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 | ||
956a4025 MS |
461 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
462 | struct block_device **bdev, | |
463 | fmode_t *mode) | |
aa129a22 | 464 | { |
66482026 | 465 | struct dm_target *tgt; |
6c182cd8 | 466 | struct dm_table *map; |
956a4025 | 467 | int srcu_idx, r; |
aa129a22 | 468 | |
6c182cd8 | 469 | retry: |
e56f81e0 | 470 | r = -ENOTTY; |
956a4025 | 471 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
472 | if (!map || !dm_table_get_size(map)) |
473 | goto out; | |
474 | ||
475 | /* We only support devices that have a single target */ | |
476 | if (dm_table_get_num_targets(map) != 1) | |
477 | goto out; | |
478 | ||
66482026 MS |
479 | tgt = dm_table_get_target(map, 0); |
480 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 481 | goto out; |
aa129a22 | 482 | |
4f186f8b | 483 | if (dm_suspended_md(md)) { |
aa129a22 MB |
484 | r = -EAGAIN; |
485 | goto out; | |
486 | } | |
487 | ||
66482026 | 488 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
489 | if (r < 0) |
490 | goto out; | |
aa129a22 | 491 | |
956a4025 MS |
492 | bdgrab(*bdev); |
493 | dm_put_live_table(md, srcu_idx); | |
e56f81e0 | 494 | return r; |
aa129a22 | 495 | |
aa129a22 | 496 | out: |
956a4025 | 497 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 498 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
499 | msleep(10); |
500 | goto retry; | |
501 | } | |
e56f81e0 CH |
502 | return r; |
503 | } | |
504 | ||
505 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
506 | unsigned int cmd, unsigned long arg) | |
507 | { | |
508 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 509 | int r; |
e56f81e0 | 510 | |
956a4025 | 511 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
512 | if (r < 0) |
513 | return r; | |
6c182cd8 | 514 | |
e56f81e0 CH |
515 | if (r > 0) { |
516 | /* | |
e980f623 CH |
517 | * Target determined this ioctl is being issued against a |
518 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 519 | */ |
e980f623 CH |
520 | if (!capable(CAP_SYS_RAWIO)) { |
521 | DMWARN_LIMIT( | |
522 | "%s: sending ioctl %x to DM device without required privilege.", | |
523 | current->comm, cmd); | |
524 | r = -ENOIOCTLCMD; | |
e56f81e0 | 525 | goto out; |
e980f623 | 526 | } |
e56f81e0 | 527 | } |
6c182cd8 | 528 | |
66482026 | 529 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 530 | out: |
956a4025 | 531 | bdput(bdev); |
aa129a22 MB |
532 | return r; |
533 | } | |
534 | ||
978e51ba MS |
535 | static void start_io_acct(struct dm_io *io); |
536 | ||
537 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) | |
1da177e4 | 538 | { |
64f52b0e MS |
539 | struct dm_io *io; |
540 | struct dm_target_io *tio; | |
541 | struct bio *clone; | |
542 | ||
543 | clone = bio_alloc_bioset(GFP_NOIO, 0, md->io_bs); | |
544 | if (!clone) | |
545 | return NULL; | |
546 | ||
547 | tio = container_of(clone, struct dm_target_io, clone); | |
548 | tio->inside_dm_io = true; | |
549 | tio->io = NULL; | |
550 | ||
551 | io = container_of(tio, struct dm_io, tio); | |
552 | io->magic = DM_IO_MAGIC; | |
978e51ba MS |
553 | io->status = 0; |
554 | atomic_set(&io->io_count, 1); | |
555 | io->orig_bio = bio; | |
556 | io->md = md; | |
557 | spin_lock_init(&io->endio_lock); | |
558 | ||
559 | start_io_acct(io); | |
64f52b0e MS |
560 | |
561 | return io; | |
1da177e4 LT |
562 | } |
563 | ||
028867ac | 564 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 | 565 | { |
64f52b0e MS |
566 | bio_put(&io->tio.clone); |
567 | } | |
568 | ||
569 | static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, | |
570 | unsigned target_bio_nr, gfp_t gfp_mask) | |
571 | { | |
572 | struct dm_target_io *tio; | |
573 | ||
574 | if (!ci->io->tio.io) { | |
575 | /* the dm_target_io embedded in ci->io is available */ | |
576 | tio = &ci->io->tio; | |
577 | } else { | |
bc02cdbe | 578 | struct bio *clone = bio_alloc_bioset(gfp_mask, 0, ci->io->md->bs); |
64f52b0e MS |
579 | if (!clone) |
580 | return NULL; | |
581 | ||
582 | tio = container_of(clone, struct dm_target_io, clone); | |
583 | tio->inside_dm_io = false; | |
584 | } | |
585 | ||
586 | tio->magic = DM_TIO_MAGIC; | |
587 | tio->io = ci->io; | |
588 | tio->ti = ti; | |
589 | tio->target_bio_nr = target_bio_nr; | |
590 | ||
591 | return tio; | |
1da177e4 LT |
592 | } |
593 | ||
cfae7529 | 594 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 595 | { |
64f52b0e MS |
596 | if (tio->inside_dm_io) |
597 | return; | |
dba14160 | 598 | bio_put(&tio->clone); |
1da177e4 LT |
599 | } |
600 | ||
4cc96131 | 601 | int md_in_flight(struct mapped_device *md) |
90abb8c4 KU |
602 | { |
603 | return atomic_read(&md->pending[READ]) + | |
604 | atomic_read(&md->pending[WRITE]); | |
605 | } | |
606 | ||
3eaf840e JNN |
607 | static void start_io_acct(struct dm_io *io) |
608 | { | |
609 | struct mapped_device *md = io->md; | |
745dc570 | 610 | struct bio *bio = io->orig_bio; |
fd2ed4d2 | 611 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
612 | |
613 | io->start_time = jiffies; | |
614 | ||
f3986374 MS |
615 | generic_start_io_acct(md->queue, rw, bio_sectors(bio), &dm_disk(md)->part0); |
616 | ||
1e9bb880 | 617 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
f3986374 | 618 | atomic_inc_return(&md->pending[rw])); |
fd2ed4d2 MP |
619 | |
620 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
621 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
622 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
623 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
624 | } |
625 | ||
d221d2e7 | 626 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
627 | { |
628 | struct mapped_device *md = io->md; | |
745dc570 | 629 | struct bio *bio = io->orig_bio; |
3eaf840e | 630 | unsigned long duration = jiffies - io->start_time; |
18c0b223 | 631 | int pending; |
3eaf840e JNN |
632 | int rw = bio_data_dir(bio); |
633 | ||
d62e26b3 | 634 | generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 635 | |
fd2ed4d2 | 636 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
637 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
638 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
639 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 640 | |
af7e466a MP |
641 | /* |
642 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 643 | * a flush. |
af7e466a | 644 | */ |
1e9bb880 SL |
645 | pending = atomic_dec_return(&md->pending[rw]); |
646 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 647 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 648 | |
d221d2e7 MP |
649 | /* nudge anyone waiting on suspend queue */ |
650 | if (!pending) | |
651 | wake_up(&md->wait); | |
3eaf840e JNN |
652 | } |
653 | ||
1da177e4 LT |
654 | /* |
655 | * Add the bio to the list of deferred io. | |
656 | */ | |
92c63902 | 657 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 658 | { |
05447420 | 659 | unsigned long flags; |
1da177e4 | 660 | |
05447420 | 661 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 662 | bio_list_add(&md->deferred, bio); |
05447420 | 663 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 664 | queue_work(md->wq, &md->work); |
1da177e4 LT |
665 | } |
666 | ||
667 | /* | |
668 | * Everyone (including functions in this file), should use this | |
669 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 670 | * dm_put_live_table() when finished. |
1da177e4 | 671 | */ |
83d5e5b0 | 672 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 673 | { |
83d5e5b0 MP |
674 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
675 | ||
676 | return srcu_dereference(md->map, &md->io_barrier); | |
677 | } | |
1da177e4 | 678 | |
83d5e5b0 MP |
679 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
680 | { | |
681 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
682 | } | |
683 | ||
684 | void dm_sync_table(struct mapped_device *md) | |
685 | { | |
686 | synchronize_srcu(&md->io_barrier); | |
687 | synchronize_rcu_expedited(); | |
688 | } | |
689 | ||
690 | /* | |
691 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
692 | * The caller must not block between these two functions. | |
693 | */ | |
694 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
695 | { | |
696 | rcu_read_lock(); | |
697 | return rcu_dereference(md->map); | |
698 | } | |
1da177e4 | 699 | |
83d5e5b0 MP |
700 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
701 | { | |
702 | rcu_read_unlock(); | |
1da177e4 LT |
703 | } |
704 | ||
86f1152b BM |
705 | /* |
706 | * Open a table device so we can use it as a map destination. | |
707 | */ | |
708 | static int open_table_device(struct table_device *td, dev_t dev, | |
709 | struct mapped_device *md) | |
710 | { | |
711 | static char *_claim_ptr = "I belong to device-mapper"; | |
712 | struct block_device *bdev; | |
713 | ||
714 | int r; | |
715 | ||
716 | BUG_ON(td->dm_dev.bdev); | |
717 | ||
718 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
719 | if (IS_ERR(bdev)) | |
720 | return PTR_ERR(bdev); | |
721 | ||
722 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
723 | if (r) { | |
724 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
725 | return r; | |
726 | } | |
727 | ||
728 | td->dm_dev.bdev = bdev; | |
817bf402 | 729 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
730 | return 0; |
731 | } | |
732 | ||
733 | /* | |
734 | * Close a table device that we've been using. | |
735 | */ | |
736 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
737 | { | |
738 | if (!td->dm_dev.bdev) | |
739 | return; | |
740 | ||
741 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
742 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 743 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 744 | td->dm_dev.bdev = NULL; |
817bf402 | 745 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
746 | } |
747 | ||
748 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
749 | fmode_t mode) { | |
750 | struct table_device *td; | |
751 | ||
752 | list_for_each_entry(td, l, list) | |
753 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
754 | return td; | |
755 | ||
756 | return NULL; | |
757 | } | |
758 | ||
759 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
760 | struct dm_dev **result) { | |
761 | int r; | |
762 | struct table_device *td; | |
763 | ||
764 | mutex_lock(&md->table_devices_lock); | |
765 | td = find_table_device(&md->table_devices, dev, mode); | |
766 | if (!td) { | |
115485e8 | 767 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
768 | if (!td) { |
769 | mutex_unlock(&md->table_devices_lock); | |
770 | return -ENOMEM; | |
771 | } | |
772 | ||
773 | td->dm_dev.mode = mode; | |
774 | td->dm_dev.bdev = NULL; | |
775 | ||
776 | if ((r = open_table_device(td, dev, md))) { | |
777 | mutex_unlock(&md->table_devices_lock); | |
778 | kfree(td); | |
779 | return r; | |
780 | } | |
781 | ||
782 | format_dev_t(td->dm_dev.name, dev); | |
783 | ||
b0b4d7c6 | 784 | refcount_set(&td->count, 1); |
86f1152b | 785 | list_add(&td->list, &md->table_devices); |
b0b4d7c6 ER |
786 | } else { |
787 | refcount_inc(&td->count); | |
86f1152b | 788 | } |
86f1152b BM |
789 | mutex_unlock(&md->table_devices_lock); |
790 | ||
791 | *result = &td->dm_dev; | |
792 | return 0; | |
793 | } | |
794 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
795 | ||
796 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
797 | { | |
798 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
799 | ||
800 | mutex_lock(&md->table_devices_lock); | |
b0b4d7c6 | 801 | if (refcount_dec_and_test(&td->count)) { |
86f1152b BM |
802 | close_table_device(td, md); |
803 | list_del(&td->list); | |
804 | kfree(td); | |
805 | } | |
806 | mutex_unlock(&md->table_devices_lock); | |
807 | } | |
808 | EXPORT_SYMBOL(dm_put_table_device); | |
809 | ||
810 | static void free_table_devices(struct list_head *devices) | |
811 | { | |
812 | struct list_head *tmp, *next; | |
813 | ||
814 | list_for_each_safe(tmp, next, devices) { | |
