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> | |
7e026c8c | 22 | #include <linux/uio.h> |
3ac51e74 | 23 | #include <linux/hdreg.h> |
3f77316d | 24 | #include <linux/delay.h> |
ffcc3936 | 25 | #include <linux/wait.h> |
71cdb697 | 26 | #include <linux/pr.h> |
b0b4d7c6 | 27 | #include <linux/refcount.h> |
55782138 | 28 | |
72d94861 AK |
29 | #define DM_MSG_PREFIX "core" |
30 | ||
60935eb2 MB |
31 | /* |
32 | * Cookies are numeric values sent with CHANGE and REMOVE | |
33 | * uevents while resuming, removing or renaming the device. | |
34 | */ | |
35 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
36 | #define DM_COOKIE_LENGTH 24 | |
37 | ||
1da177e4 LT |
38 | static const char *_name = DM_NAME; |
39 | ||
40 | static unsigned int major = 0; | |
41 | static unsigned int _major = 0; | |
42 | ||
d15b774c AK |
43 | static DEFINE_IDR(_minor_idr); |
44 | ||
f32c10b0 | 45 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
46 | |
47 | static void do_deferred_remove(struct work_struct *w); | |
48 | ||
49 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
50 | ||
acfe0ad7 MP |
51 | static struct workqueue_struct *deferred_remove_workqueue; |
52 | ||
93e6442c MP |
53 | atomic_t dm_global_event_nr = ATOMIC_INIT(0); |
54 | DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq); | |
55 | ||
62e08243 MP |
56 | void dm_issue_global_event(void) |
57 | { | |
58 | atomic_inc(&dm_global_event_nr); | |
59 | wake_up(&dm_global_eventq); | |
60 | } | |
61 | ||
1da177e4 | 62 | /* |
64f52b0e | 63 | * One of these is allocated (on-stack) per original bio. |
1da177e4 | 64 | */ |
64f52b0e | 65 | struct clone_info { |
64f52b0e MS |
66 | struct dm_table *map; |
67 | struct bio *bio; | |
68 | struct dm_io *io; | |
69 | sector_t sector; | |
70 | unsigned sector_count; | |
71 | }; | |
72 | ||
73 | /* | |
74 | * One of these is allocated per clone bio. | |
75 | */ | |
76 | #define DM_TIO_MAGIC 7282014 | |
77 | struct dm_target_io { | |
78 | unsigned magic; | |
79 | struct dm_io *io; | |
80 | struct dm_target *ti; | |
81 | unsigned target_bio_nr; | |
82 | unsigned *len_ptr; | |
83 | bool inside_dm_io; | |
84 | struct bio clone; | |
85 | }; | |
86 | ||
1da177e4 | 87 | /* |
745dc570 | 88 | * One of these is allocated per original bio. |
64f52b0e | 89 | * It contains the first clone used for that original. |
1da177e4 | 90 | */ |
64f52b0e | 91 | #define DM_IO_MAGIC 5191977 |
1da177e4 | 92 | struct dm_io { |
64f52b0e | 93 | unsigned magic; |
1da177e4 | 94 | struct mapped_device *md; |
4e4cbee9 | 95 | blk_status_t status; |
1da177e4 | 96 | atomic_t io_count; |
745dc570 | 97 | struct bio *orig_bio; |
3eaf840e | 98 | unsigned long start_time; |
f88fb981 | 99 | spinlock_t endio_lock; |
fd2ed4d2 | 100 | struct dm_stats_aux stats_aux; |
64f52b0e MS |
101 | /* last member of dm_target_io is 'struct bio' */ |
102 | struct dm_target_io tio; | |
1da177e4 LT |
103 | }; |
104 | ||
64f52b0e MS |
105 | void *dm_per_bio_data(struct bio *bio, size_t data_size) |
106 | { | |
107 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
108 | if (!tio->inside_dm_io) | |
109 | return (char *)bio - offsetof(struct dm_target_io, clone) - data_size; | |
110 | return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size; | |
111 | } | |
112 | EXPORT_SYMBOL_GPL(dm_per_bio_data); | |
113 | ||
114 | struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size) | |
115 | { | |
116 | struct dm_io *io = (struct dm_io *)((char *)data + data_size); | |
117 | if (io->magic == DM_IO_MAGIC) | |
118 | return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone)); | |
119 | BUG_ON(io->magic != DM_TIO_MAGIC); | |
120 | return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone)); | |
121 | } | |
122 | EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data); | |
123 | ||
124 | unsigned dm_bio_get_target_bio_nr(const struct bio *bio) | |
125 | { | |
126 | return container_of(bio, struct dm_target_io, clone)->target_bio_nr; | |
127 | } | |
128 | EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr); | |
129 | ||
ba61fdd1 JM |
130 | #define MINOR_ALLOCED ((void *)-1) |
131 | ||
1da177e4 LT |
132 | /* |
133 | * Bits for the md->flags field. | |
134 | */ | |
1eb787ec | 135 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 136 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 137 | #define DMF_FROZEN 2 |
fba9f90e | 138 | #define DMF_FREEING 3 |
5c6bd75d | 139 | #define DMF_DELETING 4 |
2e93ccc1 | 140 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
141 | #define DMF_DEFERRED_REMOVE 6 |
142 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 143 | |
115485e8 | 144 | #define DM_NUMA_NODE NUMA_NO_NODE |
115485e8 | 145 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 146 | |
e6ee8c0b KU |
147 | /* |
148 | * For mempools pre-allocation at the table loading time. | |
149 | */ | |
150 | struct dm_md_mempools { | |
e6ee8c0b | 151 | struct bio_set *bs; |
64f52b0e | 152 | struct bio_set *io_bs; |
e6ee8c0b KU |
153 | }; |
154 | ||
86f1152b BM |
155 | struct table_device { |
156 | struct list_head list; | |
b0b4d7c6 | 157 | refcount_t count; |
86f1152b BM |
158 | struct dm_dev dm_dev; |
159 | }; | |
160 | ||
8fbf26ad | 161 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 162 | static struct kmem_cache *_rq_cache; |
94818742 | 163 | |
e8603136 MS |
164 | /* |
165 | * Bio-based DM's mempools' reserved IOs set by the user. | |
166 | */ | |
4cc96131 | 167 | #define RESERVED_BIO_BASED_IOS 16 |
e8603136 MS |
168 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
169 | ||
115485e8 MS |
170 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
171 | { | |
6aa7de05 | 172 | int param = READ_ONCE(*module_param); |
115485e8 MS |
173 | int modified_param = 0; |
174 | bool modified = true; | |
175 | ||
176 | if (param < min) | |
177 | modified_param = min; | |
178 | else if (param > max) | |
179 | modified_param = max; | |
180 | else | |
181 | modified = false; | |
182 | ||
183 | if (modified) { | |
184 | (void)cmpxchg(module_param, param, modified_param); | |
185 | param = modified_param; | |
186 | } | |
187 | ||
188 | return param; | |
189 | } | |
190 | ||
4cc96131 MS |
191 | unsigned __dm_get_module_param(unsigned *module_param, |
192 | unsigned def, unsigned max) | |
f4790826 | 193 | { |
6aa7de05 | 194 | unsigned param = READ_ONCE(*module_param); |
09c2d531 | 195 | unsigned modified_param = 0; |
f4790826 | 196 | |
09c2d531 MS |
197 | if (!param) |
198 | modified_param = def; | |
199 | else if (param > max) | |
200 | modified_param = max; | |
f4790826 | 201 | |
09c2d531 MS |
202 | if (modified_param) { |
203 | (void)cmpxchg(module_param, param, modified_param); | |
204 | param = modified_param; | |
f4790826 MS |
205 | } |
206 | ||
09c2d531 | 207 | return param; |
f4790826 MS |
208 | } |
209 | ||
e8603136 MS |
210 | unsigned dm_get_reserved_bio_based_ios(void) |
211 | { | |
09c2d531 | 212 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 213 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
214 | } |
215 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
216 | ||
115485e8 MS |
217 | static unsigned dm_get_numa_node(void) |
218 | { | |
219 | return __dm_get_module_param_int(&dm_numa_node, | |
220 | DM_NUMA_NODE, num_online_nodes() - 1); | |
221 | } | |
222 | ||
1da177e4 LT |
223 | static int __init local_init(void) |
224 | { | |
51157b4a | 225 | int r = -ENOMEM; |
1da177e4 | 226 | |
8fbf26ad KU |
227 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
228 | if (!_rq_tio_cache) | |
dde1e1ec | 229 | return r; |
8fbf26ad | 230 | |
eca7ee6d | 231 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
1ae49ea2 MS |
232 | __alignof__(struct request), 0, NULL); |
233 | if (!_rq_cache) | |
234 | goto out_free_rq_tio_cache; | |
235 | ||
51e5b2bd | 236 | r = dm_uevent_init(); |
51157b4a | 237 | if (r) |
1ae49ea2 | 238 | goto out_free_rq_cache; |
51e5b2bd | 239 | |
acfe0ad7 MP |
240 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
241 | if (!deferred_remove_workqueue) { | |
242 | r = -ENOMEM; | |
243 | goto out_uevent_exit; | |
244 | } | |
245 | ||
1da177e4 LT |
246 | _major = major; |
247 | r = register_blkdev(_major, _name); | |
51157b4a | 248 | if (r < 0) |
acfe0ad7 | 249 | goto out_free_workqueue; |
1da177e4 LT |
250 | |
251 | if (!_major) | |
252 | _major = r; | |
253 | ||
254 | return 0; | |
51157b4a | 255 | |
acfe0ad7 MP |
256 | out_free_workqueue: |
257 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
258 | out_uevent_exit: |
259 | dm_uevent_exit(); | |
1ae49ea2 MS |
260 | out_free_rq_cache: |
261 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
262 | out_free_rq_tio_cache: |
263 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
264 | |
265 | return r; | |
1da177e4 LT |
266 | } |
267 | ||
268 | static void local_exit(void) | |
269 | { | |
2c140a24 | 270 | flush_scheduled_work(); |
acfe0ad7 | 271 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 272 | |
1ae49ea2 | 273 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 274 | kmem_cache_destroy(_rq_tio_cache); |
00d59405 | 275 | unregister_blkdev(_major, _name); |
51e5b2bd | 276 | dm_uevent_exit(); |
1da177e4 LT |
277 | |
278 | _major = 0; | |
279 | ||
280 | DMINFO("cleaned up"); | |
281 | } | |
282 | ||
b9249e55 | 283 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
284 | local_init, |
285 | dm_target_init, | |
286 | dm_linear_init, | |
287 | dm_stripe_init, | |
952b3557 | 288 | dm_io_init, |
945fa4d2 | 289 | dm_kcopyd_init, |
1da177e4 | 290 | dm_interface_init, |
fd2ed4d2 | 291 | dm_statistics_init, |
1da177e4 LT |
292 | }; |
293 | ||
b9249e55 | 294 | static void (*_exits[])(void) = { |
1da177e4 LT |
295 | local_exit, |
296 | dm_target_exit, | |
297 | dm_linear_exit, | |
298 | dm_stripe_exit, | |
952b3557 | 299 | dm_io_exit, |
945fa4d2 | 300 | dm_kcopyd_exit, |
1da177e4 | 301 | dm_interface_exit, |
fd2ed4d2 | 302 | dm_statistics_exit, |
1da177e4 LT |
303 | }; |
304 | ||
305 | static int __init dm_init(void) | |
306 | { | |
307 | const int count = ARRAY_SIZE(_inits); | |
308 | ||
309 | int r, i; | |
310 | ||
311 | for (i = 0; i < count; i++) { | |
312 | r = _inits[i](); | |
313 | if (r) | |
314 | goto bad; | |
315 | } | |
316 | ||
317 | return 0; | |
318 | ||
319 | bad: | |
320 | while (i--) | |
321 | _exits[i](); | |
322 | ||
323 | return r; | |
324 | } | |
325 | ||
326 | static void __exit dm_exit(void) | |
327 | { | |
328 | int i = ARRAY_SIZE(_exits); | |
329 | ||
330 | while (i--) | |
331 | _exits[i](); | |
d15b774c AK |
332 | |
333 | /* | |
334 | * Should be empty by this point. | |
335 | */ | |
d15b774c | 336 | idr_destroy(&_minor_idr); |
1da177e4 LT |
337 | } |
338 | ||
339 | /* | |
340 | * Block device functions | |
341 | */ | |
432a212c MA |
342 | int dm_deleting_md(struct mapped_device *md) |
343 | { | |
344 | return test_bit(DMF_DELETING, &md->flags); | |
345 | } | |
346 | ||
fe5f9f2c | 347 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
348 | { |
349 | struct mapped_device *md; | |
350 | ||
fba9f90e JM |
351 | spin_lock(&_minor_lock); |
352 | ||
fe5f9f2c | 353 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
354 | if (!md) |
355 | goto out; | |
356 | ||
5c6bd75d | 357 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 358 | dm_deleting_md(md)) { |
fba9f90e JM |
359 | md = NULL; |
360 | goto out; | |
361 | } | |
362 | ||
1da177e4 | 363 | dm_get(md); |
5c6bd75d | 364 | atomic_inc(&md->open_count); |
fba9f90e JM |
365 | out: |
366 | spin_unlock(&_minor_lock); | |
367 | ||
368 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
369 | } |
370 | ||
db2a144b | 371 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 372 | { |
63a4f065 | 373 | struct mapped_device *md; |
6e9624b8 | 374 | |
4a1aeb98 MB |
375 | spin_lock(&_minor_lock); |
376 | ||
63a4f065 MS |
377 | md = disk->private_data; |
378 | if (WARN_ON(!md)) | |
379 | goto out; | |
380 | ||
2c140a24 MP |
381 | if (atomic_dec_and_test(&md->open_count) && |
382 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 383 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 384 | |
1da177e4 | 385 | dm_put(md); |
63a4f065 | 386 | out: |
4a1aeb98 | 387 | spin_unlock(&_minor_lock); |
1da177e4 LT |
388 | } |
389 | ||
5c6bd75d AK |
390 | int dm_open_count(struct mapped_device *md) |
391 | { | |
392 | return atomic_read(&md->open_count); | |
393 | } | |
394 | ||
395 | /* | |
396 | * Guarantees nothing is using the device before it's deleted. | |
397 | */ | |
2c140a24 | 398 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
399 | { |
400 | int r = 0; | |
401 | ||
402 | spin_lock(&_minor_lock); | |
403 | ||
2c140a24 | 404 | if (dm_open_count(md)) { |
5c6bd75d | 405 | r = -EBUSY; |
2c140a24 MP |
406 | if (mark_deferred) |
407 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
408 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
409 | r = -EEXIST; | |
5c6bd75d AK |
410 | else |
411 | set_bit(DMF_DELETING, &md->flags); | |
412 | ||
413 | spin_unlock(&_minor_lock); | |
414 | ||
415 | return r; | |
416 | } | |
417 | ||
2c140a24 MP |
418 | int dm_cancel_deferred_remove(struct mapped_device *md) |
419 | { | |
420 | int r = 0; | |
421 | ||
422 | spin_lock(&_minor_lock); | |
423 | ||
424 | if (test_bit(DMF_DELETING, &md->flags)) | |
425 | r = -EBUSY; | |
426 | else | |
427 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
428 | ||
429 | spin_unlock(&_minor_lock); | |
430 | ||
431 | return r; | |
432 | } | |
433 | ||
434 | static void do_deferred_remove(struct work_struct *w) | |
435 | { | |
436 | dm_deferred_remove(); | |
437 | } | |
438 | ||
fd2ed4d2 MP |
439 | sector_t dm_get_size(struct mapped_device *md) |
440 | { | |
441 | return get_capacity(md->disk); | |
442 | } | |
443 | ||
9974fa2c MS |
444 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
445 | { | |
446 | return md->queue; | |
447 | } | |
448 | ||
fd2ed4d2 MP |
449 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
