]>
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 | ||
971888c4 | 461 | static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, |
5bd5e8d8 | 462 | struct block_device **bdev) |
971888c4 | 463 | __acquires(md->io_barrier) |
aa129a22 | 464 | { |
66482026 | 465 | struct dm_target *tgt; |
6c182cd8 | 466 | struct dm_table *map; |
971888c4 | 467 | int r; |
aa129a22 | 468 | |
6c182cd8 | 469 | retry: |
e56f81e0 | 470 | r = -ENOTTY; |
971888c4 | 471 | map = dm_get_live_table(md, srcu_idx); |
aa129a22 | 472 | if (!map || !dm_table_get_size(map)) |
971888c4 | 473 | return r; |
aa129a22 MB |
474 | |
475 | /* We only support devices that have a single target */ | |
476 | if (dm_table_get_num_targets(map) != 1) | |
971888c4 | 477 | return r; |
aa129a22 | 478 | |
66482026 MS |
479 | tgt = dm_table_get_target(map, 0); |
480 | if (!tgt->type->prepare_ioctl) | |
971888c4 | 481 | return r; |
519049af | 482 | |
971888c4 MS |
483 | if (dm_suspended_md(md)) |
484 | return -EAGAIN; | |
aa129a22 | 485 | |
5bd5e8d8 | 486 | r = tgt->type->prepare_ioctl(tgt, bdev); |
5bbbfdf6 | 487 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
971888c4 | 488 | dm_put_live_table(md, *srcu_idx); |
6c182cd8 HR |
489 | msleep(10); |
490 | goto retry; | |
491 | } | |
971888c4 | 492 | |
e56f81e0 CH |
493 | return r; |
494 | } | |
495 | ||
971888c4 MS |
496 | static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) |
497 | __releases(md->io_barrier) | |
498 | { | |
499 | dm_put_live_table(md, srcu_idx); | |
500 | } | |
501 | ||
e56f81e0 CH |
502 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
503 | unsigned int cmd, unsigned long arg) | |
504 | { | |
505 | struct mapped_device *md = bdev->bd_disk->private_data; | |
971888c4 | 506 | int r, srcu_idx; |
e56f81e0 | 507 | |
5bd5e8d8 | 508 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
e56f81e0 | 509 | if (r < 0) |
971888c4 | 510 | goto out; |
6c182cd8 | 511 | |
e56f81e0 CH |
512 | if (r > 0) { |
513 | /* | |
e980f623 CH |
514 | * Target determined this ioctl is being issued against a |
515 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 516 | */ |
e980f623 CH |
517 | if (!capable(CAP_SYS_RAWIO)) { |
518 | DMWARN_LIMIT( | |
519 | "%s: sending ioctl %x to DM device without required privilege.", | |
520 | current->comm, cmd); | |
521 | r = -ENOIOCTLCMD; | |
e56f81e0 | 522 | goto out; |
e980f623 | 523 | } |
e56f81e0 | 524 | } |
6c182cd8 | 525 | |
66482026 | 526 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 527 | out: |
971888c4 | 528 | dm_unprepare_ioctl(md, srcu_idx); |
aa129a22 MB |
529 | return r; |
530 | } | |
531 | ||
978e51ba MS |
532 | static void start_io_acct(struct dm_io *io); |
533 | ||
534 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) | |
1da177e4 | 535 | { |
64f52b0e MS |
536 | struct dm_io *io; |
537 | struct dm_target_io *tio; | |
538 | struct bio *clone; | |
539 | ||
540 | clone = bio_alloc_bioset(GFP_NOIO, 0, md->io_bs); | |
541 | if (!clone) | |
542 | return NULL; | |
543 | ||
544 | tio = container_of(clone, struct dm_target_io, clone); | |
545 | tio->inside_dm_io = true; | |
546 | tio->io = NULL; | |
547 | ||
548 | io = container_of(tio, struct dm_io, tio); | |
549 | io->magic = DM_IO_MAGIC; | |
978e51ba MS |
550 | io->status = 0; |
551 | atomic_set(&io->io_count, 1); | |
552 | io->orig_bio = bio; | |
553 | io->md = md; | |
554 | spin_lock_init(&io->endio_lock); | |
555 | ||
556 | start_io_acct(io); | |
64f52b0e MS |
557 | |
558 | return io; | |
1da177e4 LT |
559 | } |
560 | ||
028867ac | 561 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 | 562 | { |
64f52b0e MS |
563 | bio_put(&io->tio.clone); |
564 | } | |
565 | ||
566 | static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, | |
567 | unsigned target_bio_nr, gfp_t gfp_mask) | |
568 | { | |
569 | struct dm_target_io *tio; | |
570 | ||
571 | if (!ci->io->tio.io) { | |
572 | /* the dm_target_io embedded in ci->io is available */ | |
573 | tio = &ci->io->tio; | |
574 | } else { | |
bc02cdbe | 575 | struct bio *clone = bio_alloc_bioset(gfp_mask, 0, ci->io->md->bs); |
64f52b0e MS |
576 | if (!clone) |
577 | return NULL; | |
578 | ||
579 | tio = container_of(clone, struct dm_target_io, clone); | |
580 | tio->inside_dm_io = false; | |
581 | } | |
582 | ||
583 | tio->magic = DM_TIO_MAGIC; | |
584 | tio->io = ci->io; | |
585 | tio->ti = ti; | |
586 | tio->target_bio_nr = target_bio_nr; | |
587 | ||
588 | return tio; | |
1da177e4 LT |
589 | } |
590 | ||
cfae7529 | 591 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 592 | { |
64f52b0e MS |
593 | if (tio->inside_dm_io) |
594 | return; | |
dba14160 | 595 | bio_put(&tio->clone); |
1da177e4 LT |
596 | } |
597 | ||
4cc96131 | 598 | int md_in_flight(struct mapped_device *md) |
90abb8c4 KU |
599 | { |
600 | return atomic_read(&md->pending[READ]) + | |
601 | atomic_read(&md->pending[WRITE]); | |
602 | } | |
603 | ||
3eaf840e JNN |
604 | static void start_io_acct(struct dm_io *io) |
605 | { | |
606 | struct mapped_device *md = io->md; | |
745dc570 | 607 | struct bio *bio = io->orig_bio; |
fd2ed4d2 | 608 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
609 | |
610 | io->start_time = jiffies; | |
611 | ||
f3986374 MS |
612 | generic_start_io_acct(md->queue, rw, bio_sectors(bio), &dm_disk(md)->part0); |
613 | ||
1e9bb880 | 614 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
f3986374 | 615 | atomic_inc_return(&md->pending[rw])); |
fd2ed4d2 MP |
616 | |
617 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
618 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
619 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
620 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
621 | } |
622 | ||
d221d2e7 | 623 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
624 | { |
625 | struct mapped_device *md = io->md; | |
745dc570 | 626 | struct bio *bio = io->orig_bio; |
3eaf840e | 627 | unsigned long duration = jiffies - io->start_time; |
18c0b223 | 628 | int pending; |
3eaf840e JNN |
629 | int rw = bio_data_dir(bio); |
630 | ||
d62e26b3 | 631 | generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 632 | |
fd2ed4d2 | 633 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
634 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
635 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
636 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 637 | |
af7e466a MP |
638 | /* |
639 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 640 | * a flush. |
af7e466a | 641 | */ |
1e9bb880 SL |
642 | pending = atomic_dec_return(&md->pending[rw]); |
643 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 644 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 645 | |
d221d2e7 MP |
646 | /* nudge anyone waiting on suspend queue */ |
647 | if (!pending) | |
648 | wake_up(&md->wait); | |
3eaf840e JNN |
649 | } |
650 | ||
1da177e4 LT |
651 | /* |
652 | * Add the bio to the list of deferred io. | |
653 | */ | |
92c63902 | 654 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 655 | { |
05447420 | 656 | unsigned long flags; |
1da177e4 | 657 | |
05447420 | 658 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 659 | bio_list_add(&md->deferred, bio); |
05447420 | 660 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 661 | queue_work(md->wq, &md->work); |
1da177e4 LT |
662 | } |
663 | ||
664 | /* | |
665 | * Everyone (including functions in this file), should use this | |
666 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 667 | * dm_put_live_table() when finished. |
1da177e4 | 668 | */ |
83d5e5b0 | 669 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 670 | { |
83d5e5b0 MP |
671 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
672 | ||
673 | return srcu_dereference(md->map, &md->io_barrier); | |
674 | } | |
1da177e4 | 675 | |
83d5e5b0 MP |
676 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
677 | { | |
678 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
679 | } | |
680 | ||
681 | void dm_sync_table(struct mapped_device *md) | |
682 | { | |
683 | synchronize_srcu(&md->io_barrier); | |
684 | synchronize_rcu_expedited(); | |
685 | } | |
686 | ||
687 | /* | |
688 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
689 | * The caller must not block between these two functions. | |
690 | */ | |
691 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
692 | { | |
693 | rcu_read_lock(); | |
694 | return rcu_dereference(md->map); | |
695 | } | |
1da177e4 | 696 | |
83d5e5b0 MP |
697 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
698 | { | |
699 | rcu_read_unlock(); | |
1da177e4 LT |
700 | } |
701 | ||
971888c4 MS |
702 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
703 | ||
86f1152b BM |
704 | /* |
705 | * Open a table device so we can use it as a map destination. | |
706 | */ | |
707 | static int open_table_device(struct table_device *td, dev_t dev, | |
708 | struct mapped_device *md) | |
709 | { | |
86f1152b BM |
710 | struct block_device *bdev; |
711 | ||
712 | int r; | |
713 | ||
714 | BUG_ON(td->dm_dev.bdev); | |
715 | ||
519049af | 716 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr); |
86f1152b BM |
717 | if (IS_ERR(bdev)) |
718 | return PTR_ERR(bdev); | |
719 | ||
720 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
721 | if (r) { | |
722 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
723 | return r; | |
724 | } | |
725 | ||
726 | td->dm_dev.bdev = bdev; | |
817bf402 | 727 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
728 | return 0; |
729 | } | |
730 | ||
731 | /* | |
732 | * Close a table device that we've been using. | |
733 | */ | |
734 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
735 | { | |
736 | if (!td->dm_dev.bdev) | |
737 | return; | |
738 | ||
739 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
740 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 741 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 742 | td->dm_dev.bdev = NULL; |
817bf402 | 743 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
744 | } |
745 | ||
746 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
747 | fmode_t mode) { | |
748 | struct table_device *td; | |
749 | ||
750 | list_for_each_entry(td, l, list) | |
751 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
752 | return td; | |
753 | ||
754 | return NULL; | |
755 | } | |
756 | ||
757 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
758 | struct dm_dev **result) { | |
759 | int r; | |
760 | struct table_device *td; | |
761 | ||
762 | mutex_lock(&md->table_devices_lock); | |
763 | td = find_table_device(&md->table_devices, dev, mode); | |
764 | if (!td) { | |
115485e8 | 765 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
766 | if (!td) { |
767 | mutex_unlock(&md->table_devices_lock); | |
768 | return -ENOMEM; | |
769 | } | |
770 | ||
771 | td->dm_dev.mode = mode; | |
772 | td->dm_dev.bdev = NULL; | |
773 | ||
774 | if ((r = open_table_device(td, dev, md))) { | |
775 | mutex_unlock(&md->table_devices_lock); | |
776 | kfree(td); | |
777 | return r; | |
778 | } | |
779 | ||
780 | format_dev_t(td->dm_dev.name, dev); | |
781 | ||
b0b4d7c6 | 782 | refcount_set(&td->count, 1); |
86f1152b | 783 | list_add(&td->list, &md->table_devices); |
b0b4d7c6 ER |
784 | } else { |
785 | refcount_inc(&td->count); | |
86f1152b | 786 | } |
86f1152b BM |
787 | mutex_unlock(&md->table_devices_lock); |
788 | ||
789 | *result = &td->dm_dev; | |
790 | return 0; | |
791 | } | |
792 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
793 | ||
794 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
795 | { | |
796 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
797 | ||
798 | mutex_lock(&md->table_devices_lock); | |
b0b4d7c6 | 799 | if (refcount_dec_and_test(&td->count)) { |
86f1152b BM |
800 | close_table_device(td, md); |
801 | list_del(&td->list); | |
802 | kfree(td); | |
803 | } | |
804 | mutex_unlock(&md->table_devices_lock); | |
805 | } | |
806 | EXPORT_SYMBOL(dm_put_table_device); | |
807 | ||
808 | static void free_table_devices(struct list_head *devices) | |
809 | { | |
810 | struct list_head *tmp, *next; | |
811 | ||
812 | list_for_each_safe(tmp, next, devices) { | |
813 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
