]>
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 { | |
6f1c819c KO |
151 | struct bio_set bs; |
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 | ||
e76239a3 CH |
461 | static int dm_blk_report_zones(struct gendisk *disk, sector_t sector, |
462 | struct blk_zone *zones, unsigned int *nr_zones, | |
463 | gfp_t gfp_mask) | |
464 | { | |
465 | #ifdef CONFIG_BLK_DEV_ZONED | |
466 | struct mapped_device *md = disk->private_data; | |
467 | struct dm_target *tgt; | |
468 | struct dm_table *map; | |
469 | int srcu_idx, ret; | |
470 | ||
471 | if (dm_suspended_md(md)) | |
472 | return -EAGAIN; | |
473 | ||
474 | map = dm_get_live_table(md, &srcu_idx); | |
475 | if (!map) | |
476 | return -EIO; | |
477 | ||
478 | tgt = dm_table_find_target(map, sector); | |
479 | if (!dm_target_is_valid(tgt)) { | |
480 | ret = -EIO; | |
481 | goto out; | |
482 | } | |
483 | ||
484 | /* | |
485 | * If we are executing this, we already know that the block device | |
486 | * is a zoned device and so each target should have support for that | |
487 | * type of drive. A missing report_zones method means that the target | |
488 | * driver has a problem. | |
489 | */ | |
490 | if (WARN_ON(!tgt->type->report_zones)) { | |
491 | ret = -EIO; | |
492 | goto out; | |
493 | } | |
494 | ||
495 | /* | |
496 | * blkdev_report_zones() will loop and call this again to cover all the | |
497 | * zones of the target, eventually moving on to the next target. | |
498 | * So there is no need to loop here trying to fill the entire array | |
499 | * of zones. | |
500 | */ | |
501 | ret = tgt->type->report_zones(tgt, sector, zones, | |
502 | nr_zones, gfp_mask); | |
503 | ||
504 | out: | |
505 | dm_put_live_table(md, srcu_idx); | |
506 | return ret; | |
507 | #else | |
508 | return -ENOTSUPP; | |
509 | #endif | |
510 | } | |
511 | ||
971888c4 | 512 | static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, |
5bd5e8d8 | 513 | struct block_device **bdev) |
971888c4 | 514 | __acquires(md->io_barrier) |
aa129a22 | 515 | { |
66482026 | 516 | struct dm_target *tgt; |
6c182cd8 | 517 | struct dm_table *map; |
971888c4 | 518 | int r; |
aa129a22 | 519 | |
6c182cd8 | 520 | retry: |
e56f81e0 | 521 | r = -ENOTTY; |
971888c4 | 522 | map = dm_get_live_table(md, srcu_idx); |
aa129a22 | 523 | if (!map || !dm_table_get_size(map)) |
971888c4 | 524 | return r; |
aa129a22 MB |
525 | |
526 | /* We only support devices that have a single target */ | |
527 | if (dm_table_get_num_targets(map) != 1) | |
971888c4 | 528 | return r; |
aa129a22 | 529 | |
66482026 MS |
530 | tgt = dm_table_get_target(map, 0); |
531 | if (!tgt->type->prepare_ioctl) | |
971888c4 | 532 | return r; |
519049af | 533 | |
971888c4 MS |
534 | if (dm_suspended_md(md)) |
535 | return -EAGAIN; | |
aa129a22 | 536 | |
5bd5e8d8 | 537 | r = tgt->type->prepare_ioctl(tgt, bdev); |
5bbbfdf6 | 538 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
971888c4 | 539 | dm_put_live_table(md, *srcu_idx); |
6c182cd8 HR |
540 | msleep(10); |
541 | goto retry; | |
542 | } | |
971888c4 | 543 | |
e56f81e0 CH |
544 | return r; |
545 | } | |
546 | ||
971888c4 MS |
547 | static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) |
548 | __releases(md->io_barrier) | |
549 | { | |
550 | dm_put_live_table(md, srcu_idx); | |
551 | } | |
552 | ||
e56f81e0 CH |
553 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
554 | unsigned int cmd, unsigned long arg) | |
555 | { | |
556 | struct mapped_device *md = bdev->bd_disk->private_data; | |
971888c4 | 557 | int r, srcu_idx; |
e56f81e0 | 558 | |
5bd5e8d8 | 559 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
e56f81e0 | 560 | if (r < 0) |
971888c4 | 561 | goto out; |
6c182cd8 | 562 | |
e56f81e0 CH |
563 | if (r > 0) { |
564 | /* | |
e980f623 CH |
565 | * Target determined this ioctl is being issued against a |
566 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 567 | */ |
e980f623 CH |
568 | if (!capable(CAP_SYS_RAWIO)) { |
569 | DMWARN_LIMIT( | |
570 | "%s: sending ioctl %x to DM device without required privilege.", | |
571 | current->comm, cmd); | |
572 | r = -ENOIOCTLCMD; | |
e56f81e0 | 573 | goto out; |
e980f623 | 574 | } |
e56f81e0 | 575 | } |
6c182cd8 | 576 | |
66482026 | 577 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 578 | out: |
971888c4 | 579 | dm_unprepare_ioctl(md, srcu_idx); |
aa129a22 MB |
580 | return r; |
581 | } | |
582 | ||
978e51ba MS |
583 | static void start_io_acct(struct dm_io *io); |
584 | ||
585 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) | |
1da177e4 | 586 | { |
64f52b0e MS |
587 | struct dm_io *io; |
588 | struct dm_target_io *tio; | |
589 | struct bio *clone; | |
590 | ||
6f1c819c | 591 | clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs); |
64f52b0e MS |
592 | if (!clone) |
593 | return NULL; | |
594 | ||
595 | tio = container_of(clone, struct dm_target_io, clone); | |
596 | tio->inside_dm_io = true; | |
597 | tio->io = NULL; | |
598 | ||
599 | io = container_of(tio, struct dm_io, tio); | |
600 | io->magic = DM_IO_MAGIC; | |
978e51ba MS |
601 | io->status = 0; |