815 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
816 | ||
817 | DMWARN("dm_destroy: %s still exists with %d references", | |
b0b4d7c6 | 818 | td->dm_dev.name, refcount_read(&td->count)); |
86f1152b BM |
819 | kfree(td); |
820 | } | |
821 | } | |
822 | ||
3ac51e74 DW |
823 | /* |
824 | * Get the geometry associated with a dm device | |
825 | */ | |
826 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
827 | { | |
828 | *geo = md->geometry; | |
829 | ||
830 | return 0; | |
831 | } | |
832 | ||
833 | /* | |
834 | * Set the geometry of a device. | |
835 | */ | |
836 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
837 | { | |
838 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
839 | ||
840 | if (geo->start > sz) { | |
841 | DMWARN("Start sector is beyond the geometry limits."); | |
842 | return -EINVAL; | |
843 | } | |
844 | ||
845 | md->geometry = *geo; | |
846 | ||
847 | return 0; | |
848 | } | |
849 | ||
2e93ccc1 KU |
850 | static int __noflush_suspending(struct mapped_device *md) |
851 | { | |
852 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
853 | } | |
854 | ||
1da177e4 LT |
855 | /* |
856 | * Decrements the number of outstanding ios that a bio has been | |
857 | * cloned into, completing the original io if necc. | |
858 | */ | |
4e4cbee9 | 859 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 860 | { |
2e93ccc1 | 861 | unsigned long flags; |
4e4cbee9 | 862 | blk_status_t io_error; |
b35f8caa MB |
863 | struct bio *bio; |
864 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
865 | |
866 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
867 | if (unlikely(error)) { |
868 | spin_lock_irqsave(&io->endio_lock, flags); | |
745dc570 | 869 | if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) |
4e4cbee9 | 870 | io->status = error; |
f88fb981 KU |
871 | spin_unlock_irqrestore(&io->endio_lock, flags); |
872 | } | |
1da177e4 LT |
873 | |
874 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 875 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
876 | /* |
877 | * Target requested pushing back the I/O. | |
2e93ccc1 | 878 | */ |
022c2611 | 879 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 | 880 | if (__noflush_suspending(md)) |
745dc570 MS |
881 | /* NOTE early return due to BLK_STS_DM_REQUEUE below */ |
882 | bio_list_add_head(&md->deferred, io->orig_bio); | |
6a8736d1 | 883 | else |
2e93ccc1 | 884 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 885 | io->status = BLK_STS_IOERR; |
022c2611 | 886 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
887 | } |
888 | ||
4e4cbee9 | 889 | io_error = io->status; |
745dc570 | 890 | bio = io->orig_bio; |
6a8736d1 TH |
891 | end_io_acct(io); |
892 | free_io(md, io); | |
893 | ||
4e4cbee9 | 894 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 895 | return; |
2e93ccc1 | 896 | |
1eff9d32 | 897 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 898 | /* |
6a8736d1 | 899 | * Preflush done for flush with data, reissue |
28a8f0d3 | 900 | * without REQ_PREFLUSH. |
af7e466a | 901 | */ |
1eff9d32 | 902 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 903 | queue_io(md, bio); |
af7e466a | 904 | } else { |
b372d360 | 905 | /* done with normal IO or empty flush */ |
8dd601fa N |
906 | if (io_error) |
907 | bio->bi_status = io_error; | |
4246a0b6 | 908 | bio_endio(bio); |
b35f8caa | 909 | } |
1da177e4 LT |
910 | } |
911 | } | |
912 | ||
4cc96131 | 913 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
914 | { |
915 | struct queue_limits *limits = dm_get_queue_limits(md); | |
916 | ||
917 | /* device doesn't really support WRITE SAME, disable it */ | |
918 | limits->max_write_same_sectors = 0; | |
919 | } | |
920 | ||
ac62d620 CH |
921 | void disable_write_zeroes(struct mapped_device *md) |
922 | { | |
923 | struct queue_limits *limits = dm_get_queue_limits(md); | |
924 | ||
925 | /* device doesn't really support WRITE ZEROES, disable it */ | |
926 | limits->max_write_zeroes_sectors = 0; | |
927 | } | |
928 | ||
4246a0b6 | 929 | static void clone_endio(struct bio *bio) |
1da177e4 | 930 | { |
4e4cbee9 | 931 | blk_status_t error = bio->bi_status; |
bfc6d41c | 932 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 933 | struct dm_io *io = tio->io; |
9faf400f | 934 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
935 | dm_endio_fn endio = tio->ti->type->end_io; |
936 | ||
978e51ba | 937 | if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) { |
ac62d620 | 938 | if (bio_op(bio) == REQ_OP_WRITE_SAME && |
74d46992 | 939 | !bio->bi_disk->queue->limits.max_write_same_sectors) |
ac62d620 CH |
940 | disable_write_same(md); |
941 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && | |
74d46992 | 942 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) |
ac62d620 CH |
943 | disable_write_zeroes(md); |
944 | } | |
7eee4ae2 | 945 | |
1be56909 | 946 | if (endio) { |
4e4cbee9 | 947 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
948 | switch (r) { |
949 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 950 | error = BLK_STS_DM_REQUEUE; |
1be56909 CH |
951 | /*FALLTHRU*/ |
952 | case DM_ENDIO_DONE: | |
953 | break; | |
954 | case DM_ENDIO_INCOMPLETE: | |
955 | /* The target will handle the io */ | |
956 | return; | |
957 | default: | |
958 | DMWARN("unimplemented target endio return value: %d", r); | |
959 | BUG(); | |
960 | } | |
961 | } | |
962 | ||
cfae7529 | 963 | free_tio(tio); |
b35f8caa | 964 | dec_pending(io, error); |
1da177e4 LT |
965 | } |
966 | ||
56a67df7 MS |
967 | /* |
968 | * Return maximum size of I/O possible at the supplied sector up to the current | |
969 | * target boundary. | |
970 | */ | |
971 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
972 | { | |
973 | sector_t target_offset = dm_target_offset(ti, sector); | |
974 | ||
975 | return ti->len - target_offset; | |
976 | } | |
977 | ||
978 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 979 | { |
56a67df7 | 980 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 981 | sector_t offset, max_len; |
1da177e4 LT |
982 | |
983 | /* | |
542f9038 | 984 | * Does the target need to split even further? |
1da177e4 | 985 | */ |
542f9038 MS |
986 | if (ti->max_io_len) { |
987 | offset = dm_target_offset(ti, sector); | |
988 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
989 | max_len = sector_div(offset, ti->max_io_len); | |
990 | else | |
991 | max_len = offset & (ti->max_io_len - 1); | |
992 | max_len = ti->max_io_len - max_len; | |
993 | ||
994 | if (len > max_len) | |
995 | len = max_len; | |
1da177e4 LT |
996 | } |
997 | ||
998 | return len; | |
999 | } | |
1000 | ||
542f9038 MS |
1001 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1002 | { | |
1003 | if (len > UINT_MAX) { | |
1004 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1005 | (unsigned long long)len, UINT_MAX); | |
1006 | ti->error = "Maximum size of target IO is too large"; | |
1007 | return -EINVAL; | |
1008 | } | |
1009 | ||
8f50e358 ML |
1010 | /* |
1011 | * BIO based queue uses its own splitting. When multipage bvecs | |
1012 | * is switched on, size of the incoming bio may be too big to | |
1013 | * be handled in some targets, such as crypt. | |
1014 | * | |
1015 | * When these targets are ready for the big bio, we can remove | |
1016 | * the limit. | |
1017 | */ | |
1018 | ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE); | |
542f9038 MS |
1019 | |
1020 | return 0; | |
1021 | } | |
1022 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1023 | ||
f26c5719 DW |
1024 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
1025 | sector_t sector, int *srcu_idx) | |
545ed20e | 1026 | { |
545ed20e TK |
1027 | struct dm_table *map; |
1028 | struct dm_target *ti; | |
545ed20e | 1029 | |
f26c5719 | 1030 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1031 | if (!map) |
f26c5719 | 1032 | return NULL; |
545ed20e TK |
1033 | |
1034 | ti = dm_table_find_target(map, sector); | |
1035 | if (!dm_target_is_valid(ti)) | |
f26c5719 | 1036 | return NULL; |
545ed20e | 1037 | |
f26c5719 DW |
1038 | return ti; |
1039 | } | |
545ed20e | 1040 | |
f26c5719 DW |
1041 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
1042 | long nr_pages, void **kaddr, pfn_t *pfn) | |
1043 | { | |
1044 | struct mapped_device *md = dax_get_private(dax_dev); | |
1045 | sector_t sector = pgoff * PAGE_SECTORS; | |
1046 | struct dm_target *ti; | |
1047 | long len, ret = -EIO; | |
1048 | int srcu_idx; | |
545ed20e | 1049 | |
f26c5719 | 1050 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1051 | |
f26c5719 DW |
1052 | if (!ti) |
1053 | goto out; | |
1054 | if (!ti->type->direct_access) | |
1055 | goto out; | |
1056 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
1057 | if (len < 1) | |
1058 | goto out; | |
1059 | nr_pages = min(len, nr_pages); | |
545ed20e | 1060 | if (ti->type->direct_access) |
817bf402 DW |
1061 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
1062 | ||
f26c5719 | 1063 | out: |
545ed20e | 1064 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1065 | |
1066 | return ret; | |
545ed20e TK |
1067 | } |
1068 | ||
7e026c8c DW |
1069 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
1070 | void *addr, size_t bytes, struct iov_iter *i) | |
1071 | { | |
1072 | struct mapped_device *md = dax_get_private(dax_dev); | |
1073 | sector_t sector = pgoff * PAGE_SECTORS; | |
1074 | struct dm_target *ti; | |
1075 | long ret = 0; | |
1076 | int srcu_idx; | |
1077 | ||
1078 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1079 | ||
1080 | if (!ti) | |
1081 | goto out; | |
1082 | if (!ti->type->dax_copy_from_iter) { | |
1083 | ret = copy_from_iter(addr, bytes, i); | |
1084 | goto out; | |
1085 | } | |
1086 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
1087 | out: | |
1088 | dm_put_live_table(md, srcu_idx); | |
1089 | ||
1090 | return ret; | |
1091 | } | |
1092 | ||
1dd40c3e MP |
1093 | /* |
1094 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
c06b3e58 | 1095 | * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET. |
1dd40c3e MP |
1096 | * |
1097 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1098 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1099 | * sent in a next bio. | |
1100 | * | |
1101 | * A diagram that explains the arithmetics: | |
1102 | * +--------------------+---------------+-------+ | |
1103 | * | 1 | 2 | 3 | | |
1104 | * +--------------------+---------------+-------+ | |
1105 | * | |
1106 | * <-------------- *tio->len_ptr ---------------> | |
1107 | * <------- bi_size -------> | |
1108 | * <-- n_sectors --> | |
1109 | * | |
1110 | * Region 1 was already iterated over with bio_advance or similar function. | |
1111 | * (it may be empty if the target doesn't use bio_advance) | |
1112 | * Region 2 is the remaining bio size that the target wants to process. | |
1113 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1114 | * to make it empty) | |
1115 | * The target requires that region 3 is to be sent in the next bio. | |
1116 | * | |
1117 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1118 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1119 | * copies of the bio. | |
1120 | */ | |
1121 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1122 | { | |
1123 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1124 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1125 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1126 | BUG_ON(bi_size > *tio->len_ptr); |
1127 | BUG_ON(n_sectors > bi_size); | |
1128 | *tio->len_ptr -= bi_size - n_sectors; | |
1129 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1130 | } | |
1131 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1132 | ||
10999307 DLM |
1133 | /* |
1134 | * The zone descriptors obtained with a zone report indicate | |
1135 | * zone positions within the target device. The zone descriptors | |
1136 | * must be remapped to match their position within the dm device. | |
1137 | * A target may call dm_remap_zone_report after completion of a | |