450 | { | |
451 | return &md->stats; | |
452 | } | |
453 | ||
3ac51e74 DW |
454 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
455 | { | |
456 | struct mapped_device *md = bdev->bd_disk->private_data; | |
457 | ||
458 | return dm_get_geometry(md, geo); | |
459 | } | |
460 | ||
519049af MS |
461 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
462 | ||
463 | static int dm_get_bdev_for_ioctl(struct mapped_device *md, | |
464 | struct block_device **bdev, | |
465 | fmode_t *mode) | |
aa129a22 | 466 | { |
66482026 | 467 | struct dm_target *tgt; |
6c182cd8 | 468 | struct dm_table *map; |
da5dadb4 | 469 | int srcu_idx, r, r2; |
aa129a22 | 470 | |
6c182cd8 | 471 | retry: |
e56f81e0 | 472 | r = -ENOTTY; |
956a4025 | 473 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
474 | if (!map || !dm_table_get_size(map)) |
475 | goto out; | |
476 | ||
477 | /* We only support devices that have a single target */ | |
478 | if (dm_table_get_num_targets(map) != 1) | |
479 | goto out; | |
480 | ||
66482026 MS |
481 | tgt = dm_table_get_target(map, 0); |
482 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 483 | goto out; |
aa129a22 | 484 | |
4f186f8b | 485 | if (dm_suspended_md(md)) { |
aa129a22 MB |
486 | r = -EAGAIN; |
487 | goto out; | |
488 | } | |
489 | ||
66482026 | 490 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
491 | if (r < 0) |
492 | goto out; | |
aa129a22 | 493 | |
956a4025 | 494 | bdgrab(*bdev); |
da5dadb4 MS |
495 | r2 = blkdev_get(*bdev, *mode, _dm_claim_ptr); |
496 | if (r2 < 0) { | |
497 | r = r2; | |
519049af | 498 | goto out; |
da5dadb4 | 499 | } |
519049af | 500 | |
956a4025 | 501 | dm_put_live_table(md, srcu_idx); |
e56f81e0 | 502 | return r; |
aa129a22 | 503 | |
aa129a22 | 504 | out: |
956a4025 | 505 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 506 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
507 | msleep(10); |
508 | goto retry; | |
509 | } | |
e56f81e0 CH |
510 | return r; |
511 | } | |
512 | ||
513 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
514 | unsigned int cmd, unsigned long arg) | |
515 | { | |
516 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 517 | int r; |
e56f81e0 | 518 | |
519049af | 519 | r = dm_get_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
520 | if (r < 0) |
521 | return r; | |
6c182cd8 | 522 | |
e56f81e0 CH |
523 | if (r > 0) { |
524 | /* | |
e980f623 CH |
525 | * Target determined this ioctl is being issued against a |
526 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 527 | */ |
e980f623 CH |
528 | if (!capable(CAP_SYS_RAWIO)) { |
529 | DMWARN_LIMIT( | |
530 | "%s: sending ioctl %x to DM device without required privilege.", | |
531 | current->comm, cmd); | |
532 | r = -ENOIOCTLCMD; | |
e56f81e0 | 533 | goto out; |
e980f623 | 534 | } |
e56f81e0 | 535 | } |
6c182cd8 | 536 | |
66482026 | 537 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 538 | out: |
519049af | 539 | blkdev_put(bdev, mode); |
aa129a22 MB |
540 | return r; |
541 | } | |
542 | ||
978e51ba MS |
543 | static void start_io_acct(struct dm_io *io); |
544 | ||
545 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) | |
1da177e4 | 546 | { |
64f52b0e MS |
547 | struct dm_io *io; |
548 | struct dm_target_io *tio; | |
549 | struct bio *clone; | |
550 | ||
551 | clone = bio_alloc_bioset(GFP_NOIO, 0, md->io_bs); | |
552 | if (!clone) | |
553 | return NULL; | |
554 | ||
555 | tio = container_of(clone, struct dm_target_io, clone); | |
556 | tio->inside_dm_io = true; | |
557 | tio->io = NULL; | |
558 | ||
559 | io = container_of(tio, struct dm_io, tio); | |
560 | io->magic = DM_IO_MAGIC; | |
978e51ba MS |
561 | io->status = 0; |
562 | atomic_set(&io->io_count, 1); | |
563 | io->orig_bio = bio; | |
564 | io->md = md; | |
565 | spin_lock_init(&io->endio_lock); | |
566 | ||
567 | start_io_acct(io); | |
64f52b0e MS |
568 | |
569 | return io; | |
1da177e4 LT |
570 | } |
571 | ||
028867ac | 572 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 | 573 | { |
64f52b0e MS |
574 | bio_put(&io->tio.clone); |
575 | } | |
576 | ||
577 | static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, | |
578 | unsigned target_bio_nr, gfp_t gfp_mask) | |
579 | { | |
580 | struct dm_target_io *tio; | |
581 | ||
582 | if (!ci->io->tio.io) { | |
583 | /* the dm_target_io embedded in ci->io is available */ | |
584 | tio = &ci->io->tio; | |
585 | } else { | |
bc02cdbe | 586 | struct bio *clone = bio_alloc_bioset(gfp_mask, 0, ci->io->md->bs); |
64f52b0e MS |
587 | if (!clone) |
588 | return NULL; | |
589 | ||
590 | tio = container_of(clone, struct dm_target_io, clone); | |
591 | tio->inside_dm_io = false; | |
592 | } | |
593 | ||
594 | tio->magic = DM_TIO_MAGIC; | |
595 | tio->io = ci->io; | |
596 | tio->ti = ti; | |
597 | tio->target_bio_nr = target_bio_nr; | |
598 | ||
599 | return tio; | |
1da177e4 LT |
600 | } |
601 | ||
cfae7529 | 602 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 603 | { |
64f52b0e MS |
604 | if (tio->inside_dm_io) |
605 | return; | |
dba14160 | 606 | bio_put(&tio->clone); |
1da177e4 LT |
607 | } |
608 | ||
4cc96131 | 609 | int md_in_flight(struct mapped_device *md) |
90abb8c4 KU |
610 | { |
611 | return atomic_read(&md->pending[READ]) + | |
612 | atomic_read(&md->pending[WRITE]); | |
613 | } | |
614 | ||
3eaf840e JNN |
615 | static void start_io_acct(struct dm_io *io) |
616 | { | |
617 | struct mapped_device *md = io->md; | |
745dc570 | 618 | struct bio *bio = io->orig_bio; |
fd2ed4d2 | 619 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
620 | |
621 | io->start_time = jiffies; | |
622 | ||
f3986374 MS |
623 | generic_start_io_acct(md->queue, rw, bio_sectors(bio), &dm_disk(md)->part0); |
624 | ||
1e9bb880 | 625 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
f3986374 | 626 | atomic_inc_return(&md->pending[rw])); |
fd2ed4d2 MP |
627 | |
628 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
629 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
630 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
631 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
632 | } |
633 | ||
d221d2e7 | 634 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
635 | { |
636 | struct mapped_device *md = io->md; | |
745dc570 | 637 | struct bio *bio = io->orig_bio; |
3eaf840e | 638 | unsigned long duration = jiffies - io->start_time; |
18c0b223 | 639 | int pending; |
3eaf840e JNN |
640 | int rw = bio_data_dir(bio); |
641 | ||
d62e26b3 | 642 | generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 643 | |
fd2ed4d2 | 644 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
645 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
646 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
647 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 648 | |
af7e466a MP |
649 | /* |
650 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 651 | * a flush. |
af7e466a | 652 | */ |
1e9bb880 SL |
653 | pending = atomic_dec_return(&md->pending[rw]); |
654 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 655 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 656 | |
d221d2e7 MP |
657 | /* nudge anyone waiting on suspend queue */ |
658 | if (!pending) | |
659 | wake_up(&md->wait); | |
3eaf840e JNN |
660 | } |
661 | ||
1da177e4 LT |
662 | /* |
663 | * Add the bio to the list of deferred io. | |
664 | */ | |
92c63902 | 665 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 666 | { |
05447420 | 667 | unsigned long flags; |
1da177e4 | 668 | |
05447420 | 669 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 670 | bio_list_add(&md->deferred, bio); |
05447420 | 671 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 672 | queue_work(md->wq, &md->work); |
1da177e4 LT |
673 | } |
674 | ||
675 | /* | |
676 | * Everyone (including functions in this file), should use this | |
677 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 678 | * dm_put_live_table() when finished. |
1da177e4 | 679 | */ |
83d5e5b0 | 680 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 681 | { |
83d5e5b0 MP |
682 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
683 | ||
684 | return srcu_dereference(md->map, &md->io_barrier); | |
685 | } | |
1da177e4 | 686 | |
83d5e5b0 MP |
687 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
688 | { | |
689 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
690 | } | |
691 | ||
692 | void dm_sync_table(struct mapped_device *md) | |
693 | { | |
694 | synchronize_srcu(&md->io_barrier); | |
695 | synchronize_rcu_expedited(); | |
696 | } | |
697 | ||
698 | /* | |
699 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
700 | * The caller must not block between these two functions. | |
701 | */ | |
702 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
703 | { | |
704 | rcu_read_lock(); | |
705 | return rcu_dereference(md->map); | |
706 | } | |
1da177e4 | 707 | |
83d5e5b0 MP |
708 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
709 | { | |
710 | rcu_read_unlock(); | |
1da177e4 LT |
711 | } |
712 | ||
86f1152b BM |
713 | /* |
714 | * Open a table device so we can use it as a map destination. | |
715 | */ | |
716 | static int open_table_device(struct table_device *td, dev_t dev, | |
717 | struct mapped_device *md) | |
718 | { | |
86f1152b BM |
719 | struct block_device *bdev; |
720 | ||
721 | int r; | |
722 | ||
723 | BUG_ON(td->dm_dev.bdev); | |
724 | ||
519049af | 725 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr); |
86f1152b BM |
726 | if (IS_ERR(bdev)) |
727 | return PTR_ERR(bdev); | |
728 | ||
729 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
730 | if (r) { | |
731 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
732 | return r; | |
733 | } | |
734 | ||
735 | td->dm_dev.bdev = bdev; | |
817bf402 | 736 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
737 | return 0; |
738 | } | |
739 | ||
740 | /* | |
741 | * Close a table device that we've been using. | |
742 | */ | |
743 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
744 | { | |
745 | if (!td->dm_dev.bdev) | |
746 | return; | |
747 | ||
748 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
749 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 750 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 751 | td->dm_dev.bdev = NULL; |
817bf402 | 752 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
753 | } |
754 | ||
755 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
756 | fmode_t mode) { | |
757 | struct table_device *td; | |
758 | ||
759 | list_for_each_entry(td, l, list) | |
760 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
761 | return td; | |
762 | ||
763 | return NULL; | |
764 | } | |
765 | ||
766 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
767 | struct dm_dev **result) { | |
768 | int r; | |
769 | struct table_device *td; | |
770 | ||
771 | mutex_lock(&md->table_devices_lock); | |
772 | td = find_table_device(&md->table_devices, dev, mode); | |
773 | if (!td) { | |
115485e8 | 774 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
775 | if (!td) { |
776 | mutex_unlock(&md->table_devices_lock); | |
777 | return -ENOMEM; | |
778 | } | |
779 | ||
780 | td->dm_dev.mode = mode; | |
781 | td->dm_dev.bdev = NULL; | |
782 | ||
783 | if ((r = open_table_device(td, dev, md))) { | |
784 | mutex_unlock(&md->table_devices_lock); | |
785 | kfree(td); | |
786 | return r; | |
787 | } | |
788 | ||
789 | format_dev_t(td->dm_dev.name, dev); | |
790 | ||
b0b4d7c6 | 791 | refcount_set(&td->count, 1); |
86f1152b | 792 | list_add(&td->list, &md->table_devices); |
b0b4d7c6 ER |
793 | } else { |
794 | refcount_inc(&td->count); | |
86f1152b | 795 | } |
86f1152b BM |
796 | mutex_unlock(&md->table_devices_lock); |
797 | ||
798 | *result = &td->dm_dev; | |
799 | return 0; | |
800 | } | |
801 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
802 | ||
803 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
804 | { | |
805 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
806 | ||
807 | mutex_lock(&md->table_devices_lock); | |
b0b4d7c6 | 808 | if (refcount_dec_and_test(&td->count)) { |
86f1152b BM |
809 | close_table_device(td, md); |
810 | list_del(&td->list); | |
811 | kfree(td); | |
812 | } | |
813 | mutex_unlock(&md->table_devices_lock); | |
814 | } | |
815 | EXPORT_SYMBOL(dm_put_table_device); | |
816 | ||
817 | static void free_table_devices(struct list_head *devices) | |
818 | { | |
819 | struct list_head *tmp, *next; | |
820 | ||
821 | list_for_each_safe(tmp, next, devices) { | |
822 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
823 | ||
824 | DMWARN("dm_destroy: %s still exists with %d references", | |
b0b4d7c6 | 825 | td->dm_dev.name, refcount_read(&td->count)); |
86f1152b BM |
826 | kfree(td); |
827 | } | |
828 | } | |
829 | ||
3ac51e74 DW |
830 | /* |
831 | * Get the geometry associated with a dm device | |
832 | */ | |
833 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
834 | { | |
835 | *geo = md->geometry; | |
836 | ||
837 | return 0; | |
838 | } | |
839 | ||
840 | /* | |
841 | * Set the geometry of a device. | |
842 | */ | |
843 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
844 | { | |
845 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
846 | ||
847 | if (geo->start > sz) { | |
848 | DMWARN("Start sector is beyond the geometry limits."); | |
849 | return -EINVAL; | |
850 | } | |
851 | ||
852 | md->geometry = *geo; | |
853 | ||
854 | return 0; | |
855 | } | |
856 | ||
2e93ccc1 KU |
857 | static int __noflush_suspending(struct mapped_device *md) |
858 | { | |
859 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
860 | } | |
861 | ||
1da177e4 LT |
862 | /* |
863 | * Decrements the number of outstanding ios that a bio has been | |
864 | * cloned into, completing the original io if necc. | |
865 | */ | |
4e4cbee9 | 866 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 867 | { |
2e93ccc1 | 868 | unsigned long flags; |