602 | atomic_set(&io->io_count, 1); | |
603 | io->orig_bio = bio; | |
604 | io->md = md; | |
605 | spin_lock_init(&io->endio_lock); | |
606 | ||
607 | start_io_acct(io); | |
64f52b0e MS |
608 | |
609 | return io; | |
1da177e4 LT |
610 | } |
611 | ||
028867ac | 612 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 | 613 | { |
64f52b0e MS |
614 | bio_put(&io->tio.clone); |
615 | } | |
616 | ||
617 | static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, | |
618 | unsigned target_bio_nr, gfp_t gfp_mask) | |
619 | { | |
620 | struct dm_target_io *tio; | |
621 | ||
622 | if (!ci->io->tio.io) { | |
623 | /* the dm_target_io embedded in ci->io is available */ | |
624 | tio = &ci->io->tio; | |
625 | } else { | |
6f1c819c | 626 | struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs); |
64f52b0e MS |
627 | if (!clone) |
628 | return NULL; | |
629 | ||
630 | tio = container_of(clone, struct dm_target_io, clone); | |
631 | tio->inside_dm_io = false; | |
632 | } | |
633 | ||
634 | tio->magic = DM_TIO_MAGIC; | |
635 | tio->io = ci->io; | |
636 | tio->ti = ti; | |
637 | tio->target_bio_nr = target_bio_nr; | |
638 | ||
639 | return tio; | |
1da177e4 LT |
640 | } |
641 | ||
cfae7529 | 642 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 643 | { |
64f52b0e MS |
644 | if (tio->inside_dm_io) |
645 | return; | |
dba14160 | 646 | bio_put(&tio->clone); |
1da177e4 LT |
647 | } |
648 | ||
c4576aed | 649 | static bool md_in_flight_bios(struct mapped_device *md) |
90abb8c4 | 650 | { |
6f757231 MP |
651 | int cpu; |
652 | struct hd_struct *part = &dm_disk(md)->part0; | |
b7934ba4 | 653 | long sum = 0; |
6f757231 MP |
654 | |
655 | for_each_possible_cpu(cpu) { | |
b7934ba4 JA |
656 | sum += part_stat_local_read_cpu(part, in_flight[0], cpu); |
657 | sum += part_stat_local_read_cpu(part, in_flight[1], cpu); | |
6f757231 MP |
658 | } |
659 | ||
b7934ba4 | 660 | return sum != 0; |
90abb8c4 KU |
661 | } |
662 | ||
c4576aed MS |
663 | static bool md_in_flight(struct mapped_device *md) |
664 | { | |
665 | if (queue_is_mq(md->queue)) | |
3c94d83c | 666 | return blk_mq_queue_inflight(md->queue); |
c4576aed MS |
667 | else |
668 | return md_in_flight_bios(md); | |
90abb8c4 KU |
669 | } |
670 | ||
3eaf840e JNN |
671 | static void start_io_acct(struct dm_io *io) |
672 | { | |
673 | struct mapped_device *md = io->md; | |
745dc570 | 674 | struct bio *bio = io->orig_bio; |
3eaf840e JNN |
675 | |
676 | io->start_time = jiffies; | |
677 | ||
ddcf35d3 MC |
678 | generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio), |
679 | &dm_disk(md)->part0); | |
f3986374 | 680 | |
fd2ed4d2 | 681 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
682 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
683 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
684 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
685 | } |
686 | ||
d221d2e7 | 687 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
688 | { |
689 | struct mapped_device *md = io->md; | |
745dc570 | 690 | struct bio *bio = io->orig_bio; |
3eaf840e | 691 | unsigned long duration = jiffies - io->start_time; |
3eaf840e | 692 | |
ddcf35d3 MC |
693 | generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0, |
694 | io->start_time); | |
3eaf840e | 695 | |
fd2ed4d2 | 696 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
697 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
698 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
699 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 700 | |
d221d2e7 | 701 | /* nudge anyone waiting on suspend queue */ |
645efa84 | 702 | if (unlikely(wq_has_sleeper(&md->wait))) |
d221d2e7 | 703 | wake_up(&md->wait); |
3eaf840e JNN |
704 | } |
705 | ||
1da177e4 LT |
706 | /* |
707 | * Add the bio to the list of deferred io. | |
708 | */ | |
92c63902 | 709 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 710 | { |
05447420 | 711 | unsigned long flags; |
1da177e4 | 712 | |
05447420 | 713 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 714 | bio_list_add(&md->deferred, bio); |
05447420 | 715 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 716 | queue_work(md->wq, &md->work); |
1da177e4 LT |
717 | } |
718 | ||
719 | /* | |
720 | * Everyone (including functions in this file), should use this | |
721 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 722 | * dm_put_live_table() when finished. |
1da177e4 | 723 | */ |
83d5e5b0 | 724 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 725 | { |
83d5e5b0 MP |
726 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
727 | ||
728 | return srcu_dereference(md->map, &md->io_barrier); | |
729 | } | |
1da177e4 | 730 | |
83d5e5b0 MP |
731 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
732 | { | |
733 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
734 | } | |
735 | ||
736 | void dm_sync_table(struct mapped_device *md) | |
737 | { | |
738 | synchronize_srcu(&md->io_barrier); | |
739 | synchronize_rcu_expedited(); | |
740 | } | |
741 | ||
742 | /* | |
743 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