1138 | * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained | |
1139 | * from the target device mapping to the dm device. | |
1140 | */ | |
1141 | void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start) | |
1142 | { | |
1143 | #ifdef CONFIG_BLK_DEV_ZONED | |
1144 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
745dc570 | 1145 | struct bio *report_bio = tio->io->orig_bio; |
10999307 DLM |
1146 | struct blk_zone_report_hdr *hdr = NULL; |
1147 | struct blk_zone *zone; | |
1148 | unsigned int nr_rep = 0; | |
1149 | unsigned int ofst; | |
1150 | struct bio_vec bvec; | |
1151 | struct bvec_iter iter; | |
1152 | void *addr; | |
1153 | ||
1154 | if (bio->bi_status) | |
1155 | return; | |
1156 | ||
1157 | /* | |
1158 | * Remap the start sector of the reported zones. For sequential zones, | |
1159 | * also remap the write pointer position. | |
1160 | */ | |
1161 | bio_for_each_segment(bvec, report_bio, iter) { | |
1162 | addr = kmap_atomic(bvec.bv_page); | |
1163 | ||
1164 | /* Remember the report header in the first page */ | |
1165 | if (!hdr) { | |
1166 | hdr = addr; | |
1167 | ofst = sizeof(struct blk_zone_report_hdr); | |
1168 | } else | |
1169 | ofst = 0; | |
1170 | ||
1171 | /* Set zones start sector */ | |
1172 | while (hdr->nr_zones && ofst < bvec.bv_len) { | |
1173 | zone = addr + ofst; | |
1174 | if (zone->start >= start + ti->len) { | |
1175 | hdr->nr_zones = 0; | |
1176 | break; | |
1177 | } | |
1178 | zone->start = zone->start + ti->begin - start; | |
1179 | if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { | |
1180 | if (zone->cond == BLK_ZONE_COND_FULL) | |
1181 | zone->wp = zone->start + zone->len; | |
1182 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
1183 | zone->wp = zone->start; | |
1184 | else | |
1185 | zone->wp = zone->wp + ti->begin - start; | |
1186 | } | |
1187 | ofst += sizeof(struct blk_zone); | |
1188 | hdr->nr_zones--; | |
1189 | nr_rep++; | |
1190 | } | |
1191 | ||
1192 | if (addr != hdr) | |
1193 | kunmap_atomic(addr); | |
1194 | ||
1195 | if (!hdr->nr_zones) | |
1196 | break; | |
1197 | } | |
1198 | ||
1199 | if (hdr) { | |
1200 | hdr->nr_zones = nr_rep; | |
1201 | kunmap_atomic(hdr); | |
1202 | } | |
1203 | ||
1204 | bio_advance(report_bio, report_bio->bi_iter.bi_size); | |
1205 | ||
1206 | #else /* !CONFIG_BLK_DEV_ZONED */ | |
1207 | bio->bi_status = BLK_STS_NOTSUPP; | |
1208 | #endif | |
1209 | } | |
1210 | EXPORT_SYMBOL_GPL(dm_remap_zone_report); | |
1211 | ||
978e51ba | 1212 | static blk_qc_t __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1213 | { |
1214 | int r; | |
2056a782 | 1215 | sector_t sector; |
dba14160 | 1216 | struct bio *clone = &tio->clone; |
64f52b0e | 1217 | struct dm_io *io = tio->io; |
978e51ba | 1218 | struct mapped_device *md = io->md; |
bd2a49b8 | 1219 | struct dm_target *ti = tio->ti; |
978e51ba | 1220 | blk_qc_t ret = BLK_QC_T_NONE; |
1da177e4 | 1221 | |
1da177e4 | 1222 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1223 | |
1224 | /* | |
1225 | * Map the clone. If r == 0 we don't need to do | |
1226 | * anything, the target has assumed ownership of | |
1227 | * this io. | |
1228 | */ | |
64f52b0e | 1229 | atomic_inc(&io->io_count); |
4f024f37 | 1230 | sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1231 | |
7de3ee57 | 1232 | r = ti->type->map(ti, clone); |
846785e6 CH |
1233 | switch (r) { |
1234 | case DM_MAPIO_SUBMITTED: | |
1235 | break; | |
1236 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1237 | /* the bio has been remapped so dispatch it */ |
74d46992 | 1238 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
64f52b0e | 1239 | bio_dev(io->orig_bio), sector); |
978e51ba MS |
1240 | if (md->type == DM_TYPE_NVME_BIO_BASED) |
1241 | ret = direct_make_request(clone); | |
1242 | else | |
1243 | ret = generic_make_request(clone); | |
846785e6 CH |
1244 | break; |
1245 | case DM_MAPIO_KILL: | |
4e4cbee9 | 1246 | free_tio(tio); |
64f52b0e | 1247 | dec_pending(io, BLK_STS_IOERR); |
4e4cbee9 | 1248 | break; |
846785e6 | 1249 | case DM_MAPIO_REQUEUE: |
cfae7529 | 1250 | free_tio(tio); |
64f52b0e | 1251 | dec_pending(io, BLK_STS_DM_REQUEUE); |
846785e6 CH |
1252 | break; |
1253 | default: | |
45cbcd79 KU |
1254 | DMWARN("unimplemented target map return value: %d", r); |
1255 | BUG(); | |
1da177e4 | 1256 | } |
1da177e4 | 1257 | |
978e51ba | 1258 | return ret; |
1da177e4 | 1259 | } |
1da177e4 | 1260 | |
e0d6609a | 1261 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1262 | { |
4f024f37 KO |
1263 | bio->bi_iter.bi_sector = sector; |
1264 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1265 | } |
1266 | ||
1267 | /* | |
1268 | * Creates a bio that consists of range of complete bvecs. | |
1269 | */ | |
c80914e8 MS |
1270 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1271 | sector_t sector, unsigned len) | |
1da177e4 | 1272 | { |
dba14160 | 1273 | struct bio *clone = &tio->clone; |
1da177e4 | 1274 | |
1c3b13e6 KO |
1275 | __bio_clone_fast(clone, bio); |
1276 | ||
e2460f2a MP |
1277 | if (unlikely(bio_integrity(bio) != NULL)) { |
1278 | int r; | |
1279 | ||
1280 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1281 | !dm_target_passes_integrity(tio->ti->type))) { | |
1282 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1283 | dm_device_name(tio->io->md), | |
1284 | tio->ti->type->name); | |
1285 | return -EIO; | |
1286 | } | |
1287 | ||
1288 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1289 | if (r < 0) |
1290 | return r; | |
1291 | } | |
bd2a49b8 | 1292 | |
264c869d DLM |
1293 | if (bio_op(bio) != REQ_OP_ZONE_REPORT) |
1294 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); | |
1c3b13e6 KO |
1295 | clone->bi_iter.bi_size = to_bytes(len); |
1296 | ||
e2460f2a | 1297 | if (unlikely(bio_integrity(bio) != NULL)) |
fbd08e76 | 1298 | bio_integrity_trim(clone); |
c80914e8 MS |
1299 | |
1300 | return 0; | |
1da177e4 LT |
1301 | } |
1302 | ||
318716dd MS |
1303 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
1304 | struct dm_target *ti, unsigned num_bios) | |
f9ab94ce | 1305 | { |
dba14160 | 1306 | struct dm_target_io *tio; |
318716dd | 1307 | int try; |
dba14160 | 1308 | |
318716dd MS |
1309 | if (!num_bios) |
1310 | return; | |
f9ab94ce | 1311 | |
318716dd MS |
1312 | if (num_bios == 1) { |
1313 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); | |
1314 | bio_list_add(blist, &tio->clone); | |
1315 | return; | |
1316 | } | |
9015df24 | 1317 | |
318716dd MS |
1318 | for (try = 0; try < 2; try++) { |
1319 | int bio_nr; | |
1320 | struct bio *bio; | |
1321 | ||
1322 | if (try) | |
bc02cdbe | 1323 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd MS |
1324 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
1325 | tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); | |
1326 | if (!tio) | |
1327 | break; | |
1328 | ||
1329 | bio_list_add(blist, &tio->clone); | |
1330 | } | |
1331 | if (try) | |
bc02cdbe | 1332 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1333 | if (bio_nr == num_bios) |
1334 | return; | |
1335 | ||
1336 | while ((bio = bio_list_pop(blist))) { | |
1337 | tio = container_of(bio, struct dm_target_io, clone); | |
1338 | free_tio(tio); | |
1339 | } | |
1340 | } | |
9015df24 AK |
1341 | } |
1342 | ||
978e51ba MS |
1343 | static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci, |
1344 | struct dm_target_io *tio, unsigned *len) | |
9015df24 | 1345 | { |
dba14160 | 1346 | struct bio *clone = &tio->clone; |
9015df24 | 1347 | |
1dd40c3e MP |
1348 | tio->len_ptr = len; |
1349 | ||
99778273 | 1350 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1351 | if (len) |
1dd40c3e | 1352 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1353 | |
978e51ba | 1354 | return __map_bio(tio); |
f9ab94ce MP |
1355 | } |
1356 | ||
14fe594d | 1357 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1358 | unsigned num_bios, unsigned *len) |
06a426ce | 1359 | { |
318716dd MS |
1360 | struct bio_list blist = BIO_EMPTY_LIST; |
1361 | struct bio *bio; | |
1362 | struct dm_target_io *tio; | |
1363 | ||
1364 | alloc_multiple_bios(&blist, ci, ti, num_bios); | |
06a426ce | 1365 | |
318716dd MS |
1366 | while ((bio = bio_list_pop(&blist))) { |
1367 | tio = container_of(bio, struct dm_target_io, clone); | |
978e51ba | 1368 | (void) __clone_and_map_simple_bio(ci, tio, len); |
318716dd | 1369 | } |
06a426ce MS |
1370 | } |
1371 | ||
14fe594d | 1372 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1373 | { |
06a426ce | 1374 | unsigned target_nr = 0; |
f9ab94ce MP |
1375 | struct dm_target *ti; |
1376 | ||
b372d360 | 1377 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1378 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1379 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1380 | |
f9ab94ce MP |
1381 | return 0; |
1382 | } | |
1383 | ||
c80914e8 | 1384 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
f31c21e4 | 1385 | sector_t sector, unsigned *len) |
5ae89a87 | 1386 | { |
dba14160 | 1387 | struct bio *bio = ci->bio; |
5ae89a87 | 1388 | struct dm_target_io *tio; |
f31c21e4 | 1389 | int r; |
5ae89a87 | 1390 | |
318716dd | 1391 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); |
f31c21e4 N |
1392 | tio->len_ptr = len; |
1393 | r = clone_bio(tio, bio, sector, *len); | |
1394 | if (r < 0) { | |
1395 | free_tio(tio); | |
1396 | return r; | |
b0d8ed4d | 1397 | } |
978e51ba | 1398 | (void) __map_bio(tio); |
c80914e8 | 1399 | |
f31c21e4 | 1400 | return 0; |
5ae89a87 MS |
1401 | } |
1402 | ||
55a62eef | 1403 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1404 | |
55a62eef | 1405 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1406 | { |
55a62eef | 1407 | return ti->num_discard_bios; |
23508a96 MS |
1408 | } |
1409 | ||
55a62eef | 1410 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1411 | { |
55a62eef | 1412 | return ti->num_write_same_bios; |
23508a96 MS |
1413 | } |
1414 | ||
ac62d620 CH |
1415 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1416 | { | |
1417 | return ti->num_write_zeroes_bios; | |
1418 | } | |
1419 | ||
23508a96 | 1420 | typedef bool (*is_split_required_fn)(struct dm_target *ti); |
9eef87da | 1421 | |
23508a96 MS |
1422 | static bool is_split_required_for_discard(struct dm_target *ti) |
1423 | { | |
55a62eef | 1424 | return ti->split_discard_bios; |
cec47e3d KU |
1425 | } |
1426 | ||
3d7f4562 | 1427 | static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
14fe594d AK |
1428 | get_num_bios_fn get_num_bios, |
1429 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1430 | { |
e0d6609a | 1431 | unsigned len; |
55a62eef | 1432 | unsigned num_bios; |
ba1cbad9 | 1433 | |
3d7f4562 MS |
1434 | /* |
1435 | * Even though the device advertised support for this type of | |
1436 | * request, that does not mean every target supports it, and | |
1437 | * reconfiguration might also have changed that since the | |
1438 | * check was performed. | |
1439 | */ | |
1440 | num_bios = get_num_bios ? get_num_bios(ti) : 0; | |
1441 | if (!num_bios) | |
1442 | return -EOPNOTSUPP; | |
ba1cbad9 | 1443 | |
3d7f4562 MS |
1444 | if (is_split_required && !is_split_required(ti)) |
1445 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); | |
1446 | else | |
1447 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); | |
de3ec86d | 1448 | |
3d7f4562 | 1449 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1450 | |
3d7f4562 MS |
1451 | ci->sector += len; |
1452 | ci->sector_count -= len; | |
5ae89a87 MS |
1453 | |
1454 | return 0; | |
ba1cbad9 MS |
1455 | } |
1456 | ||
3d7f4562 | 1457 | static int __send_discard(struct clone_info *ci, struct dm_target *ti) |
23508a96 | 1458 | { |
3d7f4562 | 1459 | return __send_changing_extent_only(ci, ti, get_num_discard_bios, |
14fe594d | 1460 | is_split_required_for_discard); |
23508a96 | 1461 | } |
0ce65797 | 1462 | |
3d7f4562 | 1463 | static int __send_write_same(struct clone_info *ci, struct dm_target *ti) |
0ce65797 | 1464 | { |
