]>
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; | |
633 | return 0; | |
634 | } | |
635 | ||
636 | /* | |
637 | * Close a table device that we've been using. | |
638 | */ | |
639 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
640 | { | |
641 | if (!td->dm_dev.bdev) | |
642 | return; | |
643 | ||
644 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
645 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
646 | td->dm_dev.bdev = NULL; | |
647 | } | |
648 | ||
649 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
650 | fmode_t mode) { | |
651 | struct table_device *td; | |
652 | ||
653 | list_for_each_entry(td, l, list) | |
654 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
655 | return td; | |
656 | ||
657 | return NULL; | |
658 | } | |
659 | ||
660 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
661 | struct dm_dev **result) { | |
662 | int r; | |
663 | struct table_device *td; | |
664 | ||
665 | mutex_lock(&md->table_devices_lock); | |
666 | td = find_table_device(&md->table_devices, dev, mode); | |
667 | if (!td) { | |
115485e8 | 668 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
669 | if (!td) { |
670 | mutex_unlock(&md->table_devices_lock); | |
671 | return -ENOMEM; | |
672 | } | |
673 | ||
674 | td->dm_dev.mode = mode; | |
675 | td->dm_dev.bdev = NULL; | |
676 | ||
677 | if ((r = open_table_device(td, dev, md))) { | |
678 | mutex_unlock(&md->table_devices_lock); | |
679 | kfree(td); | |
680 | return r; | |
681 | } | |
682 | ||
683 | format_dev_t(td->dm_dev.name, dev); | |
684 | ||
685 | atomic_set(&td->count, 0); | |
686 | list_add(&td->list, &md->table_devices); | |
687 | } | |
688 | atomic_inc(&td->count); | |
689 | mutex_unlock(&md->table_devices_lock); | |
690 | ||
691 | *result = &td->dm_dev; | |
692 | return 0; | |
693 | } | |
694 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
695 | ||
696 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
697 | { | |
698 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
699 | ||
700 | mutex_lock(&md->table_devices_lock); | |
701 | if (atomic_dec_and_test(&td->count)) { | |
702 | close_table_device(td, md); | |
703 | list_del(&td->list); | |
704 | kfree(td); | |
705 | } | |
706 | mutex_unlock(&md->table_devices_lock); | |
707 | } | |
708 | EXPORT_SYMBOL(dm_put_table_device); | |
709 | ||
710 | static void free_table_devices(struct list_head *devices) | |
711 | { | |
712 | struct list_head *tmp, *next; | |
713 | ||
714 | list_for_each_safe(tmp, next, devices) { | |
715 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
716 | ||
717 | DMWARN("dm_destroy: %s still exists with %d references", | |
718 | td->dm_dev.name, atomic_read(&td->count)); | |
719 | kfree(td); | |
720 | } | |
721 | } | |
722 | ||
3ac51e74 DW |
723 | /* |
724 | * Get the geometry associated with a dm device | |
725 | */ | |
726 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
727 | { | |
728 | *geo = md->geometry; | |
729 | ||
730 | return 0; | |
731 | } | |
732 | ||
733 | /* | |
734 | * Set the geometry of a device. | |
735 | */ | |
736 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
737 | { | |
738 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
739 | ||
740 | if (geo->start > sz) { | |
741 | DMWARN("Start sector is beyond the geometry limits."); | |
742 | return -EINVAL; | |
743 | } | |
744 | ||
745 | md->geometry = *geo; | |
746 | ||
747 | return 0; | |
748 | } | |
749 | ||
1da177e4 LT |
750 | /*----------------------------------------------------------------- |
751 | * CRUD START: | |
752 | * A more elegant soln is in the works that uses the queue | |
753 | * merge fn, unfortunately there are a couple of changes to | |
754 | * the block layer that I want to make for this. So in the | |
755 | * interests of getting something for people to use I give | |
756 | * you this clearly demarcated crap. | |
757 | *---------------------------------------------------------------*/ | |
758 | ||
2e93ccc1 KU |
759 | static int __noflush_suspending(struct mapped_device *md) |
760 | { | |
761 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
762 | } | |
763 | ||
1da177e4 LT |
764 | /* |
765 | * Decrements the number of outstanding ios that a bio has been | |
766 | * cloned into, completing the original io if necc. | |
767 | */ | |
858119e1 | 768 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 769 | { |
2e93ccc1 | 770 | unsigned long flags; |
b35f8caa MB |
771 | int io_error; |
772 | struct bio *bio; | |
773 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
774 | |
775 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
776 | if (unlikely(error)) { |
777 | spin_lock_irqsave(&io->endio_lock, flags); | |
778 | if (!(io->error > 0 && __noflush_suspending(md))) | |
779 | io->error = error; | |
780 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
781 | } | |
1da177e4 LT |
782 | |
783 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
784 | if (io->error == DM_ENDIO_REQUEUE) { |
785 | /* | |
786 | * Target requested pushing back the I/O. | |
2e93ccc1 | 787 | */ |
022c2611 | 788 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
789 | if (__noflush_suspending(md)) |
790 | bio_list_add_head(&md->deferred, io->bio); | |
791 | else | |
2e93ccc1 KU |
792 | /* noflush suspend was interrupted. */ |
793 | io->error = -EIO; | |
022c2611 | 794 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
795 | } |
796 | ||
b35f8caa MB |
797 | io_error = io->error; |
798 | bio = io->bio; | |
6a8736d1 TH |
799 | end_io_acct(io); |
800 | free_io(md, io); | |
801 | ||
802 | if (io_error == DM_ENDIO_REQUEUE) | |
803 | return; | |
2e93ccc1 | 804 | |
1eff9d32 | 805 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 806 | /* |
6a8736d1 | 807 | * Preflush done for flush with data, reissue |
28a8f0d3 | 808 | * without REQ_PREFLUSH. |
af7e466a | 809 | */ |
1eff9d32 | 810 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 811 | queue_io(md, bio); |
af7e466a | 812 | } else { |
b372d360 | 813 | /* done with normal IO or empty flush */ |
0a82a8d1 | 814 | trace_block_bio_complete(md->queue, bio, io_error); |
4246a0b6 CH |
815 | bio->bi_error = io_error; |
816 | bio_endio(bio); | |
b35f8caa | 817 | } |
1da177e4 LT |
818 | } |
819 | } | |
820 | ||
4cc96131 | 821 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
822 | { |
823 | struct queue_limits *limits = dm_get_queue_limits(md); | |
824 | ||
825 | /* device doesn't really support WRITE SAME, disable it */ | |
826 | limits->max_write_same_sectors = 0; | |
827 | } | |
828 | ||
4246a0b6 | 829 | static void clone_endio(struct bio *bio) |
1da177e4 | 830 | { |
4246a0b6 | 831 | int error = bio->bi_error; |
5164bece | 832 | int r = error; |
bfc6d41c | 833 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 834 | struct dm_io *io = tio->io; |
9faf400f | 835 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
836 | dm_endio_fn endio = tio->ti->type->end_io; |
837 | ||
1da177e4 | 838 | if (endio) { |
7de3ee57 | 839 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
840 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
841 | /* | |
842 | * error and requeue request are handled | |
843 | * in dec_pending(). | |
844 | */ | |
1da177e4 | 845 | error = r; |
45cbcd79 KU |
846 | else if (r == DM_ENDIO_INCOMPLETE) |
847 | /* The target will handle the io */ | |
6712ecf8 | 848 | return; |
45cbcd79 KU |
849 | else if (r) { |
850 | DMWARN("unimplemented target endio return value: %d", r); | |
851 | BUG(); | |
852 | } | |
1da177e4 LT |
853 | } |
854 | ||
e6047149 | 855 | if (unlikely(r == -EREMOTEIO && (bio_op(bio) == REQ_OP_WRITE_SAME) && |
7eee4ae2 MS |
856 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) |
857 | disable_write_same(md); | |
858 | ||
cfae7529 | 859 | free_tio(tio); |
b35f8caa | 860 | dec_pending(io, error); |
1da177e4 LT |
861 | } |
862 | ||
56a67df7 MS |
863 | /* |
864 | * Return maximum size of I/O possible at the supplied sector up to the current | |
865 | * target boundary. | |
866 | */ | |
867 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
868 | { | |
869 | sector_t target_offset = dm_target_offset(ti, sector); | |
870 | ||
871 | return ti->len - target_offset; | |
872 | } | |
873 | ||
874 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 875 | { |
56a67df7 | 876 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 877 | sector_t offset, max_len; |
1da177e4 LT |
878 | |
879 | /* | |
542f9038 | 880 | * Does the target need to split even further? |
1da177e4 | 881 | */ |
542f9038 MS |
882 | if (ti->max_io_len) { |
883 | offset = dm_target_offset(ti, sector); | |
884 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
885 | max_len = sector_div(offset, ti->max_io_len); | |
886 | else | |
887 | max_len = offset & (ti->max_io_len - 1); | |
888 | max_len = ti->max_io_len - max_len; | |
889 | ||
890 | if (len > max_len) | |
891 | len = max_len; | |
1da177e4 LT |
892 | } |
893 | ||
894 | return len; | |
895 | } | |
896 | ||
542f9038 MS |
897 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
898 | { | |
899 | if (len > UINT_MAX) { | |
900 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
901 | (unsigned long long)len, UINT_MAX); | |
902 | ti->error = "Maximum size of target IO is too large"; | |
903 | return -EINVAL; | |
904 | } | |
905 | ||
906 | ti->max_io_len = (uint32_t) len; | |
907 | ||
908 | return 0; | |
909 | } | |
910 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
911 | ||
f26c5719 DW |
912 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
913 | sector_t sector, int *srcu_idx) | |
545ed20e | 914 | { |
545ed20e TK |
915 | struct dm_table *map; |
916 | struct dm_target *ti; | |
545ed20e | 917 | |
f26c5719 | 918 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 919 | if (!map) |
f26c5719 | 920 | return NULL; |
545ed20e TK |
921 | |
922 | ti = dm_table_find_target(map, sector); | |
923 | if (!dm_target_is_valid(ti)) | |
f26c5719 | 924 | return NULL; |
545ed20e | 925 | |