744 | * The caller must not block between these two functions. | |
745 | */ | |
746 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
747 | { | |
748 | rcu_read_lock(); | |
749 | return rcu_dereference(md->map); | |
750 | } | |
1da177e4 | 751 | |
83d5e5b0 MP |
752 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
753 | { | |
754 | rcu_read_unlock(); | |
1da177e4 LT |
755 | } |
756 | ||
971888c4 MS |
757 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
758 | ||
86f1152b BM |
759 | /* |
760 | * Open a table device so we can use it as a map destination. | |
761 | */ | |
762 | static int open_table_device(struct table_device *td, dev_t dev, | |
763 | struct mapped_device *md) | |
764 | { | |
86f1152b BM |
765 | struct block_device *bdev; |
766 | ||
767 | int r; | |
768 | ||
769 | BUG_ON(td->dm_dev.bdev); | |
770 | ||
519049af | 771 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr); |
86f1152b BM |
772 | if (IS_ERR(bdev)) |
773 | return PTR_ERR(bdev); | |
774 | ||
775 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
776 | if (r) { | |
777 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
778 | return r; | |
779 | } | |
780 | ||
781 | td->dm_dev.bdev = bdev; | |
817bf402 | 782 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
783 | return 0; |
784 | } | |
785 | ||
786 | /* | |
787 | * Close a table device that we've been using. | |
788 | */ | |
789 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
790 | { | |
791 | if (!td->dm_dev.bdev) | |
792 | return; | |
793 | ||
794 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
795 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 796 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 797 | td->dm_dev.bdev = NULL; |
817bf402 | 798 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
799 | } |
800 | ||
801 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
802 | fmode_t mode) { | |
803 | struct table_device *td; | |
804 | ||
805 | list_for_each_entry(td, l, list) | |
806 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
807 | return td; | |
808 | ||
809 | return NULL; | |
810 | } | |
811 | ||
812 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
813 | struct dm_dev **result) { | |
814 | int r; | |
815 | struct table_device *td; | |
816 | ||
817 | mutex_lock(&md->table_devices_lock); | |
818 | td = find_table_device(&md->table_devices, dev, mode); | |
819 | if (!td) { | |
115485e8 | 820 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
821 | if (!td) { |
822 | mutex_unlock(&md->table_devices_lock); | |
823 | return -ENOMEM; | |
824 | } | |
825 | ||
826 | td->dm_dev.mode = mode; | |
827 | td->dm_dev.bdev = NULL; | |
828 | ||
829 | if ((r = open_table_device(td, dev, md))) { | |
830 | mutex_unlock(&md->table_devices_lock); | |
831 | kfree(td); | |
832 | return r; | |
833 | } | |
834 | ||
835 | format_dev_t(td->dm_dev.name, dev); | |
836 | ||
b0b4d7c6 | 837 | refcount_set(&td->count, 1); |
86f1152b | 838 | list_add(&td->list, &md->table_devices); |
b0b4d7c6 ER |
839 | } else { |
840 | refcount_inc(&td->count); | |
86f1152b | 841 | } |
86f1152b BM |
842 | mutex_unlock(&md->table_devices_lock); |
843 | ||
844 | *result = &td->dm_dev; | |
845 | return 0; | |
846 | } | |
847 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
848 | ||
849 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
850 | { | |
851 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
852 | ||
853 | mutex_lock(&md->table_devices_lock); | |
b0b4d7c6 | 854 | if (refcount_dec_and_test(&td->count)) { |
86f1152b BM |
855 | close_table_device(td, md); |
856 | list_del(&td->list); | |
857 | kfree(td); | |
858 | } | |
859 | mutex_unlock(&md->table_devices_lock); | |
860 | } | |
861 | EXPORT_SYMBOL(dm_put_table_device); | |
862 | ||
863 | static void free_table_devices(struct list_head *devices) | |
864 | { | |
865 | struct list_head *tmp, *next; | |
866 | ||
867 | list_for_each_safe(tmp, next, devices) { | |
868 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
869 | ||
870 | DMWARN("dm_destroy: %s still exists with %d references", | |
b0b4d7c6 | 871 | td->dm_dev.name, refcount_read(&td->count)); |
86f1152b BM |
872 | kfree(td); |
873 | } | |
874 | } | |
875 | ||
3ac51e74 DW |
876 | /* |
877 | * Get the geometry associated with a dm device | |
878 | */ | |
879 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
880 | { | |
881 | *geo = md->geometry; | |
882 | ||
883 | return 0; | |
884 | } | |
885 | ||
886 | /* | |
887 | * Set the geometry of a device. | |
888 | */ | |
889 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
890 | { | |
891 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
892 | ||
893 | if (geo->start > sz) { | |
894 | DMWARN("Start sector is beyond the geometry limits."); | |
895 | return -EINVAL; | |
896 | } | |
897 | ||
898 | md->geometry = *geo; | |
899 | ||
900 | return 0; | |
901 | } | |
902 | ||
2e93ccc1 KU |
903 | static int __noflush_suspending(struct mapped_device *md) |
904 | { | |
905 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
906 | } | |
907 | ||
1da177e4 LT |
908 | /* |