3d7f4562 | 1465 | return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL); |
0ce65797 MS |
1466 | } |
1467 | ||
3d7f4562 | 1468 | static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti) |
ac62d620 | 1469 | { |
3d7f4562 | 1470 | return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL); |
ac62d620 CH |
1471 | } |
1472 | ||
e4c93811 AK |
1473 | /* |
1474 | * Select the correct strategy for processing a non-flush bio. | |
1475 | */ | |
14fe594d | 1476 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1477 | { |
dba14160 | 1478 | struct bio *bio = ci->bio; |
512875bd | 1479 | struct dm_target *ti; |
1c3b13e6 | 1480 | unsigned len; |
c80914e8 | 1481 | int r; |
0ce65797 | 1482 | |
512875bd JN |
1483 | ti = dm_table_find_target(ci->map, ci->sector); |
1484 | if (!dm_target_is_valid(ti)) | |
1485 | return -EIO; | |
1486 | ||
3d7f4562 MS |
1487 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
1488 | return __send_discard(ci, ti); | |
1489 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) | |
1490 | return __send_write_same(ci, ti); | |
1491 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES)) | |
1492 | return __send_write_zeroes(ci, ti); | |
1493 | ||
264c869d DLM |
1494 | if (bio_op(bio) == REQ_OP_ZONE_REPORT) |
1495 | len = ci->sector_count; | |
1496 | else | |
1497 | len = min_t(sector_t, max_io_len(ci->sector, ti), | |
1498 | ci->sector_count); | |
0ce65797 | 1499 | |
c80914e8 MS |
1500 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1501 | if (r < 0) | |
1502 | return r; | |
0ce65797 | 1503 | |
1c3b13e6 KO |
1504 | ci->sector += len; |
1505 | ci->sector_count -= len; | |
0ce65797 | 1506 | |
1c3b13e6 | 1507 | return 0; |
0ce65797 MS |
1508 | } |
1509 | ||
978e51ba MS |
1510 | static void init_clone_info(struct clone_info *ci, struct mapped_device *md, |
1511 | struct dm_table *map, struct bio *bio) | |
1512 | { | |
1513 | ci->map = map; | |
1514 | ci->io = alloc_io(md, bio); | |
1515 | ci->sector = bio->bi_iter.bi_sector; | |
1516 | } | |
1517 | ||
1da177e4 | 1518 | /* |
14fe594d | 1519 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1520 | */ |
978e51ba MS |
1521 | static blk_qc_t __split_and_process_bio(struct mapped_device *md, |
1522 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1523 | { |
1da177e4 | 1524 | struct clone_info ci; |
978e51ba | 1525 | blk_qc_t ret = BLK_QC_T_NONE; |
512875bd | 1526 | int error = 0; |
1da177e4 | 1527 | |
83d5e5b0 | 1528 | if (unlikely(!map)) { |
6a8736d1 | 1529 | bio_io_error(bio); |
978e51ba | 1530 | return ret; |
f0b9a450 | 1531 | } |
692d0eb9 | 1532 | |
978e51ba | 1533 | init_clone_info(&ci, md, map, bio); |
0ce65797 | 1534 | |
1eff9d32 | 1535 | if (bio->bi_opf & REQ_PREFLUSH) { |
bc02cdbe | 1536 | ci.bio = &ci.io->md->flush_bio; |
b372d360 | 1537 | ci.sector_count = 0; |
14fe594d | 1538 | error = __send_empty_flush(&ci); |
b372d360 | 1539 | /* dec_pending submits any data associated with flush */ |
a4aa5e56 DLM |
1540 | } else if (bio_op(bio) == REQ_OP_ZONE_RESET) { |
1541 | ci.bio = bio; | |
1542 | ci.sector_count = 0; | |
1543 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1544 | } else { |
6a8736d1 | 1545 | ci.bio = bio; |
d87f4c14 | 1546 | ci.sector_count = bio_sectors(bio); |
18a25da8 | 1547 | while (ci.sector_count && !error) { |
14fe594d | 1548 | error = __split_and_process_non_flush(&ci); |
18a25da8 N |
1549 | if (current->bio_list && ci.sector_count && !error) { |
1550 | /* | |
1551 | * Remainder must be passed to generic_make_request() | |
1552 | * so that it gets handled *after* bios already submitted | |
1553 | * have been completely processed. | |
1554 | * We take a clone of the original to store in | |
745dc570 | 1555 | * ci.io->orig_bio to be used by end_io_acct() and |
18a25da8 N |
1556 | * for dec_pending to use for completion handling. |
1557 | * As this path is not used for REQ_OP_ZONE_REPORT, | |
745dc570 | 1558 | * the usage of io->orig_bio in dm_remap_zone_report() |
18a25da8 N |
1559 | * won't be affected by this reassignment. |
1560 | */ | |
1561 | struct bio *b = bio_clone_bioset(bio, GFP_NOIO, | |
1562 | md->queue->bio_split); | |
745dc570 | 1563 | ci.io->orig_bio = b; |
18a25da8 N |
1564 | bio_advance(bio, (bio_sectors(bio) - ci.sector_count) << 9); |
1565 | bio_chain(b, bio); | |
978e51ba | 1566 | ret = generic_make_request(bio); |
18a25da8 N |
1567 | break; |
1568 | } | |
1569 | } | |
d87f4c14 | 1570 | } |
0ce65797 | 1571 | |
1da177e4 | 1572 | /* drop the extra reference count */ |
54385bf7 | 1573 | dec_pending(ci.io, errno_to_blk_status(error)); |
978e51ba | 1574 | return ret; |
0ce65797 MS |
1575 | } |
1576 | ||
cec47e3d | 1577 | /* |
978e51ba MS |
1578 | * Optimized variant of __split_and_process_bio that leverages the |
1579 | * fact that targets that use it do _not_ have a need to split bios. | |
cec47e3d | 1580 | */ |
978e51ba MS |
1581 | static blk_qc_t __process_bio(struct mapped_device *md, |
1582 | struct dm_table *map, struct bio *bio) | |
1583 | { | |
1584 | struct clone_info ci; | |
1585 | blk_qc_t ret = BLK_QC_T_NONE; | |
1586 | int error = 0; | |
1587 | ||
1588 | if (unlikely(!map)) { | |
1589 | bio_io_error(bio); | |
1590 | return ret; | |
1591 | } | |
1592 | ||
1593 | init_clone_info(&ci, md, map, bio); | |
1594 | ||
1595 | if (bio->bi_opf & REQ_PREFLUSH) { | |
1596 | ci.bio = &ci.io->md->flush_bio; | |
1597 | ci.sector_count = 0; | |
1598 | error = __send_empty_flush(&ci); | |
1599 | /* dec_pending submits any data associated with flush */ | |
1600 | } else { | |
1601 | struct dm_target *ti = md->immutable_target; | |
1602 | struct dm_target_io *tio; | |
1603 | ||
1604 | /* | |
1605 | * Defend against IO still getting in during teardown | |
1606 | * - as was seen for a time with nvme-fcloop | |
1607 | */ | |
1608 | if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) { | |
1609 | error = -EIO; | |
1610 | goto out; | |
1611 | } | |
1612 | ||
1613 | tio = alloc_tio(&ci, ti, 0, GFP_NOIO); | |
1614 | ci.bio = bio; | |
1615 | ci.sector_count = bio_sectors(bio); | |
1616 | ret = __clone_and_map_simple_bio(&ci, tio, NULL); | |
1617 | } | |
1618 | out: | |
1619 | /* drop the extra reference count */ | |
1620 | dec_pending(ci.io, errno_to_blk_status(error)); | |
1621 | return ret; | |
1622 | } | |
1623 | ||
1624 | typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *); | |
1625 | ||
1626 | static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio, | |
1627 | process_bio_fn process_bio) | |
cec47e3d KU |
1628 | { |
1629 | struct mapped_device *md = q->queuedata; | |
978e51ba | 1630 | blk_qc_t ret = BLK_QC_T_NONE; |
83d5e5b0 MP |
1631 | int srcu_idx; |
1632 | struct dm_table *map; | |
cec47e3d | 1633 | |
83d5e5b0 | 1634 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1635 | |
6a8736d1 TH |
1636 | /* if we're suspended, we have to queue this io for later */ |
1637 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1638 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1639 | |
1eff9d32 | 1640 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1641 | queue_io(md, bio); |
1642 | else | |
54d9a1b4 | 1643 | bio_io_error(bio); |
978e51ba | 1644 | return ret; |
cec47e3d | 1645 | } |
1da177e4 | 1646 | |
978e51ba MS |
1647 | ret = process_bio(md, map, bio); |
1648 | ||
83d5e5b0 | 1649 | dm_put_live_table(md, srcu_idx); |
978e51ba MS |
1650 | return ret; |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * The request function that remaps the bio to one target and | |
1655 | * splits off any remainder. | |
1656 | */ | |
1657 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) | |
1658 | { | |
1659 | return __dm_make_request(q, bio, __split_and_process_bio); | |
1660 | } | |
1661 | ||
1662 | static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio) | |
1663 | { | |
1664 | return __dm_make_request(q, bio, __process_bio); | |
cec47e3d KU |
1665 | } |
1666 | ||
1da177e4 LT |
1667 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1668 | { | |
8a57dfc6 CS |
1669 | int r = bdi_bits; |
1670 | struct mapped_device *md = congested_data; | |
1671 | struct dm_table *map; | |
1da177e4 | 1672 | |
1eb787ec | 1673 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1674 | if (dm_request_based(md)) { |
cec47e3d | 1675 | /* |
e522c039 MS |
1676 | * With request-based DM we only need to check the |
1677 | * top-level queue for congestion. | |
cec47e3d | 1678 | */ |
dc3b17cc | 1679 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1680 | } else { |
1681 | map = dm_get_live_table_fast(md); | |
1682 | if (map) | |
cec47e3d | 1683 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1684 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1685 | } |
1686 | } | |
1687 | ||
1da177e4 LT |
1688 | return r; |
1689 | } | |
1690 | ||
1691 | /*----------------------------------------------------------------- | |
1692 | * An IDR is used to keep track of allocated minor numbers. | |
1693 | *---------------------------------------------------------------*/ | |
2b06cfff | 1694 | static void free_minor(int minor) |
1da177e4 | 1695 | { |
f32c10b0 | 1696 | spin_lock(&_minor_lock); |
1da177e4 | 1697 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1698 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1699 | } |
1700 | ||
1701 | /* | |
1702 | * See if the device with a specific minor # is free. | |
1703 | */ | |
cf13ab8e | 1704 | static int specific_minor(int minor) |
1da177e4 | 1705 | { |
c9d76be6 | 1706 | int r; |
1da177e4 LT |
1707 | |
1708 | if (minor >= (1 << MINORBITS)) | |
1709 | return -EINVAL; | |
1710 | ||
c9d76be6 | 1711 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1712 | spin_lock(&_minor_lock); |
1da177e4 | 1713 | |
c9d76be6 | 1714 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1715 | |
f32c10b0 | 1716 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1717 | idr_preload_end(); |
1718 | if (r < 0) | |
1719 | return r == -ENOSPC ? -EBUSY : r; | |
1720 | return 0; | |
1da177e4 LT |
1721 | } |
1722 | ||
cf13ab8e | 1723 | static int next_free_minor(int *minor) |
1da177e4 | 1724 | { |
c9d76be6 | 1725 | int r; |
62f75c2f | 1726 | |
c9d76be6 | 1727 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1728 | spin_lock(&_minor_lock); |
1da177e4 | 1729 | |
c9d76be6 | 1730 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1731 | |
f32c10b0 | 1732 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1733 | idr_preload_end(); |
1734 | if (r < 0) | |
1735 | return r; | |
1736 | *minor = r; | |
1737 | return 0; | |
1da177e4 LT |
1738 | } |
1739 | ||
83d5cde4 | 1740 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1741 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1742 | |
53d5914f MP |
1743 | static void dm_wq_work(struct work_struct *work); |
1744 | ||
c12c9a3c | 1745 | static void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1746 | { |
17e149b8 | 1747 | md->use_blk_mq = false; |
bfebd1cd MS |
1748 | |
1749 | /* | |
1750 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1751 | */ | |
dc3b17cc | 1752 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf MS |
1753 | } |
1754 | ||
0f20972f MS |
1755 | static void cleanup_mapped_device(struct mapped_device *md) |
1756 | { | |
0f20972f MS |
1757 | if (md->wq) |
1758 | destroy_workqueue(md->wq); | |
1759 | if (md->kworker_task) | |
1760 | kthread_stop(md->kworker_task); | |
0f20972f MS |
1761 | if (md->bs) |
1762 | bioset_free(md->bs); | |
64f52b0e MS |
1763 | if (md->io_bs) |
1764 | bioset_free(md->io_bs); | |
0f20972f | 1765 | |
f26c5719 DW |
1766 | if (md->dax_dev) { |
1767 | kill_dax(md->dax_dev); | |
1768 | put_dax(md->dax_dev); | |
1769 | md->dax_dev = NULL; | |
1770 | } | |
1771 | ||
0f20972f MS |
1772 | if (md->disk) { |
1773 | spin_lock(&_minor_lock); | |
1774 | md->disk->private_data = NULL; | |
1775 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1776 | del_gendisk(md->disk); |
1777 | put_disk(md->disk); | |
1778 | } | |
1779 | ||
1780 | if (md->queue) | |
1781 | blk_cleanup_queue(md->queue); | |
1782 | ||
d09960b0 TE |