814 | ||
815 | DMWARN("dm_destroy: %s still exists with %d references", | |
b0b4d7c6 | 816 | td->dm_dev.name, refcount_read(&td->count)); |
86f1152b BM |
817 | kfree(td); |
818 | } | |
819 | } | |
820 | ||
3ac51e74 DW |
821 | /* |
822 | * Get the geometry associated with a dm device | |
823 | */ | |
824 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
825 | { | |
826 | *geo = md->geometry; | |
827 | ||
828 | return 0; | |
829 | } | |
830 | ||
831 | /* | |
832 | * Set the geometry of a device. | |
833 | */ | |
834 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
835 | { | |
836 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
837 | ||
838 | if (geo->start > sz) { | |
839 | DMWARN("Start sector is beyond the geometry limits."); | |
840 | return -EINVAL; | |
841 | } | |
842 | ||
843 | md->geometry = *geo; | |
844 | ||
845 | return 0; | |
846 | } | |
847 | ||
2e93ccc1 KU |
848 | static int __noflush_suspending(struct mapped_device *md) |
849 | { | |
850 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
851 | } | |
852 | ||
1da177e4 LT |
853 | /* |
854 | * Decrements the number of outstanding ios that a bio has been | |
855 | * cloned into, completing the original io if necc. | |
856 | */ | |
4e4cbee9 | 857 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 858 | { |
2e93ccc1 | 859 | unsigned long flags; |
4e4cbee9 | 860 | blk_status_t io_error; |
b35f8caa MB |
861 | struct bio *bio; |
862 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
863 | |
864 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
865 | if (unlikely(error)) { |
866 | spin_lock_irqsave(&io->endio_lock, flags); | |
745dc570 | 867 | if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) |
4e4cbee9 | 868 | io->status = error; |
f88fb981 KU |
869 | spin_unlock_irqrestore(&io->endio_lock, flags); |
870 | } | |
1da177e4 LT |
871 | |
872 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 873 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
874 | /* |
875 | * Target requested pushing back the I/O. | |
2e93ccc1 | 876 | */ |
022c2611 | 877 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 | 878 | if (__noflush_suspending(md)) |
745dc570 MS |
879 | /* NOTE early return due to BLK_STS_DM_REQUEUE below */ |
880 | bio_list_add_head(&md->deferred, io->orig_bio); | |
6a8736d1 | 881 | else |
2e93ccc1 | 882 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 883 | io->status = BLK_STS_IOERR; |
022c2611 | 884 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
885 | } |
886 | ||
4e4cbee9 | 887 | io_error = io->status; |
745dc570 | 888 | bio = io->orig_bio; |
6a8736d1 TH |
889 | end_io_acct(io); |
890 | free_io(md, io); | |
891 | ||
4e4cbee9 | 892 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 893 | return; |
2e93ccc1 | 894 | |
1eff9d32 | 895 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 896 | /* |
6a8736d1 | 897 | * Preflush done for flush with data, reissue |
28a8f0d3 | 898 | * without REQ_PREFLUSH. |
af7e466a | 899 | */ |
1eff9d32 | 900 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 901 | queue_io(md, bio); |
af7e466a | 902 | } else { |
b372d360 | 903 | /* done with normal IO or empty flush */ |
8dd601fa N |
904 | if (io_error) |
905 | bio->bi_status = io_error; | |
4246a0b6 | 906 | bio_endio(bio); |
b35f8caa | 907 | } |
1da177e4 LT |
908 | } |
909 | } | |
910 | ||
4cc96131 | 911 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
912 | { |
913 | struct queue_limits *limits = dm_get_queue_limits(md); | |
914 | ||
915 | /* device doesn't really support WRITE SAME, disable it */ | |
916 | limits->max_write_same_sectors = 0; | |
917 | } | |
918 | ||
ac62d620 CH |
919 | void disable_write_zeroes(struct mapped_device *md) |
920 | { | |
921 | struct queue_limits *limits = dm_get_queue_limits(md); | |
922 | ||
923 | /* device doesn't really support WRITE ZEROES, disable it */ | |
924 | limits->max_write_zeroes_sectors = 0; | |
925 | } | |
926 | ||
4246a0b6 | 927 | static void clone_endio(struct bio *bio) |
1da177e4 | 928 | { |
4e4cbee9 | 929 | blk_status_t error = bio->bi_status; |
bfc6d41c | 930 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 931 | struct dm_io *io = tio->io; |
9faf400f | 932 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
933 | dm_endio_fn endio = tio->ti->type->end_io; |
934 | ||
978e51ba | 935 | if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) { |
ac62d620 | 936 | if (bio_op(bio) == REQ_OP_WRITE_SAME && |
74d46992 | 937 | !bio->bi_disk->queue->limits.max_write_same_sectors) |
ac62d620 CH |
938 | disable_write_same(md); |
939 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && | |
74d46992 | 940 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) |
ac62d620 CH |
941 | disable_write_zeroes(md); |
942 | } | |
7eee4ae2 | 943 | |
1be56909 | 944 | if (endio) { |
4e4cbee9 | 945 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
946 | switch (r) { |
947 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 948 | error = BLK_STS_DM_REQUEUE; |
1be56909 CH |
949 | /*FALLTHRU*/ |
950 | case DM_ENDIO_DONE: | |
951 | break; | |
952 | case DM_ENDIO_INCOMPLETE: | |
953 | /* The target will handle the io */ | |
954 | return; | |
955 | default: | |
956 | DMWARN("unimplemented target endio return value: %d", r); | |
957 | BUG(); | |
958 | } | |
959 | } | |
960 | ||
cfae7529 | 961 | free_tio(tio); |
b35f8caa | 962 | dec_pending(io, error); |
1da177e4 LT |
963 | } |
964 | ||
56a67df7 MS |
965 | /* |
966 | * Return maximum size of I/O possible at the supplied sector up to the current | |
967 | * target boundary. | |
968 | */ | |
969 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
970 | { | |
971 | sector_t target_offset = dm_target_offset(ti, sector); | |
972 | ||
973 | return ti->len - target_offset; | |
974 | } | |
975 | ||
976 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 977 | { |
56a67df7 | 978 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 979 | sector_t offset, max_len; |
1da177e4 LT |
980 | |
981 | /* | |
542f9038 | 982 | * Does the target need to split even further? |
1da177e4 | 983 | */ |
542f9038 MS |
984 | if (ti->max_io_len) { |
985 | offset = dm_target_offset(ti, sector); | |
986 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
987 | max_len = sector_div(offset, ti->max_io_len); | |
988 | else | |
989 | max_len = offset & (ti->max_io_len - 1); | |
990 | max_len = ti->max_io_len - max_len; | |
991 | ||
992 | if (len > max_len) | |
993 | len = max_len; | |
1da177e4 LT |
994 | } |
995 | ||
996 | return len; | |
997 | } | |
998 | ||
542f9038 MS |
999 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1000 | { | |
1001 | if (len > UINT_MAX) { | |
1002 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1003 | (unsigned long long)len, UINT_MAX); | |
1004 | ti->error = "Maximum size of target IO is too large"; | |
1005 | return -EINVAL; | |
1006 | } | |
1007 | ||
8f50e358 ML |
1008 | /* |
1009 | * BIO based queue uses its own splitting. When multipage bvecs | |
1010 | * is switched on, size of the incoming bio may be too big to | |
1011 | * be handled in some targets, such as crypt. | |
1012 | * | |
1013 | * When these targets are ready for the big bio, we can remove | |
1014 | * the limit. | |
1015 | */ | |
1016 | ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE); | |
542f9038 MS |
1017 | |
1018 | return 0; | |
1019 | } | |
1020 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1021 | ||
f26c5719 | 1022 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
3d97c829 MS |
1023 | sector_t sector, int *srcu_idx) |
1024 | __acquires(md->io_barrier) | |
545ed20e | 1025 | { |
545ed20e TK |
1026 | struct dm_table *map; |
1027 | struct dm_target *ti; | |
545ed20e | 1028 | |
f26c5719 | 1029 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1030 | if (!map) |
f26c5719 | 1031 | return NULL; |
545ed20e TK |
1032 | |
1033 | ti = dm_table_find_target(map, sector); | |
1034 | if (!dm_target_is_valid(ti)) | |
f26c5719 | 1035 | return NULL; |
545ed20e | 1036 | |
f26c5719 DW |
1037 | return ti; |
1038 | } | |
545ed20e | 1039 | |
f26c5719 | 1040 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1041 | long nr_pages, void **kaddr, pfn_t *pfn) |
f26c5719 DW |
1042 | { |
1043 | struct mapped_device *md = dax_get_private(dax_dev); | |
1044 | sector_t sector = pgoff * PAGE_SECTORS; | |
1045 | struct dm_target *ti; | |
1046 | long len, ret = -EIO; | |
1047 | int srcu_idx; | |
545ed20e | 1048 | |
f26c5719 | 1049 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1050 | |
f26c5719 DW |
1051 | if (!ti) |
1052 | goto out; | |
1053 | if (!ti->type->direct_access) | |
1054 | goto out; | |
1055 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
1056 | if (len < 1) | |
1057 | goto out; | |
1058 | nr_pages = min(len, nr_pages); | |
545ed20e | 1059 | if (ti->type->direct_access) |
817bf402 DW |
1060 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
1061 | ||
f26c5719 | 1062 | out: |
545ed20e | 1063 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1064 | |
1065 | return ret; | |
545ed20e TK |
1066 | } |
1067 | ||
7e026c8c | 1068 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1069 | void *addr, size_t bytes, struct iov_iter *i) |
7e026c8c DW |
1070 | { |
1071 | struct mapped_device *md = dax_get_private(dax_dev); | |
1072 | sector_t sector = pgoff * PAGE_SECTORS; | |
1073 | struct dm_target *ti; | |
1074 | long ret = 0; | |
1075 | int srcu_idx; | |
1076 | ||
1077 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1078 | ||
1079 | if (!ti) | |
1080 | goto out; | |
1081 | if (!ti->type->dax_copy_from_iter) { | |
1082 | ret = copy_from_iter(addr, bytes, i); | |
1083 | goto out; | |
1084 | } | |
1085 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
1086 | out: | |
1087 | dm_put_live_table(md, srcu_idx); | |
1088 | ||
1089 | return ret; | |
1090 | } | |
1091 | ||
1dd40c3e MP |
1092 | /* |
1093 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
c06b3e58 | 1094 | * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET. |
1dd40c3e MP |
1095 | * |
1096 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1097 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1098 | * sent in a next bio. | |
1099 | * | |
1100 | * A diagram that explains the arithmetics: | |
1101 | * +--------------------+---------------+-------+ | |
1102 | * | 1 | 2 | 3 | | |
1103 | * +--------------------+---------------+-------+ | |
1104 | * | |
1105 | * <-------------- *tio->len_ptr ---------------> | |
1106 | * <------- bi_size -------> | |
1107 | * <-- n_sectors --> | |
1108 | * | |
1109 | * Region 1 was already iterated over with bio_advance or similar function. | |
1110 | * (it may be empty if the target doesn't use bio_advance) | |
1111 | * Region 2 is the remaining bio size that the target wants to process. | |
1112 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1113 | * to make it empty) | |
1114 | * The target requires that region 3 is to be sent in the next bio. | |
1115 | * | |
1116 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1117 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1118 | * copies of the bio. | |
1119 | */ | |
1120 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1121 | { | |
1122 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1123 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1124 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1125 | BUG_ON(bi_size > *tio->len_ptr); |
1126 | BUG_ON(n_sectors > bi_size); | |
1127 | *tio->len_ptr -= bi_size - n_sectors; | |
1128 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1129 | } | |
1130 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1131 | ||
10999307 DLM |
1132 | /* |
1133 | * The zone descriptors obtained with a zone report indicate | |
1134 | * zone positions within the target device. The zone descriptors | |
1135 | * must be remapped to match their position within the dm device. | |
1136 | * A target may call dm_remap_zone_report after completion of a | |
1137 | * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained | |
1138 | * from the target device mapping to the dm device. | |
1139 | */ | |
1140 | void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start) | |