909 | * Decrements the number of outstanding ios that a bio has been | |
910 | * cloned into, completing the original io if necc. | |
911 | */ | |
4e4cbee9 | 912 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 913 | { |
2e93ccc1 | 914 | unsigned long flags; |
4e4cbee9 | 915 | blk_status_t io_error; |
b35f8caa MB |
916 | struct bio *bio; |
917 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
918 | |
919 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
920 | if (unlikely(error)) { |
921 | spin_lock_irqsave(&io->endio_lock, flags); | |
745dc570 | 922 | if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) |
4e4cbee9 | 923 | io->status = error; |
f88fb981 KU |
924 | spin_unlock_irqrestore(&io->endio_lock, flags); |
925 | } | |
1da177e4 LT |
926 | |
927 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 928 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
929 | /* |
930 | * Target requested pushing back the I/O. | |
2e93ccc1 | 931 | */ |
022c2611 | 932 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 | 933 | if (__noflush_suspending(md)) |
745dc570 MS |
934 | /* NOTE early return due to BLK_STS_DM_REQUEUE below */ |
935 | bio_list_add_head(&md->deferred, io->orig_bio); | |
6a8736d1 | 936 | else |
2e93ccc1 | 937 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 938 | io->status = BLK_STS_IOERR; |
022c2611 | 939 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
940 | } |
941 | ||
4e4cbee9 | 942 | io_error = io->status; |
745dc570 | 943 | bio = io->orig_bio; |
6a8736d1 TH |
944 | end_io_acct(io); |
945 | free_io(md, io); | |
946 | ||
4e4cbee9 | 947 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 948 | return; |
2e93ccc1 | 949 | |
1eff9d32 | 950 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 951 | /* |
6a8736d1 | 952 | * Preflush done for flush with data, reissue |
28a8f0d3 | 953 | * without REQ_PREFLUSH. |
af7e466a | 954 | */ |
1eff9d32 | 955 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 956 | queue_io(md, bio); |
af7e466a | 957 | } else { |
b372d360 | 958 | /* done with normal IO or empty flush */ |
8dd601fa N |
959 | if (io_error) |
960 | bio->bi_status = io_error; | |
4246a0b6 | 961 | bio_endio(bio); |
b35f8caa | 962 | } |
1da177e4 LT |
963 | } |
964 | } | |
965 | ||
4cc96131 | 966 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
967 | { |
968 | struct queue_limits *limits = dm_get_queue_limits(md); | |
969 | ||
970 | /* device doesn't really support WRITE SAME, disable it */ | |
971 | limits->max_write_same_sectors = 0; | |
972 | } | |
973 | ||
ac62d620 CH |
974 | void disable_write_zeroes(struct mapped_device *md) |
975 | { | |
976 | struct queue_limits *limits = dm_get_queue_limits(md); | |
977 | ||
978 | /* device doesn't really support WRITE ZEROES, disable it */ | |
979 | limits->max_write_zeroes_sectors = 0; | |
980 | } | |
981 | ||
4246a0b6 | 982 | static void clone_endio(struct bio *bio) |
1da177e4 | 983 | { |
4e4cbee9 | 984 | blk_status_t error = bio->bi_status; |
bfc6d41c | 985 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 986 | struct dm_io *io = tio->io; |
9faf400f | 987 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
988 | dm_endio_fn endio = tio->ti->type->end_io; |
989 | ||
978e51ba | 990 | if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) { |
ac62d620 | 991 | if (bio_op(bio) == REQ_OP_WRITE_SAME && |
74d46992 | 992 | !bio->bi_disk->queue->limits.max_write_same_sectors) |
ac62d620 CH |
993 | disable_write_same(md); |
994 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && | |
74d46992 | 995 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) |
ac62d620 CH |
996 | disable_write_zeroes(md); |
997 | } | |
7eee4ae2 | 998 | |
1be56909 | 999 | if (endio) { |
4e4cbee9 | 1000 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
1001 | switch (r) { |
1002 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 1003 | error = BLK_STS_DM_REQUEUE; |
1be56909 CH |
1004 | /*FALLTHRU*/ |
1005 | case DM_ENDIO_DONE: | |
1006 | break; | |
1007 | case DM_ENDIO_INCOMPLETE: | |
1008 | /* The target will handle the io */ | |
1009 | return; | |
1010 | default: | |
1011 | DMWARN("unimplemented target endio return value: %d", r); | |
1012 | BUG(); | |
1013 | } | |
1014 | } | |
1015 | ||
cfae7529 | 1016 | free_tio(tio); |
b35f8caa | 1017 | dec_pending(io, error); |
1da177e4 LT |
1018 | } |
1019 | ||
56a67df7 MS |
1020 | /* |
1021 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1022 | * target boundary. | |
1023 | */ | |
1024 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1025 | { | |
1026 | sector_t target_offset = dm_target_offset(ti, sector); | |
1027 | ||
1028 | return ti->len - target_offset; | |
1029 | } | |
1030 | ||
1031 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1032 | { |
56a67df7 | 1033 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1034 | sector_t offset, max_len; |
1da177e4 LT |
1035 | |
1036 | /* | |
542f9038 | 1037 | * Does the target need to split even further? |
1da177e4 | 1038 | */ |
542f9038 MS |
1039 | if (ti->max_io_len) { |
