | /* | |
2498 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2499 | * which prevents interaction with userspace-driven suspend. | |
2500 | */ | |
2501 | ||
2502 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2503 | { | |
2504 | mutex_lock(&md->suspend_lock); | |
2505 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2506 | return; |
2507 | ||
2508 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2509 | synchronize_srcu(&md->io_barrier); | |
2510 | flush_workqueue(md->wq); | |
2511 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2512 | } | |
b735fede | 2513 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2514 | |
ffcc3936 | 2515 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2516 | { |
ffcc3936 | 2517 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2518 | goto done; |
2519 | ||
2520 | dm_queue_flush(md); | |
2521 | ||
2522 | done: | |
2523 | mutex_unlock(&md->suspend_lock); | |
2524 | } | |
b735fede | 2525 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2526 | |
1da177e4 LT |
2527 | /*----------------------------------------------------------------- |
2528 | * Event notification. | |
2529 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2530 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2531 | unsigned cookie) |
69267a30 | 2532 | { |
60935eb2 MB |
2533 | char udev_cookie[DM_COOKIE_LENGTH]; |
2534 | char *envp[] = { udev_cookie, NULL }; | |
2535 | ||
2536 | if (!cookie) | |
3abf85b5 | 2537 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2538 | else { |
2539 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2540 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2541 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2542 | action, envp); | |
60935eb2 | 2543 | } |
69267a30 AK |
2544 | } |
2545 | ||
7a8c3d3b MA |
2546 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2547 | { | |
2548 | return atomic_add_return(1, &md->uevent_seq); | |
2549 | } | |
2550 | ||
1da177e4 LT |
2551 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2552 | { | |
2553 | return atomic_read(&md->event_nr); | |
2554 | } | |
2555 | ||
2556 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2557 | { | |
2558 | return wait_event_interruptible(md->eventq, | |
2559 | (event_nr != atomic_read(&md->event_nr))); | |
2560 | } | |
2561 | ||
7a8c3d3b MA |
2562 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2563 | { | |
2564 | unsigned long flags; | |
2565 | ||
2566 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2567 | list_add(elist, &md->uevent_list); | |
2568 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2569 | } | |
2570 | ||
1da177e4 LT |
2571 | /* |
2572 | * The gendisk is only valid as long as you have a reference | |
2573 | * count on 'md'. | |
2574 | */ | |
2575 | struct gendisk *dm_disk(struct mapped_device *md) | |
2576 | { | |
2577 | return md->disk; | |
2578 | } | |
65ff5b7d | 2579 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2580 | |
784aae73 MB |
2581 | struct kobject *dm_kobject(struct mapped_device *md) |
2582 | { | |
2995fa78 | 2583 | return &md->kobj_holder.kobj; |
784aae73 MB |
2584 | } |
2585 | ||
784aae73 MB |
2586 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2587 | { | |
2588 | struct mapped_device *md; | |
2589 | ||
2995fa78 | 2590 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2591 | |
4d89b7b4 | 2592 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2593 | dm_deleting_md(md)) |
4d89b7b4 MB |
2594 | return NULL; |
2595 | ||
784aae73 MB |
2596 | dm_get(md); |
2597 | return md; | |
2598 | } | |
2599 | ||
4f186f8b | 2600 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2601 | { |
2602 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2603 | } | |
2604 | ||
ffcc3936 MS |
2605 | int dm_suspended_internally_md(struct mapped_device *md) |
2606 | { | |
2607 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2608 | } | |
2609 | ||
2c140a24 MP |
2610 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2611 | { | |
2612 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2613 | } | |
2614 | ||
64dbce58 KU |
2615 | int dm_suspended(struct dm_target *ti) |
2616 | { | |
ecdb2e25 | 2617 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2618 | } |
2619 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2620 | ||
2e93ccc1 KU |