f26c5719 DW |
926 | return ti; |
927 | } | |
545ed20e | 928 | |
f26c5719 DW |
929 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
930 | long nr_pages, void **kaddr, pfn_t *pfn) | |
931 | { | |
932 | struct mapped_device *md = dax_get_private(dax_dev); | |
933 | sector_t sector = pgoff * PAGE_SECTORS; | |
934 | struct dm_target *ti; | |
935 | long len, ret = -EIO; | |
936 | int srcu_idx; | |
937 | ||
938 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
939 | ||
940 | if (!ti) | |
941 | goto out; | |
942 | if (!ti->type->direct_access) | |
943 | goto out; | |
944 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
945 | if (len < 1) | |
946 | goto out; | |
947 | nr_pages = min(len, nr_pages); | |
948 | if (ti->type->direct_access) { | |
949 | ret = ti->type->direct_access(ti, sector, kaddr, pfn, | |
950 | nr_pages * PAGE_SIZE); | |
951 | /* | |
952 | * FIXME: convert ti->type->direct_access to return | |
953 | * nr_pages directly. | |
954 | */ | |
955 | if (ret >= 0) | |
956 | ret /= PAGE_SIZE; | |
957 | } | |
958 | out: | |
545ed20e | 959 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
960 | |
961 | return ret; | |
962 | } | |
963 | ||
964 | static long dm_blk_direct_access(struct block_device *bdev, sector_t sector, | |
965 | void **kaddr, pfn_t *pfn, long size) | |
966 | { | |
967 | struct mapped_device *md = bdev->bd_disk->private_data; | |
968 | struct dax_device *dax_dev = md->dax_dev; | |
969 | long nr_pages = size / PAGE_SIZE; | |
970 | ||
971 | nr_pages = dm_dax_direct_access(dax_dev, sector / PAGE_SECTORS, | |
972 | nr_pages, kaddr, pfn); | |
973 | return nr_pages < 0 ? nr_pages : nr_pages * PAGE_SIZE; | |
545ed20e TK |
974 | } |
975 | ||
1dd40c3e MP |
976 | /* |
977 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
28a8f0d3 | 978 | * allowed for all bio types except REQ_PREFLUSH. |
1dd40c3e MP |
979 | * |
980 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
981 | * additional n_sectors sectors of the bio and the rest of the data should be | |
982 | * sent in a next bio. | |
983 | * | |
984 | * A diagram that explains the arithmetics: | |
985 | * +--------------------+---------------+-------+ | |
986 | * | 1 | 2 | 3 | | |
987 | * +--------------------+---------------+-------+ | |
988 | * | |
989 | * <-------------- *tio->len_ptr ---------------> | |
990 | * <------- bi_size -------> | |
991 | * <-- n_sectors --> | |
992 | * | |
993 | * Region 1 was already iterated over with bio_advance or similar function. | |
994 | * (it may be empty if the target doesn't use bio_advance) | |
995 | * Region 2 is the remaining bio size that the target wants to process. | |
996 | * (it may be empty if region 1 is non-empty, although there is no reason | |
997 | * to make it empty) | |
998 | * The target requires that region 3 is to be sent in the next bio. | |
999 | * | |
1000 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1001 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1002 | * copies of the bio. | |
1003 | */ | |
1004 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1005 | { | |
1006 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1007 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1008 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1009 | BUG_ON(bi_size > *tio->len_ptr); |
1010 | BUG_ON(n_sectors > bi_size); | |
1011 | *tio->len_ptr -= bi_size - n_sectors; | |
1012 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1013 | } | |
1014 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1015 | ||
d67a5f4b MP |
1016 | /* |
1017 | * Flush current->bio_list when the target map method blocks. | |
1018 | * This fixes deadlocks in snapshot and possibly in other targets. | |
1019 | */ | |
1020 | struct dm_offload { | |
1021 | struct blk_plug plug; | |
1022 | struct blk_plug_cb cb; | |
1023 | }; | |
1024 | ||
1025 | static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule) | |
1026 | { | |
1027 | struct dm_offload *o = container_of(cb, struct dm_offload, cb); | |
1028 | struct bio_list list; | |
1029 | struct bio *bio; | |
f5fe1b51 | 1030 | int i; |
d67a5f4b MP |
1031 | |
1032 | INIT_LIST_HEAD(&o->cb.list); | |
1033 | ||
1034 | if (unlikely(!current->bio_list)) | |
1035 | return; | |
1036 | ||
f5fe1b51 N |
1037 | for (i = 0; i < 2; i++) { |
1038 | list = current->bio_list[i]; | |
1039 | bio_list_init(¤t->bio_list[i]); | |
1040 | ||
1041 | while ((bio = bio_list_pop(&list))) { | |
1042 | struct bio_set *bs = bio->bi_pool; | |
1043 | if (unlikely(!bs) || bs == fs_bio_set) { | |
1044 | bio_list_add(¤t->bio_list[i], bio); | |
1045 | continue; | |
1046 | } | |
1047 | ||
1048 | spin_lock(&bs->rescue_lock); | |
1049 | bio_list_add(&bs->rescue_list, bio); | |
1050 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
1051 | spin_unlock(&bs->rescue_lock); | |
d67a5f4b | 1052 | } |
d67a5f4b MP |
1053 | } |
1054 | } | |
1055 | ||
1056 | static void dm_offload_start(struct dm_offload *o) | |
1057 | { | |
1058 | blk_start_plug(&o->plug); | |
1059 | o->cb.callback = flush_current_bio_list; | |
1060 | list_add(&o->cb.list, ¤t->plug->cb_list); | |
1061 | } | |
1062 | ||
1063 | static void dm_offload_end(struct dm_offload *o) | |
1064 | { | |
1065 | list_del(&o->cb.list); | |
1066 | blk_finish_plug(&o->plug); | |
1067 | } | |
1068 | ||
bd2a49b8 | 1069 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1070 | { |
1071 | int r; | |
2056a782 | 1072 | sector_t sector; |
d67a5f4b | 1073 | struct dm_offload o; |
dba14160 | 1074 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1075 | struct dm_target *ti = tio->ti; |
1da177e4 | 1076 | |
1da177e4 | 1077 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1078 | |
1079 | /* | |
1080 | * Map the clone. If r == 0 we don't need to do | |
1081 | * anything, the target has assumed ownership of | |
1082 | * this io. | |
1083 | */ | |
1084 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1085 | sector = clone->bi_iter.bi_sector; |
d67a5f4b MP |
1086 | |
1087 | dm_offload_start(&o); | |
7de3ee57 | 1088 | r = ti->type->map(ti, clone); |
d67a5f4b MP |
1089 | dm_offload_end(&o); |
1090 | ||
45cbcd79 | 1091 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1092 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1093 | |
d07335e5 MS |
1094 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1095 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1096 | |
1da177e4 | 1097 | generic_make_request(clone); |
2e93ccc1 KU |
1098 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1099 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f | 1100 | dec_pending(tio->io, r); |
cfae7529 | 1101 | free_tio(tio); |
ab37844d | 1102 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1103 | DMWARN("unimplemented target map return value: %d", r); |
1104 | BUG(); | |
1da177e4 LT |
1105 | } |
1106 | } | |
1107 | ||
1108 | struct clone_info { | |
1109 | struct mapped_device *md; | |
1110 | struct dm_table *map; | |
1111 | struct bio *bio; | |
1112 | struct dm_io *io; | |
1113 | sector_t sector; | |
e0d6609a | 1114 | unsigned sector_count; |
1da177e4 LT |
1115 | }; |
1116 | ||
e0d6609a | 1117 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1118 | { |
4f024f37 KO |
1119 | bio->bi_iter.bi_sector = sector; |
1120 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1121 | } |
1122 | ||
1123 | /* | |
1124 | * Creates a bio that consists of range of complete bvecs. | |
1125 | */ | |
c80914e8 MS |
1126 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1127 | sector_t sector, unsigned len) | |
1da177e4 | 1128 | { |
dba14160 | 1129 | struct bio *clone = &tio->clone; |
1da177e4 | 1130 | |
1c3b13e6 KO |
1131 | __bio_clone_fast(clone, bio); |
1132 | ||
c80914e8 MS |
1133 | if (bio_integrity(bio)) { |
1134 | int r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
1135 | if (r < 0) | |
1136 | return r; | |
1137 | } | |
bd2a49b8 | 1138 | |
1c3b13e6 KO |
1139 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1140 | clone->bi_iter.bi_size = to_bytes(len); | |
1141 | ||
1142 | if (bio_integrity(bio)) | |
1143 | bio_integrity_trim(clone, 0, len); | |
c80914e8 MS |
1144 | |
1145 | return 0; | |
1da177e4 LT |
1146 | } |
1147 | ||
9015df24 | 1148 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1149 | struct dm_target *ti, |
55a62eef | 1150 | unsigned target_bio_nr) |
f9ab94ce | 1151 | { |
dba14160 MP |
1152 | struct dm_target_io *tio; |
1153 | struct bio *clone; | |
1154 | ||
99778273 | 1155 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1156 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1157 | |
1158 | tio->io = ci->io; | |
1159 | tio->ti = ti; | |
55a62eef | 1160 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1161 | |
1162 | return tio; | |
1163 | } | |
1164 | ||
14fe594d AK |
1165 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1166 | struct dm_target *ti, | |
1dd40c3e | 1167 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1168 | { |
99778273 | 1169 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1170 | struct bio *clone = &tio->clone; |
9015df24 | 1171 | |
1dd40c3e MP |
1172 | tio->len_ptr = len; |
1173 | ||
99778273 | 1174 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1175 | if (len) |
1dd40c3e | 1176 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1177 | |
bd2a49b8 | 1178 | __map_bio(tio); |
f9ab94ce MP |
1179 | } |
1180 | ||
14fe594d | 1181 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1182 | unsigned num_bios, unsigned *len) |
06a426ce | 1183 | { |
55a62eef | 1184 | unsigned target_bio_nr; |
06a426ce | 1185 | |
55a62eef | 1186 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1187 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1188 | } |
1189 | ||
14fe594d | 1190 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1191 | { |
06a426ce | 1192 | unsigned target_nr = 0; |
f9ab94ce MP |
1193 | struct dm_target *ti; |
1194 | ||