1783 | cleanup_srcu_struct(&md->io_barrier); |
1784 | ||
0f20972f MS |
1785 | if (md->bdev) { |
1786 | bdput(md->bdev); | |
1787 | md->bdev = NULL; | |
1788 | } | |
4cc96131 | 1789 | |
d5ffebdd MS |
1790 | mutex_destroy(&md->suspend_lock); |
1791 | mutex_destroy(&md->type_lock); | |
1792 | mutex_destroy(&md->table_devices_lock); | |
1793 | ||
4cc96131 | 1794 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
1795 | } |
1796 | ||
1da177e4 LT |
1797 | /* |
1798 | * Allocate and initialise a blank device with a given minor. | |
1799 | */ | |
2b06cfff | 1800 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1801 | { |
115485e8 | 1802 | int r, numa_node_id = dm_get_numa_node(); |
976431b0 | 1803 | struct dax_device *dax_dev = NULL; |
115485e8 | 1804 | struct mapped_device *md; |
ba61fdd1 | 1805 | void *old_md; |
1da177e4 | 1806 | |
856eb091 | 1807 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1808 | if (!md) { |
1809 | DMWARN("unable to allocate device, out of memory."); | |
1810 | return NULL; | |
1811 | } | |
1812 | ||
10da4f79 | 1813 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1814 | goto bad_module_get; |
10da4f79 | 1815 | |
1da177e4 | 1816 | /* get a minor number for the dev */ |
2b06cfff | 1817 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1818 | r = next_free_minor(&minor); |
2b06cfff | 1819 | else |
cf13ab8e | 1820 | r = specific_minor(minor); |
1da177e4 | 1821 | if (r < 0) |
6ed7ade8 | 1822 | goto bad_minor; |
1da177e4 | 1823 | |
83d5e5b0 MP |
1824 | r = init_srcu_struct(&md->io_barrier); |
1825 | if (r < 0) | |
1826 | goto bad_io_barrier; | |
1827 | ||
115485e8 | 1828 | md->numa_node_id = numa_node_id; |
4cc96131 | 1829 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1830 | md->init_tio_pdu = false; |
a5664dad | 1831 | md->type = DM_TYPE_NONE; |
e61290a4 | 1832 | mutex_init(&md->suspend_lock); |
a5664dad | 1833 | mutex_init(&md->type_lock); |
86f1152b | 1834 | mutex_init(&md->table_devices_lock); |
022c2611 | 1835 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1836 | atomic_set(&md->holders, 1); |
5c6bd75d | 1837 | atomic_set(&md->open_count, 0); |
1da177e4 | 1838 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1839 | atomic_set(&md->uevent_seq, 0); |
1840 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1841 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1842 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1843 | |
115485e8 | 1844 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1845 | if (!md->queue) |
0f20972f | 1846 | goto bad; |
c12c9a3c MS |
1847 | md->queue->queuedata = md; |
1848 | md->queue->backing_dev_info->congested_data = md; | |
1da177e4 | 1849 | |
c12c9a3c | 1850 | md->disk = alloc_disk_node(1, md->numa_node_id); |
1da177e4 | 1851 | if (!md->disk) |
0f20972f | 1852 | goto bad; |
1da177e4 | 1853 | |
316d315b NK |
1854 | atomic_set(&md->pending[0], 0); |
1855 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1856 | init_waitqueue_head(&md->wait); |
53d5914f | 1857 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1858 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1859 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1860 | md->kworker_task = NULL; |
f0b04115 | 1861 | |
1da177e4 LT |
1862 | md->disk->major = _major; |
1863 | md->disk->first_minor = minor; | |
1864 | md->disk->fops = &dm_blk_dops; | |
1865 | md->disk->queue = md->queue; | |
1866 | md->disk->private_data = md; | |
1867 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 | 1868 | |
976431b0 DW |
1869 | if (IS_ENABLED(CONFIG_DAX_DRIVER)) { |
1870 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); | |
1871 | if (!dax_dev) | |
1872 | goto bad; | |
1873 | } | |
f26c5719 DW |
1874 | md->dax_dev = dax_dev; |
1875 | ||
c100ec49 | 1876 | add_disk_no_queue_reg(md->disk); |
7e51f257 | 1877 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1878 | |
670368a8 | 1879 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1880 | if (!md->wq) |
0f20972f | 1881 | goto bad; |
304f3f6a | 1882 | |
32a926da MP |
1883 | md->bdev = bdget_disk(md->disk, 0); |
1884 | if (!md->bdev) | |
0f20972f | 1885 | goto bad; |
32a926da | 1886 | |
3a83f467 | 1887 | bio_init(&md->flush_bio, NULL, 0); |
74d46992 | 1888 | bio_set_dev(&md->flush_bio, md->bdev); |
ff0361b3 | 1889 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; |
6a8736d1 | 1890 | |
fd2ed4d2 MP |
1891 | dm_stats_init(&md->stats); |
1892 | ||
ba61fdd1 | 1893 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1894 | spin_lock(&_minor_lock); |
ba61fdd1 | 1895 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1896 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1897 | |
1898 | BUG_ON(old_md != MINOR_ALLOCED); | |
1899 | ||
1da177e4 LT |
1900 | return md; |
1901 | ||
0f20972f MS |
1902 | bad: |
1903 | cleanup_mapped_device(md); | |
83d5e5b0 | 1904 | bad_io_barrier: |
1da177e4 | 1905 | free_minor(minor); |
6ed7ade8 | 1906 | bad_minor: |
10da4f79 | 1907 | module_put(THIS_MODULE); |
6ed7ade8 | 1908 | bad_module_get: |
856eb091 | 1909 | kvfree(md); |
1da177e4 LT |
1910 | return NULL; |
1911 | } | |
1912 | ||
ae9da83f JN |
1913 | static void unlock_fs(struct mapped_device *md); |
1914 | ||
1da177e4 LT |
1915 | static void free_dev(struct mapped_device *md) |
1916 | { | |
f331c029 | 1917 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1918 | |
32a926da | 1919 | unlock_fs(md); |
2eb6e1e3 | 1920 | |
0f20972f | 1921 | cleanup_mapped_device(md); |
63a4f065 | 1922 | |
86f1152b | 1923 | free_table_devices(&md->table_devices); |
63a4f065 | 1924 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1925 | free_minor(minor); |
1926 | ||
10da4f79 | 1927 | module_put(THIS_MODULE); |
856eb091 | 1928 | kvfree(md); |
1da177e4 LT |
1929 | } |
1930 | ||
e6ee8c0b KU |
1931 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1932 | { | |
c0820cf5 | 1933 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1934 | |
0776aa0e | 1935 | if (dm_table_bio_based(t)) { |
64f52b0e MS |
1936 | /* |
1937 | * The md may already have mempools that need changing. | |
1938 | * If so, reload bioset because front_pad may have changed | |
1939 | * because a different table was loaded. | |
1940 | */ | |
0776aa0e | 1941 | if (md->bs) { |
16245bdc | 1942 | bioset_free(md->bs); |
0776aa0e | 1943 | md->bs = NULL; |
16245bdc | 1944 | } |
64f52b0e MS |
1945 | if (md->io_bs) { |
1946 | bioset_free(md->io_bs); | |
1947 | md->io_bs = NULL; | |
16245bdc | 1948 | } |
0776aa0e MS |
1949 | |
1950 | } else if (md->bs) { | |
4e6e36c3 MS |
1951 | /* |
1952 | * There's no need to reload with request-based dm | |
1953 | * because the size of front_pad doesn't change. | |
1954 | * Note for future: If you are to reload bioset, | |
1955 | * prep-ed requests in the queue may refer | |
1956 | * to bio from the old bioset, so you must walk | |
1957 | * through the queue to unprep. | |
1958 | */ | |
1959 | goto out; | |
c0820cf5 | 1960 | } |
e6ee8c0b | 1961 | |
dde1e1ec | 1962 | BUG_ON(!p || md->bs || md->io_bs); |
cbc4e3c1 | 1963 | |
e6ee8c0b KU |
1964 | md->bs = p->bs; |
1965 | p->bs = NULL; | |
64f52b0e MS |
1966 | md->io_bs = p->io_bs; |
1967 | p->io_bs = NULL; | |
e6ee8c0b | 1968 | out: |
02233342 | 1969 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1970 | dm_table_free_md_mempools(t); |
1971 | } | |
1972 | ||
1da177e4 LT |
1973 | /* |
1974 | * Bind a table to the device. | |
1975 | */ | |
1976 | static void event_callback(void *context) | |
1977 | { | |
7a8c3d3b MA |
1978 | unsigned long flags; |
1979 | LIST_HEAD(uevents); | |
1da177e4 LT |
1980 | struct mapped_device *md = (struct mapped_device *) context; |
1981 | ||
7a8c3d3b MA |
1982 | spin_lock_irqsave(&md->uevent_lock, flags); |
1983 | list_splice_init(&md->uevent_list, &uevents); | |
1984 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1985 | ||
ed9e1982 | 1986 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1987 | |
1da177e4 LT |
1988 | atomic_inc(&md->event_nr); |
1989 | wake_up(&md->eventq); | |
62e08243 | 1990 | dm_issue_global_event(); |
1da177e4 LT |
1991 | } |
1992 | ||
c217649b MS |
1993 | /* |
1994 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1995 | */ | |
4e90188b | 1996 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1997 | { |
1ea0654e BVA |
1998 | lockdep_assert_held(&md->suspend_lock); |
1999 | ||
4e90188b | 2000 | set_capacity(md->disk, size); |
1da177e4 | 2001 | |
db8fef4f | 2002 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2003 | } |
2004 | ||
042d2a9b AK |
2005 | /* |
2006 | * Returns old map, which caller must destroy. | |
2007 | */ | |
2008 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2009 | struct queue_limits *limits) | |
1da177e4 | 2010 | { |
042d2a9b | 2011 | struct dm_table *old_map; |
165125e1 | 2012 | struct request_queue *q = md->queue; |
978e51ba | 2013 | bool request_based = dm_table_request_based(t); |
1da177e4 LT |
2014 | sector_t size; |
2015 | ||
5a8f1f80 BVA |
2016 | lockdep_assert_held(&md->suspend_lock); |
2017 | ||
1da177e4 | 2018 | size = dm_table_get_size(t); |
3ac51e74 DW |
2019 | |
2020 | /* | |
2021 | * Wipe any geometry if the size of the table changed. | |
2022 | */ | |
fd2ed4d2 | 2023 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2024 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2025 | ||
32a926da | 2026 | __set_size(md, size); |
d5816876 | 2027 | |
2ca3310e AK |
2028 | dm_table_event_callback(t, event_callback, md); |
2029 | ||
e6ee8c0b KU |
2030 | /* |
2031 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2032 | * for request-based during suspension. So stop it to prevent | |
2033 | * I/O mapping before resume. | |
2034 | * This must be done before setting the queue restrictions, | |
2035 | * because request-based dm may be run just after the setting. | |
2036 | */ | |
978e51ba | 2037 | if (request_based) |
eca7ee6d | 2038 | dm_stop_queue(q); |
978e51ba MS |
2039 | |
2040 | if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) { | |
16f12266 | 2041 | /* |
978e51ba MS |
2042 | * Leverage the fact that request-based DM targets and |
2043 | * NVMe bio based targets are immutable singletons | |
2044 | * - used to optimize both dm_request_fn and dm_mq_queue_rq; | |
2045 | * and __process_bio. | |
16f12266 MS |
2046 | */ |
2047 | md->immutable_target = dm_table_get_immutable_target(t); | |
2048 | } | |
e6ee8c0b KU |
2049 | |
2050 | __bind_mempools(md, t); | |
2051 | ||
a12f5d48 | 2052 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2053 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2054 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2055 | ||
754c5fc7 | 2056 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2057 | if (old_map) |
2058 | dm_sync_table(md); | |
1da177e4 | 2059 | |
042d2a9b | 2060 | return old_map; |
1da177e4 LT |
2061 | } |
2062 | ||
a7940155 AK |
2063 | /* |
2064 | * Returns unbound table for the caller to free. | |
2065 | */ | |
2066 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2067 | { |
a12f5d48 | 2068 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2069 | |
2070 | if (!map) | |
a7940155 | 2071 | return NULL; |
1da177e4 LT |
2072 | |
2073 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2074 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2075 | dm_sync_table(md); |
a7940155 AK |
2076 | |
2077 | return map; | |
1da177e4 LT |
2078 | } |
2079 | ||
2080 | /* | |
2081 | * Constructor for a new device. | |
2082 | */ | |
2b06cfff | 2083 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 | 2084 | { |
c12c9a3c | 2085 | int r; |
1da177e4 LT |
2086 | struct mapped_device *md; |
2087 | ||
2b06cfff | 2088 | md = alloc_dev(minor); |
1da177e4 LT |
2089 | if (!md) |
2090 | return -ENXIO; | |
2091 | ||
c12c9a3c MS |
2092 | r = dm_sysfs_init(md); |
2093 | if (r) { | |
2094 | free_dev(md); | |
2095 | return r; | |
2096 | } | |
784aae73 | 2097 | |
1da177e4 LT |
2098 | *result = md; |
2099 | return 0; | |
2100 | } | |
2101 | ||
a5664dad MS |
2102 | /* |