2621 | int dm_noflush_suspending(struct dm_target *ti) |
2622 | { | |
ecdb2e25 | 2623 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2624 | } |
2625 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2626 | ||
7e0d574f | 2627 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
30187e1d | 2628 | unsigned integrity, unsigned per_io_data_size) |
e6ee8c0b | 2629 | { |
115485e8 | 2630 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2631 | unsigned int pool_size = 0; |
5f015204 | 2632 | unsigned int front_pad; |
e6ee8c0b KU |
2633 | |
2634 | if (!pools) | |
4e6e36c3 | 2635 | return NULL; |
e6ee8c0b | 2636 | |
78d8e58a MS |
2637 | switch (type) { |
2638 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2639 | case DM_TYPE_DAX_BIO_BASED: |
78d8e58a | 2640 | pool_size = dm_get_reserved_bio_based_ios(); |
30187e1d | 2641 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
eb8db831 CH |
2642 | |
2643 | pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); | |
2644 | if (!pools->io_pool) | |
2645 | goto out; | |
78d8e58a MS |
2646 | break; |
2647 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2648 | case DM_TYPE_MQ_REQUEST_BASED: |
eb8db831 | 2649 | pool_size = dm_get_reserved_rq_based_ios(); |
78d8e58a | 2650 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2651 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2652 | break; |
2653 | default: | |
2654 | BUG(); | |
2655 | } | |
2656 | ||
3d8aab2d | 2657 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 2658 | if (!pools->bs) |
5f015204 | 2659 | goto out; |
e6ee8c0b | 2660 | |
a91a2785 | 2661 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2662 | goto out; |
a91a2785 | 2663 | |
e6ee8c0b | 2664 | return pools; |
5f1b670d | 2665 | |
5f1b670d CH |
2666 | out: |
2667 | dm_free_md_mempools(pools); | |
78d8e58a | 2668 | |
4e6e36c3 | 2669 | return NULL; |
e6ee8c0b KU |
2670 | } |
2671 | ||
2672 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2673 | { | |
2674 | if (!pools) | |
2675 | return; | |
2676 | ||
6f65985e | 2677 | mempool_destroy(pools->io_pool); |
1ae49ea2 | 2678 | |
e6ee8c0b KU |
2679 | if (pools->bs) |
2680 | bioset_free(pools->bs); | |
2681 | ||
2682 | kfree(pools); | |
2683 | } | |
2684 | ||
9c72bad1 CH |
2685 | struct dm_pr { |
2686 | u64 old_key; | |
2687 | u64 new_key; | |
2688 | u32 flags; | |
2689 | bool fail_early; | |
2690 | }; | |
2691 | ||
2692 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2693 | void *data) | |
71cdb697 CH |
2694 | { |
2695 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2696 | struct dm_table *table; |
2697 | struct dm_target *ti; | |
2698 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2699 | |
9c72bad1 CH |
2700 | table = dm_get_live_table(md, &srcu_idx); |
2701 | if (!table || !dm_table_get_size(table)) | |
2702 | goto out; | |
71cdb697 | 2703 | |
9c72bad1 CH |
2704 | /* We only support devices that have a single target */ |
2705 | if (dm_table_get_num_targets(table) != 1) | |
2706 | goto out; | |
2707 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2708 | |
9c72bad1 CH |
2709 | ret = -EINVAL; |
2710 | if (!ti->type->iterate_devices) | |
2711 | goto out; | |
2712 | ||
2713 | ret = ti->type->iterate_devices(ti, fn, data); | |
2714 | out: | |
2715 | dm_put_live_table(md, srcu_idx); | |
2716 | return ret; | |
2717 | } | |
2718 | ||
2719 | /* | |
2720 | * For register / unregister we need to manually call out to every path. | |
2721 | */ | |
2722 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
2723 | sector_t start, sector_t len, void *data) | |
2724 | { | |
2725 | struct dm_pr *pr = data; | |
2726 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
2727 | ||
2728 | if (!ops || !ops->pr_register) | |
2729 | return -EOPNOTSUPP; | |
2730 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
2731 | } | |
2732 | ||
2733 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
2734 | u32 flags) | |
2735 | { | |
2736 | struct dm_pr pr = { | |
2737 | .old_key = old_key, | |
2738 | .new_key = new_key, | |
2739 | .flags = flags, | |
2740 | .fail_early = true, | |
2741 | }; | |
2742 | int ret; | |
2743 | ||
2744 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
2745 | if (ret && new_key) { | |
2746 | /* unregister all paths if we failed to register any path */ | |
2747 | pr.old_key = new_key; | |
2748 | pr.new_key = 0; | |