1040 | offset = dm_target_offset(ti, sector); | |
1041 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1042 | max_len = sector_div(offset, ti->max_io_len); | |
1043 | else | |
1044 | max_len = offset & (ti->max_io_len - 1); | |
1045 | max_len = ti->max_io_len - max_len; | |
1046 | ||
1047 | if (len > max_len) | |
1048 | len = max_len; | |
1da177e4 LT |
1049 | } |
1050 | ||
1051 | return len; | |
1052 | } | |
1053 | ||
542f9038 MS |
1054 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1055 | { | |
1056 | if (len > UINT_MAX) { | |
1057 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1058 | (unsigned long long)len, UINT_MAX); | |
1059 | ti->error = "Maximum size of target IO is too large"; | |
1060 | return -EINVAL; | |
1061 | } | |
1062 | ||
8f50e358 ML |
1063 | /* |
1064 | * BIO based queue uses its own splitting. When multipage bvecs | |
1065 | * is switched on, size of the incoming bio may be too big to | |
1066 | * be handled in some targets, such as crypt. | |
1067 | * | |
1068 | * When these targets are ready for the big bio, we can remove | |
1069 | * the limit. | |
1070 | */ | |
1071 | ti->max_io_len = min_t(uint32_t, len, BIO_MAX_PAGES * PAGE_SIZE); | |
542f9038 MS |
1072 | |
1073 | return 0; | |
1074 | } | |
1075 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1076 | ||
f26c5719 | 1077 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
3d97c829 MS |
1078 | sector_t sector, int *srcu_idx) |
1079 | __acquires(md->io_barrier) | |
545ed20e | 1080 | { |
545ed20e TK |
1081 | struct dm_table *map; |
1082 | struct dm_target *ti; | |
545ed20e | 1083 | |
f26c5719 | 1084 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1085 | if (!map) |
f26c5719 | 1086 | return NULL; |
545ed20e TK |
1087 | |
1088 | ti = dm_table_find_target(map, sector); | |
1089 | if (!dm_target_is_valid(ti)) | |
f26c5719 | 1090 | return NULL; |
545ed20e | 1091 | |
f26c5719 DW |
1092 | return ti; |
1093 | } | |
545ed20e | 1094 | |
f26c5719 | 1095 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1096 | long nr_pages, void **kaddr, pfn_t *pfn) |
f26c5719 DW |
1097 | { |
1098 | struct mapped_device *md = dax_get_private(dax_dev); | |
1099 | sector_t sector = pgoff * PAGE_SECTORS; | |
1100 | struct dm_target *ti; | |
1101 | long len, ret = -EIO; | |
1102 | int srcu_idx; | |
545ed20e | 1103 | |
f26c5719 | 1104 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1105 | |
f26c5719 DW |
1106 | if (!ti) |
1107 | goto out; | |
1108 | if (!ti->type->direct_access) | |
1109 | goto out; | |
1110 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
1111 | if (len < 1) | |
1112 | goto out; | |
1113 | nr_pages = min(len, nr_pages); | |
dbc62659 | 1114 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
817bf402 | 1115 | |
f26c5719 | 1116 | out: |
545ed20e | 1117 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1118 | |
1119 | return ret; | |
545ed20e TK |
1120 | } |
1121 | ||
7e026c8c | 1122 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1123 | void *addr, size_t bytes, struct iov_iter *i) |
7e026c8c DW |
1124 | { |
1125 | struct mapped_device *md = dax_get_private(dax_dev); | |
1126 | sector_t sector = pgoff * PAGE_SECTORS; | |
1127 | struct dm_target *ti; | |
1128 | long ret = 0; | |
1129 | int srcu_idx; | |
1130 | ||
1131 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1132 | ||
1133 | if (!ti) | |
1134 | goto out; | |
1135 | if (!ti->type->dax_copy_from_iter) { | |
1136 | ret = copy_from_iter(addr, bytes, i); | |
1137 | goto out; | |
1138 | } | |
1139 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
1140 | out: | |
1141 | dm_put_live_table(md, srcu_idx); | |
1142 | ||
1143 | return ret; | |
1144 | } | |
1145 | ||
b3a9a0c3 DW |
1146 | static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
1147 | void *addr, size_t bytes, struct iov_iter *i) | |
1148 | { | |
1149 | struct mapped_device *md = dax_get_private(dax_dev); | |
1150 | sector_t sector = pgoff * PAGE_SECTORS; | |
1151 | struct dm_target *ti; | |
1152 | long ret = 0; | |
1153 | int srcu_idx; | |
1154 | ||
1155 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1156 | ||
1157 | if (!ti) | |
1158 | goto out; | |
1159 | if (!ti->type->dax_copy_to_iter) { | |
1160 | ret = copy_to_iter(addr, bytes, i); | |
1161 | goto out; | |
1162 | } | |
1163 | ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i); | |
1164 | out: | |
1165 | dm_put_live_table(md, srcu_idx); | |
1166 | ||
1167 | return ret; | |
1168 | } | |
1169 | ||
1dd40c3e MP |
1170 | /* |
1171 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
c06b3e58 | 1172 | * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET. |
1dd40c3e MP |
1173 | * |
1174 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1175 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1176 | * sent in a next bio. | |
1177 | * | |
1178 | * A diagram that explains the arithmetics: | |
1179 | * +--------------------+---------------+-------+ | |
1180 | * | 1 | 2 | 3 | | |
1181 | * +--------------------+---------------+-------+ | |
1182 | * | |
1183 | * <-------------- *tio->len_ptr ---------------> | |