b372d360 | 1195 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1196 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1197 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1198 | |
f9ab94ce MP |
1199 | return 0; |
1200 | } | |
1201 | ||
c80914e8 | 1202 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1203 | sector_t sector, unsigned *len) |
5ae89a87 | 1204 | { |
dba14160 | 1205 | struct bio *bio = ci->bio; |
5ae89a87 | 1206 | struct dm_target_io *tio; |
b0d8ed4d AK |
1207 | unsigned target_bio_nr; |
1208 | unsigned num_target_bios = 1; | |
c80914e8 | 1209 | int r = 0; |
5ae89a87 | 1210 | |
b0d8ed4d AK |
1211 | /* |
1212 | * Does the target want to receive duplicate copies of the bio? | |
1213 | */ | |
1214 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1215 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1216 | |
b0d8ed4d | 1217 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1218 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e | 1219 | tio->len_ptr = len; |
c80914e8 | 1220 | r = clone_bio(tio, bio, sector, *len); |
072623de | 1221 | if (r < 0) { |
cfae7529 | 1222 | free_tio(tio); |
c80914e8 | 1223 | break; |
072623de | 1224 | } |
b0d8ed4d AK |
1225 | __map_bio(tio); |
1226 | } | |
c80914e8 MS |
1227 | |
1228 | return r; | |
5ae89a87 MS |
1229 | } |
1230 | ||
55a62eef | 1231 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1232 | |
55a62eef | 1233 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1234 | { |
55a62eef | 1235 | return ti->num_discard_bios; |
23508a96 MS |
1236 | } |
1237 | ||
55a62eef | 1238 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1239 | { |
55a62eef | 1240 | return ti->num_write_same_bios; |
23508a96 MS |
1241 | } |
1242 | ||
1243 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
9eef87da | 1244 | |
23508a96 MS |
1245 | static bool is_split_required_for_discard(struct dm_target *ti) |
1246 | { | |
55a62eef | 1247 | return ti->split_discard_bios; |
cec47e3d KU |
1248 | } |
1249 | ||
14fe594d AK |
1250 | static int __send_changing_extent_only(struct clone_info *ci, |
1251 | get_num_bios_fn get_num_bios, | |
1252 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1253 | { |
5ae89a87 | 1254 | struct dm_target *ti; |
e0d6609a | 1255 | unsigned len; |
55a62eef | 1256 | unsigned num_bios; |
ba1cbad9 | 1257 | |
a79245b3 MS |
1258 | do { |
1259 | ti = dm_table_find_target(ci->map, ci->sector); | |
1260 | if (!dm_target_is_valid(ti)) | |
1261 | return -EIO; | |
2eb6e1e3 | 1262 | |
5ae89a87 | 1263 | /* |
23508a96 MS |
1264 | * Even though the device advertised support for this type of |
1265 | * request, that does not mean every target supports it, and | |
936688d7 | 1266 | * reconfiguration might also have changed that since the |
a79245b3 | 1267 | * check was performed. |
5ae89a87 | 1268 | */ |
55a62eef AK |
1269 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1270 | if (!num_bios) | |
a79245b3 | 1271 | return -EOPNOTSUPP; |
ba1cbad9 | 1272 | |
23508a96 | 1273 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1274 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1275 | else |
e0d6609a | 1276 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
de3ec86d | 1277 | |
1dd40c3e | 1278 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1279 | |
a79245b3 MS |
1280 | ci->sector += len; |
1281 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1282 | |
1283 | return 0; | |
ba1cbad9 MS |
1284 | } |
1285 | ||
14fe594d | 1286 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1287 | { |
14fe594d AK |
1288 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1289 | is_split_required_for_discard); | |
23508a96 | 1290 | } |
0ce65797 | 1291 | |
14fe594d | 1292 | static int __send_write_same(struct clone_info *ci) |
0ce65797 | 1293 | { |
14fe594d | 1294 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
0ce65797 MS |
1295 | } |
1296 | ||
e4c93811 AK |
1297 | /* |
1298 | * Select the correct strategy for processing a non-flush bio. | |
1299 | */ | |
14fe594d | 1300 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1301 | { |
dba14160 | 1302 | struct bio *bio = ci->bio; |
512875bd | 1303 | struct dm_target *ti; |
1c3b13e6 | 1304 | unsigned len; |
c80914e8 | 1305 | int r; |
0ce65797 | 1306 | |
e6047149 | 1307 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
14fe594d | 1308 | return __send_discard(ci); |
e6047149 | 1309 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
14fe594d | 1310 | return __send_write_same(ci); |
0ce65797 | 1311 | |
512875bd JN |
1312 | ti = dm_table_find_target(ci->map, ci->sector); |
1313 | if (!dm_target_is_valid(ti)) | |
1314 | return -EIO; | |
1315 | ||
1c3b13e6 | 1316 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
0ce65797 | 1317 | |
c80914e8 MS |
1318 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1319 | if (r < 0) | |
1320 | return r; | |
0ce65797 | 1321 | |
1c3b13e6 KO |
1322 | ci->sector += len; |
1323 | ci->sector_count -= len; | |
0ce65797 | 1324 | |
1c3b13e6 | 1325 | return 0; |
0ce65797 MS |
1326 | } |
1327 | ||
1da177e4 | 1328 | /* |
14fe594d | 1329 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1330 | */ |
83d5e5b0 MP |
1331 | static void __split_and_process_bio(struct mapped_device *md, |
1332 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1333 | { |
1da177e4 | 1334 | struct clone_info ci; |
512875bd | 1335 | int error = 0; |
1da177e4 | 1336 | |
83d5e5b0 | 1337 | if (unlikely(!map)) { |
6a8736d1 | 1338 | bio_io_error(bio); |
f0b9a450 MP |
1339 | return; |
1340 | } | |
692d0eb9 | 1341 | |
83d5e5b0 | 1342 | ci.map = map; |
1da177e4 | 1343 | ci.md = md; |
1da177e4 LT |
1344 | ci.io = alloc_io(md); |
1345 | ci.io->error = 0; | |
1346 | atomic_set(&ci.io->io_count, 1); | |
1347 | ci.io->bio = bio; | |
1348 | ci.io->md = md; | |
f88fb981 | 1349 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1350 | ci.sector = bio->bi_iter.bi_sector; |
0ce65797 | 1351 | |
3eaf840e | 1352 | start_io_acct(ci.io); |
0ce65797 | 1353 | |
1eff9d32 | 1354 | if (bio->bi_opf & REQ_PREFLUSH) { |
b372d360 MS |
1355 | ci.bio = &ci.md->flush_bio; |
1356 | ci.sector_count = 0; | |
14fe594d | 1357 | error = __send_empty_flush(&ci); |
b372d360 MS |
1358 | /* dec_pending submits any data associated with flush */ |
1359 | } else { | |
6a8736d1 | 1360 | ci.bio = bio; |
d87f4c14 | 1361 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1362 | while (ci.sector_count && !error) |
14fe594d | 1363 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1364 | } |
0ce65797 | 1365 | |
1da177e4 | 1366 | /* drop the extra reference count */ |
512875bd | 1367 | dec_pending(ci.io, error); |
0ce65797 | 1368 | } |
1da177e4 LT |
1369 | /*----------------------------------------------------------------- |
1370 | * CRUD END | |
1371 | *---------------------------------------------------------------*/ | |
0ce65797 | 1372 | |
cec47e3d | 1373 | /* |
1da177e4 LT |
1374 | * The request function that just remaps the bio built up by |
1375 | * dm_merge_bvec. | |
cec47e3d | 1376 | */ |
dece1635 | 1377 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
cec47e3d | 1378 | { |
12f03a49 | 1379 | int rw = bio_data_dir(bio); |
cec47e3d | 1380 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1381 | int srcu_idx; |
1382 | struct dm_table *map; | |
cec47e3d | 1383 | |
83d5e5b0 | 1384 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1385 | |
18c0b223 | 1386 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
d0bcb878 | 1387 | |
6a8736d1 TH |
1388 | /* if we're suspended, we have to queue this io for later */ |
1389 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1390 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1391 | |
1eff9d32 | 1392 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1393 | queue_io(md, bio); |
1394 | else | |
54d9a1b4 | 1395 | bio_io_error(bio); |
dece1635 | 1396 | return BLK_QC_T_NONE; |
cec47e3d | 1397 | } |
1da177e4 | 1398 | |
83d5e5b0 MP |
1399 | __split_and_process_bio(md, map, bio); |
1400 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1401 | return BLK_QC_T_NONE; |
cec47e3d KU |
1402 | } |
1403 | ||
1da177e4 LT |
1404 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1405 | { | |
8a57dfc6 CS |
1406 | int r = bdi_bits; |
1407 | struct mapped_device *md = congested_data; | |
1408 | struct dm_table *map; | |
1da177e4 | 1409 | |
1eb787ec | 1410 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1411 | if (dm_request_based(md)) { |
cec47e3d | 1412 | /* |
e522c039 MS |
1413 | * With request-based DM we only need to check the |
1414 | * top-level queue for congestion. | |
cec47e3d | 1415 | */ |
dc3b17cc | 1416 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1417 | } else { |
1418 | map = dm_get_live_table_fast(md); | |
1419 | if (map) | |
cec47e3d | 1420 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1421 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1422 | } |
1423 | } | |
1424 | ||
1da177e4 LT |
1425 | return r; |
1426 | } | |
1427 | ||
1428 | /*----------------------------------------------------------------- | |
1429 | * An IDR is used to keep track of allocated minor numbers. | |
1430 | *---------------------------------------------------------------*/ | |
2b06cfff | 1431 | static void free_minor(int minor) |
1da177e4 | 1432 | { |
f32c10b0 | 1433 | spin_lock(&_minor_lock); |
1da177e4 | 1434 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1435 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1436 | } |
1437 | ||
1438 | /* | |
1439 | * See if the device with a specific minor # is free. | |
1440 | */ | |
cf13ab8e | 1441 | static int specific_minor(int minor) |
1da177e4 | 1442 | { |
c9d76be6 | 1443 | int r; |
1da177e4 LT |
1444 | |
1445 | if (minor >= (1 << MINORBITS)) | |
1446 | return -EINVAL; | |
1447 | ||
c9d76be6 | 1448 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1449 | spin_lock(&_minor_lock); |
1da177e4 | 1450 | |
c9d76be6 | 1451 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1452 | |
f32c10b0 | 1453 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1454 | idr_preload_end(); |