4e4cbee9 | 869 | blk_status_t io_error; |
b35f8caa MB |
870 | struct bio *bio; |
871 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
872 | |
873 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
874 | if (unlikely(error)) { |
875 | spin_lock_irqsave(&io->endio_lock, flags); | |
745dc570 | 876 | if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) |
4e4cbee9 | 877 | io->status = error; |
f88fb981 KU |
878 | spin_unlock_irqrestore(&io->endio_lock, flags); |
879 | } | |
1da177e4 LT |
880 | |
881 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 882 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
883 | /* |
884 | * Target requested pushing back the I/O. | |
2e93ccc1 | 885 | */ |
022c2611 | 886 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 | 887 | if (__noflush_suspending(md)) |
745dc570 MS |
888 | /* NOTE early return due to BLK_STS_DM_REQUEUE below */ |
889 | bio_list_add_head(&md->deferred, io->orig_bio); | |
6a8736d1 | 890 | else |
2e93ccc1 | 891 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 892 | io->status = BLK_STS_IOERR; |
022c2611 | 893 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
894 | } |
895 | ||
4e4cbee9 | 896 | io_error = io->status; |
745dc570 | 897 | bio = io->orig_bio; |
6a8736d1 TH |
898 | end_io_acct(io); |
899 | free_io(md, io); | |
900 | ||
4e4cbee9 | 901 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 902 | return; |
2e93ccc1 | 903 | |
1eff9d32 | 904 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 905 | /* |
6a8736d1 | 906 | * Preflush done for flush with data, reissue |
28a8f0d3 | 907 | * without REQ_PREFLUSH. |
af7e466a | 908 | */ |
1eff9d32 | 909 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 910 | queue_io(md, bio); |
af7e466a | 911 | } else { |
b372d360 | 912 | /* done with normal IO or empty flush */ |
8dd601fa N |
913 | if (io_error) |
914 | bio->bi_status = io_error; | |
4246a0b6 | 915 | bio_endio(bio); |
b35f8caa | 916 | } |
1da177e4 LT |
917 | } |
918 | } | |
919 | ||
4cc96131 | 920 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
921 | { |
922 | struct queue_limits *limits = dm_get_queue_limits(md); | |
923 | ||
924 | /* device doesn't really support WRITE SAME, disable it */ | |
925 | limits->max_write_same_sectors = 0; | |
926 | } | |
927 | ||
ac62d620 CH |
928 | void disable_write_zeroes(struct mapped_device *md) |
929 | { | |
930 | struct queue_limits *limits = dm_get_queue_limits(md); | |
931 | ||
932 | /* device doesn't really support WRITE ZEROES, disable it */ | |
933 | limits->max_write_zeroes_sectors = 0; | |
934 | } | |
935 | ||
4246a0b6 | 936 | static void clone_endio(struct bio *bio) |
1da177e4 | 937 | { |
4e4cbee9 | 938 | blk_status_t error = bio->bi_status; |
bfc6d41c | 939 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 940 | struct dm_io *io = tio->io; |
9faf400f | 941 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
942 | dm_endio_fn endio = tio->ti->type->end_io; |
943 | ||
978e51ba | 944 | if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) { |
ac62d620 | 945 | if (bio_op(bio) == REQ_OP_WRITE_SAME && |
74d46992 | 946 | !bio->bi_disk->queue->limits.max_write_same_sectors) |
ac62d620 CH |
947 | disable_write_same(md); |
948 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && | |
74d46992 | 949 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) |
ac62d620 CH |
950 | disable_write_zeroes(md); |
951 | } | |
7eee4ae2 | 952 | |
1be56909 | 953 | if (endio) { |
4e4cbee9 | 954 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
955 | switch (r) { |
956 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 957 | error = BLK_STS_DM_REQUEUE; |
1be56909 CH |
958 | /*FALLTHRU*/ |
959 | case DM_ENDIO_DONE: | |
960 | break; | |
961 | case DM_ENDIO_INCOMPLETE: | |
962 | /* The target will handle the io */ | |
963 | return; | |
964 | default: | |
965 | DMWARN("unimplemented target endio return value: %d", r); | |
966 | BUG(); | |
967 | } | |
968 | } | |
969 | ||
cfae7529 | 970 | free_tio(tio); |
b35f8caa | 971 | dec_pending(io, error); |
1da177e4 LT |
972 | } |
973 | ||
56a67df7 MS |
974 | /* |
975 | * Return maximum size of I/O possible at the supplied sector up to the current | |
976 | * target boundary. | |
977 | */ | |
978 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
979 | { | |
980 | sector_t target_offset = dm_target_offset(ti, sector); | |
981 | ||
982 | return ti->len - target_offset; | |
983 | } | |
984 | ||
985 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 986 | { |
56a67df7 | 987 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 988 | sector_t offset, max_len; |
1da177e4 LT |
989 | |
990 | /* | |
542f9038 | 991 | * Does the target need to split even further? |
1da177e4 | 992 | */ |
542f9038 MS |
993 | if (ti->max_io_len) { |
994 | offset = dm_target_offset(ti, sector); | |
995 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
996 | max_len = sector_div(offset, ti->max_io_len); | |
997 | else | |
998 | max_len = offset & (ti->max_io_len - 1); | |
999 | max_len = ti->max_io_len - max_len; | |
1000 | ||
1001 | if (len > max_len) | |
1002 | len = max_len; | |
1da177e4 LT |
1003 | } |
1004 | ||
1005 | return len; | |
1006 | } | |
1007 | ||
542f9038 MS |
1008 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1009 | { | |
1010 | if (len > UINT_MAX) { | |
1011 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1012 | (unsigned long long)len, UINT_MAX); | |
1013 | ti->error = "Maximum size of target IO is too large"; | |
1014 | return -EINVAL; | |
1015 | } | |
1016 | ||
8f50e358 ML |
1017 | /* |
1018 | * BIO based queue uses its own splitting. When multipage bvecs | |
1019 | * is switched on, size of the incoming bio may be too big to | |
1020 | * be handled in some targets, such as crypt. | |
1021 | * | |
1022 | * When these targets are ready for the big bio, we can remove | |
1023 | * the limit. | |
1024 | */ | |
1025 | ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE); | |
542f9038 MS |
1026 | |
1027 | return 0; | |
1028 | } | |
1029 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1030 | ||
f26c5719 DW |
1031 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
1032 | sector_t sector, int *srcu_idx) | |
545ed20e | 1033 | { |
545ed20e TK |
1034 | struct dm_table *map; |
1035 | struct dm_target *ti; | |
545ed20e | 1036 | |
f26c5719 | 1037 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1038 | if (!map) |
f26c5719 | 1039 | return NULL; |
545ed20e TK |
1040 | |
1041 | ti = dm_table_find_target(map, sector); | |
1042 | if (!dm_target_is_valid(ti)) | |
f26c5719 | 1043 | return NULL; |
545ed20e | 1044 | |
f26c5719 DW |
1045 | return ti; |
1046 | } | |
545ed20e | 1047 | |
f26c5719 DW |
1048 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
1049 | long nr_pages, void **kaddr, pfn_t *pfn) | |
1050 | { | |
1051 | struct mapped_device *md = dax_get_private(dax_dev); | |
1052 | sector_t sector = pgoff * PAGE_SECTORS; | |
1053 | struct dm_target *ti; | |
1054 | long len, ret = -EIO; | |
1055 | int srcu_idx; | |
545ed20e | 1056 | |
f26c5719 | 1057 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1058 | |
f26c5719 DW |
1059 | if (!ti) |
1060 | goto out; | |
1061 | if (!ti->type->direct_access) | |
1062 | goto out; | |
1063 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
1064 | if (len < 1) | |
1065 | goto out; | |
1066 | nr_pages = min(len, nr_pages); | |
545ed20e | 1067 | if (ti->type->direct_access) |
817bf402 DW |
1068 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
1069 | ||
f26c5719 | 1070 | out: |
545ed20e | 1071 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1072 | |
1073 | return ret; | |
545ed20e TK |
1074 | } |
1075 | ||
7e026c8c DW |
1076 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
1077 | void *addr, size_t bytes, struct iov_iter *i) | |
1078 | { | |
1079 | struct mapped_device *md = dax_get_private(dax_dev); | |
1080 | sector_t sector = pgoff * PAGE_SECTORS; | |
1081 | struct dm_target *ti; | |
1082 | long ret = 0; | |
1083 | int srcu_idx; | |
1084 | ||
1085 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1086 | ||
1087 | if (!ti) | |
1088 | goto out; | |
1089 | if (!ti->type->dax_copy_from_iter) { | |
1090 | ret = copy_from_iter(addr, bytes, i); | |
1091 | goto out; | |
1092 | } | |
1093 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
1094 | out: | |
1095 | dm_put_live_table(md, srcu_idx); | |
1096 | ||
1097 | return ret; | |
1098 | } | |
1099 | ||
1dd40c3e MP |
1100 | /* |
1101 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
c06b3e58 | 1102 | * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET. |
1dd40c3e MP |
1103 | * |
1104 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1105 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1106 | * sent in a next bio. | |
1107 | * | |
1108 | * A diagram that explains the arithmetics: | |
1109 | * +--------------------+---------------+-------+ | |
1110 | * | 1 | 2 | 3 | | |
1111 | * +--------------------+---------------+-------+ | |
1112 | * | |
1113 | * <-------------- *tio->len_ptr ---------------> | |
1114 | * <------- bi_size -------> | |
1115 | * <-- n_sectors --> | |
1116 | * | |
1117 | * Region 1 was already iterated over with bio_advance or similar function. | |
1118 | * (it may be empty if the target doesn't use bio_advance) | |
1119 | * Region 2 is the remaining bio size that the target wants to process. | |
1120 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1121 | * to make it empty) | |
1122 | * The target requires that region 3 is to be sent in the next bio. | |
1123 | * | |
1124 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1125 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1126 | * copies of the bio. | |
1127 | */ | |
1128 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1129 | { | |
1130 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1131 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1132 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1133 | BUG_ON(bi_size > *tio->len_ptr); |
1134 | BUG_ON(n_sectors > bi_size); | |
1135 | *tio->len_ptr -= bi_size - n_sectors; | |
1136 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1137 | } | |
1138 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1139 | ||
10999307 DLM |
1140 | /* |
1141 | * The zone descriptors obtained with a zone report indicate | |
1142 | * zone positions within the target device. The zone descriptors | |
1143 | * must be remapped to match their position within the dm device. | |
1144 | * A target may call dm_remap_zone_report after completion of a | |
1145 | * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained | |
1146 | * from the target device mapping to the dm device. | |
1147 | */ | |
1148 | void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start) | |
1149 | { | |
1150 | #ifdef CONFIG_BLK_DEV_ZONED | |
1151 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
745dc570 | 1152 | struct bio *report_bio = tio->io->orig_bio; |
10999307 DLM |
1153 | struct blk_zone_report_hdr *hdr = NULL; |
1154 | struct blk_zone *zone; | |
1155 | unsigned int nr_rep = 0; | |
1156 | unsigned int ofst; | |
1157 | struct bio_vec bvec; | |
1158 | struct bvec_iter iter; | |
1159 | void *addr; | |
1160 | ||
1161 | if (bio->bi_status) | |
1162 | return; | |
1163 | ||
1164 | /* | |
1165 | * Remap the start sector of the reported zones. For sequential zones, | |
1166 | * also remap the write pointer position. | |
1167 | */ | |
1168 | bio_for_each_segment(bvec, report_bio, iter) { | |
1169 | addr = kmap_atomic(bvec.bv_page); | |
1170 | ||
1171 | /* Remember the report header in the first page */ | |
1172 | if (!hdr) { | |
1173 | hdr = addr; | |
1174 | ofst = sizeof(struct blk_zone_report_hdr); | |
1175 | } else | |
1176 | ofst = 0; | |
1177 | ||
1178 | /* Set zones start sector */ | |
1179 | while (hdr->nr_zones && ofst < bvec.bv_len) { | |
1180 | zone = addr + ofst; | |
1181 | if (zone->start >= start + ti->len) { | |
1182 | hdr->nr_zones = 0; | |
1183 | break; | |
1184 | } | |
1185 | zone->start = zone->start + ti->begin - start; | |
1186 | if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { | |
1187 | if (zone->cond == BLK_ZONE_COND_FULL) | |
1188 | zone->wp = zone->start + zone->len; | |
1189 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
1190 | zone->wp = zone->start; | |
1191 | else | |
1192 | zone->wp = zone->wp + ti->begin - start; | |
1193 | } | |
1194 | ofst += sizeof(struct blk_zone); | |
1195 | hdr->nr_zones--; | |
1196 | nr_rep++; | |
1197 | } | |
1198 | ||
1199 | if (addr != hdr) | |
1200 | kunmap_atomic(addr); | |
1201 | ||
1202 | if (!hdr->nr_zones) | |
1203 | break; | |
1204 | } | |
1205 | ||
1206 | if (hdr) { | |
1207 | hdr->nr_zones = nr_rep; | |
1208 | kunmap_atomic(hdr); | |
1209 | } | |
1210 | ||
1211 | bio_advance(report_bio, report_bio->bi_iter.bi_size); | |
1212 | ||
1213 | #else /* !CONFIG_BLK_DEV_ZONED */ | |
1214 | bio->bi_status = BLK_STS_NOTSUPP; | |
1215 | #endif | |
1216 | } | |
1217 | EXPORT_SYMBOL_GPL(dm_remap_zone_report); | |
1218 | ||
978e51ba | 1219 | static blk_qc_t __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1220 | { |
1221 | int r; | |
2056a782 | 1222 | sector_t sector; |
dba14160 | 1223 | struct bio *clone = &tio->clone; |
64f52b0e | 1224 | struct dm_io *io = tio->io; |
978e51ba | 1225 | struct mapped_device *md = io->md; |
bd2a49b8 | 1226 | struct dm_target *ti = tio->ti; |
978e51ba | 1227 | blk_qc_t ret = BLK_QC_T_NONE; |
1da177e4 | 1228 | |
1da177e4 | 1229 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1230 | |
1231 | /* | |
1232 | * Map the clone. If r == 0 we don't need to do | |
1233 | * anything, the target has assumed ownership of | |
1234 | * this io. | |
1235 | */ | |
64f52b0e | 1236 | atomic_inc(&io->io_count); |
4f024f37 | 1237 | sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1238 | |
7de3ee57 | 1239 | r = ti->type->map(ti, clone); |