2103 | * Functions to manage md->type. | |
2104 | * All are required to hold md->type_lock. | |
2105 | */ | |
2106 | void dm_lock_md_type(struct mapped_device *md) | |
2107 | { | |
2108 | mutex_lock(&md->type_lock); | |
2109 | } | |
2110 | ||
2111 | void dm_unlock_md_type(struct mapped_device *md) | |
2112 | { | |
2113 | mutex_unlock(&md->type_lock); | |
2114 | } | |
2115 | ||
7e0d574f | 2116 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2117 | { |
00c4fc3b | 2118 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2119 | md->type = type; |
2120 | } | |
2121 | ||
7e0d574f | 2122 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2123 | { |
2124 | return md->type; | |
2125 | } | |
2126 | ||
36a0456f AK |
2127 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2128 | { | |
2129 | return md->immutable_target_type; | |
2130 | } | |
2131 | ||
f84cb8a4 MS |
2132 | /* |
2133 | * The queue_limits are only valid as long as you have a reference | |
2134 | * count on 'md'. | |
2135 | */ | |
2136 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2137 | { | |
2138 | BUG_ON(!atomic_read(&md->holders)); | |
2139 | return &md->queue->limits; | |
2140 | } | |
2141 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2142 | ||
4a0b4ddf MS |
2143 | /* |
2144 | * Setup the DM device's queue based on md's type | |
2145 | */ | |
591ddcfc | 2146 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2147 | { |
bfebd1cd | 2148 | int r; |
c100ec49 | 2149 | struct queue_limits limits; |
7e0d574f | 2150 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2151 | |
545ed20e | 2152 | switch (type) { |
bfebd1cd | 2153 | case DM_TYPE_REQUEST_BASED: |
c12c9a3c | 2154 | dm_init_normal_md_queue(md); |
eb8db831 | 2155 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 2156 | if (r) { |
eca7ee6d | 2157 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2158 | return r; |
ff36ab34 | 2159 | } |
bfebd1cd MS |
2160 | break; |
2161 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 2162 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2163 | if (r) { |
eca7ee6d | 2164 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2165 | return r; |
2166 | } | |
2167 | break; | |
2168 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2169 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 2170 | dm_init_normal_md_queue(md); |
ff36ab34 | 2171 | blk_queue_make_request(md->queue, dm_make_request); |
bfebd1cd | 2172 | break; |
978e51ba MS |
2173 | case DM_TYPE_NVME_BIO_BASED: |
2174 | dm_init_normal_md_queue(md); | |
2175 | blk_queue_make_request(md->queue, dm_make_request_nvme); | |
bfebd1cd | 2176 | break; |
7e0d574f BVA |
2177 | case DM_TYPE_NONE: |
2178 | WARN_ON_ONCE(true); | |
2179 | break; | |
4a0b4ddf MS |
2180 | } |
2181 | ||
c100ec49 MS |
2182 | r = dm_calculate_queue_limits(t, &limits); |
2183 | if (r) { | |
2184 | DMERR("Cannot calculate initial queue limits"); | |
2185 | return r; | |
2186 | } | |
2187 | dm_table_set_restrictions(t, md->queue, &limits); | |
2188 | blk_register_queue(md->disk); | |
2189 | ||
4a0b4ddf MS |
2190 | return 0; |
2191 | } | |
2192 | ||
2bec1f4a | 2193 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2194 | { |
2195 | struct mapped_device *md; | |
1da177e4 LT |
2196 | unsigned minor = MINOR(dev); |
2197 | ||
2198 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2199 | return NULL; | |
2200 | ||
f32c10b0 | 2201 | spin_lock(&_minor_lock); |
1da177e4 LT |
2202 | |
2203 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2204 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2205 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2206 | md = NULL; | |
2207 | goto out; | |
fba9f90e | 2208 | } |
49de5769 | 2209 | dm_get(md); |
fba9f90e | 2210 | out: |
f32c10b0 | 2211 | spin_unlock(&_minor_lock); |
1da177e4 | 2212 | |
637842cf DT |
2213 | return md; |
2214 | } | |
3cf2e4ba | 2215 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2216 | |
9ade92a9 | 2217 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2218 | { |
9ade92a9 | 2219 | return md->interface_ptr; |
1da177e4 LT |
2220 | } |
2221 | ||
2222 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2223 | { | |
2224 | md->interface_ptr = ptr; | |
2225 | } | |
2226 | ||
2227 | void dm_get(struct mapped_device *md) | |
2228 | { | |
2229 | atomic_inc(&md->holders); | |
3f77316d | 2230 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2231 | } |
2232 | ||
09ee96b2 MP |
2233 | int dm_hold(struct mapped_device *md) |
2234 | { | |
2235 | spin_lock(&_minor_lock); | |
2236 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2237 | spin_unlock(&_minor_lock); | |
2238 | return -EBUSY; | |
2239 | } | |
2240 | dm_get(md); | |
2241 | spin_unlock(&_minor_lock); | |
2242 | return 0; | |
2243 | } | |
2244 | EXPORT_SYMBOL_GPL(dm_hold); | |
2245 | ||
72d94861 AK |
2246 | const char *dm_device_name(struct mapped_device *md) |
2247 | { | |
2248 | return md->name; | |
2249 | } | |
2250 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2251 | ||
3f77316d | 2252 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2253 | { |
1134e5ae | 2254 | struct dm_table *map; |
83d5e5b0 | 2255 | int srcu_idx; |
1da177e4 | 2256 | |
3f77316d | 2257 | might_sleep(); |
fba9f90e | 2258 | |
63a4f065 | 2259 | spin_lock(&_minor_lock); |
3f77316d KU |
2260 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2261 | set_bit(DMF_FREEING, &md->flags); | |
2262 | spin_unlock(&_minor_lock); | |
3b785fbc | 2263 | |
c12c9a3c | 2264 | blk_set_queue_dying(md->queue); |
3f77316d | 2265 | |
02233342 | 2266 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 2267 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2268 | |
ab7c7bb6 MP |
2269 | /* |
2270 | * Take suspend_lock so that presuspend and postsuspend methods | |
2271 | * do not race with internal suspend. | |
2272 | */ | |
2273 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2274 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2275 | if (!dm_suspended_md(md)) { |
2276 | dm_table_presuspend_targets(map); | |
2277 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2278 | } |
83d5e5b0 MP |
2279 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2280 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2281 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2282 | |
3f77316d KU |
2283 | /* |
2284 | * Rare, but there may be I/O requests still going to complete, | |
2285 | * for example. Wait for all references to disappear. | |
2286 | * No one should increment the reference count of the mapped_device, | |
2287 | * after the mapped_device state becomes DMF_FREEING. | |
2288 | */ | |
2289 | if (wait) | |
2290 | while (atomic_read(&md->holders)) | |
2291 | msleep(1); | |
2292 | else if (atomic_read(&md->holders)) | |
2293 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2294 | dm_device_name(md), atomic_read(&md->holders)); | |
2295 | ||
2296 | dm_sysfs_exit(md); | |
3f77316d KU |
2297 | dm_table_destroy(__unbind(md)); |
2298 | free_dev(md); | |
2299 | } | |
2300 | ||
2301 | void dm_destroy(struct mapped_device *md) | |
2302 | { | |
2303 | __dm_destroy(md, true); | |
2304 | } | |
2305 | ||
2306 | void dm_destroy_immediate(struct mapped_device *md) | |
2307 | { | |
2308 | __dm_destroy(md, false); | |
2309 | } | |
2310 | ||
2311 | void dm_put(struct mapped_device *md) | |
2312 | { | |
2313 | atomic_dec(&md->holders); | |
1da177e4 | 2314 | } |
79eb885c | 2315 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2316 | |
b48633f8 | 2317 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2318 | { |
2319 | int r = 0; | |
9f4c3f87 | 2320 | DEFINE_WAIT(wait); |
46125c1c MB |
2321 | |
2322 | while (1) { | |
9f4c3f87 | 2323 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2324 | |
b4324fee | 2325 | if (!md_in_flight(md)) |
46125c1c MB |
2326 | break; |
2327 | ||
e3fabdfd | 2328 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2329 | r = -EINTR; |
2330 | break; | |
2331 | } | |
2332 | ||
2333 | io_schedule(); | |
2334 | } | |
9f4c3f87 | 2335 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2336 | |
46125c1c MB |
2337 | return r; |
2338 | } | |
2339 | ||
1da177e4 LT |
2340 | /* |
2341 | * Process the deferred bios | |
2342 | */ | |
ef208587 | 2343 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2344 | { |
ef208587 MP |
2345 | struct mapped_device *md = container_of(work, struct mapped_device, |
2346 | work); | |
6d6f10df | 2347 | struct bio *c; |
83d5e5b0 MP |
2348 | int srcu_idx; |
2349 | struct dm_table *map; | |
1da177e4 | 2350 | |
83d5e5b0 | 2351 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2352 | |
3b00b203 | 2353 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2354 | spin_lock_irq(&md->deferred_lock); |
2355 | c = bio_list_pop(&md->deferred); | |
2356 | spin_unlock_irq(&md->deferred_lock); | |
2357 | ||
6a8736d1 | 2358 | if (!c) |
df12ee99 | 2359 | break; |
022c2611 | 2360 | |
e6ee8c0b KU |
2361 | if (dm_request_based(md)) |
2362 | generic_make_request(c); | |
6a8736d1 | 2363 | else |
83d5e5b0 | 2364 | __split_and_process_bio(md, map, c); |
022c2611 | 2365 | } |
73d410c0 | 2366 | |
83d5e5b0 | 2367 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2368 | } |
2369 | ||
9a1fb464 | 2370 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2371 | { |
3b00b203 | 2372 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2373 | smp_mb__after_atomic(); |
53d5914f | 2374 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2375 | } |
2376 | ||
1da177e4 | 2377 | /* |
042d2a9b | 2378 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2379 | */ |
042d2a9b | 2380 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2381 | { |
87eb5b21 | 2382 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2383 | struct queue_limits limits; |
042d2a9b | 2384 | int r; |
1da177e4 | 2385 | |
e61290a4 | 2386 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2387 | |
2388 | /* device must be suspended */ | |
4f186f8b | 2389 | if (!dm_suspended_md(md)) |
93c534ae | 2390 | goto out; |
1da177e4 | 2391 | |
3ae70656 MS |
2392 | /* |
2393 | * If the new table has no data devices, retain the existing limits. | |
2394 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2395 | * then new I/O is queued based on these limits, and then some paths | |
2396 | * reappear. | |
2397 | */ | |
2398 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2399 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2400 | if (live_map) |
2401 | limits = md->queue->limits; | |
83d5e5b0 | 2402 | dm_put_live_table_fast(md); |
3ae70656 MS |
2403 | } |
2404 | ||
87eb5b21 MC |
2405 | if (!live_map) { |
2406 | r = dm_calculate_queue_limits(table, &limits); | |
2407 | if (r) { | |
2408 | map = ERR_PTR(r); | |
2409 | goto out; | |
2410 | } | |
042d2a9b | 2411 | } |
754c5fc7 | 2412 | |
042d2a9b | 2413 | map = __bind(md, table, &limits); |
62e08243 | 2414 | dm_issue_global_event(); |
1da177e4 | 2415 | |
93c534ae | 2416 | out: |
e61290a4 | 2417 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2418 | return map; |
1da177e4 LT |
2419 | } |
2420 | ||
2421 | /* | |
2422 | * Functions to lock and unlock any filesystem running on the | |
2423 | * device. | |
2424 | */ | |
2ca3310e | 2425 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2426 | { |
e39e2e95 | 2427 | int r; |
1da177e4 LT |
2428 | |
2429 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2430 | |
db8fef4f | 2431 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2432 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2433 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2434 | md->frozen_sb = NULL; |
2435 | return r; | |
dfbe03f6 AK |
2436 | } |
2437 | ||
aa8d7c2f AK |
2438 | set_bit(DMF_FROZEN, &md->flags); |
2439 | ||
1da177e4 LT |
2440 | return 0; |
2441 | } | |
2442 | ||
2ca3310e | 2443 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2444 | { |
aa8d7c2f AK |
2445 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2446 | return; | |
2447 | ||