1455 | if (r < 0) | |
1456 | return r == -ENOSPC ? -EBUSY : r; | |
1457 | return 0; | |
1da177e4 LT |
1458 | } |
1459 | ||
cf13ab8e | 1460 | static int next_free_minor(int *minor) |
1da177e4 | 1461 | { |
c9d76be6 | 1462 | int r; |
62f75c2f | 1463 | |
c9d76be6 | 1464 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1465 | spin_lock(&_minor_lock); |
1da177e4 | 1466 | |
c9d76be6 | 1467 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1468 | |
f32c10b0 | 1469 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1470 | idr_preload_end(); |
1471 | if (r < 0) | |
1472 | return r; | |
1473 | *minor = r; | |
1474 | return 0; | |
1da177e4 LT |
1475 | } |
1476 | ||
83d5cde4 | 1477 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1478 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1479 | |
53d5914f MP |
1480 | static void dm_wq_work(struct work_struct *work); |
1481 | ||
4cc96131 | 1482 | void dm_init_md_queue(struct mapped_device *md) |
4a0b4ddf MS |
1483 | { |
1484 | /* | |
1485 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 1486 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
1487 | * The type is decided at the first table loading time. |
1488 | * To prevent problematic device stacking, clear the queue flag | |
1489 | * for request stacking support until then. | |
1490 | * | |
1491 | * This queue is new, so no concurrency on the queue_flags. | |
1492 | */ | |
1493 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
1494 | |
1495 | /* | |
1496 | * Initialize data that will only be used by a non-blk-mq DM queue | |
1497 | * - must do so here (in alloc_dev callchain) before queue is used | |
1498 | */ | |
1499 | md->queue->queuedata = md; | |
dc3b17cc | 1500 | md->queue->backing_dev_info->congested_data = md; |
bfebd1cd | 1501 | } |
4a0b4ddf | 1502 | |
4cc96131 | 1503 | void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1504 | { |
17e149b8 | 1505 | md->use_blk_mq = false; |
bfebd1cd MS |
1506 | dm_init_md_queue(md); |
1507 | ||
1508 | /* | |
1509 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1510 | */ | |
dc3b17cc | 1511 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf | 1512 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
1513 | } |
1514 | ||
0f20972f MS |
1515 | static void cleanup_mapped_device(struct mapped_device *md) |
1516 | { | |
0f20972f MS |
1517 | if (md->wq) |
1518 | destroy_workqueue(md->wq); | |
1519 | if (md->kworker_task) | |
1520 | kthread_stop(md->kworker_task); | |
6f65985e | 1521 | mempool_destroy(md->io_pool); |
0f20972f MS |
1522 | if (md->bs) |
1523 | bioset_free(md->bs); | |
1524 | ||
f26c5719 DW |
1525 | if (md->dax_dev) { |
1526 | kill_dax(md->dax_dev); | |
1527 | put_dax(md->dax_dev); | |
1528 | md->dax_dev = NULL; | |
1529 | } | |
1530 | ||
0f20972f MS |
1531 | if (md->disk) { |
1532 | spin_lock(&_minor_lock); | |
1533 | md->disk->private_data = NULL; | |
1534 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1535 | del_gendisk(md->disk); |
1536 | put_disk(md->disk); | |
1537 | } | |
1538 | ||
1539 | if (md->queue) | |
1540 | blk_cleanup_queue(md->queue); | |
1541 | ||
d09960b0 TE |
1542 | cleanup_srcu_struct(&md->io_barrier); |
1543 | ||
0f20972f MS |
1544 | if (md->bdev) { |
1545 | bdput(md->bdev); | |
1546 | md->bdev = NULL; | |
1547 | } | |
4cc96131 MS |
1548 | |
1549 | dm_mq_cleanup_mapped_device(md); | |
0f20972f MS |
1550 | } |
1551 | ||
1da177e4 LT |
1552 | /* |
1553 | * Allocate and initialise a blank device with a given minor. | |
1554 | */ | |
2b06cfff | 1555 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1556 | { |
115485e8 | 1557 | int r, numa_node_id = dm_get_numa_node(); |
f26c5719 | 1558 | struct dax_device *dax_dev; |
115485e8 | 1559 | struct mapped_device *md; |
ba61fdd1 | 1560 | void *old_md; |
1da177e4 | 1561 | |
115485e8 | 1562 | md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1563 | if (!md) { |
1564 | DMWARN("unable to allocate device, out of memory."); | |
1565 | return NULL; | |
1566 | } | |
1567 | ||
10da4f79 | 1568 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1569 | goto bad_module_get; |
10da4f79 | 1570 | |
1da177e4 | 1571 | /* get a minor number for the dev */ |
2b06cfff | 1572 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1573 | r = next_free_minor(&minor); |
2b06cfff | 1574 | else |
cf13ab8e | 1575 | r = specific_minor(minor); |
1da177e4 | 1576 | if (r < 0) |
6ed7ade8 | 1577 | goto bad_minor; |
1da177e4 | 1578 | |
83d5e5b0 MP |
1579 | r = init_srcu_struct(&md->io_barrier); |
1580 | if (r < 0) | |
1581 | goto bad_io_barrier; | |
1582 | ||
115485e8 | 1583 | md->numa_node_id = numa_node_id; |
4cc96131 | 1584 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1585 | md->init_tio_pdu = false; |
a5664dad | 1586 | md->type = DM_TYPE_NONE; |
e61290a4 | 1587 | mutex_init(&md->suspend_lock); |
a5664dad | 1588 | mutex_init(&md->type_lock); |
86f1152b | 1589 | mutex_init(&md->table_devices_lock); |
022c2611 | 1590 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1591 | atomic_set(&md->holders, 1); |
5c6bd75d | 1592 | atomic_set(&md->open_count, 0); |
1da177e4 | 1593 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1594 | atomic_set(&md->uevent_seq, 0); |
1595 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1596 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1597 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1598 | |
115485e8 | 1599 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1600 | if (!md->queue) |
0f20972f | 1601 | goto bad; |
1da177e4 | 1602 | |
4a0b4ddf | 1603 | dm_init_md_queue(md); |
9faf400f | 1604 | |
115485e8 | 1605 | md->disk = alloc_disk_node(1, numa_node_id); |
1da177e4 | 1606 | if (!md->disk) |
0f20972f | 1607 | goto bad; |
1da177e4 | 1608 | |
316d315b NK |
1609 | atomic_set(&md->pending[0], 0); |
1610 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1611 | init_waitqueue_head(&md->wait); |
53d5914f | 1612 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1613 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1614 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1615 | md->kworker_task = NULL; |
f0b04115 | 1616 | |
1da177e4 LT |
1617 | md->disk->major = _major; |
1618 | md->disk->first_minor = minor; | |
1619 | md->disk->fops = &dm_blk_dops; | |
1620 | md->disk->queue = md->queue; | |
1621 | md->disk->private_data = md; | |
1622 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 DW |
1623 | |
1624 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); | |
1625 | if (!dax_dev) | |
1626 | goto bad; | |
1627 | md->dax_dev = dax_dev; | |
1628 | ||
1da177e4 | 1629 | add_disk(md->disk); |
7e51f257 | 1630 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1631 | |
670368a8 | 1632 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1633 | if (!md->wq) |
0f20972f | 1634 | goto bad; |
304f3f6a | 1635 | |
32a926da MP |
1636 | md->bdev = bdget_disk(md->disk, 0); |
1637 | if (!md->bdev) | |
0f20972f | 1638 | goto bad; |
32a926da | 1639 | |
3a83f467 | 1640 | bio_init(&md->flush_bio, NULL, 0); |
6a8736d1 | 1641 | md->flush_bio.bi_bdev = md->bdev; |
70fd7614 | 1642 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
6a8736d1 | 1643 | |
fd2ed4d2 MP |
1644 | dm_stats_init(&md->stats); |
1645 | ||
ba61fdd1 | 1646 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1647 | spin_lock(&_minor_lock); |
ba61fdd1 | 1648 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1649 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1650 | |
1651 | BUG_ON(old_md != MINOR_ALLOCED); | |
1652 | ||
1da177e4 LT |
1653 | return md; |
1654 | ||
0f20972f MS |
1655 | bad: |
1656 | cleanup_mapped_device(md); | |
83d5e5b0 | 1657 | bad_io_barrier: |
1da177e4 | 1658 | free_minor(minor); |
6ed7ade8 | 1659 | bad_minor: |
10da4f79 | 1660 | module_put(THIS_MODULE); |
6ed7ade8 | 1661 | bad_module_get: |
1da177e4 LT |
1662 | kfree(md); |
1663 | return NULL; | |
1664 | } | |
1665 | ||
ae9da83f JN |
1666 | static void unlock_fs(struct mapped_device *md); |
1667 | ||
1da177e4 LT |
1668 | static void free_dev(struct mapped_device *md) |
1669 | { | |
f331c029 | 1670 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1671 | |
32a926da | 1672 | unlock_fs(md); |
2eb6e1e3 | 1673 | |
0f20972f | 1674 | cleanup_mapped_device(md); |
63a4f065 | 1675 | |
86f1152b | 1676 | free_table_devices(&md->table_devices); |
63a4f065 | 1677 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1678 | free_minor(minor); |
1679 | ||
10da4f79 | 1680 | module_put(THIS_MODULE); |
1da177e4 LT |
1681 | kfree(md); |
1682 | } | |
1683 | ||
e6ee8c0b KU |
1684 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1685 | { | |
c0820cf5 | 1686 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1687 | |
4e6e36c3 MS |
1688 | if (md->bs) { |
1689 | /* The md already has necessary mempools. */ | |
545ed20e | 1690 | if (dm_table_bio_based(t)) { |
16245bdc JN |
1691 | /* |
1692 | * Reload bioset because front_pad may have changed | |
1693 | * because a different table was loaded. | |
1694 | */ | |
1695 | bioset_free(md->bs); | |
1696 | md->bs = p->bs; | |
1697 | p->bs = NULL; | |
16245bdc | 1698 | } |
4e6e36c3 MS |
1699 | /* |
1700 | * There's no need to reload with request-based dm | |
1701 | * because the size of front_pad doesn't change. | |
1702 | * Note for future: If you are to reload bioset, | |
1703 | * prep-ed requests in the queue may refer | |
1704 | * to bio from the old bioset, so you must walk | |
1705 | * through the queue to unprep. | |
1706 | */ | |
1707 | goto out; | |
c0820cf5 | 1708 | } |
e6ee8c0b | 1709 | |
eb8db831 | 1710 | BUG_ON(!p || md->io_pool || md->bs); |
cbc4e3c1 | 1711 | |
e6ee8c0b KU |
1712 | md->io_pool = p->io_pool; |
1713 | p->io_pool = NULL; | |
e6ee8c0b KU |
1714 | md->bs = p->bs; |
1715 | p->bs = NULL; | |
4e6e36c3 | 1716 | |
e6ee8c0b | 1717 | out: |
02233342 | 1718 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1719 | dm_table_free_md_mempools(t); |
1720 | } | |
1721 | ||