846785e6 CH |
1240 | switch (r) { |
1241 | case DM_MAPIO_SUBMITTED: | |
1242 | break; | |
1243 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1244 | /* the bio has been remapped so dispatch it */ |
74d46992 | 1245 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
64f52b0e | 1246 | bio_dev(io->orig_bio), sector); |
978e51ba MS |
1247 | if (md->type == DM_TYPE_NVME_BIO_BASED) |
1248 | ret = direct_make_request(clone); | |
1249 | else | |
1250 | ret = generic_make_request(clone); | |
846785e6 CH |
1251 | break; |
1252 | case DM_MAPIO_KILL: | |
4e4cbee9 | 1253 | free_tio(tio); |
64f52b0e | 1254 | dec_pending(io, BLK_STS_IOERR); |
4e4cbee9 | 1255 | break; |
846785e6 | 1256 | case DM_MAPIO_REQUEUE: |
cfae7529 | 1257 | free_tio(tio); |
64f52b0e | 1258 | dec_pending(io, BLK_STS_DM_REQUEUE); |
846785e6 CH |
1259 | break; |
1260 | default: | |
45cbcd79 KU |
1261 | DMWARN("unimplemented target map return value: %d", r); |
1262 | BUG(); | |
1da177e4 | 1263 | } |
1da177e4 | 1264 | |
978e51ba | 1265 | return ret; |
1da177e4 | 1266 | } |
1da177e4 | 1267 | |
e0d6609a | 1268 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1269 | { |
4f024f37 KO |
1270 | bio->bi_iter.bi_sector = sector; |
1271 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1272 | } |
1273 | ||
1274 | /* | |
1275 | * Creates a bio that consists of range of complete bvecs. | |
1276 | */ | |
c80914e8 MS |
1277 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1278 | sector_t sector, unsigned len) | |
1da177e4 | 1279 | { |
dba14160 | 1280 | struct bio *clone = &tio->clone; |
1da177e4 | 1281 | |
1c3b13e6 KO |
1282 | __bio_clone_fast(clone, bio); |
1283 | ||
e2460f2a MP |
1284 | if (unlikely(bio_integrity(bio) != NULL)) { |
1285 | int r; | |
1286 | ||
1287 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1288 | !dm_target_passes_integrity(tio->ti->type))) { | |
1289 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1290 | dm_device_name(tio->io->md), | |
1291 | tio->ti->type->name); | |
1292 | return -EIO; | |
1293 | } | |
1294 | ||
1295 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1296 | if (r < 0) |
1297 | return r; | |
1298 | } | |
bd2a49b8 | 1299 | |
264c869d DLM |
1300 | if (bio_op(bio) != REQ_OP_ZONE_REPORT) |
1301 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); | |
1c3b13e6 KO |
1302 | clone->bi_iter.bi_size = to_bytes(len); |
1303 | ||
e2460f2a | 1304 | if (unlikely(bio_integrity(bio) != NULL)) |
fbd08e76 | 1305 | bio_integrity_trim(clone); |
c80914e8 MS |
1306 | |
1307 | return 0; | |
1da177e4 LT |
1308 | } |
1309 | ||
318716dd MS |
1310 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
1311 | struct dm_target *ti, unsigned num_bios) | |
f9ab94ce | 1312 | { |
dba14160 | 1313 | struct dm_target_io *tio; |
318716dd | 1314 | int try; |
dba14160 | 1315 | |
318716dd MS |
1316 | if (!num_bios) |
1317 | return; | |
f9ab94ce | 1318 | |
318716dd MS |
1319 | if (num_bios == 1) { |
1320 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); | |
1321 | bio_list_add(blist, &tio->clone); | |
1322 | return; | |
1323 | } | |
9015df24 | 1324 | |
318716dd MS |
1325 | for (try = 0; try < 2; try++) { |
1326 | int bio_nr; | |
1327 | struct bio *bio; | |
1328 | ||
1329 | if (try) | |
bc02cdbe | 1330 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd MS |
1331 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
1332 | tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); | |
1333 | if (!tio) | |
1334 | break; | |
1335 | ||
1336 | bio_list_add(blist, &tio->clone); | |
1337 | } | |
1338 | if (try) | |
bc02cdbe | 1339 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1340 | if (bio_nr == num_bios) |
1341 | return; | |
1342 | ||
1343 | while ((bio = bio_list_pop(blist))) { | |
1344 | tio = container_of(bio, struct dm_target_io, clone); | |
1345 | free_tio(tio); | |
1346 | } | |
1347 | } | |
9015df24 AK |
1348 | } |
1349 | ||
978e51ba MS |
1350 | static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci, |
1351 | struct dm_target_io *tio, unsigned *len) | |
9015df24 | 1352 | { |
dba14160 | 1353 | struct bio *clone = &tio->clone; |
9015df24 | 1354 | |
1dd40c3e MP |
1355 | tio->len_ptr = len; |
1356 | ||
99778273 | 1357 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1358 | if (len) |
1dd40c3e | 1359 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1360 | |
978e51ba | 1361 | return __map_bio(tio); |
f9ab94ce MP |
1362 | } |
1363 | ||
14fe594d | 1364 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1365 | unsigned num_bios, unsigned *len) |
06a426ce | 1366 | { |
318716dd MS |
1367 | struct bio_list blist = BIO_EMPTY_LIST; |
1368 | struct bio *bio; | |
1369 | struct dm_target_io *tio; | |
1370 | ||
1371 | alloc_multiple_bios(&blist, ci, ti, num_bios); | |
06a426ce | 1372 | |
318716dd MS |
1373 | while ((bio = bio_list_pop(&blist))) { |
1374 | tio = container_of(bio, struct dm_target_io, clone); | |
978e51ba | 1375 | (void) __clone_and_map_simple_bio(ci, tio, len); |
318716dd | 1376 | } |
06a426ce MS |
1377 | } |
1378 | ||
14fe594d | 1379 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1380 | { |
06a426ce | 1381 | unsigned target_nr = 0; |
f9ab94ce MP |
1382 | struct dm_target *ti; |
1383 | ||
b372d360 | 1384 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1385 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1386 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1387 | |
f9ab94ce MP |
1388 | return 0; |
1389 | } | |
1390 | ||
c80914e8 | 1391 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
f31c21e4 | 1392 | sector_t sector, unsigned *len) |
5ae89a87 | 1393 | { |
dba14160 | 1394 | struct bio *bio = ci->bio; |
5ae89a87 | 1395 | struct dm_target_io *tio; |
f31c21e4 | 1396 | int r; |
5ae89a87 | 1397 | |
318716dd | 1398 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); |
f31c21e4 N |
1399 | tio->len_ptr = len; |
1400 | r = clone_bio(tio, bio, sector, *len); | |
1401 | if (r < 0) { | |
1402 | free_tio(tio); | |
1403 | return r; | |
b0d8ed4d | 1404 | } |
978e51ba | 1405 | (void) __map_bio(tio); |
c80914e8 | 1406 | |
f31c21e4 | 1407 | return 0; |
5ae89a87 MS |
1408 | } |
1409 | ||
55a62eef | 1410 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1411 | |
55a62eef | 1412 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1413 | { |
55a62eef | 1414 | return ti->num_discard_bios; |
23508a96 MS |
1415 | } |
1416 | ||
55a62eef | 1417 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1418 | { |
55a62eef | 1419 | return ti->num_write_same_bios; |
23508a96 MS |
1420 | } |
1421 | ||
ac62d620 CH |
1422 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1423 | { | |
1424 | return ti->num_write_zeroes_bios; | |
1425 | } | |
1426 | ||
23508a96 | 1427 | typedef bool (*is_split_required_fn)(struct dm_target *ti); |
9eef87da | 1428 | |
23508a96 MS |
1429 | static bool is_split_required_for_discard(struct dm_target *ti) |
1430 | { | |
55a62eef | 1431 | return ti->split_discard_bios; |
cec47e3d KU |
1432 | } |
1433 | ||
3d7f4562 | 1434 | static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
14fe594d AK |
1435 | get_num_bios_fn get_num_bios, |
1436 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1437 | { |
e0d6609a | 1438 | unsigned len; |
55a62eef | 1439 | unsigned num_bios; |
ba1cbad9 | 1440 | |
3d7f4562 MS |
1441 | /* |
1442 | * Even though the device advertised support for this type of | |
1443 | * request, that does not mean every target supports it, and | |
1444 | * reconfiguration might also have changed that since the | |
1445 | * check was performed. | |
1446 | */ | |
1447 | num_bios = get_num_bios ? get_num_bios(ti) : 0; | |
1448 | if (!num_bios) | |
1449 | return -EOPNOTSUPP; | |
ba1cbad9 | 1450 | |
3d7f4562 MS |
1451 | if (is_split_required && !is_split_required(ti)) |
1452 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); | |
1453 | else | |
1454 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); | |
de3ec86d | 1455 | |
3d7f4562 | 1456 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1457 | |
3d7f4562 MS |
1458 | ci->sector += len; |
1459 | ci->sector_count -= len; | |
5ae89a87 MS |
1460 | |
1461 | return 0; | |
ba1cbad9 MS |
1462 | } |
1463 | ||
3d7f4562 | 1464 | static int __send_discard(struct clone_info *ci, struct dm_target *ti) |
23508a96 | 1465 | { |
3d7f4562 | 1466 | return __send_changing_extent_only(ci, ti, get_num_discard_bios, |
14fe594d | 1467 | is_split_required_for_discard); |
23508a96 | 1468 | } |
0ce65797 | 1469 | |
3d7f4562 | 1470 | static int __send_write_same(struct clone_info *ci, struct dm_target *ti) |
0ce65797 | 1471 | { |
3d7f4562 | 1472 | return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL); |
0ce65797 MS |
1473 | } |
1474 | ||
3d7f4562 | 1475 | static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti) |
ac62d620 | 1476 | { |
3d7f4562 | 1477 | return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL); |
ac62d620 CH |
1478 | } |
1479 | ||
0519c71e MS |
1480 | static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti, |
1481 | int *result) | |
1482 | { | |
1483 | struct bio *bio = ci->bio; | |
1484 | ||
1485 | if (bio_op(bio) == REQ_OP_DISCARD) | |
1486 | *result = __send_discard(ci, ti); | |
1487 | else if (bio_op(bio) == REQ_OP_WRITE_SAME) | |
1488 | *result = __send_write_same(ci, ti); | |
1489 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES) | |
1490 | *result = __send_write_zeroes(ci, ti); | |
1491 | else | |
1492 | return false; | |
1493 | ||
1494 | return true; | |
1495 | } | |
1496 | ||
e4c93811 AK |
1497 | /* |
1498 | * Select the correct strategy for processing a non-flush bio. | |
1499 | */ | |
14fe594d | 1500 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1501 | { |
dba14160 | 1502 | struct bio *bio = ci->bio; |
512875bd | 1503 | struct dm_target *ti; |
1c3b13e6 | 1504 | unsigned len; |
c80914e8 | 1505 | int r; |
0ce65797 | 1506 | |
512875bd JN |
1507 | ti = dm_table_find_target(ci->map, ci->sector); |
1508 | if (!dm_target_is_valid(ti)) | |
1509 | return -EIO; | |
1510 | ||
0519c71e MS |
1511 | if (unlikely(__process_abnormal_io(ci, ti, &r))) |
1512 | return r; | |
3d7f4562 | 1513 | |
264c869d DLM |
1514 | if (bio_op(bio) == REQ_OP_ZONE_REPORT) |
1515 | len = ci->sector_count; | |
1516 | else | |
1517 | len = min_t(sector_t, max_io_len(ci->sector, ti), | |
1518 | ci->sector_count); | |
0ce65797 | 1519 | |
c80914e8 MS |
1520 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1521 | if (r < 0) | |
1522 | return r; | |
0ce65797 | 1523 | |
1c3b13e6 KO |
1524 | ci->sector += len; |
1525 | ci->sector_count -= len; | |
0ce65797 | 1526 | |
1c3b13e6 | 1527 | return 0; |
0ce65797 MS |
1528 | } |
1529 | ||
978e51ba MS |
1530 | static void init_clone_info(struct clone_info *ci, struct mapped_device *md, |
1531 | struct dm_table *map, struct bio *bio) | |
1532 | { | |
1533 | ci->map = map; | |
1534 | ci->io = alloc_io(md, bio); | |
1535 | ci->sector = bio->bi_iter.bi_sector; | |
1536 | } | |
1537 | ||
1da177e4 | 1538 | /* |
14fe594d | 1539 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1540 | */ |
978e51ba MS |
1541 | static blk_qc_t __split_and_process_bio(struct mapped_device *md, |
1542 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1543 | { |
1da177e4 | 1544 | struct clone_info ci; |
978e51ba | 1545 | blk_qc_t ret = BLK_QC_T_NONE; |
512875bd | 1546 | int error = 0; |
1da177e4 | 1547 | |
83d5e5b0 | 1548 | if (unlikely(!map)) { |
6a8736d1 | 1549 | bio_io_error(bio); |
978e51ba | 1550 | return ret; |
f0b9a450 | 1551 | } |
692d0eb9 | 1552 | |
978e51ba | 1553 | init_clone_info(&ci, md, map, bio); |
0ce65797 | 1554 | |
1eff9d32 | 1555 | if (bio->bi_opf & REQ_PREFLUSH) { |
bc02cdbe | 1556 | ci.bio = &ci.io->md->flush_bio; |
b372d360 | 1557 | ci.sector_count = 0; |
14fe594d | 1558 | error = __send_empty_flush(&ci); |
b372d360 | 1559 | /* dec_pending submits any data associated with flush */ |
a4aa5e56 DLM |
1560 | } else if (bio_op(bio) == REQ_OP_ZONE_RESET) { |
1561 | ci.bio = bio; | |
1562 | ci.sector_count = 0; | |
1563 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1564 | } else { |
6a8736d1 | 1565 | ci.bio = bio; |
d87f4c14 | 1566 | ci.sector_count = bio_sectors(bio); |
18a25da8 | 1567 | while (ci.sector_count && !error) { |
14fe594d | 1568 | error = __split_and_process_non_flush(&ci); |
18a25da8 N |
1569 | if (current->bio_list && ci.sector_count && !error) { |
1570 | /* | |
1571 | * Remainder must be passed to generic_make_request() | |
1572 | * so that it gets handled *after* bios already submitted | |
1573 | * have been completely processed. | |
1574 | * We take a clone of the original to store in | |
745dc570 | 1575 | * ci.io->orig_bio to be used by end_io_acct() and |
18a25da8 N |
1576 | * for dec_pending to use for completion handling. |
1577 | * As this path is not used for REQ_OP_ZONE_REPORT, | |
745dc570 | 1578 | * the usage of io->orig_bio in dm_remap_zone_report() |
18a25da8 N |
1579 | * won't be affected by this reassignment. |
1580 | */ | |
1581 | struct bio *b = bio_clone_bioset(bio, GFP_NOIO, | |
1582 | md->queue->bio_split); | |
745dc570 | 1583 | ci.io->orig_bio = b; |
18a25da8 N |
1584 | bio_advance(bio, (bio_sectors(bio) - ci.sector_count) << 9); |
1585 | bio_chain(b, bio); | |
978e51ba | 1586 | ret = generic_make_request(bio); |
18a25da8 N |
1587 | break; |
1588 | } | |
1589 | } | |
d87f4c14 | 1590 | } |
0ce65797 | 1591 | |
1da177e4 | 1592 | /* drop the extra reference count */ |
54385bf7 | 1593 | dec_pending(ci.io, errno_to_blk_status(error)); |
978e51ba | 1594 | return ret; |
0ce65797 MS |
1595 | } |
1596 | ||
cec47e3d | 1597 | /* |
978e51ba MS |
1598 | * Optimized variant of __split_and_process_bio that leverages the |
1599 | * fact that targets that use it do _not_ have a need to split bios. | |
cec47e3d | 1600 | */ |
978e51ba MS |
1601 | static blk_qc_t __process_bio(struct mapped_device *md, |