1141 | { | |
1142 | #ifdef CONFIG_BLK_DEV_ZONED | |
1143 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
745dc570 | 1144 | struct bio *report_bio = tio->io->orig_bio; |
10999307 DLM |
1145 | struct blk_zone_report_hdr *hdr = NULL; |
1146 | struct blk_zone *zone; | |
1147 | unsigned int nr_rep = 0; | |
1148 | unsigned int ofst; | |
1149 | struct bio_vec bvec; | |
1150 | struct bvec_iter iter; | |
1151 | void *addr; | |
1152 | ||
1153 | if (bio->bi_status) | |
1154 | return; | |
1155 | ||
1156 | /* | |
1157 | * Remap the start sector of the reported zones. For sequential zones, | |
1158 | * also remap the write pointer position. | |
1159 | */ | |
1160 | bio_for_each_segment(bvec, report_bio, iter) { | |
1161 | addr = kmap_atomic(bvec.bv_page); | |
1162 | ||
1163 | /* Remember the report header in the first page */ | |
1164 | if (!hdr) { | |
1165 | hdr = addr; | |
1166 | ofst = sizeof(struct blk_zone_report_hdr); | |
1167 | } else | |
1168 | ofst = 0; | |
1169 | ||
1170 | /* Set zones start sector */ | |
1171 | while (hdr->nr_zones && ofst < bvec.bv_len) { | |
1172 | zone = addr + ofst; | |
1173 | if (zone->start >= start + ti->len) { | |
1174 | hdr->nr_zones = 0; | |
1175 | break; | |
1176 | } | |
1177 | zone->start = zone->start + ti->begin - start; | |
1178 | if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { | |
1179 | if (zone->cond == BLK_ZONE_COND_FULL) | |
1180 | zone->wp = zone->start + zone->len; | |
1181 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
1182 | zone->wp = zone->start; | |
1183 | else | |
1184 | zone->wp = zone->wp + ti->begin - start; | |
1185 | } | |
1186 | ofst += sizeof(struct blk_zone); | |
1187 | hdr->nr_zones--; | |
1188 | nr_rep++; | |
1189 | } | |
1190 | ||
1191 | if (addr != hdr) | |
1192 | kunmap_atomic(addr); | |
1193 | ||
1194 | if (!hdr->nr_zones) | |
1195 | break; | |
1196 | } | |
1197 | ||
1198 | if (hdr) { | |
1199 | hdr->nr_zones = nr_rep; | |
1200 | kunmap_atomic(hdr); | |
1201 | } | |
1202 | ||
1203 | bio_advance(report_bio, report_bio->bi_iter.bi_size); | |
1204 | ||
1205 | #else /* !CONFIG_BLK_DEV_ZONED */ | |
1206 | bio->bi_status = BLK_STS_NOTSUPP; | |
1207 | #endif | |
1208 | } | |
1209 | EXPORT_SYMBOL_GPL(dm_remap_zone_report); | |
1210 | ||
978e51ba | 1211 | static blk_qc_t __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1212 | { |
1213 | int r; | |
2056a782 | 1214 | sector_t sector; |
dba14160 | 1215 | struct bio *clone = &tio->clone; |
64f52b0e | 1216 | struct dm_io *io = tio->io; |
978e51ba | 1217 | struct mapped_device *md = io->md; |
bd2a49b8 | 1218 | struct dm_target *ti = tio->ti; |
978e51ba | 1219 | blk_qc_t ret = BLK_QC_T_NONE; |
1da177e4 | 1220 | |
1da177e4 | 1221 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1222 | |
1223 | /* | |
1224 | * Map the clone. If r == 0 we don't need to do | |
1225 | * anything, the target has assumed ownership of | |
1226 | * this io. | |
1227 | */ | |
64f52b0e | 1228 | atomic_inc(&io->io_count); |
4f024f37 | 1229 | sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1230 | |
7de3ee57 | 1231 | r = ti->type->map(ti, clone); |
846785e6 CH |
1232 | switch (r) { |
1233 | case DM_MAPIO_SUBMITTED: | |
1234 | break; | |
1235 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1236 | /* the bio has been remapped so dispatch it */ |
74d46992 | 1237 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
64f52b0e | 1238 | bio_dev(io->orig_bio), sector); |
978e51ba MS |
1239 | if (md->type == DM_TYPE_NVME_BIO_BASED) |
1240 | ret = direct_make_request(clone); | |
1241 | else | |
1242 | ret = generic_make_request(clone); | |
846785e6 CH |
1243 | break; |
1244 | case DM_MAPIO_KILL: | |
4e4cbee9 | 1245 | free_tio(tio); |
64f52b0e | 1246 | dec_pending(io, BLK_STS_IOERR); |
4e4cbee9 | 1247 | break; |
846785e6 | 1248 | case DM_MAPIO_REQUEUE: |
cfae7529 | 1249 | free_tio(tio); |
64f52b0e | 1250 | dec_pending(io, BLK_STS_DM_REQUEUE); |
846785e6 CH |
1251 | break; |
1252 | default: | |
45cbcd79 KU |
1253 | DMWARN("unimplemented target map return value: %d", r); |
1254 | BUG(); | |
1da177e4 | 1255 | } |
1da177e4 | 1256 | |
978e51ba | 1257 | return ret; |
1da177e4 | 1258 | } |
1da177e4 | 1259 | |
e0d6609a | 1260 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1261 | { |
4f024f37 KO |
1262 | bio->bi_iter.bi_sector = sector; |
1263 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1264 | } |
1265 | ||
1266 | /* | |
1267 | * Creates a bio that consists of range of complete bvecs. | |
1268 | */ | |
c80914e8 MS |
1269 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1270 | sector_t sector, unsigned len) | |
1da177e4 | 1271 | { |
dba14160 | 1272 | struct bio *clone = &tio->clone; |
1da177e4 | 1273 | |
1c3b13e6 KO |
1274 | __bio_clone_fast(clone, bio); |
1275 | ||
e2460f2a MP |
1276 | if (unlikely(bio_integrity(bio) != NULL)) { |
1277 | int r; | |
1278 | ||
1279 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1280 | !dm_target_passes_integrity(tio->ti->type))) { | |
1281 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1282 | dm_device_name(tio->io->md), | |
1283 | tio->ti->type->name); | |
1284 | return -EIO; | |
1285 | } | |
1286 | ||
1287 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1288 | if (r < 0) |
1289 | return r; | |
1290 | } | |
bd2a49b8 | 1291 | |
264c869d DLM |
1292 | if (bio_op(bio) != REQ_OP_ZONE_REPORT) |
1293 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); | |
1c3b13e6 KO |
1294 | clone->bi_iter.bi_size = to_bytes(len); |
1295 | ||
e2460f2a | 1296 | if (unlikely(bio_integrity(bio) != NULL)) |
fbd08e76 | 1297 | bio_integrity_trim(clone); |
c80914e8 MS |
1298 | |
1299 | return 0; | |
1da177e4 LT |
1300 | } |
1301 | ||
318716dd MS |
1302 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
1303 | struct dm_target *ti, unsigned num_bios) | |
f9ab94ce | 1304 | { |
dba14160 | 1305 | struct dm_target_io *tio; |
318716dd | 1306 | int try; |
dba14160 | 1307 | |
318716dd MS |
1308 | if (!num_bios) |
1309 | return; | |
f9ab94ce | 1310 | |
318716dd MS |
1311 | if (num_bios == 1) { |
1312 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); | |
1313 | bio_list_add(blist, &tio->clone); | |
1314 | return; | |
1315 | } | |
9015df24 | 1316 | |
318716dd MS |
1317 | for (try = 0; try < 2; try++) { |
1318 | int bio_nr; | |
1319 | struct bio *bio; | |
1320 | ||
1321 | if (try) | |
bc02cdbe | 1322 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd MS |
1323 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
1324 | tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); | |
1325 | if (!tio) | |
1326 | break; | |
1327 | ||
1328 | bio_list_add(blist, &tio->clone); | |
1329 | } | |
1330 | if (try) | |
bc02cdbe | 1331 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1332 | if (bio_nr == num_bios) |
1333 | return; | |
1334 | ||
1335 | while ((bio = bio_list_pop(blist))) { | |
1336 | tio = container_of(bio, struct dm_target_io, clone); | |
1337 | free_tio(tio); | |
1338 | } | |
1339 | } | |
9015df24 AK |
1340 | } |
1341 | ||
978e51ba MS |
1342 | static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci, |
1343 | struct dm_target_io *tio, unsigned *len) | |
9015df24 | 1344 | { |
dba14160 | 1345 | struct bio *clone = &tio->clone; |
9015df24 | 1346 | |
1dd40c3e MP |
1347 | tio->len_ptr = len; |
1348 | ||
99778273 | 1349 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1350 | if (len) |
1dd40c3e | 1351 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1352 | |
978e51ba | 1353 | return __map_bio(tio); |
f9ab94ce MP |
1354 | } |
1355 | ||
14fe594d | 1356 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1357 | unsigned num_bios, unsigned *len) |
06a426ce | 1358 | { |
318716dd MS |
1359 | struct bio_list blist = BIO_EMPTY_LIST; |
1360 | struct bio *bio; | |
1361 | struct dm_target_io *tio; | |
1362 | ||
1363 | alloc_multiple_bios(&blist, ci, ti, num_bios); | |
06a426ce | 1364 | |
318716dd MS |
1365 | while ((bio = bio_list_pop(&blist))) { |
1366 | tio = container_of(bio, struct dm_target_io, clone); | |
978e51ba | 1367 | (void) __clone_and_map_simple_bio(ci, tio, len); |
318716dd | 1368 | } |
06a426ce MS |
1369 | } |
1370 | ||
14fe594d | 1371 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1372 | { |
06a426ce | 1373 | unsigned target_nr = 0; |
f9ab94ce MP |
1374 | struct dm_target *ti; |
1375 | ||
b372d360 | 1376 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1377 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1378 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1379 | |
f9ab94ce MP |
1380 | return 0; |
1381 | } | |
1382 | ||
c80914e8 | 1383 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
f31c21e4 | 1384 | sector_t sector, unsigned *len) |
5ae89a87 | 1385 | { |
dba14160 | 1386 | struct bio *bio = ci->bio; |
5ae89a87 | 1387 | struct dm_target_io *tio; |
f31c21e4 | 1388 | int r; |
5ae89a87 | 1389 | |
318716dd | 1390 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); |
f31c21e4 N |
1391 | tio->len_ptr = len; |
1392 | r = clone_bio(tio, bio, sector, *len); | |
1393 | if (r < 0) { | |
1394 | free_tio(tio); | |
1395 | return r; | |
b0d8ed4d | 1396 | } |
978e51ba | 1397 | (void) __map_bio(tio); |
c80914e8 | 1398 | |
f31c21e4 | 1399 | return 0; |
5ae89a87 MS |
1400 | } |
1401 | ||
55a62eef | 1402 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1403 | |
55a62eef | 1404 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1405 | { |
55a62eef | 1406 | return ti->num_discard_bios; |
23508a96 MS |
1407 | } |
1408 | ||
00716545 DS |
1409 | static unsigned get_num_secure_erase_bios(struct dm_target *ti) |
1410 | { | |
1411 | return ti->num_secure_erase_bios; | |
1412 | } | |
1413 | ||
55a62eef | 1414 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1415 | { |
55a62eef | 1416 | return ti->num_write_same_bios; |
23508a96 MS |
1417 | } |
1418 | ||
ac62d620 CH |
1419 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1420 | { | |
1421 | return ti->num_write_zeroes_bios; | |
1422 | } | |
1423 | ||
23508a96 | 1424 | typedef bool (*is_split_required_fn)(struct dm_target *ti); |
9eef87da | 1425 | |
23508a96 MS |
1426 | static bool is_split_required_for_discard(struct dm_target *ti) |
1427 | { | |
55a62eef | 1428 | return ti->split_discard_bios; |
cec47e3d KU |
1429 | } |
1430 | ||
3d7f4562 | 1431 | static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
14fe594d AK |
1432 | get_num_bios_fn get_num_bios, |
1433 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1434 | { |
e0d6609a | 1435 | unsigned len; |
55a62eef | 1436 | unsigned num_bios; |
ba1cbad9 | 1437 | |
3d7f4562 MS |
1438 | /* |
1439 | * Even though the device advertised support for this type of | |
1440 | * request, that does not mean every target supports it, and | |
1441 | * reconfiguration might also have changed that since the | |
1442 | * check was performed. | |
1443 | */ | |
1444 | num_bios = get_num_bios ? get_num_bios(ti) : 0; | |
1445 | if (!num_bios) | |
1446 | return -EOPNOTSUPP; | |
ba1cbad9 | 1447 | |
3d7f4562 MS |
1448 | if (is_split_required && !is_split_required(ti)) |
1449 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); | |
1450 | else | |
1451 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); | |
de3ec86d | 1452 | |
3d7f4562 | 1453 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1454 | |
3d7f4562 MS |
1455 | ci->sector += len; |
1456 | ci->sector_count -= len; | |
5ae89a87 MS |
1457 | |
1458 | return 0; | |
ba1cbad9 MS |
1459 | } |
1460 | ||
3d7f4562 | 1461 | static int __send_discard(struct clone_info *ci, struct dm_target *ti) |
23508a96 | 1462 | { |
3d7f4562 | 1463 | return __send_changing_extent_only(ci, ti, get_num_discard_bios, |
14fe594d | 1464 | is_split_required_for_discard); |
23508a96 | 1465 | } |
0ce65797 | 1466 | |
00716545 DS |
1467 | static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti) |
1468 | { | |
1469 | return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios, NULL); | |
1470 | } | |
1471 | ||
3d7f4562 | 1472 | static int __send_write_same(struct clone_info *ci, struct dm_target *ti) |
0ce65797 | 1473 | { |
3d7f4562 | 1474 | return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL); |
0ce65797 MS |
1475 | } |
1476 | ||