1da177e4 LT |
1722 | /* |
1723 | * Bind a table to the device. | |
1724 | */ | |
1725 | static void event_callback(void *context) | |
1726 | { | |
7a8c3d3b MA |
1727 | unsigned long flags; |
1728 | LIST_HEAD(uevents); | |
1da177e4 LT |
1729 | struct mapped_device *md = (struct mapped_device *) context; |
1730 | ||
7a8c3d3b MA |
1731 | spin_lock_irqsave(&md->uevent_lock, flags); |
1732 | list_splice_init(&md->uevent_list, &uevents); | |
1733 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1734 | ||
ed9e1982 | 1735 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1736 | |
1da177e4 LT |
1737 | atomic_inc(&md->event_nr); |
1738 | wake_up(&md->eventq); | |
1739 | } | |
1740 | ||
c217649b MS |
1741 | /* |
1742 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1743 | */ | |
4e90188b | 1744 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1745 | { |
4e90188b | 1746 | set_capacity(md->disk, size); |
1da177e4 | 1747 | |
db8fef4f | 1748 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
1749 | } |
1750 | ||
042d2a9b AK |
1751 | /* |
1752 | * Returns old map, which caller must destroy. | |
1753 | */ | |
1754 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
1755 | struct queue_limits *limits) | |
1da177e4 | 1756 | { |
042d2a9b | 1757 | struct dm_table *old_map; |
165125e1 | 1758 | struct request_queue *q = md->queue; |
1da177e4 LT |
1759 | sector_t size; |
1760 | ||
5a8f1f80 BVA |
1761 | lockdep_assert_held(&md->suspend_lock); |
1762 | ||
1da177e4 | 1763 | size = dm_table_get_size(t); |
3ac51e74 DW |
1764 | |
1765 | /* | |
1766 | * Wipe any geometry if the size of the table changed. | |
1767 | */ | |
fd2ed4d2 | 1768 | if (size != dm_get_size(md)) |
3ac51e74 DW |
1769 | memset(&md->geometry, 0, sizeof(md->geometry)); |
1770 | ||
32a926da | 1771 | __set_size(md, size); |
d5816876 | 1772 | |
2ca3310e AK |
1773 | dm_table_event_callback(t, event_callback, md); |
1774 | ||
e6ee8c0b KU |
1775 | /* |
1776 | * The queue hasn't been stopped yet, if the old table type wasn't | |
1777 | * for request-based during suspension. So stop it to prevent | |
1778 | * I/O mapping before resume. | |
1779 | * This must be done before setting the queue restrictions, | |
1780 | * because request-based dm may be run just after the setting. | |
1781 | */ | |
16f12266 | 1782 | if (dm_table_request_based(t)) { |
eca7ee6d | 1783 | dm_stop_queue(q); |
16f12266 MS |
1784 | /* |
1785 | * Leverage the fact that request-based DM targets are | |
1786 | * immutable singletons and establish md->immutable_target | |
1787 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
1788 | */ | |
1789 | md->immutable_target = dm_table_get_immutable_target(t); | |
1790 | } | |
e6ee8c0b KU |
1791 | |
1792 | __bind_mempools(md, t); | |
1793 | ||
a12f5d48 | 1794 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 1795 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
1796 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
1797 | ||
754c5fc7 | 1798 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
1799 | if (old_map) |
1800 | dm_sync_table(md); | |
1da177e4 | 1801 | |
042d2a9b | 1802 | return old_map; |
1da177e4 LT |
1803 | } |
1804 | ||
a7940155 AK |
1805 | /* |
1806 | * Returns unbound table for the caller to free. | |
1807 | */ | |
1808 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 1809 | { |
a12f5d48 | 1810 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
1811 | |
1812 | if (!map) | |
a7940155 | 1813 | return NULL; |
1da177e4 LT |
1814 | |
1815 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 1816 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 1817 | dm_sync_table(md); |
a7940155 AK |
1818 | |
1819 | return map; | |
1da177e4 LT |
1820 | } |
1821 | ||
1822 | /* | |
1823 | * Constructor for a new device. | |
1824 | */ | |
2b06cfff | 1825 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1826 | { |
1827 | struct mapped_device *md; | |
1828 | ||
2b06cfff | 1829 | md = alloc_dev(minor); |
1da177e4 LT |
1830 | if (!md) |
1831 | return -ENXIO; | |
1832 | ||
784aae73 MB |
1833 | dm_sysfs_init(md); |
1834 | ||
1da177e4 LT |
1835 | *result = md; |
1836 | return 0; | |
1837 | } | |
1838 | ||
a5664dad MS |
1839 | /* |
1840 | * Functions to manage md->type. | |
1841 | * All are required to hold md->type_lock. | |
1842 | */ | |
1843 | void dm_lock_md_type(struct mapped_device *md) | |
1844 | { | |
1845 | mutex_lock(&md->type_lock); | |
1846 | } | |
1847 | ||
1848 | void dm_unlock_md_type(struct mapped_device *md) | |
1849 | { | |
1850 | mutex_unlock(&md->type_lock); | |
1851 | } | |
1852 | ||
1853 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
1854 | { | |
00c4fc3b | 1855 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
1856 | md->type = type; |
1857 | } | |
1858 | ||
1859 | unsigned dm_get_md_type(struct mapped_device *md) | |
1860 | { | |
1861 | return md->type; | |
1862 | } | |
1863 | ||
36a0456f AK |
1864 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
1865 | { | |
1866 | return md->immutable_target_type; | |
1867 | } | |
1868 | ||
f84cb8a4 MS |
1869 | /* |
1870 | * The queue_limits are only valid as long as you have a reference | |
1871 | * count on 'md'. | |
1872 | */ | |
1873 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
1874 | { | |
1875 | BUG_ON(!atomic_read(&md->holders)); | |
1876 | return &md->queue->limits; | |
1877 | } | |
1878 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
1879 | ||
4a0b4ddf MS |
1880 | /* |
1881 | * Setup the DM device's queue based on md's type | |
1882 | */ | |
591ddcfc | 1883 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 1884 | { |
bfebd1cd | 1885 | int r; |
545ed20e | 1886 | unsigned type = dm_get_md_type(md); |
bfebd1cd | 1887 | |
545ed20e | 1888 | switch (type) { |
bfebd1cd | 1889 | case DM_TYPE_REQUEST_BASED: |
eb8db831 | 1890 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 1891 | if (r) { |
eca7ee6d | 1892 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 1893 | return r; |
ff36ab34 | 1894 | } |
bfebd1cd MS |
1895 | break; |
1896 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 1897 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 1898 | if (r) { |
eca7ee6d | 1899 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
1900 | return r; |
1901 | } | |
1902 | break; | |
1903 | case DM_TYPE_BIO_BASED: | |
545ed20e | 1904 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 1905 | dm_init_normal_md_queue(md); |
ff36ab34 | 1906 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
1907 | /* |
1908 | * DM handles splitting bios as needed. Free the bio_split bioset | |
1909 | * since it won't be used (saves 1 process per bio-based DM device). | |
1910 | */ | |
1911 | bioset_free(md->queue->bio_split); | |
1912 | md->queue->bio_split = NULL; | |
545ed20e TK |
1913 | |
1914 | if (type == DM_TYPE_DAX_BIO_BASED) | |
1915 | queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue); | |
bfebd1cd | 1916 | break; |
4a0b4ddf MS |
1917 | } |
1918 | ||
1919 | return 0; | |
1920 | } | |
1921 | ||
2bec1f4a | 1922 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
1923 | { |
1924 | struct mapped_device *md; | |
1da177e4 LT |
1925 | unsigned minor = MINOR(dev); |
1926 | ||
1927 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1928 | return NULL; | |
1929 | ||
f32c10b0 | 1930 | spin_lock(&_minor_lock); |
1da177e4 LT |
1931 | |
1932 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
1933 | if (md) { |
1934 | if ((md == MINOR_ALLOCED || | |
1935 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
1936 | dm_deleting_md(md) || | |
1937 | test_bit(DMF_FREEING, &md->flags))) { | |
1938 | md = NULL; | |
1939 | goto out; | |
1940 | } | |
1941 | dm_get(md); | |
fba9f90e | 1942 | } |
1da177e4 | 1943 | |
fba9f90e | 1944 | out: |
f32c10b0 | 1945 | spin_unlock(&_minor_lock); |
1da177e4 | 1946 | |
637842cf DT |
1947 | return md; |
1948 | } | |
3cf2e4ba | 1949 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 1950 | |
9ade92a9 | 1951 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 1952 | { |
9ade92a9 | 1953 | return md->interface_ptr; |
1da177e4 LT |
1954 | } |
1955 | ||
1956 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
1957 | { | |
1958 | md->interface_ptr = ptr; | |
1959 | } | |
1960 | ||
1961 | void dm_get(struct mapped_device *md) | |
1962 | { | |
1963 | atomic_inc(&md->holders); | |
3f77316d | 1964 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
1965 | } |
1966 | ||
09ee96b2 MP |
1967 | int dm_hold(struct mapped_device *md) |
1968 | { | |
1969 | spin_lock(&_minor_lock); | |
1970 | if (test_bit(DMF_FREEING, &md->flags)) { | |
1971 | spin_unlock(&_minor_lock); | |
1972 | return -EBUSY; | |
1973 | } | |
1974 | dm_get(md); | |
1975 | spin_unlock(&_minor_lock); | |
1976 | return 0; | |
1977 | } | |
1978 | EXPORT_SYMBOL_GPL(dm_hold); | |
1979 | ||
72d94861 AK |
1980 | const char *dm_device_name(struct mapped_device *md) |
1981 | { | |
1982 | return md->name; | |
1983 | } | |
1984 | EXPORT_SYMBOL_GPL(dm_device_name); | |
1985 | ||
3f77316d | 1986 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 1987 | { |
3b785fbc | 1988 | struct request_queue *q = dm_get_md_queue(md); |
1134e5ae | 1989 | struct dm_table *map; |
83d5e5b0 | 1990 | int srcu_idx; |
1da177e4 | 1991 | |
3f77316d | 1992 | might_sleep(); |
fba9f90e | 1993 | |
63a4f065 | 1994 | spin_lock(&_minor_lock); |
3f77316d KU |
1995 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
1996 | set_bit(DMF_FREEING, &md->flags); | |
1997 | spin_unlock(&_minor_lock); | |
3b785fbc | 1998 | |
2e91c369 | 1999 | blk_set_queue_dying(q); |
3f77316d | 2000 | |
02233342 | 2001 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 2002 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2003 | |
ab7c7bb6 MP |
2004 | /* |
2005 | * Take suspend_lock so that presuspend and postsuspend methods | |
2006 | * do not race with internal suspend. | |
2007 | */ | |