db8fef4f | 2448 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2449 | md->frozen_sb = NULL; |
aa8d7c2f | 2450 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2451 | } |
2452 | ||
2453 | /* | |
b48633f8 BVA |
2454 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2455 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2456 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2457 | * | |
ffcc3936 MS |
2458 | * If __dm_suspend returns 0, the device is completely quiescent |
2459 | * now. There is no request-processing activity. All new requests | |
2460 | * are being added to md->deferred list. | |
cec47e3d | 2461 | */ |
ffcc3936 | 2462 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2463 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2464 | int dmf_suspended_flag) |
1da177e4 | 2465 | { |
ffcc3936 MS |
2466 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2467 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2468 | int r; | |
1da177e4 | 2469 | |
5a8f1f80 BVA |
2470 | lockdep_assert_held(&md->suspend_lock); |
2471 | ||
2e93ccc1 KU |
2472 | /* |
2473 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2474 | * This flag is cleared before dm_suspend returns. | |
2475 | */ | |
2476 | if (noflush) | |
2477 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 BVA |
2478 | else |
2479 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); | |
2e93ccc1 | 2480 | |
d67ee213 MS |
2481 | /* |
2482 | * This gets reverted if there's an error later and the targets | |
2483 | * provide the .presuspend_undo hook. | |
2484 | */ | |
cf222b37 AK |
2485 | dm_table_presuspend_targets(map); |
2486 | ||
32a926da | 2487 | /* |
9f518b27 KU |
2488 | * Flush I/O to the device. |
2489 | * Any I/O submitted after lock_fs() may not be flushed. | |
2490 | * noflush takes precedence over do_lockfs. | |
2491 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2492 | */ |
2493 | if (!noflush && do_lockfs) { | |
2494 | r = lock_fs(md); | |
d67ee213 MS |
2495 | if (r) { |
2496 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2497 | return r; |
d67ee213 | 2498 | } |
aa8d7c2f | 2499 | } |
1da177e4 LT |
2500 | |
2501 | /* | |
3b00b203 MP |
2502 | * Here we must make sure that no processes are submitting requests |
2503 | * to target drivers i.e. no one may be executing | |
2504 | * __split_and_process_bio. This is called from dm_request and | |
2505 | * dm_wq_work. | |
2506 | * | |
2507 | * To get all processes out of __split_and_process_bio in dm_request, | |
2508 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2509 | * __split_and_process_bio from dm_request and quiesce the thread |
2510 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2511 | * flush_workqueue(md->wq). | |
1da177e4 | 2512 | */ |
1eb787ec | 2513 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2514 | if (map) |
2515 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2516 | |
d0bcb878 | 2517 | /* |
29e4013d TH |
2518 | * Stop md->queue before flushing md->wq in case request-based |
2519 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2520 | */ |
2eb6e1e3 | 2521 | if (dm_request_based(md)) { |
eca7ee6d | 2522 | dm_stop_queue(md->queue); |
02233342 | 2523 | if (md->kworker_task) |
3989144f | 2524 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2525 | } |
cec47e3d | 2526 | |
d0bcb878 KU |
2527 | flush_workqueue(md->wq); |
2528 | ||
1da177e4 | 2529 | /* |
3b00b203 MP |
2530 | * At this point no more requests are entering target request routines. |
2531 | * We call dm_wait_for_completion to wait for all existing requests | |
2532 | * to finish. | |
1da177e4 | 2533 | */ |
b48633f8 | 2534 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2535 | if (!r) |
2536 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2537 | |
6d6f10df | 2538 | if (noflush) |
022c2611 | 2539 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2540 | if (map) |
2541 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2542 | |
1da177e4 | 2543 | /* were we interrupted ? */ |
46125c1c | 2544 | if (r < 0) { |
9a1fb464 | 2545 | dm_queue_flush(md); |
73d410c0 | 2546 | |
cec47e3d | 2547 | if (dm_request_based(md)) |
eca7ee6d | 2548 | dm_start_queue(md->queue); |
cec47e3d | 2549 | |
2ca3310e | 2550 | unlock_fs(md); |
d67ee213 | 2551 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2552 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2553 | } |
1da177e4 | 2554 | |
ffcc3936 MS |
2555 | return r; |
2556 | } | |
2557 | ||
2558 | /* | |
2559 | * We need to be able to change a mapping table under a mounted | |
2560 | * filesystem. For example we might want to move some data in | |
2561 | * the background. Before the table can be swapped with | |
2562 | * dm_bind_table, dm_suspend must be called to flush any in | |
2563 | * flight bios and ensure that any further io gets deferred. | |
2564 | */ | |
2565 | /* | |
2566 | * Suspend mechanism in request-based dm. | |
2567 | * | |
2568 | * 1. Flush all I/Os by lock_fs() if needed. | |
2569 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2570 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2571 | * | |
2572 | * To abort suspend, start the request_queue. | |
2573 | */ | |
2574 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2575 | { | |
2576 | struct dm_table *map = NULL; | |
2577 | int r = 0; | |
2578 | ||
2579 | retry: | |
2580 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2581 | ||
2582 | if (dm_suspended_md(md)) { | |
2583 | r = -EINVAL; | |
2584 | goto out_unlock; | |
2585 | } | |
2586 | ||
2587 | if (dm_suspended_internally_md(md)) { | |
2588 | /* already internally suspended, wait for internal resume */ | |
2589 | mutex_unlock(&md->suspend_lock); | |
2590 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2591 | if (r) | |
2592 | return r; | |
2593 | goto retry; | |
2594 | } | |
2595 | ||
a12f5d48 | 2596 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2597 | |
eaf9a736 | 2598 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2599 | if (r) |
2600 | goto out_unlock; | |
3b00b203 | 2601 | |
4d4471cb KU |
2602 | dm_table_postsuspend_targets(map); |
2603 | ||
d287483d | 2604 | out_unlock: |
e61290a4 | 2605 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2606 | return r; |
1da177e4 LT |
2607 | } |
2608 | ||
ffcc3936 MS |
2609 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2610 | { | |
2611 | if (map) { | |
2612 | int r = dm_table_resume_targets(map); | |
2613 | if (r) | |
2614 | return r; | |
2615 | } | |
2616 | ||
2617 | dm_queue_flush(md); | |
2618 | ||
2619 | /* | |
2620 | * Flushing deferred I/Os must be done after targets are resumed | |
2621 | * so that mapping of targets can work correctly. | |
2622 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2623 | */ | |
2624 | if (dm_request_based(md)) | |
eca7ee6d | 2625 | dm_start_queue(md->queue); |
ffcc3936 MS |
2626 | |
2627 | unlock_fs(md); | |
2628 | ||
2629 | return 0; | |
2630 | } | |
2631 | ||
1da177e4 LT |
2632 | int dm_resume(struct mapped_device *md) |
2633 | { | |
8dc23658 | 2634 | int r; |
cf222b37 | 2635 | struct dm_table *map = NULL; |
1da177e4 | 2636 | |
ffcc3936 | 2637 | retry: |
8dc23658 | 2638 | r = -EINVAL; |
ffcc3936 MS |
2639 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2640 | ||
4f186f8b | 2641 | if (!dm_suspended_md(md)) |
cf222b37 | 2642 | goto out; |
cf222b37 | 2643 | |
ffcc3936 MS |
2644 | if (dm_suspended_internally_md(md)) { |
2645 | /* already internally suspended, wait for internal resume */ | |
2646 | mutex_unlock(&md->suspend_lock); | |
2647 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2648 | if (r) | |
2649 | return r; | |
2650 | goto retry; | |
2651 | } | |
2652 | ||
a12f5d48 | 2653 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2654 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2655 | goto out; |
1da177e4 | 2656 | |
ffcc3936 | 2657 | r = __dm_resume(md, map); |
8757b776 MB |
2658 | if (r) |
2659 | goto out; | |
2ca3310e | 2660 | |
2ca3310e | 2661 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2662 | out: |
e61290a4 | 2663 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2664 | |
cf222b37 | 2665 | return r; |
1da177e4 LT |
2666 | } |
2667 | ||
fd2ed4d2 MP |
2668 | /* |
2669 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2670 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2671 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2672 | */ |
2673 | ||
ffcc3936 | 2674 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2675 | { |
ffcc3936 MS |
2676 | struct dm_table *map = NULL; |
2677 | ||
1ea0654e BVA |
2678 | lockdep_assert_held(&md->suspend_lock); |
2679 | ||
96b26c8c | 2680 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2681 | return; /* nested internal suspend */ |
2682 | ||
2683 | if (dm_suspended_md(md)) { | |
2684 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2685 | return; /* nest suspend */ | |
2686 | } | |
2687 | ||
a12f5d48 | 2688 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2689 | |
2690 | /* | |
2691 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2692 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2693 | * would require changing .presuspend to return an error -- avoid this | |
2694 | * until there is a need for more elaborate variants of internal suspend. | |
2695 | */ | |
eaf9a736 MS |
2696 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2697 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2698 | |
2699 | dm_table_postsuspend_targets(map); | |
2700 | } | |
2701 | ||
2702 | static void __dm_internal_resume(struct mapped_device *md) | |
2703 | { | |
96b26c8c MP |
2704 | BUG_ON(!md->internal_suspend_count); |
2705 | ||
2706 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2707 | return; /* resume from nested internal suspend */ |
2708 | ||
fd2ed4d2 | 2709 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2710 | goto done; /* resume from nested suspend */ |
2711 | ||
2712 | /* | |
2713 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2714 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2715 | */ | |
2716 | (void) __dm_resume(md, NULL); | |
2717 | ||
2718 | done: | |
2719 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2720 | smp_mb__after_atomic(); | |
2721 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2722 | } | |
2723 | ||
2724 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2725 | { | |
2726 | mutex_lock(&md->suspend_lock); | |
2727 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2728 | mutex_unlock(&md->suspend_lock); | |
2729 | } | |
2730 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2731 | ||
2732 | void dm_internal_resume(struct mapped_device *md) | |
2733 | { | |
2734 | mutex_lock(&md->suspend_lock); | |
2735 | __dm_internal_resume(md); | |
2736 | mutex_unlock(&md->suspend_lock); | |
2737 | } | |
2738 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2739 | ||
2740 | /* | |
2741 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2742 | * which prevents interaction with userspace-driven suspend. | |
2743 | */ | |
2744 | ||
2745 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2746 | { | |
2747 | mutex_lock(&md->suspend_lock); | |
2748 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2749 | return; |
2750 | ||
2751 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2752 | synchronize_srcu(&md->io_barrier); | |
2753 | flush_workqueue(md->wq); | |
2754 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2755 | } | |
b735fede | 2756 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2757 | |
ffcc3936 | 2758 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2759 | { |
ffcc3936 | 2760 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2761 | goto done; |
2762 | ||
2763 | dm_queue_flush(md); | |
2764 | ||
2765 | done: | |
2766 | mutex_unlock(&md->suspend_lock); | |
2767 | } | |
b735fede | 2768 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2769 | |
1da177e4 LT |
2770 | /*----------------------------------------------------------------- |