1602 | struct dm_table *map, struct bio *bio) | |
1603 | { | |
1604 | struct clone_info ci; | |
1605 | blk_qc_t ret = BLK_QC_T_NONE; | |
1606 | int error = 0; | |
1607 | ||
1608 | if (unlikely(!map)) { | |
1609 | bio_io_error(bio); | |
1610 | return ret; | |
1611 | } | |
1612 | ||
1613 | init_clone_info(&ci, md, map, bio); | |
1614 | ||
1615 | if (bio->bi_opf & REQ_PREFLUSH) { | |
1616 | ci.bio = &ci.io->md->flush_bio; | |
1617 | ci.sector_count = 0; | |
1618 | error = __send_empty_flush(&ci); | |
1619 | /* dec_pending submits any data associated with flush */ | |
1620 | } else { | |
1621 | struct dm_target *ti = md->immutable_target; | |
1622 | struct dm_target_io *tio; | |
1623 | ||
1624 | /* | |
1625 | * Defend against IO still getting in during teardown | |
1626 | * - as was seen for a time with nvme-fcloop | |
1627 | */ | |
1628 | if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) { | |
1629 | error = -EIO; | |
1630 | goto out; | |
1631 | } | |
1632 | ||
978e51ba MS |
1633 | ci.bio = bio; |
1634 | ci.sector_count = bio_sectors(bio); | |
0519c71e MS |
1635 | if (unlikely(__process_abnormal_io(&ci, ti, &error))) |
1636 | goto out; | |
1637 | ||
1638 | tio = alloc_tio(&ci, ti, 0, GFP_NOIO); | |
978e51ba MS |
1639 | ret = __clone_and_map_simple_bio(&ci, tio, NULL); |
1640 | } | |
1641 | out: | |
1642 | /* drop the extra reference count */ | |
1643 | dec_pending(ci.io, errno_to_blk_status(error)); | |
1644 | return ret; | |
1645 | } | |
1646 | ||
1647 | typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *); | |
1648 | ||
1649 | static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio, | |
1650 | process_bio_fn process_bio) | |
cec47e3d KU |
1651 | { |
1652 | struct mapped_device *md = q->queuedata; | |
978e51ba | 1653 | blk_qc_t ret = BLK_QC_T_NONE; |
83d5e5b0 MP |
1654 | int srcu_idx; |
1655 | struct dm_table *map; | |
cec47e3d | 1656 | |
83d5e5b0 | 1657 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1658 | |
6a8736d1 TH |
1659 | /* if we're suspended, we have to queue this io for later */ |
1660 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1661 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1662 | |
1eff9d32 | 1663 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1664 | queue_io(md, bio); |
1665 | else | |
54d9a1b4 | 1666 | bio_io_error(bio); |
978e51ba | 1667 | return ret; |
cec47e3d | 1668 | } |
1da177e4 | 1669 | |
978e51ba MS |
1670 | ret = process_bio(md, map, bio); |
1671 | ||
83d5e5b0 | 1672 | dm_put_live_table(md, srcu_idx); |
978e51ba MS |
1673 | return ret; |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * The request function that remaps the bio to one target and | |
1678 | * splits off any remainder. | |
1679 | */ | |
1680 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) | |
1681 | { | |
1682 | return __dm_make_request(q, bio, __split_and_process_bio); | |
1683 | } | |
1684 | ||
1685 | static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio) | |
1686 | { | |
1687 | return __dm_make_request(q, bio, __process_bio); | |
cec47e3d KU |
1688 | } |
1689 | ||
1da177e4 LT |
1690 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1691 | { | |
8a57dfc6 CS |
1692 | int r = bdi_bits; |
1693 | struct mapped_device *md = congested_data; | |
1694 | struct dm_table *map; | |
1da177e4 | 1695 | |
1eb787ec | 1696 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1697 | if (dm_request_based(md)) { |
cec47e3d | 1698 | /* |
e522c039 MS |
1699 | * With request-based DM we only need to check the |
1700 | * top-level queue for congestion. | |
cec47e3d | 1701 | */ |
dc3b17cc | 1702 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1703 | } else { |
1704 | map = dm_get_live_table_fast(md); | |
1705 | if (map) | |
cec47e3d | 1706 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1707 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1708 | } |
1709 | } | |
1710 | ||
1da177e4 LT |
1711 | return r; |
1712 | } | |
1713 | ||
1714 | /*----------------------------------------------------------------- | |
1715 | * An IDR is used to keep track of allocated minor numbers. | |
1716 | *---------------------------------------------------------------*/ | |
2b06cfff | 1717 | static void free_minor(int minor) |
1da177e4 | 1718 | { |
f32c10b0 | 1719 | spin_lock(&_minor_lock); |
1da177e4 | 1720 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1721 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1722 | } |
1723 | ||
1724 | /* | |
1725 | * See if the device with a specific minor # is free. | |
1726 | */ | |
cf13ab8e | 1727 | static int specific_minor(int minor) |
1da177e4 | 1728 | { |
c9d76be6 | 1729 | int r; |
1da177e4 LT |
1730 | |
1731 | if (minor >= (1 << MINORBITS)) | |
1732 | return -EINVAL; | |
1733 | ||
c9d76be6 | 1734 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1735 | spin_lock(&_minor_lock); |
1da177e4 | 1736 | |
c9d76be6 | 1737 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1738 | |
f32c10b0 | 1739 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1740 | idr_preload_end(); |
1741 | if (r < 0) | |
1742 | return r == -ENOSPC ? -EBUSY : r; | |
1743 | return 0; | |
1da177e4 LT |
1744 | } |
1745 | ||
cf13ab8e | 1746 | static int next_free_minor(int *minor) |
1da177e4 | 1747 | { |
c9d76be6 | 1748 | int r; |
62f75c2f | 1749 | |
c9d76be6 | 1750 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1751 | spin_lock(&_minor_lock); |
1da177e4 | 1752 | |
c9d76be6 | 1753 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1754 | |
f32c10b0 | 1755 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1756 | idr_preload_end(); |
1757 | if (r < 0) | |
1758 | return r; | |
1759 | *minor = r; | |
1760 | return 0; | |
1da177e4 LT |
1761 | } |
1762 | ||
83d5cde4 | 1763 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1764 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1765 | |
53d5914f MP |
1766 | static void dm_wq_work(struct work_struct *work); |
1767 | ||
c12c9a3c | 1768 | static void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1769 | { |
17e149b8 | 1770 | md->use_blk_mq = false; |
bfebd1cd MS |
1771 | |
1772 | /* | |
1773 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1774 | */ | |
dc3b17cc | 1775 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf MS |
1776 | } |
1777 | ||
0f20972f MS |
1778 | static void cleanup_mapped_device(struct mapped_device *md) |
1779 | { | |
0f20972f MS |
1780 | if (md->wq) |
1781 | destroy_workqueue(md->wq); | |
1782 | if (md->kworker_task) | |
1783 | kthread_stop(md->kworker_task); | |
0f20972f MS |
1784 | if (md->bs) |
1785 | bioset_free(md->bs); | |
64f52b0e MS |
1786 | if (md->io_bs) |
1787 | bioset_free(md->io_bs); | |
0f20972f | 1788 | |
f26c5719 DW |
1789 | if (md->dax_dev) { |
1790 | kill_dax(md->dax_dev); | |
1791 | put_dax(md->dax_dev); | |
1792 | md->dax_dev = NULL; | |
1793 | } | |
1794 | ||
0f20972f MS |
1795 | if (md->disk) { |
1796 | spin_lock(&_minor_lock); | |
1797 | md->disk->private_data = NULL; | |
1798 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1799 | del_gendisk(md->disk); |
1800 | put_disk(md->disk); | |
1801 | } | |
1802 | ||
1803 | if (md->queue) | |
1804 | blk_cleanup_queue(md->queue); | |
1805 | ||
d09960b0 TE |
1806 | cleanup_srcu_struct(&md->io_barrier); |
1807 | ||
0f20972f MS |
1808 | if (md->bdev) { |
1809 | bdput(md->bdev); | |
1810 | md->bdev = NULL; | |
1811 | } | |
4cc96131 | 1812 | |
d5ffebdd MS |
1813 | mutex_destroy(&md->suspend_lock); |
1814 | mutex_destroy(&md->type_lock); | |
1815 | mutex_destroy(&md->table_devices_lock); | |
1816 | ||
4cc96131 | 1817 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
1818 | } |
1819 | ||
1da177e4 LT |
1820 | /* |
1821 | * Allocate and initialise a blank device with a given minor. | |
1822 | */ | |
2b06cfff | 1823 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1824 | { |
115485e8 | 1825 | int r, numa_node_id = dm_get_numa_node(); |
f26c5719 | 1826 | struct dax_device *dax_dev; |
115485e8 | 1827 | struct mapped_device *md; |
ba61fdd1 | 1828 | void *old_md; |
1da177e4 | 1829 | |
856eb091 | 1830 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1831 | if (!md) { |
1832 | DMWARN("unable to allocate device, out of memory."); | |
1833 | return NULL; | |
1834 | } | |
1835 | ||
10da4f79 | 1836 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1837 | goto bad_module_get; |
10da4f79 | 1838 | |
1da177e4 | 1839 | /* get a minor number for the dev */ |
2b06cfff | 1840 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1841 | r = next_free_minor(&minor); |
2b06cfff | 1842 | else |
cf13ab8e | 1843 | r = specific_minor(minor); |
1da177e4 | 1844 | if (r < 0) |
6ed7ade8 | 1845 | goto bad_minor; |
1da177e4 | 1846 | |
83d5e5b0 MP |
1847 | r = init_srcu_struct(&md->io_barrier); |
1848 | if (r < 0) | |
1849 | goto bad_io_barrier; | |
1850 | ||
115485e8 | 1851 | md->numa_node_id = numa_node_id; |
4cc96131 | 1852 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1853 | md->init_tio_pdu = false; |
a5664dad | 1854 | md->type = DM_TYPE_NONE; |
e61290a4 | 1855 | mutex_init(&md->suspend_lock); |
a5664dad | 1856 | mutex_init(&md->type_lock); |
86f1152b | 1857 | mutex_init(&md->table_devices_lock); |
022c2611 | 1858 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1859 | atomic_set(&md->holders, 1); |
5c6bd75d | 1860 | atomic_set(&md->open_count, 0); |
1da177e4 | 1861 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1862 | atomic_set(&md->uevent_seq, 0); |
1863 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1864 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1865 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1866 | |
115485e8 | 1867 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1868 | if (!md->queue) |
0f20972f | 1869 | goto bad; |
c12c9a3c MS |
1870 | md->queue->queuedata = md; |
1871 | md->queue->backing_dev_info->congested_data = md; | |
1da177e4 | 1872 | |
c12c9a3c | 1873 | md->disk = alloc_disk_node(1, md->numa_node_id); |
1da177e4 | 1874 | if (!md->disk) |
0f20972f | 1875 | goto bad; |
1da177e4 | 1876 | |
316d315b NK |
1877 | atomic_set(&md->pending[0], 0); |
1878 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1879 | init_waitqueue_head(&md->wait); |
53d5914f | 1880 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1881 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1882 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1883 | md->kworker_task = NULL; |
f0b04115 | 1884 | |
1da177e4 LT |
1885 | md->disk->major = _major; |
1886 | md->disk->first_minor = minor; | |
1887 | md->disk->fops = &dm_blk_dops; | |
1888 | md->disk->queue = md->queue; | |
1889 | md->disk->private_data = md; | |
1890 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 DW |
1891 | |
1892 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); | |
1893 | if (!dax_dev) | |
1894 | goto bad; | |
1895 | md->dax_dev = dax_dev; | |
1896 | ||
c100ec49 | 1897 | add_disk_no_queue_reg(md->disk); |
7e51f257 | 1898 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1899 | |
670368a8 | 1900 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1901 | if (!md->wq) |
0f20972f | 1902 | goto bad; |
304f3f6a | 1903 | |
32a926da MP |
1904 | md->bdev = bdget_disk(md->disk, 0); |
1905 | if (!md->bdev) | |
0f20972f | 1906 | goto bad; |
32a926da | 1907 | |
3a83f467 | 1908 | bio_init(&md->flush_bio, NULL, 0); |
74d46992 | 1909 | bio_set_dev(&md->flush_bio, md->bdev); |
ff0361b3 | 1910 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; |
6a8736d1 | 1911 | |
fd2ed4d2 MP |
1912 | dm_stats_init(&md->stats); |
1913 | ||
ba61fdd1 | 1914 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1915 | spin_lock(&_minor_lock); |
ba61fdd1 | 1916 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1917 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1918 | |
1919 | BUG_ON(old_md != MINOR_ALLOCED); | |
1920 | ||
1da177e4 LT |
1921 | return md; |
1922 | ||
0f20972f MS |
1923 | bad: |
1924 | cleanup_mapped_device(md); | |
83d5e5b0 | 1925 | bad_io_barrier: |
1da177e4 | 1926 | free_minor(minor); |
6ed7ade8 | 1927 | bad_minor: |
10da4f79 | 1928 | module_put(THIS_MODULE); |
6ed7ade8 | 1929 | bad_module_get: |
856eb091 | 1930 | kvfree(md); |
1da177e4 LT |
1931 | return NULL; |
1932 | } | |
1933 | ||
ae9da83f JN |
1934 | static void unlock_fs(struct mapped_device *md); |
1935 | ||
1da177e4 LT |
1936 | static void free_dev(struct mapped_device *md) |
1937 | { | |
f331c029 | 1938 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1939 | |
32a926da | 1940 | unlock_fs(md); |
2eb6e1e3 | 1941 | |
0f20972f | 1942 | cleanup_mapped_device(md); |
63a4f065 | 1943 | |
86f1152b | 1944 | free_table_devices(&md->table_devices); |
63a4f065 | 1945 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1946 | free_minor(minor); |
1947 | ||
10da4f79 | 1948 | module_put(THIS_MODULE); |
856eb091 | 1949 | kvfree(md); |
1da177e4 LT |
1950 | } |
1951 | ||
e6ee8c0b KU |
1952 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1953 | { | |
c0820cf5 | 1954 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1955 | |
0776aa0e | 1956 | if (dm_table_bio_based(t)) { |
64f52b0e MS |
1957 | /* |
1958 | * The md may already have mempools that need changing. | |
1959 | * If so, reload bioset because front_pad may have changed | |
1960 | * because a different table was loaded. | |
1961 | */ | |
0776aa0e | 1962 | if (md->bs) { |
16245bdc | 1963 | bioset_free(md->bs); |
0776aa0e | 1964 | md->bs = NULL; |
16245bdc | 1965 | } |
64f52b0e MS |
1966 | if (md->io_bs) { |
1967 | bioset_free(md->io_bs); | |
1968 | md->io_bs = NULL; | |
16245bdc | 1969 | } |
0776aa0e MS |
1970 | |
1971 | } else if (md->bs) { | |
4e6e36c3 MS |
1972 | /* |
1973 | * There's no need to reload with request-based dm | |
1974 | * because the size of front_pad doesn't change. | |
1975 | * Note for future: If you are to reload bioset, | |
1976 | * prep-ed requests in the queue may refer | |