2008 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2009 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2010 | if (!dm_suspended_md(md)) { |
2011 | dm_table_presuspend_targets(map); | |
2012 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2013 | } |
83d5e5b0 MP |
2014 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2015 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2016 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2017 | |
3f77316d KU |
2018 | /* |
2019 | * Rare, but there may be I/O requests still going to complete, | |
2020 | * for example. Wait for all references to disappear. | |
2021 | * No one should increment the reference count of the mapped_device, | |
2022 | * after the mapped_device state becomes DMF_FREEING. | |
2023 | */ | |
2024 | if (wait) | |
2025 | while (atomic_read(&md->holders)) | |
2026 | msleep(1); | |
2027 | else if (atomic_read(&md->holders)) | |
2028 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2029 | dm_device_name(md), atomic_read(&md->holders)); | |
2030 | ||
2031 | dm_sysfs_exit(md); | |
3f77316d KU |
2032 | dm_table_destroy(__unbind(md)); |
2033 | free_dev(md); | |
2034 | } | |
2035 | ||
2036 | void dm_destroy(struct mapped_device *md) | |
2037 | { | |
2038 | __dm_destroy(md, true); | |
2039 | } | |
2040 | ||
2041 | void dm_destroy_immediate(struct mapped_device *md) | |
2042 | { | |
2043 | __dm_destroy(md, false); | |
2044 | } | |
2045 | ||
2046 | void dm_put(struct mapped_device *md) | |
2047 | { | |
2048 | atomic_dec(&md->holders); | |
1da177e4 | 2049 | } |
79eb885c | 2050 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2051 | |
b48633f8 | 2052 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2053 | { |
2054 | int r = 0; | |
9f4c3f87 | 2055 | DEFINE_WAIT(wait); |
46125c1c MB |
2056 | |
2057 | while (1) { | |
9f4c3f87 | 2058 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2059 | |
b4324fee | 2060 | if (!md_in_flight(md)) |
46125c1c MB |
2061 | break; |
2062 | ||
e3fabdfd | 2063 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2064 | r = -EINTR; |
2065 | break; | |
2066 | } | |
2067 | ||
2068 | io_schedule(); | |
2069 | } | |
9f4c3f87 | 2070 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2071 | |
46125c1c MB |
2072 | return r; |
2073 | } | |
2074 | ||
1da177e4 LT |
2075 | /* |
2076 | * Process the deferred bios | |
2077 | */ | |
ef208587 | 2078 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2079 | { |
ef208587 MP |
2080 | struct mapped_device *md = container_of(work, struct mapped_device, |
2081 | work); | |
6d6f10df | 2082 | struct bio *c; |
83d5e5b0 MP |
2083 | int srcu_idx; |
2084 | struct dm_table *map; | |
1da177e4 | 2085 | |
83d5e5b0 | 2086 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2087 | |
3b00b203 | 2088 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2089 | spin_lock_irq(&md->deferred_lock); |
2090 | c = bio_list_pop(&md->deferred); | |
2091 | spin_unlock_irq(&md->deferred_lock); | |
2092 | ||
6a8736d1 | 2093 | if (!c) |
df12ee99 | 2094 | break; |
022c2611 | 2095 | |
e6ee8c0b KU |
2096 | if (dm_request_based(md)) |
2097 | generic_make_request(c); | |
6a8736d1 | 2098 | else |
83d5e5b0 | 2099 | __split_and_process_bio(md, map, c); |
022c2611 | 2100 | } |
73d410c0 | 2101 | |
83d5e5b0 | 2102 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2103 | } |
2104 | ||
9a1fb464 | 2105 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2106 | { |
3b00b203 | 2107 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2108 | smp_mb__after_atomic(); |
53d5914f | 2109 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2110 | } |
2111 | ||
1da177e4 | 2112 | /* |
042d2a9b | 2113 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2114 | */ |
042d2a9b | 2115 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2116 | { |
87eb5b21 | 2117 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2118 | struct queue_limits limits; |
042d2a9b | 2119 | int r; |
1da177e4 | 2120 | |
e61290a4 | 2121 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2122 | |
2123 | /* device must be suspended */ | |
4f186f8b | 2124 | if (!dm_suspended_md(md)) |
93c534ae | 2125 | goto out; |
1da177e4 | 2126 | |
3ae70656 MS |
2127 | /* |
2128 | * If the new table has no data devices, retain the existing limits. | |
2129 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2130 | * then new I/O is queued based on these limits, and then some paths | |
2131 | * reappear. | |
2132 | */ | |
2133 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2134 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2135 | if (live_map) |
2136 | limits = md->queue->limits; | |
83d5e5b0 | 2137 | dm_put_live_table_fast(md); |
3ae70656 MS |
2138 | } |
2139 | ||
87eb5b21 MC |
2140 | if (!live_map) { |
2141 | r = dm_calculate_queue_limits(table, &limits); | |
2142 | if (r) { | |
2143 | map = ERR_PTR(r); | |
2144 | goto out; | |
2145 | } | |
042d2a9b | 2146 | } |
754c5fc7 | 2147 | |
042d2a9b | 2148 | map = __bind(md, table, &limits); |
1da177e4 | 2149 | |
93c534ae | 2150 | out: |
e61290a4 | 2151 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2152 | return map; |
1da177e4 LT |
2153 | } |
2154 | ||
2155 | /* | |
2156 | * Functions to lock and unlock any filesystem running on the | |
2157 | * device. | |
2158 | */ | |
2ca3310e | 2159 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2160 | { |
e39e2e95 | 2161 | int r; |
1da177e4 LT |
2162 | |
2163 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2164 | |
db8fef4f | 2165 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2166 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2167 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2168 | md->frozen_sb = NULL; |
2169 | return r; | |
dfbe03f6 AK |
2170 | } |
2171 | ||
aa8d7c2f AK |
2172 | set_bit(DMF_FROZEN, &md->flags); |
2173 | ||
1da177e4 LT |
2174 | return 0; |
2175 | } | |
2176 | ||
2ca3310e | 2177 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2178 | { |
aa8d7c2f AK |
2179 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2180 | return; | |
2181 | ||
db8fef4f | 2182 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2183 | md->frozen_sb = NULL; |
aa8d7c2f | 2184 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2185 | } |
2186 | ||
2187 | /* | |
b48633f8 BVA |
2188 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2189 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2190 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2191 | * | |
ffcc3936 MS |
2192 | * If __dm_suspend returns 0, the device is completely quiescent |
2193 | * now. There is no request-processing activity. All new requests | |
2194 | * are being added to md->deferred list. | |
cec47e3d | 2195 | * |
ffcc3936 | 2196 | * Caller must hold md->suspend_lock |
cec47e3d | 2197 | */ |
ffcc3936 | 2198 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2199 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2200 | int dmf_suspended_flag) |
1da177e4 | 2201 | { |
ffcc3936 MS |
2202 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2203 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2204 | int r; | |
1da177e4 | 2205 | |
5a8f1f80 BVA |
2206 | lockdep_assert_held(&md->suspend_lock); |
2207 | ||
2e93ccc1 KU |
2208 | /* |
2209 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2210 | * This flag is cleared before dm_suspend returns. | |
2211 | */ | |
2212 | if (noflush) | |
2213 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2214 | ||
d67ee213 MS |
2215 | /* |
2216 | * This gets reverted if there's an error later and the targets | |
2217 | * provide the .presuspend_undo hook. | |
2218 | */ | |
cf222b37 AK |
2219 | dm_table_presuspend_targets(map); |
2220 | ||
32a926da | 2221 | /* |
9f518b27 KU |
2222 | * Flush I/O to the device. |
2223 | * Any I/O submitted after lock_fs() may not be flushed. | |
2224 | * noflush takes precedence over do_lockfs. | |
2225 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2226 | */ |
2227 | if (!noflush && do_lockfs) { | |
2228 | r = lock_fs(md); | |
d67ee213 MS |
2229 | if (r) { |
2230 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2231 | return r; |
d67ee213 | 2232 | } |
aa8d7c2f | 2233 | } |
1da177e4 LT |
2234 | |
2235 | /* | |
3b00b203 MP |
2236 | * Here we must make sure that no processes are submitting requests |
2237 | * to target drivers i.e. no one may be executing | |
2238 | * __split_and_process_bio. This is called from dm_request and | |
2239 | * dm_wq_work. | |
2240 | * | |
2241 | * To get all processes out of __split_and_process_bio in dm_request, | |
2242 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2243 | * __split_and_process_bio from dm_request and quiesce the thread |
2244 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2245 | * flush_workqueue(md->wq). | |
1da177e4 | 2246 | */ |
1eb787ec | 2247 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2248 | if (map) |
2249 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2250 | |
d0bcb878 | 2251 | /* |
29e4013d TH |
2252 | * Stop md->queue before flushing md->wq in case request-based |
2253 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2254 | */ |
2eb6e1e3 | 2255 | if (dm_request_based(md)) { |
eca7ee6d | 2256 | dm_stop_queue(md->queue); |
02233342 | 2257 | if (md->kworker_task) |
3989144f | 2258 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2259 | } |
cec47e3d | 2260 | |
d0bcb878 KU |
2261 | flush_workqueue(md->wq); |
2262 | ||
1da177e4 | 2263 | /* |
3b00b203 MP |
2264 | * At this point no more requests are entering target request routines. |
2265 | * We call dm_wait_for_completion to wait for all existing requests | |
2266 | * to finish. | |
1da177e4 | 2267 | */ |
b48633f8 | 2268 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2269 | if (!r) |
2270 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2271 | |
6d6f10df | 2272 | if (noflush) |
022c2611 | 2273 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2274 | if (map) |
2275 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2276 | |