3d7f4562 | 1477 | static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti) |
ac62d620 | 1478 | { |
3d7f4562 | 1479 | return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL); |
ac62d620 CH |
1480 | } |
1481 | ||
0519c71e MS |
1482 | static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti, |
1483 | int *result) | |
1484 | { | |
1485 | struct bio *bio = ci->bio; | |
1486 | ||
1487 | if (bio_op(bio) == REQ_OP_DISCARD) | |
1488 | *result = __send_discard(ci, ti); | |
00716545 DS |
1489 | else if (bio_op(bio) == REQ_OP_SECURE_ERASE) |
1490 | *result = __send_secure_erase(ci, ti); | |
0519c71e MS |
1491 | else if (bio_op(bio) == REQ_OP_WRITE_SAME) |
1492 | *result = __send_write_same(ci, ti); | |
1493 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES) | |
1494 | *result = __send_write_zeroes(ci, ti); | |
1495 | else | |
1496 | return false; | |
1497 | ||
1498 | return true; | |
1499 | } | |
1500 | ||
e4c93811 AK |
1501 | /* |
1502 | * Select the correct strategy for processing a non-flush bio. | |
1503 | */ | |
14fe594d | 1504 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1505 | { |
dba14160 | 1506 | struct bio *bio = ci->bio; |
512875bd | 1507 | struct dm_target *ti; |
1c3b13e6 | 1508 | unsigned len; |
c80914e8 | 1509 | int r; |
0ce65797 | 1510 | |
512875bd JN |
1511 | ti = dm_table_find_target(ci->map, ci->sector); |
1512 | if (!dm_target_is_valid(ti)) | |
1513 | return -EIO; | |
1514 | ||
0519c71e MS |
1515 | if (unlikely(__process_abnormal_io(ci, ti, &r))) |
1516 | return r; | |
3d7f4562 | 1517 | |
264c869d DLM |
1518 | if (bio_op(bio) == REQ_OP_ZONE_REPORT) |
1519 | len = ci->sector_count; | |
1520 | else | |
1521 | len = min_t(sector_t, max_io_len(ci->sector, ti), | |
1522 | ci->sector_count); | |
0ce65797 | 1523 | |
c80914e8 MS |
1524 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1525 | if (r < 0) | |
1526 | return r; | |
0ce65797 | 1527 | |
1c3b13e6 KO |
1528 | ci->sector += len; |
1529 | ci->sector_count -= len; | |
0ce65797 | 1530 | |
1c3b13e6 | 1531 | return 0; |
0ce65797 MS |
1532 | } |
1533 | ||
978e51ba MS |
1534 | static void init_clone_info(struct clone_info *ci, struct mapped_device *md, |
1535 | struct dm_table *map, struct bio *bio) | |
1536 | { | |
1537 | ci->map = map; | |
1538 | ci->io = alloc_io(md, bio); | |
1539 | ci->sector = bio->bi_iter.bi_sector; | |
1540 | } | |
1541 | ||
1da177e4 | 1542 | /* |
14fe594d | 1543 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1544 | */ |
978e51ba MS |
1545 | static blk_qc_t __split_and_process_bio(struct mapped_device *md, |
1546 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1547 | { |
1da177e4 | 1548 | struct clone_info ci; |
978e51ba | 1549 | blk_qc_t ret = BLK_QC_T_NONE; |
512875bd | 1550 | int error = 0; |
1da177e4 | 1551 | |
83d5e5b0 | 1552 | if (unlikely(!map)) { |
6a8736d1 | 1553 | bio_io_error(bio); |
978e51ba | 1554 | return ret; |
f0b9a450 | 1555 | } |
692d0eb9 | 1556 | |
978e51ba | 1557 | init_clone_info(&ci, md, map, bio); |
0ce65797 | 1558 | |
1eff9d32 | 1559 | if (bio->bi_opf & REQ_PREFLUSH) { |
bc02cdbe | 1560 | ci.bio = &ci.io->md->flush_bio; |
b372d360 | 1561 | ci.sector_count = 0; |
14fe594d | 1562 | error = __send_empty_flush(&ci); |
b372d360 | 1563 | /* dec_pending submits any data associated with flush */ |
a4aa5e56 DLM |
1564 | } else if (bio_op(bio) == REQ_OP_ZONE_RESET) { |
1565 | ci.bio = bio; | |
1566 | ci.sector_count = 0; | |
1567 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1568 | } else { |
6a8736d1 | 1569 | ci.bio = bio; |
d87f4c14 | 1570 | ci.sector_count = bio_sectors(bio); |
18a25da8 | 1571 | while (ci.sector_count && !error) { |
14fe594d | 1572 | error = __split_and_process_non_flush(&ci); |
18a25da8 N |
1573 | if (current->bio_list && ci.sector_count && !error) { |
1574 | /* | |
1575 | * Remainder must be passed to generic_make_request() | |
1576 | * so that it gets handled *after* bios already submitted | |
1577 | * have been completely processed. | |
1578 | * We take a clone of the original to store in | |
745dc570 | 1579 | * ci.io->orig_bio to be used by end_io_acct() and |
18a25da8 N |
1580 | * for dec_pending to use for completion handling. |
1581 | * As this path is not used for REQ_OP_ZONE_REPORT, | |
745dc570 | 1582 | * the usage of io->orig_bio in dm_remap_zone_report() |
18a25da8 N |
1583 | * won't be affected by this reassignment. |
1584 | */ | |
1585 | struct bio *b = bio_clone_bioset(bio, GFP_NOIO, | |
1586 | md->queue->bio_split); | |
745dc570 | 1587 | ci.io->orig_bio = b; |
18a25da8 N |
1588 | bio_advance(bio, (bio_sectors(bio) - ci.sector_count) << 9); |
1589 | bio_chain(b, bio); | |
978e51ba | 1590 | ret = generic_make_request(bio); |
18a25da8 N |
1591 | break; |
1592 | } | |
1593 | } | |
d87f4c14 | 1594 | } |
0ce65797 | 1595 | |
1da177e4 | 1596 | /* drop the extra reference count */ |
54385bf7 | 1597 | dec_pending(ci.io, errno_to_blk_status(error)); |
978e51ba | 1598 | return ret; |
0ce65797 MS |
1599 | } |
1600 | ||
cec47e3d | 1601 | /* |
978e51ba MS |
1602 | * Optimized variant of __split_and_process_bio that leverages the |
1603 | * fact that targets that use it do _not_ have a need to split bios. | |
cec47e3d | 1604 | */ |
978e51ba MS |
1605 | static blk_qc_t __process_bio(struct mapped_device *md, |
1606 | struct dm_table *map, struct bio *bio) | |
1607 | { | |
1608 | struct clone_info ci; | |
1609 | blk_qc_t ret = BLK_QC_T_NONE; | |
1610 | int error = 0; | |
1611 | ||
1612 | if (unlikely(!map)) { | |
1613 | bio_io_error(bio); | |
1614 | return ret; | |
1615 | } | |
1616 | ||
1617 | init_clone_info(&ci, md, map, bio); | |
1618 | ||
1619 | if (bio->bi_opf & REQ_PREFLUSH) { | |
1620 | ci.bio = &ci.io->md->flush_bio; | |
1621 | ci.sector_count = 0; | |
1622 | error = __send_empty_flush(&ci); | |
1623 | /* dec_pending submits any data associated with flush */ | |
1624 | } else { | |
1625 | struct dm_target *ti = md->immutable_target; | |
1626 | struct dm_target_io *tio; | |
1627 | ||
1628 | /* | |
1629 | * Defend against IO still getting in during teardown | |
1630 | * - as was seen for a time with nvme-fcloop | |
1631 | */ | |
1632 | if (unlikely(WARN_ON_ONCE(!ti || !dm_target_is_valid(ti)))) { | |
1633 | error = -EIO; | |
1634 | goto out; | |
1635 | } | |
1636 | ||
978e51ba MS |
1637 | ci.bio = bio; |
1638 | ci.sector_count = bio_sectors(bio); | |
0519c71e MS |
1639 | if (unlikely(__process_abnormal_io(&ci, ti, &error))) |
1640 | goto out; | |
1641 | ||
1642 | tio = alloc_tio(&ci, ti, 0, GFP_NOIO); | |
978e51ba MS |
1643 | ret = __clone_and_map_simple_bio(&ci, tio, NULL); |
1644 | } | |
1645 | out: | |
1646 | /* drop the extra reference count */ | |
1647 | dec_pending(ci.io, errno_to_blk_status(error)); | |
1648 | return ret; | |
1649 | } | |
1650 | ||
1651 | typedef blk_qc_t (process_bio_fn)(struct mapped_device *, struct dm_table *, struct bio *); | |
1652 | ||
1653 | static blk_qc_t __dm_make_request(struct request_queue *q, struct bio *bio, | |
1654 | process_bio_fn process_bio) | |
cec47e3d KU |
1655 | { |
1656 | struct mapped_device *md = q->queuedata; | |
978e51ba | 1657 | blk_qc_t ret = BLK_QC_T_NONE; |
83d5e5b0 MP |
1658 | int srcu_idx; |
1659 | struct dm_table *map; | |
cec47e3d | 1660 | |
83d5e5b0 | 1661 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1662 | |
6a8736d1 TH |
1663 | /* if we're suspended, we have to queue this io for later */ |
1664 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1665 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1666 | |
1eff9d32 | 1667 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1668 | queue_io(md, bio); |
1669 | else | |
54d9a1b4 | 1670 | bio_io_error(bio); |
978e51ba | 1671 | return ret; |
cec47e3d | 1672 | } |
1da177e4 | 1673 | |
978e51ba MS |
1674 | ret = process_bio(md, map, bio); |
1675 | ||
83d5e5b0 | 1676 | dm_put_live_table(md, srcu_idx); |
978e51ba MS |
1677 | return ret; |
1678 | } | |
1679 | ||
1680 | /* | |
1681 | * The request function that remaps the bio to one target and | |
1682 | * splits off any remainder. | |
1683 | */ | |
1684 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) | |
1685 | { | |
1686 | return __dm_make_request(q, bio, __split_and_process_bio); | |
1687 | } | |
1688 | ||
1689 | static blk_qc_t dm_make_request_nvme(struct request_queue *q, struct bio *bio) | |
1690 | { | |
1691 | return __dm_make_request(q, bio, __process_bio); | |
cec47e3d KU |
1692 | } |
1693 | ||
1da177e4 LT |
1694 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1695 | { | |
8a57dfc6 CS |
1696 | int r = bdi_bits; |
1697 | struct mapped_device *md = congested_data; | |
1698 | struct dm_table *map; | |
1da177e4 | 1699 | |
1eb787ec | 1700 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1701 | if (dm_request_based(md)) { |
cec47e3d | 1702 | /* |
e522c039 MS |
1703 | * With request-based DM we only need to check the |
1704 | * top-level queue for congestion. | |
cec47e3d | 1705 | */ |
dc3b17cc | 1706 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1707 | } else { |
1708 | map = dm_get_live_table_fast(md); | |
1709 | if (map) | |
cec47e3d | 1710 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1711 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1712 | } |
1713 | } | |
1714 | ||
1da177e4 LT |
1715 | return r; |
1716 | } | |
1717 | ||
1718 | /*----------------------------------------------------------------- | |
1719 | * An IDR is used to keep track of allocated minor numbers. | |
1720 | *---------------------------------------------------------------*/ | |
2b06cfff | 1721 | static void free_minor(int minor) |
1da177e4 | 1722 | { |
f32c10b0 | 1723 | spin_lock(&_minor_lock); |
1da177e4 | 1724 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1725 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1726 | } |
1727 | ||
1728 | /* | |
1729 | * See if the device with a specific minor # is free. | |
1730 | */ | |
cf13ab8e | 1731 | static int specific_minor(int minor) |
1da177e4 | 1732 | { |
c9d76be6 | 1733 | int r; |
1da177e4 LT |
1734 | |
1735 | if (minor >= (1 << MINORBITS)) | |
1736 | return -EINVAL; | |
1737 | ||
c9d76be6 | 1738 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1739 | spin_lock(&_minor_lock); |
1da177e4 | 1740 | |
c9d76be6 | 1741 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1742 | |
f32c10b0 | 1743 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1744 | idr_preload_end(); |
1745 | if (r < 0) | |
1746 | return r == -ENOSPC ? -EBUSY : r; | |
1747 | return 0; | |
1da177e4 LT |
1748 | } |
1749 | ||
cf13ab8e | 1750 | static int next_free_minor(int *minor) |
1da177e4 | 1751 | { |
c9d76be6 | 1752 | int r; |
62f75c2f | 1753 | |
c9d76be6 | 1754 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1755 | spin_lock(&_minor_lock); |
1da177e4 | 1756 | |
c9d76be6 | 1757 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1758 | |
f32c10b0 | 1759 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1760 | idr_preload_end(); |
1761 | if (r < 0) | |
1762 | return r; | |
1763 | *minor = r; | |
1764 | return 0; | |
1da177e4 LT |
1765 | } |
1766 | ||
83d5cde4 | 1767 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1768 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1769 | |
53d5914f MP |
1770 | static void dm_wq_work(struct work_struct *work); |
1771 | ||
c12c9a3c | 1772 | static void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1773 | { |
17e149b8 | 1774 | md->use_blk_mq = false; |
bfebd1cd MS |
1775 | |
1776 | /* | |
1777 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1778 | */ | |
dc3b17cc | 1779 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf MS |
1780 | } |
1781 | ||
0f20972f MS |
1782 | static void cleanup_mapped_device(struct mapped_device *md) |
1783 | { | |
0f20972f MS |
1784 | if (md->wq) |
1785 | destroy_workqueue(md->wq); | |
1786 | if (md->kworker_task) | |
1787 | kthread_stop(md->kworker_task); | |
0f20972f MS |
1788 | if (md->bs) |
1789 | bioset_free(md->bs); | |
64f52b0e MS |
1790 | if (md->io_bs) |
1791 | bioset_free(md->io_bs); | |
0f20972f | 1792 | |
f26c5719 DW |
1793 | if (md->dax_dev) { |
1794 | kill_dax(md->dax_dev); | |
1795 | put_dax(md->dax_dev); | |
1796 | md->dax_dev = NULL; | |
1797 | } | |
1798 | ||
0f20972f MS |