1184 | * <------- bi_size -------> | |
1185 | * <-- n_sectors --> | |
1186 | * | |
1187 | * Region 1 was already iterated over with bio_advance or similar function. | |
1188 | * (it may be empty if the target doesn't use bio_advance) | |
1189 | * Region 2 is the remaining bio size that the target wants to process. | |
1190 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1191 | * to make it empty) | |
1192 | * The target requires that region 3 is to be sent in the next bio. | |
1193 | * | |
1194 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1195 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1196 | * copies of the bio. | |
1197 | */ | |
1198 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1199 | { | |
1200 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1201 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1202 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1203 | BUG_ON(bi_size > *tio->len_ptr); |
1204 | BUG_ON(n_sectors > bi_size); | |
1205 | *tio->len_ptr -= bi_size - n_sectors; | |
1206 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1207 | } | |
1208 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1209 | ||
10999307 | 1210 | /* |
e76239a3 CH |
1211 | * The zone descriptors obtained with a zone report indicate |
1212 | * zone positions within the underlying device of the target. The zone | |
1213 | * descriptors must be remapped to match their position within the dm device. | |
1214 | * The caller target should obtain the zones information using | |
1215 | * blkdev_report_zones() to ensure that remapping for partition offset is | |
1216 | * already handled. | |
10999307 | 1217 | */ |
e76239a3 CH |
1218 | void dm_remap_zone_report(struct dm_target *ti, sector_t start, |
1219 | struct blk_zone *zones, unsigned int *nr_zones) | |
10999307 DLM |
1220 | { |
1221 | #ifdef CONFIG_BLK_DEV_ZONED | |
10999307 | 1222 | struct blk_zone *zone; |
e76239a3 CH |
1223 | unsigned int nrz = *nr_zones; |
1224 | int i; | |
9864cd5d | 1225 | |
10999307 | 1226 | /* |
e76239a3 CH |
1227 | * Remap the start sector and write pointer position of the zones in |
1228 | * the array. Since we may have obtained from the target underlying | |
1229 | * device more zones that the target size, also adjust the number | |
1230 | * of zones. | |
10999307 | 1231 | */ |
e76239a3 CH |
1232 | for (i = 0; i < nrz; i++) { |
1233 | zone = zones + i; | |
1234 | if (zone->start >= start + ti->len) { | |
1235 | memset(zone, 0, sizeof(struct blk_zone) * (nrz - i)); | |
1236 | break; | |
10999307 DLM |
1237 | } |
1238 | ||
e76239a3 CH |
1239 | zone->start = zone->start + ti->begin - start; |
1240 | if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL) | |
1241 | continue; | |
10999307 | 1242 | |
e76239a3 CH |
1243 | if (zone->cond == BLK_ZONE_COND_FULL) |
1244 | zone->wp = zone->start + zone->len; | |
1245 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
1246 | zone->wp = zone->start; | |
1247 | else | |
1248 | zone->wp = zone->wp + ti->begin - start; | |
10999307 DLM |
1249 | } |
1250 | ||
e76239a3 | 1251 | *nr_zones = i; |
10999307 | 1252 | #else /* !CONFIG_BLK_DEV_ZONED */ |
e76239a3 | 1253 | *nr_zones = 0; |
10999307 DLM |
1254 | #endif |
1255 | } | |
1256 | EXPORT_SYMBOL_GPL(dm_remap_zone_report); | |
1257 | ||
978e51ba | 1258 | static blk_qc_t __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1259 | { |
1260 | int r; | |
2056a782 | 1261 | sector_t sector; |
dba14160 | 1262 | struct bio *clone = &tio->clone; |
64f52b0e | 1263 | struct dm_io *io = tio->io; |
978e51ba | 1264 | struct mapped_device *md = io->md; |
bd2a49b8 | 1265 | struct dm_target *ti = tio->ti; |
978e51ba | 1266 | blk_qc_t ret = BLK_QC_T_NONE; |
1da177e4 | 1267 | |
1da177e4 | 1268 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1269 | |
1270 | /* | |
1271 | * Map the clone. If r == 0 we don't need to do | |
1272 | * anything, the target has assumed ownership of | |
1273 | * this io. | |
1274 | */ | |
64f52b0e | 1275 | atomic_inc(&io->io_count); |
4f024f37 | 1276 | sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1277 | |
7de3ee57 | 1278 | r = ti->type->map(ti, clone); |
846785e6 CH |
1279 | switch (r) { |
1280 | case DM_MAPIO_SUBMITTED: | |
1281 | break; | |
1282 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1283 | /* the bio has been remapped so dispatch it */ |
74d46992 | 1284 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
64f52b0e | 1285 | bio_dev(io->orig_bio), sector); |
978e51ba MS |
1286 | if (md->type == DM_TYPE_NVME_BIO_BASED) |
1287 | ret = direct_make_request(clone); | |
1288 | else | |
1289 | ret = generic_make_request(clone); | |
846785e6 CH |
1290 | break; |
1291 | case DM_MAPIO_KILL: | |
4e4cbee9 | 1292 | free_tio(tio); |
64f52b0e | 1293 | dec_pending(io, BLK_STS_IOERR); |
4e4cbee9 | 1294 | break; |
846785e6 | 1295 | case DM_MAPIO_REQUEUE: |
cfae7529 | 1296 | free_tio(tio); |
64f52b0e | 1297 | dec_pending(io, BLK_STS_DM_REQUEUE); |
846785e6 CH |
1298 | break; |
1299 | default: | |
45cbcd79 KU |