2771 | * Event notification. | |
2772 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2773 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2774 | unsigned cookie) |
69267a30 | 2775 | { |
60935eb2 MB |
2776 | char udev_cookie[DM_COOKIE_LENGTH]; |
2777 | char *envp[] = { udev_cookie, NULL }; | |
2778 | ||
2779 | if (!cookie) | |
3abf85b5 | 2780 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2781 | else { |
2782 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2783 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2784 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2785 | action, envp); | |
60935eb2 | 2786 | } |
69267a30 AK |
2787 | } |
2788 | ||
7a8c3d3b MA |
2789 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2790 | { | |
2791 | return atomic_add_return(1, &md->uevent_seq); | |
2792 | } | |
2793 | ||
1da177e4 LT |
2794 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2795 | { | |
2796 | return atomic_read(&md->event_nr); | |
2797 | } | |
2798 | ||
2799 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2800 | { | |
2801 | return wait_event_interruptible(md->eventq, | |
2802 | (event_nr != atomic_read(&md->event_nr))); | |
2803 | } | |
2804 | ||
7a8c3d3b MA |
2805 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2806 | { | |
2807 | unsigned long flags; | |
2808 | ||
2809 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2810 | list_add(elist, &md->uevent_list); | |
2811 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2812 | } | |
2813 | ||
1da177e4 LT |
2814 | /* |
2815 | * The gendisk is only valid as long as you have a reference | |
2816 | * count on 'md'. | |
2817 | */ | |
2818 | struct gendisk *dm_disk(struct mapped_device *md) | |
2819 | { | |
2820 | return md->disk; | |
2821 | } | |
65ff5b7d | 2822 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2823 | |
784aae73 MB |
2824 | struct kobject *dm_kobject(struct mapped_device *md) |
2825 | { | |
2995fa78 | 2826 | return &md->kobj_holder.kobj; |
784aae73 MB |
2827 | } |
2828 | ||
784aae73 MB |
2829 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2830 | { | |
2831 | struct mapped_device *md; | |
2832 | ||
2995fa78 | 2833 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2834 | |
b9a41d21 HT |
2835 | spin_lock(&_minor_lock); |
2836 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2837 | md = NULL; | |
2838 | goto out; | |
2839 | } | |
784aae73 | 2840 | dm_get(md); |
b9a41d21 HT |
2841 | out: |
2842 | spin_unlock(&_minor_lock); | |
2843 | ||
784aae73 MB |
2844 | return md; |
2845 | } | |
2846 | ||
4f186f8b | 2847 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2848 | { |
2849 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2850 | } | |
2851 | ||
ffcc3936 MS |
2852 | int dm_suspended_internally_md(struct mapped_device *md) |
2853 | { | |
2854 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2855 | } | |
2856 | ||
2c140a24 MP |
2857 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2858 | { | |
2859 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2860 | } | |
2861 | ||
64dbce58 KU |
2862 | int dm_suspended(struct dm_target *ti) |
2863 | { | |
ecdb2e25 | 2864 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2865 | } |
2866 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2867 | ||
2e93ccc1 KU |
2868 | int dm_noflush_suspending(struct dm_target *ti) |
2869 | { | |
ecdb2e25 | 2870 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2871 | } |
2872 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2873 | ||
7e0d574f | 2874 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
0776aa0e MS |
2875 | unsigned integrity, unsigned per_io_data_size, |
2876 | unsigned min_pool_size) | |
e6ee8c0b | 2877 | { |
115485e8 | 2878 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2879 | unsigned int pool_size = 0; |
64f52b0e | 2880 | unsigned int front_pad, io_front_pad; |
e6ee8c0b KU |
2881 | |
2882 | if (!pools) | |
4e6e36c3 | 2883 | return NULL; |
e6ee8c0b | 2884 | |
78d8e58a MS |
2885 | switch (type) { |
2886 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2887 | case DM_TYPE_DAX_BIO_BASED: |
22c11858 | 2888 | case DM_TYPE_NVME_BIO_BASED: |
0776aa0e | 2889 | pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); |
30187e1d | 2890 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
64f52b0e MS |
2891 | io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio); |
2892 | pools->io_bs = bioset_create(pool_size, io_front_pad, 0); | |
2893 | if (!pools->io_bs) | |
2894 | goto out; | |
2895 | if (integrity && bioset_integrity_create(pools->io_bs, pool_size)) | |
eb8db831 | 2896 | goto out; |
78d8e58a MS |
2897 | break; |
2898 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2899 | case DM_TYPE_MQ_REQUEST_BASED: |
0776aa0e | 2900 | pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); |
78d8e58a | 2901 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2902 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2903 | break; |
2904 | default: | |
2905 | BUG(); | |
2906 | } | |
2907 | ||
4a3f54d9 | 2908 | pools->bs = bioset_create(pool_size, front_pad, 0); |
e6ee8c0b | 2909 | if (!pools->bs) |
5f015204 | 2910 | goto out; |
e6ee8c0b | 2911 | |
a91a2785 | 2912 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2913 | goto out; |
a91a2785 | 2914 | |
e6ee8c0b | 2915 | return pools; |
5f1b670d | 2916 | |
5f1b670d CH |
2917 | out: |
2918 | dm_free_md_mempools(pools); | |
78d8e58a | 2919 | |
4e6e36c3 | 2920 | return NULL; |
e6ee8c0b KU |
2921 | } |
2922 | ||
2923 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2924 | { | |
2925 | if (!pools) | |
2926 | return; | |
2927 | ||
e6ee8c0b KU |
2928 | if (pools->bs) |
2929 | bioset_free(pools->bs); | |
64f52b0e MS |
2930 | if (pools->io_bs) |
2931 | bioset_free(pools->io_bs); | |
e6ee8c0b KU |
2932 | |
2933 | kfree(pools); | |
2934 | } | |
2935 | ||
9c72bad1 CH |
2936 | struct dm_pr { |
2937 | u64 old_key; | |
2938 | u64 new_key; | |
2939 | u32 flags; | |
2940 | bool fail_early; | |
2941 | }; | |
2942 | ||
2943 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2944 | void *data) | |
71cdb697 CH |
2945 | { |
2946 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2947 | struct dm_table *table; |
2948 | struct dm_target *ti; | |
2949 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2950 | |
9c72bad1 CH |
2951 | table = dm_get_live_table(md, &srcu_idx); |
2952 | if (!table || !dm_table_get_size(table)) | |
2953 | goto out; | |
71cdb697 | 2954 | |
9c72bad1 CH |
2955 | /* We only support devices that have a single target */ |
2956 | if (dm_table_get_num_targets(table) != 1) | |
2957 | goto out; | |
2958 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2959 | |
9c72bad1 CH |
2960 | ret = -EINVAL; |
2961 | if (!ti->type->iterate_devices) | |
2962 | goto out; | |
2963 | ||
2964 | ret = ti->type->iterate_devices(ti, fn, data); | |
2965 | out: | |
2966 | dm_put_live_table(md, srcu_idx); | |
2967 | return ret; | |
2968 | } | |
2969 | ||
2970 | /* | |
2971 | * For register / unregister we need to manually call out to every path. | |
2972 | */ | |
2973 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
2974 | sector_t start, sector_t len, void *data) | |
2975 | { | |
2976 | struct dm_pr *pr = data; | |
2977 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
2978 | ||
2979 | if (!ops || !ops->pr_register) | |
2980 | return -EOPNOTSUPP; | |
2981 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
2982 | } | |
2983 | ||
2984 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
2985 | u32 flags) | |
2986 | { | |
2987 | struct dm_pr pr = { | |
2988 | .old_key = old_key, | |
2989 | .new_key = new_key, | |
2990 | .flags = flags, | |
2991 | .fail_early = true, | |
2992 | }; | |
2993 | int ret; | |
2994 | ||
2995 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
2996 | if (ret && new_key) { | |
2997 | /* unregister all paths if we failed to register any path */ | |
2998 | pr.old_key = new_key; | |
2999 | pr.new_key = 0; | |
3000 | pr.flags = 0; | |
3001 | pr.fail_early = false; | |
3002 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
3003 | } | |
3004 | ||
3005 | return ret; | |
71cdb697 CH |
3006 | } |
3007 | ||
3008 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3009 | u32 flags) |
71cdb697 CH |
3010 | { |
3011 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3012 | const struct pr_ops *ops; | |
71cdb697 | 3013 | fmode_t mode; |
956a4025 | 3014 | int r; |
71cdb697 | 3015 | |
956a4025 | 3016 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3017 | if (r < 0) |
3018 | return r; | |
3019 | ||
3020 | ops = bdev->bd_disk->fops->pr_ops; | |
3021 | if (ops && ops->pr_reserve) | |
3022 | r = ops->pr_reserve(bdev, key, type, flags); | |
3023 | else | |
3024 | r = -EOPNOTSUPP; | |
3025 | ||
956a4025 | 3026 | bdput(bdev); |
71cdb697 CH |
3027 | return r; |
3028 | } | |
3029 | ||
3030 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
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 | |
956a4025 | 3037 | r = dm_grab_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_release) | |
3043 | r = ops->pr_release(bdev, key, type); | |
3044 | else | |
3045 | r = -EOPNOTSUPP; | |
3046 | ||
956a4025 | 3047 | bdput(bdev); |
71cdb697 CH |
3048 | return r; |
3049 | } | |
3050 | ||
3051 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3052 | enum pr_type type, bool abort) |
71cdb697 CH |
3053 | { |
3054 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3055 | const struct pr_ops *ops; | |
71cdb697 | 3056 | fmode_t mode; |
956a4025 | 3057 | int r; |
71cdb697 | 3058 | |
956a4025 | 3059 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3060 | if (r < 0) |
3061 | return r; | |
3062 | ||
3063 | ops = bdev->bd_disk->fops->pr_ops; | |
3064 | if (ops && ops->pr_preempt) | |
3065 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3066 | else | |
3067 | r = -EOPNOTSUPP; | |
3068 | ||
956a4025 | 3069 | bdput(bdev); |
71cdb697 CH |
3070 | return r; |
3071 | } | |
3072 | ||
3073 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
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 | |
956a4025 | 3080 | r = dm_grab_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_clear) | |
3086 | r = ops->pr_clear(bdev, key); | |
3087 | else | |
3088 | r = -EOPNOTSUPP; | |
3089 | ||
956a4025 | 3090 | bdput(bdev); |
71cdb697 CH |
3091 | return r; |
3092 | } | |
3093 | ||
3094 | static const struct pr_ops dm_pr_ops = { | |
3095 | .pr_register = dm_pr_register, | |
3096 | .pr_reserve = dm_pr_reserve, | |
3097 | .pr_release = dm_pr_release, | |
3098 | .pr_preempt = dm_pr_preempt, | |
3099 | .pr_clear = dm_pr_clear, | |
3100 | }; | |
3101 | ||
83d5cde4 | 3102 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3103 | .open = dm_blk_open, |
3104 | .release = dm_blk_close, | |
aa129a22 | 3105 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3106 | .getgeo = dm_blk_getgeo, |
71cdb697 | 3107 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3108 | .owner = THIS_MODULE |
3109 | }; | |
3110 | ||
f26c5719 DW |
3111 | static const struct dax_operations dm_dax_ops = { |
3112 | .direct_access = dm_dax_direct_access, | |
7e026c8c | 3113 | .copy_from_iter = dm_dax_copy_from_iter, |
f26c5719 DW |
3114 | }; |
3115 | ||
1da177e4 LT |
3116 | /* |
3117 | * module hooks | |
3118 | */ | |
3119 | module_init(dm_init); | |
3120 | module_exit(dm_exit); | |
3121 | ||
3122 | module_param(major, uint, 0); | |
3123 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3124 | |
e8603136 MS |
3125 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3126 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3127 | ||
115485e8 MS |
3128 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
3129 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
3130 | ||
1da177e4 LT |
3131 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3132 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3133 | MODULE_LICENSE("GPL"); |