1799 | if (md->disk) { |
1800 | spin_lock(&_minor_lock); | |
1801 | md->disk->private_data = NULL; | |
1802 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1803 | del_gendisk(md->disk); |
1804 | put_disk(md->disk); | |
1805 | } | |
1806 | ||
1807 | if (md->queue) | |
1808 | blk_cleanup_queue(md->queue); | |
1809 | ||
d09960b0 TE |
1810 | cleanup_srcu_struct(&md->io_barrier); |
1811 | ||
0f20972f MS |
1812 | if (md->bdev) { |
1813 | bdput(md->bdev); | |
1814 | md->bdev = NULL; | |
1815 | } | |
4cc96131 | 1816 | |
d5ffebdd MS |
1817 | mutex_destroy(&md->suspend_lock); |
1818 | mutex_destroy(&md->type_lock); | |
1819 | mutex_destroy(&md->table_devices_lock); | |
1820 | ||
4cc96131 | 1821 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
1822 | } |
1823 | ||
1da177e4 LT |
1824 | /* |
1825 | * Allocate and initialise a blank device with a given minor. | |
1826 | */ | |
2b06cfff | 1827 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1828 | { |
115485e8 | 1829 | int r, numa_node_id = dm_get_numa_node(); |
976431b0 | 1830 | struct dax_device *dax_dev = NULL; |
115485e8 | 1831 | struct mapped_device *md; |
ba61fdd1 | 1832 | void *old_md; |
1da177e4 | 1833 | |
856eb091 | 1834 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1835 | if (!md) { |
1836 | DMWARN("unable to allocate device, out of memory."); | |
1837 | return NULL; | |
1838 | } | |
1839 | ||
10da4f79 | 1840 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1841 | goto bad_module_get; |
10da4f79 | 1842 | |
1da177e4 | 1843 | /* get a minor number for the dev */ |
2b06cfff | 1844 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1845 | r = next_free_minor(&minor); |
2b06cfff | 1846 | else |
cf13ab8e | 1847 | r = specific_minor(minor); |
1da177e4 | 1848 | if (r < 0) |
6ed7ade8 | 1849 | goto bad_minor; |
1da177e4 | 1850 | |
83d5e5b0 MP |
1851 | r = init_srcu_struct(&md->io_barrier); |
1852 | if (r < 0) | |
1853 | goto bad_io_barrier; | |
1854 | ||
115485e8 | 1855 | md->numa_node_id = numa_node_id; |
4cc96131 | 1856 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1857 | md->init_tio_pdu = false; |
a5664dad | 1858 | md->type = DM_TYPE_NONE; |
e61290a4 | 1859 | mutex_init(&md->suspend_lock); |
a5664dad | 1860 | mutex_init(&md->type_lock); |
86f1152b | 1861 | mutex_init(&md->table_devices_lock); |
022c2611 | 1862 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1863 | atomic_set(&md->holders, 1); |
5c6bd75d | 1864 | atomic_set(&md->open_count, 0); |
1da177e4 | 1865 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1866 | atomic_set(&md->uevent_seq, 0); |
1867 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1868 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1869 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1870 | |
5ee0524b | 1871 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id, NULL); |
1da177e4 | 1872 | if (!md->queue) |
0f20972f | 1873 | goto bad; |
c12c9a3c MS |
1874 | md->queue->queuedata = md; |
1875 | md->queue->backing_dev_info->congested_data = md; | |
1da177e4 | 1876 | |
c12c9a3c | 1877 | md->disk = alloc_disk_node(1, md->numa_node_id); |
1da177e4 | 1878 | if (!md->disk) |
0f20972f | 1879 | goto bad; |
1da177e4 | 1880 | |
316d315b NK |
1881 | atomic_set(&md->pending[0], 0); |
1882 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1883 | init_waitqueue_head(&md->wait); |
53d5914f | 1884 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1885 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1886 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1887 | md->kworker_task = NULL; |
f0b04115 | 1888 | |
1da177e4 LT |
1889 | md->disk->major = _major; |
1890 | md->disk->first_minor = minor; | |
1891 | md->disk->fops = &dm_blk_dops; | |
1892 | md->disk->queue = md->queue; | |
1893 | md->disk->private_data = md; | |
1894 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 | 1895 | |
976431b0 DW |
1896 | if (IS_ENABLED(CONFIG_DAX_DRIVER)) { |
1897 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); | |
1898 | if (!dax_dev) | |
1899 | goto bad; | |
1900 | } | |
f26c5719 DW |
1901 | md->dax_dev = dax_dev; |
1902 | ||
c100ec49 | 1903 | add_disk_no_queue_reg(md->disk); |
7e51f257 | 1904 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1905 | |
670368a8 | 1906 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1907 | if (!md->wq) |
0f20972f | 1908 | goto bad; |
304f3f6a | 1909 | |
32a926da MP |
1910 | md->bdev = bdget_disk(md->disk, 0); |
1911 | if (!md->bdev) | |
0f20972f | 1912 | goto bad; |
32a926da | 1913 | |
3a83f467 | 1914 | bio_init(&md->flush_bio, NULL, 0); |
74d46992 | 1915 | bio_set_dev(&md->flush_bio, md->bdev); |
ff0361b3 | 1916 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; |
6a8736d1 | 1917 | |
fd2ed4d2 MP |
1918 | dm_stats_init(&md->stats); |
1919 | ||
ba61fdd1 | 1920 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1921 | spin_lock(&_minor_lock); |
ba61fdd1 | 1922 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1923 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1924 | |
1925 | BUG_ON(old_md != MINOR_ALLOCED); | |
1926 | ||
1da177e4 LT |
1927 | return md; |
1928 | ||
0f20972f MS |
1929 | bad: |
1930 | cleanup_mapped_device(md); | |
83d5e5b0 | 1931 | bad_io_barrier: |
1da177e4 | 1932 | free_minor(minor); |
6ed7ade8 | 1933 | bad_minor: |
10da4f79 | 1934 | module_put(THIS_MODULE); |
6ed7ade8 | 1935 | bad_module_get: |
856eb091 | 1936 | kvfree(md); |
1da177e4 LT |
1937 | return NULL; |
1938 | } | |
1939 | ||
ae9da83f JN |
1940 | static void unlock_fs(struct mapped_device *md); |
1941 | ||
1da177e4 LT |
1942 | static void free_dev(struct mapped_device *md) |
1943 | { | |
f331c029 | 1944 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1945 | |
32a926da | 1946 | unlock_fs(md); |
2eb6e1e3 | 1947 | |
0f20972f | 1948 | cleanup_mapped_device(md); |
63a4f065 | 1949 | |
86f1152b | 1950 | free_table_devices(&md->table_devices); |
63a4f065 | 1951 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1952 | free_minor(minor); |
1953 | ||
10da4f79 | 1954 | module_put(THIS_MODULE); |
856eb091 | 1955 | kvfree(md); |
1da177e4 LT |
1956 | } |
1957 | ||
e6ee8c0b KU |
1958 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1959 | { | |
c0820cf5 | 1960 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1961 | |
0776aa0e | 1962 | if (dm_table_bio_based(t)) { |
64f52b0e MS |
1963 | /* |
1964 | * The md may already have mempools that need changing. | |
1965 | * If so, reload bioset because front_pad may have changed | |
1966 | * because a different table was loaded. | |
1967 | */ | |
0776aa0e | 1968 | if (md->bs) { |
16245bdc | 1969 | bioset_free(md->bs); |
0776aa0e | 1970 | md->bs = NULL; |
16245bdc | 1971 | } |
64f52b0e MS |
1972 | if (md->io_bs) { |
1973 | bioset_free(md->io_bs); | |
1974 | md->io_bs = NULL; | |
16245bdc | 1975 | } |
0776aa0e MS |
1976 | |
1977 | } else if (md->bs) { | |
4e6e36c3 MS |
1978 | /* |
1979 | * There's no need to reload with request-based dm | |
1980 | * because the size of front_pad doesn't change. | |
1981 | * Note for future: If you are to reload bioset, | |
1982 | * prep-ed requests in the queue may refer | |
1983 | * to bio from the old bioset, so you must walk | |
1984 | * through the queue to unprep. | |
1985 | */ | |
1986 | goto out; | |
c0820cf5 | 1987 | } |
e6ee8c0b | 1988 | |
dde1e1ec | 1989 | BUG_ON(!p || md->bs || md->io_bs); |
cbc4e3c1 | 1990 | |
e6ee8c0b KU |
1991 | md->bs = p->bs; |
1992 | p->bs = NULL; | |
64f52b0e MS |
1993 | md->io_bs = p->io_bs; |
1994 | p->io_bs = NULL; | |
e6ee8c0b | 1995 | out: |
02233342 | 1996 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1997 | dm_table_free_md_mempools(t); |
1998 | } | |
1999 | ||
1da177e4 LT |
2000 | /* |
2001 | * Bind a table to the device. | |
2002 | */ | |
2003 | static void event_callback(void *context) | |
2004 | { | |
7a8c3d3b MA |
2005 | unsigned long flags; |
2006 | LIST_HEAD(uevents); | |
1da177e4 LT |
2007 | struct mapped_device *md = (struct mapped_device *) context; |
2008 | ||
7a8c3d3b MA |
2009 | spin_lock_irqsave(&md->uevent_lock, flags); |
2010 | list_splice_init(&md->uevent_list, &uevents); | |
2011 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2012 | ||
ed9e1982 | 2013 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2014 | |
1da177e4 LT |
2015 | atomic_inc(&md->event_nr); |
2016 | wake_up(&md->eventq); | |
62e08243 | 2017 | dm_issue_global_event(); |
1da177e4 LT |
2018 | } |
2019 | ||
c217649b MS |
2020 | /* |
2021 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2022 | */ | |
4e90188b | 2023 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2024 | { |
1ea0654e BVA |
2025 | lockdep_assert_held(&md->suspend_lock); |
2026 | ||
4e90188b | 2027 | set_capacity(md->disk, size); |
1da177e4 | 2028 | |
db8fef4f | 2029 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2030 | } |
2031 | ||
042d2a9b AK |
2032 | /* |
2033 | * Returns old map, which caller must destroy. | |
2034 | */ | |
2035 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2036 | struct queue_limits *limits) | |
1da177e4 | 2037 | { |
042d2a9b | 2038 | struct dm_table *old_map; |
165125e1 | 2039 | struct request_queue *q = md->queue; |
978e51ba | 2040 | bool request_based = dm_table_request_based(t); |
1da177e4 LT |
2041 | sector_t size; |
2042 | ||
5a8f1f80 BVA |
2043 | lockdep_assert_held(&md->suspend_lock); |
2044 | ||
1da177e4 | 2045 | size = dm_table_get_size(t); |
3ac51e74 DW |
2046 | |
2047 | /* | |
2048 | * Wipe any geometry if the size of the table changed. | |
2049 | */ | |
fd2ed4d2 | 2050 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2051 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2052 | ||
32a926da | 2053 | __set_size(md, size); |
d5816876 | 2054 | |
2ca3310e AK |
2055 | dm_table_event_callback(t, event_callback, md); |
2056 | ||
e6ee8c0b KU |
2057 | /* |
2058 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2059 | * for request-based during suspension. So stop it to prevent | |
2060 | * I/O mapping before resume. | |
2061 | * This must be done before setting the queue restrictions, | |
2062 | * because request-based dm may be run just after the setting. | |
2063 | */ | |
978e51ba | 2064 | if (request_based) |
eca7ee6d | 2065 | dm_stop_queue(q); |
978e51ba MS |
2066 | |
2067 | if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) { | |
16f12266 | 2068 | /* |
978e51ba MS |
2069 | * Leverage the fact that request-based DM targets and |
2070 | * NVMe bio based targets are immutable singletons | |
2071 | * - used to optimize both dm_request_fn and dm_mq_queue_rq; | |
2072 | * and __process_bio. | |
16f12266 MS |
2073 | */ |
2074 | md->immutable_target = dm_table_get_immutable_target(t); | |
2075 | } | |
e6ee8c0b KU |
2076 | |
2077 | __bind_mempools(md, t); | |
2078 | ||
a12f5d48 | 2079 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2080 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2081 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2082 | ||
754c5fc7 | 2083 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2084 | if (old_map) |
2085 | dm_sync_table(md); | |
1da177e4 | 2086 | |
042d2a9b | 2087 | return old_map; |
1da177e4 LT |
2088 | } |
2089 | ||
a7940155 AK |
2090 | /* |
2091 | * Returns unbound table for the caller to free. | |
2092 | */ | |
2093 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2094 | { |
a12f5d48 | 2095 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2096 | |
2097 | if (!map) | |
a7940155 | 2098 | return NULL; |
1da177e4 LT |
2099 | |
2100 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2101 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2102 | dm_sync_table(md); |
a7940155 AK |
2103 | |
2104 | return map; | |
1da177e4 LT |
2105 | } |
2106 | ||
2107 | /* | |
2108 | * Constructor for a new device. | |
2109 | */ | |
2b06cfff | 2110 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 | 2111 | { |
c12c9a3c | 2112 | int r; |
1da177e4 LT |
2113 | struct mapped_device *md; |
2114 | ||
2b06cfff | 2115 | md = alloc_dev(minor); |
1da177e4 LT |
2116 | if (!md) |
2117 | return -ENXIO; | |
2118 | ||
c12c9a3c MS |
2119 | r = dm_sysfs_init(md); |
2120 | if (r) { | |
2121 | free_dev(md); | |
2122 | return r; | |
2123 | } | |