1977 | * to bio from the old bioset, so you must walk | |
1978 | * through the queue to unprep. | |
1979 | */ | |
1980 | goto out; | |
c0820cf5 | 1981 | } |
e6ee8c0b | 1982 | |
dde1e1ec | 1983 | BUG_ON(!p || md->bs || md->io_bs); |
cbc4e3c1 | 1984 | |
e6ee8c0b KU |
1985 | md->bs = p->bs; |
1986 | p->bs = NULL; | |
64f52b0e MS |
1987 | md->io_bs = p->io_bs; |
1988 | p->io_bs = NULL; | |
e6ee8c0b | 1989 | out: |
02233342 | 1990 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1991 | dm_table_free_md_mempools(t); |
1992 | } | |
1993 | ||
1da177e4 LT |
1994 | /* |
1995 | * Bind a table to the device. | |
1996 | */ | |
1997 | static void event_callback(void *context) | |
1998 | { | |
7a8c3d3b MA |
1999 | unsigned long flags; |
2000 | LIST_HEAD(uevents); | |
1da177e4 LT |
2001 | struct mapped_device *md = (struct mapped_device *) context; |
2002 | ||
7a8c3d3b MA |
2003 | spin_lock_irqsave(&md->uevent_lock, flags); |
2004 | list_splice_init(&md->uevent_list, &uevents); | |
2005 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2006 | ||
ed9e1982 | 2007 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2008 | |
1da177e4 LT |
2009 | atomic_inc(&md->event_nr); |
2010 | wake_up(&md->eventq); | |
62e08243 | 2011 | dm_issue_global_event(); |
1da177e4 LT |
2012 | } |
2013 | ||
c217649b MS |
2014 | /* |
2015 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2016 | */ | |
4e90188b | 2017 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2018 | { |
1ea0654e BVA |
2019 | lockdep_assert_held(&md->suspend_lock); |
2020 | ||
4e90188b | 2021 | set_capacity(md->disk, size); |
1da177e4 | 2022 | |
db8fef4f | 2023 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2024 | } |
2025 | ||
042d2a9b AK |
2026 | /* |
2027 | * Returns old map, which caller must destroy. | |
2028 | */ | |
2029 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2030 | struct queue_limits *limits) | |
1da177e4 | 2031 | { |
042d2a9b | 2032 | struct dm_table *old_map; |
165125e1 | 2033 | struct request_queue *q = md->queue; |
978e51ba | 2034 | bool request_based = dm_table_request_based(t); |
1da177e4 LT |
2035 | sector_t size; |
2036 | ||
5a8f1f80 BVA |
2037 | lockdep_assert_held(&md->suspend_lock); |
2038 | ||
1da177e4 | 2039 | size = dm_table_get_size(t); |
3ac51e74 DW |
2040 | |
2041 | /* | |
2042 | * Wipe any geometry if the size of the table changed. | |
2043 | */ | |
fd2ed4d2 | 2044 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2045 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2046 | ||
32a926da | 2047 | __set_size(md, size); |
d5816876 | 2048 | |
2ca3310e AK |
2049 | dm_table_event_callback(t, event_callback, md); |
2050 | ||
e6ee8c0b KU |
2051 | /* |
2052 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2053 | * for request-based during suspension. So stop it to prevent | |
2054 | * I/O mapping before resume. | |
2055 | * This must be done before setting the queue restrictions, | |
2056 | * because request-based dm may be run just after the setting. | |
2057 | */ | |
978e51ba | 2058 | if (request_based) |
eca7ee6d | 2059 | dm_stop_queue(q); |
978e51ba MS |
2060 | |
2061 | if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) { | |
16f12266 | 2062 | /* |
978e51ba MS |
2063 | * Leverage the fact that request-based DM targets and |
2064 | * NVMe bio based targets are immutable singletons | |
2065 | * - used to optimize both dm_request_fn and dm_mq_queue_rq; | |
2066 | * and __process_bio. | |
16f12266 MS |
2067 | */ |
2068 | md->immutable_target = dm_table_get_immutable_target(t); | |
2069 | } | |
e6ee8c0b KU |
2070 | |
2071 | __bind_mempools(md, t); | |
2072 | ||
a12f5d48 | 2073 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2074 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2075 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2076 | ||
754c5fc7 | 2077 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2078 | if (old_map) |
2079 | dm_sync_table(md); | |
1da177e4 | 2080 | |
042d2a9b | 2081 | return old_map; |
1da177e4 LT |
2082 | } |
2083 | ||
a7940155 AK |
2084 | /* |
2085 | * Returns unbound table for the caller to free. | |
2086 | */ | |
2087 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2088 | { |
a12f5d48 | 2089 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2090 | |
2091 | if (!map) | |
a7940155 | 2092 | return NULL; |
1da177e4 LT |
2093 | |
2094 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2095 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2096 | dm_sync_table(md); |
a7940155 AK |
2097 | |
2098 | return map; | |
1da177e4 LT |
2099 | } |
2100 | ||
2101 | /* | |
2102 | * Constructor for a new device. | |
2103 | */ | |
2b06cfff | 2104 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 | 2105 | { |
c12c9a3c | 2106 | int r; |
1da177e4 LT |
2107 | struct mapped_device *md; |
2108 | ||
2b06cfff | 2109 | md = alloc_dev(minor); |
1da177e4 LT |
2110 | if (!md) |
2111 | return -ENXIO; | |
2112 | ||
c12c9a3c MS |
2113 | r = dm_sysfs_init(md); |
2114 | if (r) { | |
2115 | free_dev(md); | |
2116 | return r; | |
2117 | } | |
784aae73 | 2118 | |
1da177e4 LT |
2119 | *result = md; |
2120 | return 0; | |
2121 | } | |
2122 | ||
a5664dad MS |
2123 | /* |
2124 | * Functions to manage md->type. | |
2125 | * All are required to hold md->type_lock. | |
2126 | */ | |
2127 | void dm_lock_md_type(struct mapped_device *md) | |
2128 | { | |
2129 | mutex_lock(&md->type_lock); | |
2130 | } | |
2131 | ||
2132 | void dm_unlock_md_type(struct mapped_device *md) | |
2133 | { | |
2134 | mutex_unlock(&md->type_lock); | |
2135 | } | |
2136 | ||
7e0d574f | 2137 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2138 | { |
00c4fc3b | 2139 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2140 | md->type = type; |
2141 | } | |
2142 | ||
7e0d574f | 2143 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2144 | { |
2145 | return md->type; | |
2146 | } | |
2147 | ||
36a0456f AK |
2148 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2149 | { | |
2150 | return md->immutable_target_type; | |
2151 | } | |
2152 | ||
f84cb8a4 MS |
2153 | /* |
2154 | * The queue_limits are only valid as long as you have a reference | |
2155 | * count on 'md'. | |
2156 | */ | |
2157 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2158 | { | |
2159 | BUG_ON(!atomic_read(&md->holders)); | |
2160 | return &md->queue->limits; | |
2161 | } | |
2162 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2163 | ||
4a0b4ddf MS |
2164 | /* |
2165 | * Setup the DM device's queue based on md's type | |
2166 | */ | |
591ddcfc | 2167 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2168 | { |
bfebd1cd | 2169 | int r; |
c100ec49 | 2170 | struct queue_limits limits; |
7e0d574f | 2171 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2172 | |
545ed20e | 2173 | switch (type) { |
bfebd1cd | 2174 | case DM_TYPE_REQUEST_BASED: |
c12c9a3c | 2175 | dm_init_normal_md_queue(md); |
eb8db831 | 2176 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 2177 | if (r) { |
eca7ee6d | 2178 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2179 | return r; |
ff36ab34 | 2180 | } |
bfebd1cd MS |
2181 | break; |
2182 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 2183 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2184 | if (r) { |
eca7ee6d | 2185 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2186 | return r; |
2187 | } | |
2188 | break; | |
2189 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2190 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 2191 | dm_init_normal_md_queue(md); |
ff36ab34 | 2192 | blk_queue_make_request(md->queue, dm_make_request); |
bfebd1cd | 2193 | break; |
978e51ba MS |
2194 | case DM_TYPE_NVME_BIO_BASED: |
2195 | dm_init_normal_md_queue(md); | |
2196 | blk_queue_make_request(md->queue, dm_make_request_nvme); | |
bfebd1cd | 2197 | break; |
7e0d574f BVA |
2198 | case DM_TYPE_NONE: |
2199 | WARN_ON_ONCE(true); | |
2200 | break; | |
4a0b4ddf MS |
2201 | } |
2202 | ||
c100ec49 MS |
2203 | r = dm_calculate_queue_limits(t, &limits); |
2204 | if (r) { | |
2205 | DMERR("Cannot calculate initial queue limits"); | |
2206 | return r; | |
2207 | } | |
2208 | dm_table_set_restrictions(t, md->queue, &limits); | |
2209 | blk_register_queue(md->disk); | |
2210 | ||
4a0b4ddf MS |
2211 | return 0; |
2212 | } | |
2213 | ||
2bec1f4a | 2214 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2215 | { |
2216 | struct mapped_device *md; | |
1da177e4 LT |
2217 | unsigned minor = MINOR(dev); |
2218 | ||
2219 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2220 | return NULL; | |
2221 | ||
f32c10b0 | 2222 | spin_lock(&_minor_lock); |
1da177e4 LT |
2223 | |
2224 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2225 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2226 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2227 | md = NULL; | |
2228 | goto out; | |
fba9f90e | 2229 | } |
49de5769 | 2230 | dm_get(md); |
fba9f90e | 2231 | out: |
f32c10b0 | 2232 | spin_unlock(&_minor_lock); |
1da177e4 | 2233 | |
637842cf DT |
2234 | return md; |
2235 | } | |
3cf2e4ba | 2236 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2237 | |
9ade92a9 | 2238 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2239 | { |
9ade92a9 | 2240 | return md->interface_ptr; |
1da177e4 LT |
2241 | } |
2242 | ||
2243 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2244 | { | |
2245 | md->interface_ptr = ptr; | |
2246 | } | |
2247 | ||
2248 | void dm_get(struct mapped_device *md) | |
2249 | { | |
2250 | atomic_inc(&md->holders); | |
3f77316d | 2251 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2252 | } |
2253 | ||
09ee96b2 MP |
2254 | int dm_hold(struct mapped_device *md) |
2255 | { | |
2256 | spin_lock(&_minor_lock); | |
2257 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2258 | spin_unlock(&_minor_lock); | |
2259 | return -EBUSY; | |
2260 | } | |
2261 | dm_get(md); | |
2262 | spin_unlock(&_minor_lock); | |
2263 | return 0; | |
2264 | } | |
2265 | EXPORT_SYMBOL_GPL(dm_hold); | |
2266 | ||
72d94861 AK |
2267 | const char *dm_device_name(struct mapped_device *md) |
2268 | { | |
2269 | return md->name; | |
2270 | } | |
2271 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2272 | ||
3f77316d | 2273 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2274 | { |
1134e5ae | 2275 | struct dm_table *map; |
83d5e5b0 | 2276 | int srcu_idx; |
1da177e4 | 2277 | |
3f77316d | 2278 | might_sleep(); |
fba9f90e | 2279 | |
63a4f065 | 2280 | spin_lock(&_minor_lock); |
3f77316d KU |
2281 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2282 | set_bit(DMF_FREEING, &md->flags); | |
2283 | spin_unlock(&_minor_lock); | |
3b785fbc | 2284 | |
c12c9a3c | 2285 | blk_set_queue_dying(md->queue); |
3f77316d | 2286 | |
02233342 | 2287 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 2288 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2289 | |
ab7c7bb6 MP |
2290 | /* |
2291 | * Take suspend_lock so that presuspend and postsuspend methods | |
2292 | * do not race with internal suspend. | |
2293 | */ | |
2294 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2295 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2296 | if (!dm_suspended_md(md)) { |
2297 | dm_table_presuspend_targets(map); | |
2298 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2299 | } |
83d5e5b0 MP |
2300 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2301 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2302 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2303 | |
3f77316d KU |
2304 | /* |
2305 | * Rare, but there may be I/O requests still going to complete, | |
2306 | * for example. Wait for all references to disappear. | |
2307 | * No one should increment the reference count of the mapped_device, | |
2308 | * after the mapped_device state becomes DMF_FREEING. | |
2309 | */ | |
2310 | if (wait) | |
2311 | while (atomic_read(&md->holders)) | |
2312 | msleep(1); | |
2313 | else if (atomic_read(&md->holders)) | |
2314 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2315 | dm_device_name(md), atomic_read(&md->holders)); | |
2316 | ||
2317 | dm_sysfs_exit(md); | |
3f77316d KU |
2318 | dm_table_destroy(__unbind(md)); |
2319 | free_dev(md); | |
2320 | } | |
2321 | ||
2322 | void dm_destroy(struct mapped_device *md) | |
2323 | { | |
2324 | __dm_destroy(md, true); | |
2325 | } | |
2326 | ||
2327 | void dm_destroy_immediate(struct mapped_device *md) | |
2328 | { | |
2329 | __dm_destroy(md, false); | |
2330 | } | |
2331 | ||
2332 | void dm_put(struct mapped_device *md) | |
2333 | { | |
2334 | atomic_dec(&md->holders); | |
1da177e4 | 2335 | } |
79eb885c | 2336 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2337 | |
b48633f8 | 2338 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2339 | { |
2340 | int r = 0; | |
9f4c3f87 | 2341 | DEFINE_WAIT(wait); |
46125c1c MB |
2342 | |
2343 | while (1) { | |
9f4c3f87 | 2344 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2345 | |
b4324fee | 2346 | if (!md_in_flight(md)) |
46125c1c MB |
2347 | break; |
2348 | ||
e3fabdfd | 2349 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2350 | r = -EINTR; |
2351 | break; | |
2352 | } | |
2353 | ||
2354 | io_schedule(); | |
2355 | } | |
9f4c3f87 | 2356 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2357 | |
46125c1c MB |
2358 | return r; |
2359 | } | |
2360 | ||
1da177e4 LT |
2361 | /* |
2362 | * Process the deferred bios | |
2363 | */ | |
ef208587 | 2364 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2365 | { |
ef208587 MP |
2366 | struct mapped_device *md = container_of(work, struct mapped_device, |
2367 | work); | |
6d6f10df | 2368 | struct bio *c; |
83d5e5b0 MP |
2369 | int srcu_idx; |
2370 | struct dm_table *map; | |
1da177e4 | 2371 | |
83d5e5b0 | 2372 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2373 | |
3b00b203 | 2374 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2375 | spin_lock_irq(&md->deferred_lock); |