2749 | pr.flags = 0; | |
2750 | pr.fail_early = false; | |
2751 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
2752 | } | |
2753 | ||
2754 | return ret; | |
71cdb697 CH |
2755 | } |
2756 | ||
2757 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 2758 | u32 flags) |
71cdb697 CH |
2759 | { |
2760 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2761 | const struct pr_ops *ops; | |
71cdb697 | 2762 | fmode_t mode; |
956a4025 | 2763 | int r; |
71cdb697 | 2764 | |
956a4025 | 2765 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2766 | if (r < 0) |
2767 | return r; | |
2768 | ||
2769 | ops = bdev->bd_disk->fops->pr_ops; | |
2770 | if (ops && ops->pr_reserve) | |
2771 | r = ops->pr_reserve(bdev, key, type, flags); | |
2772 | else | |
2773 | r = -EOPNOTSUPP; | |
2774 | ||
956a4025 | 2775 | bdput(bdev); |
71cdb697 CH |
2776 | return r; |
2777 | } | |
2778 | ||
2779 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
2780 | { | |
2781 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2782 | const struct pr_ops *ops; | |
71cdb697 | 2783 | fmode_t mode; |
956a4025 | 2784 | int r; |
71cdb697 | 2785 | |
956a4025 | 2786 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2787 | if (r < 0) |
2788 | return r; | |
2789 | ||
2790 | ops = bdev->bd_disk->fops->pr_ops; | |
2791 | if (ops && ops->pr_release) | |
2792 | r = ops->pr_release(bdev, key, type); | |
2793 | else | |
2794 | r = -EOPNOTSUPP; | |
2795 | ||
956a4025 | 2796 | bdput(bdev); |
71cdb697 CH |
2797 | return r; |
2798 | } | |
2799 | ||
2800 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 2801 | enum pr_type type, bool abort) |
71cdb697 CH |
2802 | { |
2803 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2804 | const struct pr_ops *ops; | |
71cdb697 | 2805 | fmode_t mode; |
956a4025 | 2806 | int r; |
71cdb697 | 2807 | |
956a4025 | 2808 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2809 | if (r < 0) |
2810 | return r; | |
2811 | ||
2812 | ops = bdev->bd_disk->fops->pr_ops; | |
2813 | if (ops && ops->pr_preempt) | |
2814 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
2815 | else | |
2816 | r = -EOPNOTSUPP; | |
2817 | ||
956a4025 | 2818 | bdput(bdev); |
71cdb697 CH |
2819 | return r; |
2820 | } | |
2821 | ||
2822 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
2823 | { | |
2824 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2825 | const struct pr_ops *ops; | |
71cdb697 | 2826 | fmode_t mode; |
956a4025 | 2827 | int r; |
71cdb697 | 2828 | |
956a4025 | 2829 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2830 | if (r < 0) |
2831 | return r; | |
2832 | ||
2833 | ops = bdev->bd_disk->fops->pr_ops; | |
2834 | if (ops && ops->pr_clear) | |
2835 | r = ops->pr_clear(bdev, key); | |
2836 | else | |
2837 | r = -EOPNOTSUPP; | |
2838 | ||
956a4025 | 2839 | bdput(bdev); |
71cdb697 CH |
2840 | return r; |
2841 | } | |
2842 | ||
2843 | static const struct pr_ops dm_pr_ops = { | |
2844 | .pr_register = dm_pr_register, | |
2845 | .pr_reserve = dm_pr_reserve, | |
2846 | .pr_release = dm_pr_release, | |
2847 | .pr_preempt = dm_pr_preempt, | |
2848 | .pr_clear = dm_pr_clear, | |
2849 | }; | |
2850 | ||
83d5cde4 | 2851 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2852 | .open = dm_blk_open, |
2853 | .release = dm_blk_close, | |
aa129a22 | 2854 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2855 | .getgeo = dm_blk_getgeo, |
71cdb697 | 2856 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
2857 | .owner = THIS_MODULE |
2858 | }; | |
2859 | ||
f26c5719 DW |
2860 | static const struct dax_operations dm_dax_ops = { |
2861 | .direct_access = dm_dax_direct_access, | |
2862 | }; | |
2863 | ||
1da177e4 LT |
2864 | /* |
2865 | * module hooks | |
2866 | */ | |
2867 | module_init(dm_init); | |
2868 | module_exit(dm_exit); | |
2869 | ||
2870 | module_param(major, uint, 0); | |
2871 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 2872 | |
e8603136 MS |
2873 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
2874 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
2875 | ||
115485e8 MS |
2876 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
2877 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
2878 | ||
1da177e4 LT |
2879 | MODULE_DESCRIPTION(DM_NAME " driver"); |
2880 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2881 | MODULE_LICENSE("GPL"); |