784aae73 | 2124 | |
1da177e4 LT |
2125 | *result = md; |
2126 | return 0; | |
2127 | } | |
2128 | ||
a5664dad MS |
2129 | /* |
2130 | * Functions to manage md->type. | |
2131 | * All are required to hold md->type_lock. | |
2132 | */ | |
2133 | void dm_lock_md_type(struct mapped_device *md) | |
2134 | { | |
2135 | mutex_lock(&md->type_lock); | |
2136 | } | |
2137 | ||
2138 | void dm_unlock_md_type(struct mapped_device *md) | |
2139 | { | |
2140 | mutex_unlock(&md->type_lock); | |
2141 | } | |
2142 | ||
7e0d574f | 2143 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2144 | { |
00c4fc3b | 2145 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2146 | md->type = type; |
2147 | } | |
2148 | ||
7e0d574f | 2149 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2150 | { |
2151 | return md->type; | |
2152 | } | |
2153 | ||
36a0456f AK |
2154 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2155 | { | |
2156 | return md->immutable_target_type; | |
2157 | } | |
2158 | ||
f84cb8a4 MS |
2159 | /* |
2160 | * The queue_limits are only valid as long as you have a reference | |
2161 | * count on 'md'. | |
2162 | */ | |
2163 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2164 | { | |
2165 | BUG_ON(!atomic_read(&md->holders)); | |
2166 | return &md->queue->limits; | |
2167 | } | |
2168 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2169 | ||
4a0b4ddf MS |
2170 | /* |
2171 | * Setup the DM device's queue based on md's type | |
2172 | */ | |
591ddcfc | 2173 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2174 | { |
bfebd1cd | 2175 | int r; |
c100ec49 | 2176 | struct queue_limits limits; |
7e0d574f | 2177 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2178 | |
545ed20e | 2179 | switch (type) { |
bfebd1cd | 2180 | case DM_TYPE_REQUEST_BASED: |
c12c9a3c | 2181 | dm_init_normal_md_queue(md); |
eb8db831 | 2182 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 2183 | if (r) { |
eca7ee6d | 2184 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2185 | return r; |
ff36ab34 | 2186 | } |
bfebd1cd MS |
2187 | break; |
2188 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 2189 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2190 | if (r) { |
eca7ee6d | 2191 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2192 | return r; |
2193 | } | |
2194 | break; | |
2195 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2196 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 2197 | dm_init_normal_md_queue(md); |
ff36ab34 | 2198 | blk_queue_make_request(md->queue, dm_make_request); |
bfebd1cd | 2199 | break; |
978e51ba MS |
2200 | case DM_TYPE_NVME_BIO_BASED: |
2201 | dm_init_normal_md_queue(md); | |
2202 | blk_queue_make_request(md->queue, dm_make_request_nvme); | |
bfebd1cd | 2203 | break; |
7e0d574f BVA |
2204 | case DM_TYPE_NONE: |
2205 | WARN_ON_ONCE(true); | |
2206 | break; | |
4a0b4ddf MS |
2207 | } |
2208 | ||
c100ec49 MS |
2209 | r = dm_calculate_queue_limits(t, &limits); |
2210 | if (r) { | |
2211 | DMERR("Cannot calculate initial queue limits"); | |
2212 | return r; | |
2213 | } | |
2214 | dm_table_set_restrictions(t, md->queue, &limits); | |
2215 | blk_register_queue(md->disk); | |
2216 | ||
4a0b4ddf MS |
2217 | return 0; |
2218 | } | |
2219 | ||
2bec1f4a | 2220 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2221 | { |
2222 | struct mapped_device *md; | |
1da177e4 LT |
2223 | unsigned minor = MINOR(dev); |
2224 | ||
2225 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2226 | return NULL; | |
2227 | ||
f32c10b0 | 2228 | spin_lock(&_minor_lock); |
1da177e4 LT |
2229 | |
2230 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2231 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2232 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2233 | md = NULL; | |
2234 | goto out; | |
fba9f90e | 2235 | } |
49de5769 | 2236 | dm_get(md); |
fba9f90e | 2237 | out: |
f32c10b0 | 2238 | spin_unlock(&_minor_lock); |
1da177e4 | 2239 | |
637842cf DT |
2240 | return md; |
2241 | } | |
3cf2e4ba | 2242 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2243 | |
9ade92a9 | 2244 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2245 | { |
9ade92a9 | 2246 | return md->interface_ptr; |
1da177e4 LT |
2247 | } |
2248 | ||
2249 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2250 | { | |
2251 | md->interface_ptr = ptr; | |
2252 | } | |
2253 | ||
2254 | void dm_get(struct mapped_device *md) | |
2255 | { | |
2256 | atomic_inc(&md->holders); | |
3f77316d | 2257 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2258 | } |
2259 | ||
09ee96b2 MP |
2260 | int dm_hold(struct mapped_device *md) |
2261 | { | |
2262 | spin_lock(&_minor_lock); | |
2263 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2264 | spin_unlock(&_minor_lock); | |
2265 | return -EBUSY; | |
2266 | } | |
2267 | dm_get(md); | |
2268 | spin_unlock(&_minor_lock); | |
2269 | return 0; | |
2270 | } | |
2271 | EXPORT_SYMBOL_GPL(dm_hold); | |
2272 | ||
72d94861 AK |
2273 | const char *dm_device_name(struct mapped_device *md) |
2274 | { | |
2275 | return md->name; | |
2276 | } | |
2277 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2278 | ||
3f77316d | 2279 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2280 | { |
1134e5ae | 2281 | struct dm_table *map; |
83d5e5b0 | 2282 | int srcu_idx; |
1da177e4 | 2283 | |
3f77316d | 2284 | might_sleep(); |
fba9f90e | 2285 | |
63a4f065 | 2286 | spin_lock(&_minor_lock); |
3f77316d KU |
2287 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2288 | set_bit(DMF_FREEING, &md->flags); | |
2289 | spin_unlock(&_minor_lock); | |
3b785fbc | 2290 | |
c12c9a3c | 2291 | blk_set_queue_dying(md->queue); |
3f77316d | 2292 | |
02233342 | 2293 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 2294 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2295 | |
ab7c7bb6 MP |
2296 | /* |
2297 | * Take suspend_lock so that presuspend and postsuspend methods | |
2298 | * do not race with internal suspend. | |
2299 | */ | |
2300 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2301 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2302 | if (!dm_suspended_md(md)) { |
2303 | dm_table_presuspend_targets(map); | |
2304 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2305 | } |
83d5e5b0 MP |
2306 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2307 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2308 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2309 | |
3f77316d KU |
2310 | /* |
2311 | * Rare, but there may be I/O requests still going to complete, | |
2312 | * for example. Wait for all references to disappear. | |
2313 | * No one should increment the reference count of the mapped_device, | |
2314 | * after the mapped_device state becomes DMF_FREEING. | |
2315 | */ | |
2316 | if (wait) | |
2317 | while (atomic_read(&md->holders)) | |
2318 | msleep(1); | |
2319 | else if (atomic_read(&md->holders)) | |
2320 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2321 | dm_device_name(md), atomic_read(&md->holders)); | |
2322 | ||
2323 | dm_sysfs_exit(md); | |
3f77316d KU |
2324 | dm_table_destroy(__unbind(md)); |
2325 | free_dev(md); | |
2326 | } | |
2327 | ||
2328 | void dm_destroy(struct mapped_device *md) | |
2329 | { | |
2330 | __dm_destroy(md, true); | |
2331 | } | |
2332 | ||
2333 | void dm_destroy_immediate(struct mapped_device *md) | |
2334 | { | |
2335 | __dm_destroy(md, false); | |
2336 | } | |
2337 | ||
2338 | void dm_put(struct mapped_device *md) | |
2339 | { | |
2340 | atomic_dec(&md->holders); | |
1da177e4 | 2341 | } |
79eb885c | 2342 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2343 | |
b48633f8 | 2344 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2345 | { |
2346 | int r = 0; | |
9f4c3f87 | 2347 | DEFINE_WAIT(wait); |
46125c1c MB |
2348 | |
2349 | while (1) { | |
9f4c3f87 | 2350 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2351 | |
b4324fee | 2352 | if (!md_in_flight(md)) |
46125c1c MB |
2353 | break; |
2354 | ||
e3fabdfd | 2355 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2356 | r = -EINTR; |
2357 | break; | |
2358 | } | |
2359 | ||
2360 | io_schedule(); | |
2361 | } | |
9f4c3f87 | 2362 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2363 | |
46125c1c MB |
2364 | return r; |
2365 | } | |
2366 | ||
1da177e4 LT |
2367 | /* |
2368 | * Process the deferred bios | |
2369 | */ | |
ef208587 | 2370 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2371 | { |
ef208587 MP |
2372 | struct mapped_device *md = container_of(work, struct mapped_device, |
2373 | work); | |
6d6f10df | 2374 | struct bio *c; |
83d5e5b0 MP |
2375 | int srcu_idx; |
2376 | struct dm_table *map; | |
1da177e4 | 2377 | |
83d5e5b0 | 2378 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2379 | |
3b00b203 | 2380 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2381 | spin_lock_irq(&md->deferred_lock); |
2382 | c = bio_list_pop(&md->deferred); | |
2383 | spin_unlock_irq(&md->deferred_lock); | |
2384 | ||
6a8736d1 | 2385 | if (!c) |
df12ee99 | 2386 | break; |
022c2611 | 2387 | |
e6ee8c0b KU |
2388 | if (dm_request_based(md)) |
2389 | generic_make_request(c); | |
6a8736d1 | 2390 | else |
83d5e5b0 | 2391 | __split_and_process_bio(md, map, c); |
022c2611 | 2392 | } |
73d410c0 | 2393 | |
83d5e5b0 | 2394 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2395 | } |
2396 | ||
9a1fb464 | 2397 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2398 | { |
3b00b203 | 2399 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2400 | smp_mb__after_atomic(); |
53d5914f | 2401 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2402 | } |
2403 | ||
1da177e4 | 2404 | /* |
042d2a9b | 2405 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2406 | */ |
042d2a9b | 2407 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2408 | { |
87eb5b21 | 2409 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2410 | struct queue_limits limits; |
042d2a9b | 2411 | int r; |
1da177e4 | 2412 | |
e61290a4 | 2413 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2414 | |
2415 | /* device must be suspended */ | |
4f186f8b | 2416 | if (!dm_suspended_md(md)) |
93c534ae | 2417 | goto out; |
1da177e4 | 2418 | |
3ae70656 MS |
2419 | /* |
2420 | * If the new table has no data devices, retain the existing limits. | |
2421 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2422 | * then new I/O is queued based on these limits, and then some paths | |
2423 | * reappear. | |
2424 | */ | |
2425 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2426 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2427 | if (live_map) |
2428 | limits = md->queue->limits; | |
83d5e5b0 | 2429 | dm_put_live_table_fast(md); |
3ae70656 MS |
2430 | } |
2431 | ||
87eb5b21 MC |
2432 | if (!live_map) { |
2433 | r = dm_calculate_queue_limits(table, &limits); | |
2434 | if (r) { | |
2435 | map = ERR_PTR(r); | |
2436 | goto out; | |
2437 | } | |
042d2a9b | 2438 | } |
754c5fc7 | 2439 | |
042d2a9b | 2440 | map = __bind(md, table, &limits); |
62e08243 | 2441 | dm_issue_global_event(); |
1da177e4 | 2442 | |
93c534ae | 2443 | out: |
e61290a4 | 2444 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2445 | return map; |
1da177e4 LT |
2446 | } |
2447 | ||
2448 | /* | |
2449 | * Functions to lock and unlock any filesystem running on the | |
2450 | * device. | |
2451 | */ | |
2ca3310e | 2452 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2453 | { |
e39e2e95 | 2454 | int r; |
1da177e4 LT |
2455 | |
2456 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2457 | |
db8fef4f | 2458 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2459 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2460 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2461 | md->frozen_sb = NULL; |
2462 | return r; | |
dfbe03f6 AK |
2463 | } |
2464 | ||
aa8d7c2f AK |
2465 | set_bit(DMF_FROZEN, &md->flags); |
2466 | ||
1da177e4 LT |
2467 | return 0; |
2468 | } | |
2469 | ||
2ca3310e | 2470 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2471 | { |
aa8d7c2f AK |
2472 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2473 | return; | |
2474 | ||
db8fef4f | 2475 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2476 | md->frozen_sb = NULL; |