1da177e4 | 2277 | /* were we interrupted ? */ |
46125c1c | 2278 | if (r < 0) { |
9a1fb464 | 2279 | dm_queue_flush(md); |
73d410c0 | 2280 | |
cec47e3d | 2281 | if (dm_request_based(md)) |
eca7ee6d | 2282 | dm_start_queue(md->queue); |
cec47e3d | 2283 | |
2ca3310e | 2284 | unlock_fs(md); |
d67ee213 | 2285 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2286 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2287 | } |
1da177e4 | 2288 | |
ffcc3936 MS |
2289 | return r; |
2290 | } | |
2291 | ||
2292 | /* | |
2293 | * We need to be able to change a mapping table under a mounted | |
2294 | * filesystem. For example we might want to move some data in | |
2295 | * the background. Before the table can be swapped with | |
2296 | * dm_bind_table, dm_suspend must be called to flush any in | |
2297 | * flight bios and ensure that any further io gets deferred. | |
2298 | */ | |
2299 | /* | |
2300 | * Suspend mechanism in request-based dm. | |
2301 | * | |
2302 | * 1. Flush all I/Os by lock_fs() if needed. | |
2303 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2304 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2305 | * | |
2306 | * To abort suspend, start the request_queue. | |
2307 | */ | |
2308 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2309 | { | |
2310 | struct dm_table *map = NULL; | |
2311 | int r = 0; | |
2312 | ||
2313 | retry: | |
2314 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2315 | ||
2316 | if (dm_suspended_md(md)) { | |
2317 | r = -EINVAL; | |
2318 | goto out_unlock; | |
2319 | } | |
2320 | ||
2321 | if (dm_suspended_internally_md(md)) { | |
2322 | /* already internally suspended, wait for internal resume */ | |
2323 | mutex_unlock(&md->suspend_lock); | |
2324 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2325 | if (r) | |
2326 | return r; | |
2327 | goto retry; | |
2328 | } | |
2329 | ||
a12f5d48 | 2330 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2331 | |
eaf9a736 | 2332 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2333 | if (r) |
2334 | goto out_unlock; | |
3b00b203 | 2335 | |
4d4471cb KU |
2336 | dm_table_postsuspend_targets(map); |
2337 | ||
d287483d | 2338 | out_unlock: |
e61290a4 | 2339 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2340 | return r; |
1da177e4 LT |
2341 | } |
2342 | ||
ffcc3936 MS |
2343 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2344 | { | |
2345 | if (map) { | |
2346 | int r = dm_table_resume_targets(map); | |
2347 | if (r) | |
2348 | return r; | |
2349 | } | |
2350 | ||
2351 | dm_queue_flush(md); | |
2352 | ||
2353 | /* | |
2354 | * Flushing deferred I/Os must be done after targets are resumed | |
2355 | * so that mapping of targets can work correctly. | |
2356 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2357 | */ | |
2358 | if (dm_request_based(md)) | |
eca7ee6d | 2359 | dm_start_queue(md->queue); |
ffcc3936 MS |
2360 | |
2361 | unlock_fs(md); | |
2362 | ||
2363 | return 0; | |
2364 | } | |
2365 | ||
1da177e4 LT |
2366 | int dm_resume(struct mapped_device *md) |
2367 | { | |
8dc23658 | 2368 | int r; |
cf222b37 | 2369 | struct dm_table *map = NULL; |
1da177e4 | 2370 | |
ffcc3936 | 2371 | retry: |
8dc23658 | 2372 | r = -EINVAL; |
ffcc3936 MS |
2373 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2374 | ||
4f186f8b | 2375 | if (!dm_suspended_md(md)) |
cf222b37 | 2376 | goto out; |
cf222b37 | 2377 | |
ffcc3936 MS |
2378 | if (dm_suspended_internally_md(md)) { |
2379 | /* already internally suspended, wait for internal resume */ | |
2380 | mutex_unlock(&md->suspend_lock); | |
2381 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2382 | if (r) | |
2383 | return r; | |
2384 | goto retry; | |
2385 | } | |
2386 | ||
a12f5d48 | 2387 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2388 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2389 | goto out; |
1da177e4 | 2390 | |
ffcc3936 | 2391 | r = __dm_resume(md, map); |
8757b776 MB |
2392 | if (r) |
2393 | goto out; | |
2ca3310e | 2394 | |
2ca3310e | 2395 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2396 | out: |
e61290a4 | 2397 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2398 | |
cf222b37 | 2399 | return r; |
1da177e4 LT |
2400 | } |
2401 | ||
fd2ed4d2 MP |
2402 | /* |
2403 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2404 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2405 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2406 | */ |
2407 | ||
ffcc3936 | 2408 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2409 | { |
ffcc3936 MS |
2410 | struct dm_table *map = NULL; |
2411 | ||
96b26c8c | 2412 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2413 | return; /* nested internal suspend */ |
2414 | ||
2415 | if (dm_suspended_md(md)) { | |
2416 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2417 | return; /* nest suspend */ | |
2418 | } | |
2419 | ||
a12f5d48 | 2420 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2421 | |
2422 | /* | |
2423 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2424 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2425 | * would require changing .presuspend to return an error -- avoid this | |
2426 | * until there is a need for more elaborate variants of internal suspend. | |
2427 | */ | |
eaf9a736 MS |
2428 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2429 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2430 | |
2431 | dm_table_postsuspend_targets(map); | |
2432 | } | |
2433 | ||
2434 | static void __dm_internal_resume(struct mapped_device *md) | |
2435 | { | |
96b26c8c MP |
2436 | BUG_ON(!md->internal_suspend_count); |
2437 | ||
2438 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2439 | return; /* resume from nested internal suspend */ |
2440 | ||
fd2ed4d2 | 2441 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2442 | goto done; /* resume from nested suspend */ |
2443 | ||
2444 | /* | |
2445 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2446 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2447 | */ | |
2448 | (void) __dm_resume(md, NULL); | |
2449 | ||
2450 | done: | |
2451 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2452 | smp_mb__after_atomic(); | |
2453 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2454 | } | |
2455 | ||
2456 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2457 | { | |
2458 | mutex_lock(&md->suspend_lock); | |
2459 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2460 | mutex_unlock(&md->suspend_lock); | |
2461 | } | |
2462 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2463 | ||
2464 | void dm_internal_resume(struct mapped_device *md) | |
2465 | { | |
2466 | mutex_lock(&md->suspend_lock); | |
2467 | __dm_internal_resume(md); | |
2468 | mutex_unlock(&md->suspend_lock); | |
2469 | } | |
2470 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2471 | ||
2472 | /* | |
2473 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2474 | * which prevents interaction with userspace-driven suspend. | |
2475 | */ | |
2476 | ||
2477 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2478 | { | |
2479 | mutex_lock(&md->suspend_lock); | |
2480 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2481 | return; |
2482 | ||
2483 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2484 | synchronize_srcu(&md->io_barrier); | |
2485 | flush_workqueue(md->wq); | |
2486 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2487 | } | |
b735fede | 2488 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2489 | |
ffcc3936 | 2490 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2491 | { |
ffcc3936 | 2492 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2493 | goto done; |
2494 | ||
2495 | dm_queue_flush(md); | |
2496 | ||
2497 | done: | |
2498 | mutex_unlock(&md->suspend_lock); | |
2499 | } | |
b735fede | 2500 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2501 | |
1da177e4 LT |
2502 | /*----------------------------------------------------------------- |
2503 | * Event notification. | |
2504 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2505 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2506 | unsigned cookie) |
69267a30 | 2507 | { |
60935eb2 MB |
2508 | char udev_cookie[DM_COOKIE_LENGTH]; |
2509 | char *envp[] = { udev_cookie, NULL }; | |
2510 | ||
2511 | if (!cookie) | |
3abf85b5 | 2512 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2513 | else { |
2514 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2515 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2516 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2517 | action, envp); | |
60935eb2 | 2518 | } |
69267a30 AK |
2519 | } |
2520 | ||
7a8c3d3b MA |
2521 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2522 | { | |
2523 | return atomic_add_return(1, &md->uevent_seq); | |
2524 | } | |
2525 | ||
1da177e4 LT |
2526 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2527 | { | |
2528 | return atomic_read(&md->event_nr); | |
2529 | } | |
2530 | ||
2531 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2532 | { | |
2533 | return wait_event_interruptible(md->eventq, | |
2534 | (event_nr != atomic_read(&md->event_nr))); | |
2535 | } | |
2536 | ||
7a8c3d3b MA |
2537 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2538 | { | |
2539 | unsigned long flags; | |
2540 | ||
2541 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2542 | list_add(elist, &md->uevent_list); | |
2543 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2544 | } | |
2545 | ||
1da177e4 LT |
2546 | /* |
2547 | * The gendisk is only valid as long as you have a reference | |
2548 | * count on 'md'. | |
2549 | */ | |
2550 | struct gendisk *dm_disk(struct mapped_device *md) | |
2551 | { | |
2552 | return md->disk; | |
2553 | } | |
65ff5b7d | 2554 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2555 | |
784aae73 MB |
2556 | struct kobject *dm_kobject(struct mapped_device *md) |
2557 | { | |
2995fa78 | 2558 | return &md->kobj_holder.kobj; |