1300 | DMWARN("unimplemented target map return value: %d", r); |
1301 | BUG(); | |
1da177e4 | 1302 | } |
1da177e4 | 1303 | |
978e51ba | 1304 | return ret; |
1da177e4 | 1305 | } |
1da177e4 | 1306 | |
e0d6609a | 1307 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1308 | { |
4f024f37 KO |
1309 | bio->bi_iter.bi_sector = sector; |
1310 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1311 | } |
1312 | ||
1313 | /* | |
1314 | * Creates a bio that consists of range of complete bvecs. | |
1315 | */ | |
c80914e8 MS |
1316 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1317 | sector_t sector, unsigned len) | |
1da177e4 | 1318 | { |
dba14160 | 1319 | struct bio *clone = &tio->clone; |
1da177e4 | 1320 | |
1c3b13e6 KO |
1321 | __bio_clone_fast(clone, bio); |
1322 | ||
57c36519 | 1323 | if (bio_integrity(bio)) { |
e2460f2a MP |
1324 | int r; |
1325 | ||
1326 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1327 | !dm_target_passes_integrity(tio->ti->type))) { | |
1328 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1329 | dm_device_name(tio->io->md), | |
1330 | tio->ti->type->name); | |
1331 | return -EIO; | |
1332 | } | |
1333 | ||
1334 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1335 | if (r < 0) |
1336 | return r; | |
1337 | } | |
bd2a49b8 | 1338 | |
fa8db494 MS |
1339 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1340 | clone->bi_iter.bi_size = to_bytes(len); | |
1341 | ||
1342 | if (bio_integrity(bio)) | |
1343 | bio_integrity_trim(clone); | |
c80914e8 MS |
1344 | |
1345 | return 0; | |
1da177e4 LT |
1346 | } |
1347 | ||
318716dd MS |
1348 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
1349 | struct dm_target *ti, unsigned num_bios) | |
f9ab94ce | 1350 | { |
dba14160 | 1351 | struct dm_target_io *tio; |
318716dd | 1352 | int try; |
dba14160 | 1353 | |
318716dd MS |
1354 | if (!num_bios) |
1355 | return; | |
f9ab94ce | 1356 | |
318716dd MS |
1357 | if (num_bios == 1) { |
1358 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); | |
1359 | bio_list_add(blist, &tio->clone); | |
1360 | return; | |
1361 | } | |
9015df24 | 1362 | |
318716dd MS |
1363 | for (try = 0; try < 2; try++) { |
1364 | int bio_nr; | |
1365 | struct bio *bio; | |
1366 | ||
1367 | if (try) | |
bc02cdbe | 1368 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd MS |
1369 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
1370 | tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); | |
1371 | if (!tio) | |
1372 | break; | |
1373 | ||
1374 | bio_list_add(blist, &tio->clone); | |
1375 | } | |
1376 | if (try) | |
bc02cdbe | 1377 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1378 | if (bio_nr == num_bios) |
1379 | return; | |
1380 | ||
1381 | while ((bio = bio_list_pop(blist))) { | |
1382 | tio = container_of(bio, struct dm_target_io, clone); | |
1383 | free_tio(tio); | |
1384 | } | |
1385 | } | |
9015df24 AK |
1386 | } |
1387 | ||
978e51ba MS |
1388 | static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci, |
1389 | struct dm_target_io *tio, unsigned *len) | |
9015df24 | 1390 | { |
dba14160 | 1391 | struct bio *clone = &tio->clone; |
9015df24 | 1392 | |
1dd40c3e MP |
1393 | tio->len_ptr = len; |
1394 | ||
99778273 | 1395 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1396 | if (len) |
1dd40c3e | 1397 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1398 | |
978e51ba | 1399 | return __map_bio(tio); |
f9ab94ce MP |
1400 | } |
1401 | ||
14fe594d | 1402 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1403 | unsigned num_bios, unsigned *len) |
06a426ce | 1404 | { |
318716dd MS |
1405 | struct bio_list blist = BIO_EMPTY_LIST; |
1406 | struct bio *bio; | |
1407 | struct dm_target_io *tio; | |
1408 | ||
1409 | alloc_multiple_bios(&blist, ci, ti, num_bios); | |
06a426ce | 1410 | |
318716dd MS |
1411 | while ((bio = bio_list_pop(&blist))) { |
1412 | tio = container_of(bio, struct dm_target_io, clone); | |
978e51ba | 1413 | (void) __clone_and_map_simple_bio(ci, tio, len); |
318716dd | 1414 | } |
06a426ce MS |
1415 | } |
1416 | ||
14fe594d | 1417 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1418 | { |
06a426ce | 1419 | unsigned target_nr = 0; |
f9ab94ce MP |
1420 | struct dm_target *ti; |
1421 | ||
892ad71f | 1422 | /* |
dbe3ece1 JA |
1423 | * Empty flush uses a statically initialized bio, as the base for |
1424 | * cloning. However, blkg association requires that a bdev is | |
1425 | * associated with a gendisk, which doesn't happen until the bdev is | |
1426 | * opened. So, blkg association is done at issue time of the flush | |
1427 | * rather than when the device is created in alloc_dev(). | |
892ad71f DZ |
1428 | */ |
1429 | bio_set_dev(ci->bio, ci->io->md->bdev); | |
1430 | ||
b372d360 | 1431 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1432 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1433 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1434 | |
892ad71f DZ |
1435 | bio_disassociate_blkg(ci->bio); |
1436 | ||
f9ab94ce MP |
1437 | return 0; |
1438 | } | |
1439 | ||
c80914e8 | 1440 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