aa8d7c2f | 2477 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2478 | } |
2479 | ||
2480 | /* | |
b48633f8 BVA |
2481 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2482 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2483 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2484 | * | |
ffcc3936 MS |
2485 | * If __dm_suspend returns 0, the device is completely quiescent |
2486 | * now. There is no request-processing activity. All new requests | |
2487 | * are being added to md->deferred list. | |
cec47e3d | 2488 | */ |
ffcc3936 | 2489 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2490 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2491 | int dmf_suspended_flag) |
1da177e4 | 2492 | { |
ffcc3936 MS |
2493 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2494 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2495 | int r; | |
1da177e4 | 2496 | |
5a8f1f80 BVA |
2497 | lockdep_assert_held(&md->suspend_lock); |
2498 | ||
2e93ccc1 KU |
2499 | /* |
2500 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2501 | * This flag is cleared before dm_suspend returns. | |
2502 | */ | |
2503 | if (noflush) | |
2504 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 BVA |
2505 | else |
2506 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); | |
2e93ccc1 | 2507 | |
d67ee213 MS |
2508 | /* |
2509 | * This gets reverted if there's an error later and the targets | |
2510 | * provide the .presuspend_undo hook. | |
2511 | */ | |
cf222b37 AK |
2512 | dm_table_presuspend_targets(map); |
2513 | ||
32a926da | 2514 | /* |
9f518b27 KU |
2515 | * Flush I/O to the device. |
2516 | * Any I/O submitted after lock_fs() may not be flushed. | |
2517 | * noflush takes precedence over do_lockfs. | |
2518 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2519 | */ |
2520 | if (!noflush && do_lockfs) { | |
2521 | r = lock_fs(md); | |
d67ee213 MS |
2522 | if (r) { |
2523 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2524 | return r; |
d67ee213 | 2525 | } |
aa8d7c2f | 2526 | } |
1da177e4 LT |
2527 | |
2528 | /* | |
3b00b203 MP |
2529 | * Here we must make sure that no processes are submitting requests |
2530 | * to target drivers i.e. no one may be executing | |
2531 | * __split_and_process_bio. This is called from dm_request and | |
2532 | * dm_wq_work. | |
2533 | * | |
2534 | * To get all processes out of __split_and_process_bio in dm_request, | |
2535 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2536 | * __split_and_process_bio from dm_request and quiesce the thread |
2537 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2538 | * flush_workqueue(md->wq). | |
1da177e4 | 2539 | */ |
1eb787ec | 2540 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2541 | if (map) |
2542 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2543 | |
d0bcb878 | 2544 | /* |
29e4013d TH |
2545 | * Stop md->queue before flushing md->wq in case request-based |
2546 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2547 | */ |
2eb6e1e3 | 2548 | if (dm_request_based(md)) { |
eca7ee6d | 2549 | dm_stop_queue(md->queue); |
02233342 | 2550 | if (md->kworker_task) |
3989144f | 2551 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2552 | } |
cec47e3d | 2553 | |
d0bcb878 KU |
2554 | flush_workqueue(md->wq); |
2555 | ||
1da177e4 | 2556 | /* |
3b00b203 MP |
2557 | * At this point no more requests are entering target request routines. |
2558 | * We call dm_wait_for_completion to wait for all existing requests | |
2559 | * to finish. | |
1da177e4 | 2560 | */ |
b48633f8 | 2561 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2562 | if (!r) |
2563 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2564 | |
6d6f10df | 2565 | if (noflush) |
022c2611 | 2566 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2567 | if (map) |
2568 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2569 | |
1da177e4 | 2570 | /* were we interrupted ? */ |
46125c1c | 2571 | if (r < 0) { |
9a1fb464 | 2572 | dm_queue_flush(md); |
73d410c0 | 2573 | |
cec47e3d | 2574 | if (dm_request_based(md)) |
eca7ee6d | 2575 | dm_start_queue(md->queue); |
cec47e3d | 2576 | |
2ca3310e | 2577 | unlock_fs(md); |
d67ee213 | 2578 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2579 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2580 | } |
1da177e4 | 2581 | |
ffcc3936 MS |
2582 | return r; |
2583 | } | |
2584 | ||
2585 | /* | |
2586 | * We need to be able to change a mapping table under a mounted | |
2587 | * filesystem. For example we might want to move some data in | |
2588 | * the background. Before the table can be swapped with | |
2589 | * dm_bind_table, dm_suspend must be called to flush any in | |
2590 | * flight bios and ensure that any further io gets deferred. | |
2591 | */ | |
2592 | /* | |
2593 | * Suspend mechanism in request-based dm. | |
2594 | * | |
2595 | * 1. Flush all I/Os by lock_fs() if needed. | |
2596 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2597 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2598 | * | |
2599 | * To abort suspend, start the request_queue. | |
2600 | */ | |
2601 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2602 | { | |
2603 | struct dm_table *map = NULL; | |
2604 | int r = 0; | |
2605 | ||
2606 | retry: | |
2607 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2608 | ||
2609 | if (dm_suspended_md(md)) { | |
2610 | r = -EINVAL; | |
2611 | goto out_unlock; | |
2612 | } | |
2613 | ||
2614 | if (dm_suspended_internally_md(md)) { | |
2615 | /* already internally suspended, wait for internal resume */ | |
2616 | mutex_unlock(&md->suspend_lock); | |
2617 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2618 | if (r) | |
2619 | return r; | |
2620 | goto retry; | |
2621 | } | |
2622 | ||
a12f5d48 | 2623 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2624 | |
eaf9a736 | 2625 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2626 | if (r) |
2627 | goto out_unlock; | |
3b00b203 | 2628 | |
4d4471cb KU |
2629 | dm_table_postsuspend_targets(map); |
2630 | ||
d287483d | 2631 | out_unlock: |
e61290a4 | 2632 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2633 | return r; |
1da177e4 LT |
2634 | } |
2635 | ||
ffcc3936 MS |
2636 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2637 | { | |
2638 | if (map) { | |
2639 | int r = dm_table_resume_targets(map); | |
2640 | if (r) | |
2641 | return r; | |
2642 | } | |
2643 | ||
2644 | dm_queue_flush(md); | |
2645 | ||
2646 | /* | |
2647 | * Flushing deferred I/Os must be done after targets are resumed | |
2648 | * so that mapping of targets can work correctly. | |
2649 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2650 | */ | |
2651 | if (dm_request_based(md)) | |
eca7ee6d | 2652 | dm_start_queue(md->queue); |
ffcc3936 MS |
2653 | |
2654 | unlock_fs(md); | |
2655 | ||
2656 | return 0; | |
2657 | } | |
2658 | ||
1da177e4 LT |
2659 | int dm_resume(struct mapped_device *md) |
2660 | { | |
8dc23658 | 2661 | int r; |
cf222b37 | 2662 | struct dm_table *map = NULL; |
1da177e4 | 2663 | |
ffcc3936 | 2664 | retry: |
8dc23658 | 2665 | r = -EINVAL; |
ffcc3936 MS |
2666 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2667 | ||
4f186f8b | 2668 | if (!dm_suspended_md(md)) |
cf222b37 | 2669 | goto out; |
cf222b37 | 2670 | |
ffcc3936 MS |
2671 | if (dm_suspended_internally_md(md)) { |
2672 | /* already internally suspended, wait for internal resume */ | |
2673 | mutex_unlock(&md->suspend_lock); | |
2674 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2675 | if (r) | |
2676 | return r; | |
2677 | goto retry; | |
2678 | } | |
2679 | ||
a12f5d48 | 2680 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2681 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2682 | goto out; |
1da177e4 | 2683 | |
ffcc3936 | 2684 | r = __dm_resume(md, map); |
8757b776 MB |
2685 | if (r) |
2686 | goto out; | |
2ca3310e | 2687 | |
2ca3310e | 2688 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2689 | out: |
e61290a4 | 2690 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2691 | |
cf222b37 | 2692 | return r; |
1da177e4 LT |
2693 | } |
2694 | ||
fd2ed4d2 MP |
2695 | /* |
2696 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2697 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2698 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2699 | */ |
2700 | ||
ffcc3936 | 2701 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2702 | { |
ffcc3936 MS |
2703 | struct dm_table *map = NULL; |
2704 | ||
1ea0654e BVA |
2705 | lockdep_assert_held(&md->suspend_lock); |
2706 | ||
96b26c8c | 2707 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2708 | return; /* nested internal suspend */ |
2709 | ||
2710 | if (dm_suspended_md(md)) { | |
2711 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2712 | return; /* nest suspend */ | |
2713 | } | |
2714 | ||
a12f5d48 | 2715 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2716 | |
2717 | /* | |
2718 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2719 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2720 | * would require changing .presuspend to return an error -- avoid this | |
2721 | * until there is a need for more elaborate variants of internal suspend. | |
2722 | */ | |
eaf9a736 MS |
2723 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2724 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2725 | |
2726 | dm_table_postsuspend_targets(map); | |
2727 | } | |
2728 | ||
2729 | static void __dm_internal_resume(struct mapped_device *md) | |
2730 | { | |
96b26c8c MP |
2731 | BUG_ON(!md->internal_suspend_count); |
2732 | ||
2733 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2734 | return; /* resume from nested internal suspend */ |
2735 | ||
fd2ed4d2 | 2736 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2737 | goto done; /* resume from nested suspend */ |
2738 | ||
2739 | /* | |
2740 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2741 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2742 | */ | |
2743 | (void) __dm_resume(md, NULL); | |
2744 | ||
2745 | done: | |
2746 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2747 | smp_mb__after_atomic(); | |
2748 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2749 | } | |
2750 | ||
2751 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2752 | { | |
2753 | mutex_lock(&md->suspend_lock); | |
2754 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2755 | mutex_unlock(&md->suspend_lock); | |
2756 | } | |
2757 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2758 | ||
2759 | void dm_internal_resume(struct mapped_device *md) | |
2760 | { | |
2761 | mutex_lock(&md->suspend_lock); | |
2762 | __dm_internal_resume(md); | |
2763 | mutex_unlock(&md->suspend_lock); | |
2764 | } | |
2765 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2766 | ||
2767 | /* | |
2768 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2769 | * which prevents interaction with userspace-driven suspend. | |
2770 | */ | |
2771 | ||
2772 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2773 | { | |
2774 | mutex_lock(&md->suspend_lock); | |
2775 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2776 | return; |
2777 | ||
2778 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2779 | synchronize_srcu(&md->io_barrier); | |
2780 | flush_workqueue(md->wq); | |
2781 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2782 | } | |
b735fede | 2783 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2784 | |
ffcc3936 | 2785 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2786 | { |
ffcc3936 | 2787 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2788 | goto done; |
2789 | ||
2790 | dm_queue_flush(md); | |
2791 | ||
2792 | done: | |
2793 | mutex_unlock(&md->suspend_lock); | |
2794 | } | |
b735fede | 2795 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2796 | |
1da177e4 LT |
2797 | /*----------------------------------------------------------------- |
2798 | * Event notification. | |
2799 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2800 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2801 | unsigned cookie) |
69267a30 | 2802 | { |
60935eb2 MB |
2803 | char udev_cookie[DM_COOKIE_LENGTH]; |
2804 | char *envp[] = { udev_cookie, NULL }; | |
2805 | ||