2376 | c = bio_list_pop(&md->deferred); | |
2377 | spin_unlock_irq(&md->deferred_lock); | |
2378 | ||
6a8736d1 | 2379 | if (!c) |
df12ee99 | 2380 | break; |
022c2611 | 2381 | |
e6ee8c0b KU |
2382 | if (dm_request_based(md)) |
2383 | generic_make_request(c); | |
6a8736d1 | 2384 | else |
83d5e5b0 | 2385 | __split_and_process_bio(md, map, c); |
022c2611 | 2386 | } |
73d410c0 | 2387 | |
83d5e5b0 | 2388 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2389 | } |
2390 | ||
9a1fb464 | 2391 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2392 | { |
3b00b203 | 2393 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2394 | smp_mb__after_atomic(); |
53d5914f | 2395 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2396 | } |
2397 | ||
1da177e4 | 2398 | /* |
042d2a9b | 2399 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2400 | */ |
042d2a9b | 2401 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2402 | { |
87eb5b21 | 2403 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2404 | struct queue_limits limits; |
042d2a9b | 2405 | int r; |
1da177e4 | 2406 | |
e61290a4 | 2407 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2408 | |
2409 | /* device must be suspended */ | |
4f186f8b | 2410 | if (!dm_suspended_md(md)) |
93c534ae | 2411 | goto out; |
1da177e4 | 2412 | |
3ae70656 MS |
2413 | /* |
2414 | * If the new table has no data devices, retain the existing limits. | |
2415 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2416 | * then new I/O is queued based on these limits, and then some paths | |
2417 | * reappear. | |
2418 | */ | |
2419 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2420 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2421 | if (live_map) |
2422 | limits = md->queue->limits; | |
83d5e5b0 | 2423 | dm_put_live_table_fast(md); |
3ae70656 MS |
2424 | } |
2425 | ||
87eb5b21 MC |
2426 | if (!live_map) { |
2427 | r = dm_calculate_queue_limits(table, &limits); | |
2428 | if (r) { | |
2429 | map = ERR_PTR(r); | |
2430 | goto out; | |
2431 | } | |
042d2a9b | 2432 | } |
754c5fc7 | 2433 | |
042d2a9b | 2434 | map = __bind(md, table, &limits); |
62e08243 | 2435 | dm_issue_global_event(); |
1da177e4 | 2436 | |
93c534ae | 2437 | out: |
e61290a4 | 2438 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2439 | return map; |
1da177e4 LT |
2440 | } |
2441 | ||
2442 | /* | |
2443 | * Functions to lock and unlock any filesystem running on the | |
2444 | * device. | |
2445 | */ | |
2ca3310e | 2446 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2447 | { |
e39e2e95 | 2448 | int r; |
1da177e4 LT |
2449 | |
2450 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2451 | |
db8fef4f | 2452 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2453 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2454 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2455 | md->frozen_sb = NULL; |
2456 | return r; | |
dfbe03f6 AK |
2457 | } |
2458 | ||
aa8d7c2f AK |
2459 | set_bit(DMF_FROZEN, &md->flags); |
2460 | ||
1da177e4 LT |
2461 | return 0; |
2462 | } | |
2463 | ||
2ca3310e | 2464 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2465 | { |
aa8d7c2f AK |
2466 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2467 | return; | |
2468 | ||
db8fef4f | 2469 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2470 | md->frozen_sb = NULL; |
aa8d7c2f | 2471 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2472 | } |
2473 | ||
2474 | /* | |
b48633f8 BVA |
2475 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2476 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2477 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2478 | * | |
ffcc3936 MS |
2479 | * If __dm_suspend returns 0, the device is completely quiescent |
2480 | * now. There is no request-processing activity. All new requests | |
2481 | * are being added to md->deferred list. | |
cec47e3d | 2482 | */ |
ffcc3936 | 2483 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2484 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2485 | int dmf_suspended_flag) |
1da177e4 | 2486 | { |
ffcc3936 MS |
2487 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2488 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2489 | int r; | |
1da177e4 | 2490 | |
5a8f1f80 BVA |
2491 | lockdep_assert_held(&md->suspend_lock); |
2492 | ||
2e93ccc1 KU |
2493 | /* |
2494 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2495 | * This flag is cleared before dm_suspend returns. | |
2496 | */ | |
2497 | if (noflush) | |
2498 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 BVA |
2499 | else |
2500 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); | |
2e93ccc1 | 2501 | |
d67ee213 MS |
2502 | /* |
2503 | * This gets reverted if there's an error later and the targets | |
2504 | * provide the .presuspend_undo hook. | |
2505 | */ | |
cf222b37 AK |
2506 | dm_table_presuspend_targets(map); |
2507 | ||
32a926da | 2508 | /* |
9f518b27 KU |
2509 | * Flush I/O to the device. |
2510 | * Any I/O submitted after lock_fs() may not be flushed. | |
2511 | * noflush takes precedence over do_lockfs. | |
2512 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2513 | */ |
2514 | if (!noflush && do_lockfs) { | |
2515 | r = lock_fs(md); | |
d67ee213 MS |
2516 | if (r) { |
2517 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2518 | return r; |
d67ee213 | 2519 | } |
aa8d7c2f | 2520 | } |
1da177e4 LT |
2521 | |
2522 | /* | |
3b00b203 MP |
2523 | * Here we must make sure that no processes are submitting requests |
2524 | * to target drivers i.e. no one may be executing | |
2525 | * __split_and_process_bio. This is called from dm_request and | |
2526 | * dm_wq_work. | |
2527 | * | |
2528 | * To get all processes out of __split_and_process_bio in dm_request, | |
2529 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2530 | * __split_and_process_bio from dm_request and quiesce the thread |
2531 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2532 | * flush_workqueue(md->wq). | |
1da177e4 | 2533 | */ |
1eb787ec | 2534 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2535 | if (map) |
2536 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2537 | |
d0bcb878 | 2538 | /* |
29e4013d TH |
2539 | * Stop md->queue before flushing md->wq in case request-based |
2540 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2541 | */ |
2eb6e1e3 | 2542 | if (dm_request_based(md)) { |
eca7ee6d | 2543 | dm_stop_queue(md->queue); |
02233342 | 2544 | if (md->kworker_task) |
3989144f | 2545 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2546 | } |
cec47e3d | 2547 | |
d0bcb878 KU |
2548 | flush_workqueue(md->wq); |
2549 | ||
1da177e4 | 2550 | /* |
3b00b203 MP |
2551 | * At this point no more requests are entering target request routines. |
2552 | * We call dm_wait_for_completion to wait for all existing requests | |
2553 | * to finish. | |
1da177e4 | 2554 | */ |
b48633f8 | 2555 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2556 | if (!r) |
2557 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2558 | |
6d6f10df | 2559 | if (noflush) |
022c2611 | 2560 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2561 | if (map) |
2562 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2563 | |
1da177e4 | 2564 | /* were we interrupted ? */ |
46125c1c | 2565 | if (r < 0) { |
9a1fb464 | 2566 | dm_queue_flush(md); |
73d410c0 | 2567 | |
cec47e3d | 2568 | if (dm_request_based(md)) |
eca7ee6d | 2569 | dm_start_queue(md->queue); |
cec47e3d | 2570 | |
2ca3310e | 2571 | unlock_fs(md); |
d67ee213 | 2572 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2573 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2574 | } |
1da177e4 | 2575 | |
ffcc3936 MS |
2576 | return r; |
2577 | } | |
2578 | ||
2579 | /* | |
2580 | * We need to be able to change a mapping table under a mounted | |
2581 | * filesystem. For example we might want to move some data in | |
2582 | * the background. Before the table can be swapped with | |
2583 | * dm_bind_table, dm_suspend must be called to flush any in | |
2584 | * flight bios and ensure that any further io gets deferred. | |
2585 | */ | |
2586 | /* | |
2587 | * Suspend mechanism in request-based dm. | |
2588 | * | |
2589 | * 1. Flush all I/Os by lock_fs() if needed. | |
2590 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2591 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2592 | * | |
2593 | * To abort suspend, start the request_queue. | |
2594 | */ | |
2595 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2596 | { | |
2597 | struct dm_table *map = NULL; | |
2598 | int r = 0; | |
2599 | ||
2600 | retry: | |
2601 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2602 | ||
2603 | if (dm_suspended_md(md)) { | |
2604 | r = -EINVAL; | |
2605 | goto out_unlock; | |
2606 | } | |
2607 | ||
2608 | if (dm_suspended_internally_md(md)) { | |
2609 | /* already internally suspended, wait for internal resume */ | |
2610 | mutex_unlock(&md->suspend_lock); | |
2611 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2612 | if (r) | |
2613 | return r; | |
2614 | goto retry; | |
2615 | } | |
2616 | ||
a12f5d48 | 2617 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2618 | |
eaf9a736 | 2619 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2620 | if (r) |
2621 | goto out_unlock; | |
3b00b203 | 2622 | |
4d4471cb KU |
2623 | dm_table_postsuspend_targets(map); |
2624 | ||
d287483d | 2625 | out_unlock: |
e61290a4 | 2626 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2627 | return r; |
1da177e4 LT |
2628 | } |
2629 | ||
ffcc3936 MS |
2630 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2631 | { | |
2632 | if (map) { | |
2633 | int r = dm_table_resume_targets(map); | |
2634 | if (r) | |
2635 | return r; | |
2636 | } | |
2637 | ||
2638 | dm_queue_flush(md); | |
2639 | ||
2640 | /* | |
2641 | * Flushing deferred I/Os must be done after targets are resumed | |
2642 | * so that mapping of targets can work correctly. | |
2643 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2644 | */ | |
2645 | if (dm_request_based(md)) | |
eca7ee6d | 2646 | dm_start_queue(md->queue); |
ffcc3936 MS |
2647 | |
2648 | unlock_fs(md); | |
2649 | ||
2650 | return 0; | |
2651 | } | |
2652 | ||
1da177e4 LT |
2653 | int dm_resume(struct mapped_device *md) |
2654 | { | |
8dc23658 | 2655 | int r; |
cf222b37 | 2656 | struct dm_table *map = NULL; |
1da177e4 | 2657 | |
ffcc3936 | 2658 | retry: |
8dc23658 | 2659 | r = -EINVAL; |
ffcc3936 MS |
2660 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2661 | ||
4f186f8b | 2662 | if (!dm_suspended_md(md)) |
cf222b37 | 2663 | goto out; |
cf222b37 | 2664 | |
ffcc3936 MS |
2665 | if (dm_suspended_internally_md(md)) { |
2666 | /* already internally suspended, wait for internal resume */ | |
2667 | mutex_unlock(&md->suspend_lock); | |
2668 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2669 | if (r) | |
2670 | return r; | |
2671 | goto retry; | |
2672 | } | |
2673 | ||
a12f5d48 | 2674 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2675 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2676 | goto out; |
1da177e4 | 2677 | |
ffcc3936 | 2678 | r = __dm_resume(md, map); |
8757b776 MB |
2679 | if (r) |
2680 | goto out; | |
2ca3310e | 2681 | |
2ca3310e | 2682 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2683 | out: |
e61290a4 | 2684 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2685 | |
cf222b37 | 2686 | return r; |
1da177e4 LT |
2687 | } |
2688 | ||
fd2ed4d2 MP |
2689 | /* |
2690 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2691 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2692 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2693 | */ |
2694 | ||
ffcc3936 | 2695 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2696 | { |
ffcc3936 MS |
2697 | struct dm_table *map = NULL; |
2698 | ||
1ea0654e BVA |
2699 | lockdep_assert_held(&md->suspend_lock); |
2700 | ||
96b26c8c | 2701 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2702 | return; /* nested internal suspend */ |
2703 | ||
2704 | if (dm_suspended_md(md)) { | |
2705 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2706 | return; /* nest suspend */ | |
2707 | } | |
2708 | ||
a12f5d48 | 2709 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2710 | |
2711 | /* | |
2712 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2713 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2714 | * would require changing .presuspend to return an error -- avoid this | |
2715 | * until there is a need for more elaborate variants of internal suspend. | |
2716 | */ | |
eaf9a736 MS |
2717 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2718 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2719 | |
2720 | dm_table_postsuspend_targets(map); | |
2721 | } | |
2722 | ||
2723 | static void __dm_internal_resume(struct mapped_device *md) | |
2724 | { | |
96b26c8c MP |
2725 | BUG_ON(!md->internal_suspend_count); |
2726 | ||
2727 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2728 | return; /* resume from nested internal suspend */ |
2729 | ||
fd2ed4d2 | 2730 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2731 | goto done; /* resume from nested suspend */ |
2732 | ||
2733 | /* | |
2734 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2735 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2736 | */ | |
2737 | (void) __dm_resume(md, NULL); | |
2738 | ||
2739 | done: | |
2740 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2741 | smp_mb__after_atomic(); | |
2742 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2743 | } | |
2744 | ||
2745 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2746 | { | |
2747 | mutex_lock(&md->suspend_lock); | |
2748 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2749 | mutex_unlock(&md->suspend_lock); | |
2750 | } | |
2751 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2752 | ||