784aae73 MB |
2559 | } |
2560 | ||
784aae73 MB |
2561 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2562 | { | |
2563 | struct mapped_device *md; | |
2564 | ||
2995fa78 | 2565 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2566 | |
4d89b7b4 | 2567 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2568 | dm_deleting_md(md)) |
4d89b7b4 MB |
2569 | return NULL; |
2570 | ||
784aae73 MB |
2571 | dm_get(md); |
2572 | return md; | |
2573 | } | |
2574 | ||
4f186f8b | 2575 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2576 | { |
2577 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2578 | } | |
2579 | ||
ffcc3936 MS |
2580 | int dm_suspended_internally_md(struct mapped_device *md) |
2581 | { | |
2582 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2583 | } | |
2584 | ||
2c140a24 MP |
2585 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2586 | { | |
2587 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2588 | } | |
2589 | ||
64dbce58 KU |
2590 | int dm_suspended(struct dm_target *ti) |
2591 | { | |
ecdb2e25 | 2592 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2593 | } |
2594 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2595 | ||
2e93ccc1 KU |
2596 | int dm_noflush_suspending(struct dm_target *ti) |
2597 | { | |
ecdb2e25 | 2598 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2599 | } |
2600 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2601 | ||
78d8e58a | 2602 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
30187e1d | 2603 | unsigned integrity, unsigned per_io_data_size) |
e6ee8c0b | 2604 | { |
115485e8 | 2605 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2606 | unsigned int pool_size = 0; |
5f015204 | 2607 | unsigned int front_pad; |
e6ee8c0b KU |
2608 | |
2609 | if (!pools) | |
4e6e36c3 | 2610 | return NULL; |
e6ee8c0b | 2611 | |
78d8e58a MS |
2612 | switch (type) { |
2613 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2614 | case DM_TYPE_DAX_BIO_BASED: |
78d8e58a | 2615 | pool_size = dm_get_reserved_bio_based_ios(); |
30187e1d | 2616 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
eb8db831 CH |
2617 | |
2618 | pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); | |
2619 | if (!pools->io_pool) | |
2620 | goto out; | |
78d8e58a MS |
2621 | break; |
2622 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2623 | case DM_TYPE_MQ_REQUEST_BASED: |
eb8db831 | 2624 | pool_size = dm_get_reserved_rq_based_ios(); |
78d8e58a | 2625 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2626 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2627 | break; |
2628 | default: | |
2629 | BUG(); | |
2630 | } | |
2631 | ||
3d8aab2d | 2632 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 2633 | if (!pools->bs) |
5f015204 | 2634 | goto out; |
e6ee8c0b | 2635 | |
a91a2785 | 2636 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2637 | goto out; |
a91a2785 | 2638 | |
e6ee8c0b | 2639 | return pools; |
5f1b670d | 2640 | |
5f1b670d CH |
2641 | out: |
2642 | dm_free_md_mempools(pools); | |
78d8e58a | 2643 | |
4e6e36c3 | 2644 | return NULL; |
e6ee8c0b KU |
2645 | } |
2646 | ||
2647 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2648 | { | |
2649 | if (!pools) | |
2650 | return; | |
2651 | ||
6f65985e | 2652 | mempool_destroy(pools->io_pool); |
1ae49ea2 | 2653 | |
e6ee8c0b KU |
2654 | if (pools->bs) |
2655 | bioset_free(pools->bs); | |
2656 | ||
2657 | kfree(pools); | |
2658 | } | |
2659 | ||
9c72bad1 CH |
2660 | struct dm_pr { |
2661 | u64 old_key; | |
2662 | u64 new_key; | |
2663 | u32 flags; | |
2664 | bool fail_early; | |
2665 | }; | |
2666 | ||
2667 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2668 | void *data) | |
71cdb697 CH |
2669 | { |
2670 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2671 | struct dm_table *table; |
2672 | struct dm_target *ti; | |
2673 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2674 | |
9c72bad1 CH |
2675 | table = dm_get_live_table(md, &srcu_idx); |
2676 | if (!table || !dm_table_get_size(table)) | |
2677 | goto out; | |
71cdb697 | 2678 | |
9c72bad1 CH |
2679 | /* We only support devices that have a single target */ |
2680 | if (dm_table_get_num_targets(table) != 1) | |
2681 | goto out; | |
2682 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2683 | |
9c72bad1 CH |
2684 | ret = -EINVAL; |
2685 | if (!ti->type->iterate_devices) | |
2686 | goto out; | |
2687 | ||
2688 | ret = ti->type->iterate_devices(ti, fn, data); | |
2689 | out: | |
2690 | dm_put_live_table(md, srcu_idx); | |
2691 | return ret; | |
2692 | } | |
2693 | ||
2694 | /* | |
2695 | * For register / unregister we need to manually call out to every path. | |
2696 | */ | |
2697 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
2698 | sector_t start, sector_t len, void *data) | |
2699 | { | |
2700 | struct dm_pr *pr = data; | |
2701 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
2702 | ||
2703 | if (!ops || !ops->pr_register) | |
2704 | return -EOPNOTSUPP; | |
2705 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
2706 | } | |
2707 | ||
2708 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
2709 | u32 flags) | |
2710 | { | |
2711 | struct dm_pr pr = { | |
2712 | .old_key = old_key, | |
2713 | .new_key = new_key, | |
2714 | .flags = flags, | |
2715 | .fail_early = true, | |
2716 | }; | |
2717 | int ret; | |
2718 | ||
2719 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
2720 | if (ret && new_key) { | |
2721 | /* unregister all paths if we failed to register any path */ | |
2722 | pr.old_key = new_key; | |
2723 | pr.new_key = 0; | |
2724 | pr.flags = 0; | |
2725 | pr.fail_early = false; | |
2726 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
2727 | } | |
2728 | ||
2729 | return ret; | |
71cdb697 CH |
2730 | } |
2731 | ||
2732 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 2733 | u32 flags) |
71cdb697 CH |
2734 | { |
2735 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2736 | const struct pr_ops *ops; | |
71cdb697 | 2737 | fmode_t mode; |
956a4025 | 2738 | int r; |
71cdb697 | 2739 | |
956a4025 | 2740 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2741 | if (r < 0) |
2742 | return r; | |
2743 | ||
2744 | ops = bdev->bd_disk->fops->pr_ops; | |
2745 | if (ops && ops->pr_reserve) | |
2746 | r = ops->pr_reserve(bdev, key, type, flags); | |
2747 | else | |
2748 | r = -EOPNOTSUPP; | |
2749 | ||
956a4025 | 2750 | bdput(bdev); |
71cdb697 CH |
2751 | return r; |
2752 | } | |
2753 | ||
2754 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
2755 | { | |
2756 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2757 | const struct pr_ops *ops; | |
71cdb697 | 2758 | fmode_t mode; |
956a4025 | 2759 | int r; |
71cdb697 | 2760 | |
956a4025 | 2761 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2762 | if (r < 0) |
2763 | return r; | |
2764 | ||
2765 | ops = bdev->bd_disk->fops->pr_ops; | |
2766 | if (ops && ops->pr_release) | |
2767 | r = ops->pr_release(bdev, key, type); | |
2768 | else | |
2769 | r = -EOPNOTSUPP; | |
2770 | ||
956a4025 | 2771 | bdput(bdev); |
71cdb697 CH |
2772 | return r; |
2773 | } | |
2774 | ||
2775 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 2776 | enum pr_type type, bool abort) |
71cdb697 CH |
2777 | { |
2778 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2779 | const struct pr_ops *ops; | |
71cdb697 | 2780 | fmode_t mode; |
956a4025 | 2781 | int r; |
71cdb697 | 2782 | |
956a4025 | 2783 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2784 | if (r < 0) |
2785 | return r; | |
2786 | ||
2787 | ops = bdev->bd_disk->fops->pr_ops; | |
2788 | if (ops && ops->pr_preempt) | |
2789 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
2790 | else | |
2791 | r = -EOPNOTSUPP; | |
2792 | ||
956a4025 | 2793 | bdput(bdev); |
71cdb697 CH |
2794 | return r; |
2795 | } | |
2796 | ||
2797 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
2798 | { | |
2799 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2800 | const struct pr_ops *ops; | |
71cdb697 | 2801 | fmode_t mode; |
956a4025 | 2802 | int r; |
71cdb697 | 2803 | |
956a4025 | 2804 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2805 | if (r < 0) |
2806 | return r; | |
2807 | ||
2808 | ops = bdev->bd_disk->fops->pr_ops; | |
2809 | if (ops && ops->pr_clear) | |
2810 | r = ops->pr_clear(bdev, key); | |
2811 | else | |
2812 | r = -EOPNOTSUPP; | |
2813 | ||
956a4025 | 2814 | bdput(bdev); |
71cdb697 CH |
2815 | return r; |
2816 | } | |
2817 | ||
2818 | static const struct pr_ops dm_pr_ops = { | |
2819 | .pr_register = dm_pr_register, | |
2820 | .pr_reserve = dm_pr_reserve, | |
2821 | .pr_release = dm_pr_release, | |
2822 | .pr_preempt = dm_pr_preempt, | |
2823 | .pr_clear = dm_pr_clear, | |
2824 | }; | |
2825 | ||
83d5cde4 | 2826 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2827 | .open = dm_blk_open, |
2828 | .release = dm_blk_close, | |
aa129a22 | 2829 | .ioctl = dm_blk_ioctl, |
545ed20e | 2830 | .direct_access = dm_blk_direct_access, |
3ac51e74 | 2831 | .getgeo = dm_blk_getgeo, |
71cdb697 | 2832 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
2833 | .owner = THIS_MODULE |
2834 | }; | |
2835 | ||
f26c5719 DW |
2836 | static const struct dax_operations dm_dax_ops = { |
2837 | .direct_access = dm_dax_direct_access, | |
2838 | }; | |
2839 | ||
1da177e4 LT |
2840 | /* |
2841 | * module hooks | |
2842 | */ | |
2843 | module_init(dm_init); | |
2844 | module_exit(dm_exit); | |
2845 | ||
2846 | module_param(major, uint, 0); | |
2847 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 2848 | |
e8603136 MS |
2849 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
2850 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
2851 | ||
115485e8 MS |
2852 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
2853 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
2854 | ||
1da177e4 LT |
2855 | MODULE_DESCRIPTION(DM_NAME " driver"); |
2856 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2857 | MODULE_LICENSE("GPL"); |