2806 | if (!cookie) | |
3abf85b5 | 2807 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2808 | else { |
2809 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2810 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2811 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2812 | action, envp); | |
60935eb2 | 2813 | } |
69267a30 AK |
2814 | } |
2815 | ||
7a8c3d3b MA |
2816 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2817 | { | |
2818 | return atomic_add_return(1, &md->uevent_seq); | |
2819 | } | |
2820 | ||
1da177e4 LT |
2821 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2822 | { | |
2823 | return atomic_read(&md->event_nr); | |
2824 | } | |
2825 | ||
2826 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2827 | { | |
2828 | return wait_event_interruptible(md->eventq, | |
2829 | (event_nr != atomic_read(&md->event_nr))); | |
2830 | } | |
2831 | ||
7a8c3d3b MA |
2832 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2833 | { | |
2834 | unsigned long flags; | |
2835 | ||
2836 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2837 | list_add(elist, &md->uevent_list); | |
2838 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2839 | } | |
2840 | ||
1da177e4 LT |
2841 | /* |
2842 | * The gendisk is only valid as long as you have a reference | |
2843 | * count on 'md'. | |
2844 | */ | |
2845 | struct gendisk *dm_disk(struct mapped_device *md) | |
2846 | { | |
2847 | return md->disk; | |
2848 | } | |
65ff5b7d | 2849 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2850 | |
784aae73 MB |
2851 | struct kobject *dm_kobject(struct mapped_device *md) |
2852 | { | |
2995fa78 | 2853 | return &md->kobj_holder.kobj; |
784aae73 MB |
2854 | } |
2855 | ||
784aae73 MB |
2856 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2857 | { | |
2858 | struct mapped_device *md; | |
2859 | ||
2995fa78 | 2860 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2861 | |
b9a41d21 HT |
2862 | spin_lock(&_minor_lock); |
2863 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2864 | md = NULL; | |
2865 | goto out; | |
2866 | } | |
784aae73 | 2867 | dm_get(md); |
b9a41d21 HT |
2868 | out: |
2869 | spin_unlock(&_minor_lock); | |
2870 | ||
784aae73 MB |
2871 | return md; |
2872 | } | |
2873 | ||
4f186f8b | 2874 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2875 | { |
2876 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2877 | } | |
2878 | ||
ffcc3936 MS |
2879 | int dm_suspended_internally_md(struct mapped_device *md) |
2880 | { | |
2881 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2882 | } | |
2883 | ||
2c140a24 MP |
2884 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2885 | { | |
2886 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2887 | } | |
2888 | ||
64dbce58 KU |
2889 | int dm_suspended(struct dm_target *ti) |
2890 | { | |
ecdb2e25 | 2891 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2892 | } |
2893 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2894 | ||
2e93ccc1 KU |
2895 | int dm_noflush_suspending(struct dm_target *ti) |
2896 | { | |
ecdb2e25 | 2897 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2898 | } |
2899 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2900 | ||
7e0d574f | 2901 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
0776aa0e MS |
2902 | unsigned integrity, unsigned per_io_data_size, |
2903 | unsigned min_pool_size) | |
e6ee8c0b | 2904 | { |
115485e8 | 2905 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2906 | unsigned int pool_size = 0; |
64f52b0e | 2907 | unsigned int front_pad, io_front_pad; |
e6ee8c0b KU |
2908 | |
2909 | if (!pools) | |
4e6e36c3 | 2910 | return NULL; |
e6ee8c0b | 2911 | |
78d8e58a MS |
2912 | switch (type) { |
2913 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2914 | case DM_TYPE_DAX_BIO_BASED: |
22c11858 | 2915 | case DM_TYPE_NVME_BIO_BASED: |
0776aa0e | 2916 | pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); |
30187e1d | 2917 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
64f52b0e MS |
2918 | io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio); |
2919 | pools->io_bs = bioset_create(pool_size, io_front_pad, 0); | |
2920 | if (!pools->io_bs) | |
2921 | goto out; | |
2922 | if (integrity && bioset_integrity_create(pools->io_bs, pool_size)) | |
eb8db831 | 2923 | goto out; |
78d8e58a MS |
2924 | break; |
2925 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2926 | case DM_TYPE_MQ_REQUEST_BASED: |
0776aa0e | 2927 | pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); |
78d8e58a | 2928 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2929 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2930 | break; |
2931 | default: | |
2932 | BUG(); | |
2933 | } | |
2934 | ||
4a3f54d9 | 2935 | pools->bs = bioset_create(pool_size, front_pad, 0); |
e6ee8c0b | 2936 | if (!pools->bs) |
5f015204 | 2937 | goto out; |
e6ee8c0b | 2938 | |
a91a2785 | 2939 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2940 | goto out; |
a91a2785 | 2941 | |
e6ee8c0b | 2942 | return pools; |
5f1b670d | 2943 | |
5f1b670d CH |
2944 | out: |
2945 | dm_free_md_mempools(pools); | |
78d8e58a | 2946 | |
4e6e36c3 | 2947 | return NULL; |
e6ee8c0b KU |
2948 | } |
2949 | ||
2950 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2951 | { | |
2952 | if (!pools) | |
2953 | return; | |
2954 | ||
e6ee8c0b KU |
2955 | if (pools->bs) |
2956 | bioset_free(pools->bs); | |
64f52b0e MS |
2957 | if (pools->io_bs) |
2958 | bioset_free(pools->io_bs); | |
e6ee8c0b KU |
2959 | |
2960 | kfree(pools); | |
2961 | } | |
2962 | ||
9c72bad1 CH |
2963 | struct dm_pr { |
2964 | u64 old_key; | |
2965 | u64 new_key; | |
2966 | u32 flags; | |
2967 | bool fail_early; | |
2968 | }; | |
2969 | ||
2970 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2971 | void *data) | |
71cdb697 CH |
2972 | { |
2973 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2974 | struct dm_table *table; |
2975 | struct dm_target *ti; | |
2976 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2977 | |
9c72bad1 CH |
2978 | table = dm_get_live_table(md, &srcu_idx); |
2979 | if (!table || !dm_table_get_size(table)) | |
2980 | goto out; | |
71cdb697 | 2981 | |
9c72bad1 CH |
2982 | /* We only support devices that have a single target */ |
2983 | if (dm_table_get_num_targets(table) != 1) | |
2984 | goto out; | |
2985 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2986 | |
9c72bad1 CH |
2987 | ret = -EINVAL; |
2988 | if (!ti->type->iterate_devices) | |
2989 | goto out; | |
2990 | ||
2991 | ret = ti->type->iterate_devices(ti, fn, data); | |
2992 | out: | |
2993 | dm_put_live_table(md, srcu_idx); | |
2994 | return ret; | |
2995 | } | |
2996 | ||
2997 | /* | |
2998 | * For register / unregister we need to manually call out to every path. | |
2999 | */ | |
3000 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
3001 | sector_t start, sector_t len, void *data) | |
3002 | { | |
3003 | struct dm_pr *pr = data; | |
3004 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3005 | ||
3006 | if (!ops || !ops->pr_register) | |
3007 | return -EOPNOTSUPP; | |
3008 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
3009 | } | |
3010 | ||
3011 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
3012 | u32 flags) | |
3013 | { | |
3014 | struct dm_pr pr = { | |
3015 | .old_key = old_key, | |
3016 | .new_key = new_key, | |
3017 | .flags = flags, | |
3018 | .fail_early = true, | |
3019 | }; | |
3020 | int ret; | |
3021 | ||
3022 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
3023 | if (ret && new_key) { | |
3024 | /* unregister all paths if we failed to register any path */ | |
3025 | pr.old_key = new_key; | |
3026 | pr.new_key = 0; | |
3027 | pr.flags = 0; | |
3028 | pr.fail_early = false; | |
3029 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
3030 | } | |
3031 | ||
3032 | return ret; | |
71cdb697 CH |
3033 | } |
3034 | ||
3035 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3036 | u32 flags) |
71cdb697 CH |
3037 | { |
3038 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3039 | const struct pr_ops *ops; | |
971888c4 | 3040 | int r, srcu_idx; |
71cdb697 | 3041 | |
5bd5e8d8 | 3042 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3043 | if (r < 0) |
971888c4 | 3044 | goto out; |
71cdb697 CH |
3045 | |
3046 | ops = bdev->bd_disk->fops->pr_ops; | |
3047 | if (ops && ops->pr_reserve) | |
3048 | r = ops->pr_reserve(bdev, key, type, flags); | |
3049 | else | |
3050 | r = -EOPNOTSUPP; | |
971888c4 MS |
3051 | out: |
3052 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3053 | return r; |
3054 | } | |
3055 | ||
3056 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3057 | { | |
3058 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3059 | const struct pr_ops *ops; | |
971888c4 | 3060 | int r, srcu_idx; |
71cdb697 | 3061 | |
5bd5e8d8 | 3062 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3063 | if (r < 0) |
971888c4 | 3064 | goto out; |
71cdb697 CH |
3065 | |
3066 | ops = bdev->bd_disk->fops->pr_ops; | |
3067 | if (ops && ops->pr_release) | |
3068 | r = ops->pr_release(bdev, key, type); | |
3069 | else | |
3070 | r = -EOPNOTSUPP; | |
971888c4 MS |
3071 | out: |
3072 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3073 | return r; |
3074 | } | |
3075 | ||
3076 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3077 | enum pr_type type, bool abort) |
71cdb697 CH |
3078 | { |
3079 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3080 | const struct pr_ops *ops; | |
971888c4 | 3081 | int r, srcu_idx; |
71cdb697 | 3082 | |
5bd5e8d8 | 3083 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3084 | if (r < 0) |
971888c4 | 3085 | goto out; |
71cdb697 CH |
3086 | |
3087 | ops = bdev->bd_disk->fops->pr_ops; | |
3088 | if (ops && ops->pr_preempt) | |
3089 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3090 | else | |
3091 | r = -EOPNOTSUPP; | |
971888c4 MS |
3092 | out: |
3093 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3094 | return r; |
3095 | } | |
3096 | ||
3097 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3098 | { | |
3099 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3100 | const struct pr_ops *ops; | |
971888c4 | 3101 | int r, srcu_idx; |
71cdb697 | 3102 | |
5bd5e8d8 | 3103 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3104 | if (r < 0) |
971888c4 | 3105 | goto out; |
71cdb697 CH |
3106 | |
3107 | ops = bdev->bd_disk->fops->pr_ops; | |
3108 | if (ops && ops->pr_clear) | |
3109 | r = ops->pr_clear(bdev, key); | |
3110 | else | |
3111 | r = -EOPNOTSUPP; | |
971888c4 MS |
3112 | out: |
3113 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3114 | return r; |
3115 | } | |
3116 | ||
3117 | static const struct pr_ops dm_pr_ops = { | |
3118 | .pr_register = dm_pr_register, | |
3119 | .pr_reserve = dm_pr_reserve, | |
3120 | .pr_release = dm_pr_release, | |
3121 | .pr_preempt = dm_pr_preempt, | |
3122 | .pr_clear = dm_pr_clear, | |
3123 | }; | |
3124 | ||
83d5cde4 | 3125 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3126 | .open = dm_blk_open, |
3127 | .release = dm_blk_close, | |
aa129a22 | 3128 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3129 | .getgeo = dm_blk_getgeo, |
71cdb697 | 3130 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3131 | .owner = THIS_MODULE |
3132 | }; | |
3133 | ||
f26c5719 DW |
3134 | static const struct dax_operations dm_dax_ops = { |
3135 | .direct_access = dm_dax_direct_access, | |
7e026c8c | 3136 | .copy_from_iter = dm_dax_copy_from_iter, |
f26c5719 DW |
3137 | }; |
3138 | ||
1da177e4 LT |
3139 | /* |
3140 | * module hooks | |
3141 | */ | |
3142 | module_init(dm_init); | |
3143 | module_exit(dm_exit); | |
3144 | ||
3145 | module_param(major, uint, 0); | |
3146 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3147 | |
e8603136 MS |
3148 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3149 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3150 | ||
115485e8 MS |
3151 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
3152 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
3153 | ||
1da177e4 LT |
3154 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3155 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3156 | MODULE_LICENSE("GPL"); |