2753 | void dm_internal_resume(struct mapped_device *md) | |
2754 | { | |
2755 | mutex_lock(&md->suspend_lock); | |
2756 | __dm_internal_resume(md); | |
2757 | mutex_unlock(&md->suspend_lock); | |
2758 | } | |
2759 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2760 | ||
2761 | /* | |
2762 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2763 | * which prevents interaction with userspace-driven suspend. | |
2764 | */ | |
2765 | ||
2766 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2767 | { | |
2768 | mutex_lock(&md->suspend_lock); | |
2769 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2770 | return; |
2771 | ||
2772 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2773 | synchronize_srcu(&md->io_barrier); | |
2774 | flush_workqueue(md->wq); | |
2775 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2776 | } | |
b735fede | 2777 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2778 | |
ffcc3936 | 2779 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2780 | { |
ffcc3936 | 2781 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2782 | goto done; |
2783 | ||
2784 | dm_queue_flush(md); | |
2785 | ||
2786 | done: | |
2787 | mutex_unlock(&md->suspend_lock); | |
2788 | } | |
b735fede | 2789 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2790 | |
1da177e4 LT |
2791 | /*----------------------------------------------------------------- |
2792 | * Event notification. | |
2793 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2794 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2795 | unsigned cookie) |
69267a30 | 2796 | { |
60935eb2 MB |
2797 | char udev_cookie[DM_COOKIE_LENGTH]; |
2798 | char *envp[] = { udev_cookie, NULL }; | |
2799 | ||
2800 | if (!cookie) | |
3abf85b5 | 2801 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2802 | else { |
2803 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2804 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2805 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2806 | action, envp); | |
60935eb2 | 2807 | } |
69267a30 AK |
2808 | } |
2809 | ||
7a8c3d3b MA |
2810 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2811 | { | |
2812 | return atomic_add_return(1, &md->uevent_seq); | |
2813 | } | |
2814 | ||
1da177e4 LT |
2815 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2816 | { | |
2817 | return atomic_read(&md->event_nr); | |
2818 | } | |
2819 | ||
2820 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2821 | { | |
2822 | return wait_event_interruptible(md->eventq, | |
2823 | (event_nr != atomic_read(&md->event_nr))); | |
2824 | } | |
2825 | ||
7a8c3d3b MA |
2826 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2827 | { | |
2828 | unsigned long flags; | |
2829 | ||
2830 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2831 | list_add(elist, &md->uevent_list); | |
2832 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2833 | } | |
2834 | ||
1da177e4 LT |
2835 | /* |
2836 | * The gendisk is only valid as long as you have a reference | |
2837 | * count on 'md'. | |
2838 | */ | |
2839 | struct gendisk *dm_disk(struct mapped_device *md) | |
2840 | { | |
2841 | return md->disk; | |
2842 | } | |
65ff5b7d | 2843 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2844 | |
784aae73 MB |
2845 | struct kobject *dm_kobject(struct mapped_device *md) |
2846 | { | |
2995fa78 | 2847 | return &md->kobj_holder.kobj; |
784aae73 MB |
2848 | } |
2849 | ||
784aae73 MB |
2850 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2851 | { | |
2852 | struct mapped_device *md; | |
2853 | ||
2995fa78 | 2854 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2855 | |
b9a41d21 HT |
2856 | spin_lock(&_minor_lock); |
2857 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2858 | md = NULL; | |
2859 | goto out; | |
2860 | } | |
784aae73 | 2861 | dm_get(md); |
b9a41d21 HT |
2862 | out: |
2863 | spin_unlock(&_minor_lock); | |
2864 | ||
784aae73 MB |
2865 | return md; |
2866 | } | |
2867 | ||
4f186f8b | 2868 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2869 | { |
2870 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2871 | } | |
2872 | ||
ffcc3936 MS |
2873 | int dm_suspended_internally_md(struct mapped_device *md) |
2874 | { | |
2875 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2876 | } | |
2877 | ||
2c140a24 MP |
2878 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2879 | { | |
2880 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2881 | } | |
2882 | ||
64dbce58 KU |
2883 | int dm_suspended(struct dm_target *ti) |
2884 | { | |
ecdb2e25 | 2885 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2886 | } |
2887 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2888 | ||
2e93ccc1 KU |
2889 | int dm_noflush_suspending(struct dm_target *ti) |
2890 | { | |
ecdb2e25 | 2891 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2892 | } |
2893 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2894 | ||
7e0d574f | 2895 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
0776aa0e MS |
2896 | unsigned integrity, unsigned per_io_data_size, |
2897 | unsigned min_pool_size) | |
e6ee8c0b | 2898 | { |
115485e8 | 2899 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2900 | unsigned int pool_size = 0; |
64f52b0e | 2901 | unsigned int front_pad, io_front_pad; |
e6ee8c0b KU |
2902 | |
2903 | if (!pools) | |
4e6e36c3 | 2904 | return NULL; |
e6ee8c0b | 2905 | |
78d8e58a MS |
2906 | switch (type) { |
2907 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2908 | case DM_TYPE_DAX_BIO_BASED: |
22c11858 | 2909 | case DM_TYPE_NVME_BIO_BASED: |
0776aa0e | 2910 | pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); |
30187e1d | 2911 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
64f52b0e MS |
2912 | io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio); |
2913 | pools->io_bs = bioset_create(pool_size, io_front_pad, 0); | |
2914 | if (!pools->io_bs) | |
2915 | goto out; | |
2916 | if (integrity && bioset_integrity_create(pools->io_bs, pool_size)) | |
eb8db831 | 2917 | goto out; |
78d8e58a MS |
2918 | break; |
2919 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2920 | case DM_TYPE_MQ_REQUEST_BASED: |
0776aa0e | 2921 | pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); |
78d8e58a | 2922 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2923 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2924 | break; |
2925 | default: | |
2926 | BUG(); | |
2927 | } | |
2928 | ||
4a3f54d9 | 2929 | pools->bs = bioset_create(pool_size, front_pad, 0); |
e6ee8c0b | 2930 | if (!pools->bs) |
5f015204 | 2931 | goto out; |
e6ee8c0b | 2932 | |
a91a2785 | 2933 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2934 | goto out; |
a91a2785 | 2935 | |
e6ee8c0b | 2936 | return pools; |
5f1b670d | 2937 | |
5f1b670d CH |
2938 | out: |
2939 | dm_free_md_mempools(pools); | |
78d8e58a | 2940 | |
4e6e36c3 | 2941 | return NULL; |
e6ee8c0b KU |
2942 | } |
2943 | ||
2944 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2945 | { | |
2946 | if (!pools) | |
2947 | return; | |
2948 | ||
e6ee8c0b KU |
2949 | if (pools->bs) |
2950 | bioset_free(pools->bs); | |
64f52b0e MS |
2951 | if (pools->io_bs) |
2952 | bioset_free(pools->io_bs); | |
e6ee8c0b KU |
2953 | |
2954 | kfree(pools); | |
2955 | } | |
2956 | ||
9c72bad1 CH |
2957 | struct dm_pr { |
2958 | u64 old_key; | |
2959 | u64 new_key; | |
2960 | u32 flags; | |
2961 | bool fail_early; | |
2962 | }; | |
2963 | ||
2964 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2965 | void *data) | |
71cdb697 CH |
2966 | { |
2967 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2968 | struct dm_table *table; |
2969 | struct dm_target *ti; | |
2970 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2971 | |
9c72bad1 CH |
2972 | table = dm_get_live_table(md, &srcu_idx); |
2973 | if (!table || !dm_table_get_size(table)) | |
2974 | goto out; | |
71cdb697 | 2975 | |
9c72bad1 CH |
2976 | /* We only support devices that have a single target */ |
2977 | if (dm_table_get_num_targets(table) != 1) | |
2978 | goto out; | |
2979 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2980 | |
9c72bad1 CH |
2981 | ret = -EINVAL; |
2982 | if (!ti->type->iterate_devices) | |
2983 | goto out; | |
2984 | ||
2985 | ret = ti->type->iterate_devices(ti, fn, data); | |
2986 | out: | |
2987 | dm_put_live_table(md, srcu_idx); | |
2988 | return ret; | |
2989 | } | |
2990 | ||
2991 | /* | |
2992 | * For register / unregister we need to manually call out to every path. | |
2993 | */ | |
2994 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
2995 | sector_t start, sector_t len, void *data) | |
2996 | { | |
2997 | struct dm_pr *pr = data; | |
2998 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
2999 | ||
3000 | if (!ops || !ops->pr_register) | |
3001 | return -EOPNOTSUPP; | |
3002 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
3003 | } | |
3004 | ||
3005 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
3006 | u32 flags) | |
3007 | { | |
3008 | struct dm_pr pr = { | |
3009 | .old_key = old_key, | |
3010 | .new_key = new_key, | |
3011 | .flags = flags, | |
3012 | .fail_early = true, | |
3013 | }; | |
3014 | int ret; | |
3015 | ||
3016 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
3017 | if (ret && new_key) { | |
3018 | /* unregister all paths if we failed to register any path */ | |
3019 | pr.old_key = new_key; | |
3020 | pr.new_key = 0; | |
3021 | pr.flags = 0; | |
3022 | pr.fail_early = false; | |
3023 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
3024 | } | |
3025 | ||
3026 | return ret; | |
71cdb697 CH |
3027 | } |
3028 | ||
3029 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3030 | u32 flags) |
71cdb697 CH |
3031 | { |
3032 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3033 | const struct pr_ops *ops; | |
71cdb697 | 3034 | fmode_t mode; |
956a4025 | 3035 | int r; |
71cdb697 | 3036 | |
519049af | 3037 | r = dm_get_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3038 | if (r < 0) |
3039 | return r; | |
3040 | ||
3041 | ops = bdev->bd_disk->fops->pr_ops; | |
3042 | if (ops && ops->pr_reserve) | |
3043 | r = ops->pr_reserve(bdev, key, type, flags); | |
3044 | else | |
3045 | r = -EOPNOTSUPP; | |
3046 | ||
519049af | 3047 | blkdev_put(bdev, mode); |
71cdb697 CH |
3048 | return r; |
3049 | } | |
3050 | ||
3051 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3052 | { | |
3053 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3054 | const struct pr_ops *ops; | |
71cdb697 | 3055 | fmode_t mode; |
956a4025 | 3056 | int r; |
71cdb697 | 3057 | |
519049af | 3058 | r = dm_get_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3059 | if (r < 0) |
3060 | return r; | |
3061 | ||
3062 | ops = bdev->bd_disk->fops->pr_ops; | |
3063 | if (ops && ops->pr_release) | |
3064 | r = ops->pr_release(bdev, key, type); | |
3065 | else | |
3066 | r = -EOPNOTSUPP; | |
3067 | ||
519049af | 3068 | blkdev_put(bdev, mode); |
71cdb697 CH |
3069 | return r; |
3070 | } | |
3071 | ||
3072 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3073 | enum pr_type type, bool abort) |
71cdb697 CH |
3074 | { |
3075 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3076 | const struct pr_ops *ops; | |
71cdb697 | 3077 | fmode_t mode; |
956a4025 | 3078 | int r; |
71cdb697 | 3079 | |
519049af | 3080 | r = dm_get_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3081 | if (r < 0) |
3082 | return r; | |
3083 | ||
3084 | ops = bdev->bd_disk->fops->pr_ops; | |
3085 | if (ops && ops->pr_preempt) | |
3086 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3087 | else | |
3088 | r = -EOPNOTSUPP; | |
3089 | ||
519049af | 3090 | blkdev_put(bdev, mode); |
71cdb697 CH |
3091 | return r; |
3092 | } | |
3093 | ||
3094 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3095 | { | |
3096 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3097 | const struct pr_ops *ops; | |
71cdb697 | 3098 | fmode_t mode; |
956a4025 | 3099 | int r; |
71cdb697 | 3100 | |
519049af | 3101 | r = dm_get_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3102 | if (r < 0) |
3103 | return r; | |
3104 | ||
3105 | ops = bdev->bd_disk->fops->pr_ops; | |
3106 | if (ops && ops->pr_clear) | |
3107 | r = ops->pr_clear(bdev, key); | |
3108 | else | |
3109 | r = -EOPNOTSUPP; | |
3110 | ||
519049af | 3111 | blkdev_put(bdev, mode); |
71cdb697 CH |
3112 | return r; |
3113 | } | |
3114 | ||
3115 | static const struct pr_ops dm_pr_ops = { | |
3116 | .pr_register = dm_pr_register, | |
3117 | .pr_reserve = dm_pr_reserve, | |
3118 | .pr_release = dm_pr_release, | |
3119 | .pr_preempt = dm_pr_preempt, | |
3120 | .pr_clear = dm_pr_clear, | |
3121 | }; | |
3122 | ||
83d5cde4 | 3123 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3124 | .open = dm_blk_open, |
3125 | .release = dm_blk_close, | |
aa129a22 | 3126 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3127 | .getgeo = dm_blk_getgeo, |
71cdb697 | 3128 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3129 | .owner = THIS_MODULE |
3130 | }; | |
3131 | ||
f26c5719 DW |
3132 | static const struct dax_operations dm_dax_ops = { |
3133 | .direct_access = dm_dax_direct_access, | |
7e026c8c | 3134 | .copy_from_iter = dm_dax_copy_from_iter, |
f26c5719 DW |
3135 | }; |
3136 | ||
1da177e4 LT |
3137 | /* |
3138 | * module hooks | |
3139 | */ | |
3140 | module_init(dm_init); | |
3141 | module_exit(dm_exit); | |
3142 | ||
3143 | module_param(major, uint, 0); | |
3144 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3145 | |
e8603136 MS |
3146 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3147 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3148 | ||
115485e8 MS |
3149 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
3150 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
3151 | ||
1da177e4 LT |
3152 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3153 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3154 | MODULE_LICENSE("GPL"); |