f31c21e4 | 1441 | sector_t sector, unsigned *len) |
5ae89a87 | 1442 | { |
dba14160 | 1443 | struct bio *bio = ci->bio; |
5ae89a87 | 1444 | struct dm_target_io *tio; |
f31c21e4 | 1445 | int r; |
5ae89a87 | 1446 | |
318716dd | 1447 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); |
f31c21e4 N |
1448 | tio->len_ptr = len; |
1449 | r = clone_bio(tio, bio, sector, *len); | |
1450 | if (r < 0) { | |
1451 | free_tio(tio); | |
1452 | return r; | |
b0d8ed4d | 1453 | } |
978e51ba | 1454 | (void) __map_bio(tio); |
c80914e8 | 1455 | |
f31c21e4 | 1456 | return 0; |
5ae89a87 MS |
1457 | } |
1458 | ||
55a62eef | 1459 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1460 | |
55a62eef | 1461 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1462 | { |
55a62eef | 1463 | return ti->num_discard_bios; |
23508a96 MS |
1464 | } |
1465 | ||
00716545 DS |
1466 | static unsigned get_num_secure_erase_bios(struct dm_target *ti) |
1467 | { | |
1468 | return ti->num_secure_erase_bios; | |
1469 | } | |
1470 | ||
55a62eef | 1471 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1472 | { |
55a62eef | 1473 | return ti->num_write_same_bios; |
23508a96 MS |
1474 | } |
1475 | ||
ac62d620 CH |
1476 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1477 | { | |
1478 | return ti->num_write_zeroes_bios; | |
1479 | } | |
1480 | ||
3d7f4562 | 1481 | static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
61697a6a | 1482 | unsigned num_bios) |
ba1cbad9 | 1483 | { |
61697a6a | 1484 | unsigned len = ci->sector_count; |
ba1cbad9 | 1485 | |
3d7f4562 MS |
1486 | /* |
1487 | * Even though the device advertised support for this type of | |
1488 | * request, that does not mean every target supports it, and | |
1489 | * reconfiguration might also have changed that since the | |
1490 | * check was performed. | |
1491 | */ | |
3d7f4562 MS |
1492 | if (!num_bios) |
1493 | return -EOPNOTSUPP; | |
ba1cbad9 | 1494 | |
3d7f4562 | 1495 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1496 | |
3d7f4562 MS |
1497 | ci->sector += len; |
1498 | ci->sector_count -= len; | |
5ae89a87 MS |
1499 | |
1500 | return 0; | |
ba1cbad9 MS |
1501 | } |
1502 | ||
3d7f4562 | 1503 | static int __send_discard(struct clone_info *ci, struct dm_target *ti) |
23508a96 | 1504 | { |
61697a6a | 1505 | return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti)); |
23508a96 | 1506 | } |
0ce65797 | 1507 | |
00716545 DS |
1508 | static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti) |
1509 | { | |
61697a6a | 1510 | return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti)); |
00716545 DS |
1511 | } |
1512 | ||
3d7f4562 | 1513 | static int __send_write_same(struct clone_info *ci, struct dm_target *ti) |
0ce65797 | 1514 | { |
61697a6a | 1515 | return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti)); |
0ce65797 MS |
1516 | } |
1517 | ||
3d7f4562 | 1518 | static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti) |
ac62d620 | 1519 | { |
61697a6a | 1520 | return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti)); |
ac62d620 CH |
1521 | } |
1522 | ||
568c73a3 MS |
1523 | static bool is_abnormal_io(struct bio *bio) |
1524 | { | |
1525 | bool r = false; | |
1526 | ||
1527 | switch (bio_op(bio)) { | |
1528 | case REQ_OP_DISCARD: | |
1529 | case REQ_OP_SECURE_ERASE: | |
1530 | case REQ_OP_WRITE_SAME: | |
1531 | case REQ_OP_WRITE_ZEROES: | |
1532 | r = true; | |
1533 | break; | |
1534 | } | |
1535 | ||
1536 | return r; | |
1537 | } | |
1538 | ||
0519c71e MS |
1539 | static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti, |
1540 | int *result) | |
1541 | { | |
1542 | struct bio *bio = ci->bio; | |
1543 | ||
1544 | if (bio_op(bio) == REQ_OP_DISCARD) | |
1545 | *result = __send_discard(ci, ti); | |
00716545 DS |
1546 | else if (bio_op(bio) == REQ_OP_SECURE_ERASE) |
1547 | *result = __send_secure_erase(ci, ti); | |
0519c71e MS |
1548 | else if (bio_op(bio) == REQ_OP_WRITE_SAME) |
1549 | *result = __send_write_same(ci, ti); | |
1550 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES) | |
1551 | *result = __send_write_zeroes(ci, ti); | |
1552 | else | |
1553 | return false; | |
1554 | ||
1555 | return true; | |
1556 | } | |
1557 | ||
e4c93811 AK |
1558 | /* |
1559 | * Select the correct strategy for processing a non-flush bio. | |
1560 | */ | |
14fe594d | 1561 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1562 | { |
512875bd | 1563 | struct dm_target *ti; |
1c3b13e6 | 1564 | unsigned len; |
c80914e8 | 1565 | int r; |
0ce65797 | 1566 | |
512875bd JN |
1567 | ti = dm_table_find_target(ci->map, ci->sector); |
1568 | if (!dm_target_is_valid(ti)) | |
1569 | return -EIO; | |
1570 | ||
568c73a3 | 1571 | if (__process_abnormal_io(ci, ti, &r)) |
0519c71e | 1572 | return r; |
3d7f4562 | 1573 | |
e76239a3 | 1574 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
0ce65797 | 1575 | |
c80914e8 MS |
1576 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1577 | if (r < 0) | |
1578 | return r; | |
0ce65797 | 1579 | |
1c3b13e6 KO |
1580 | ci->sector += len; |
1581 | ci->sector_count -= len; | |
0ce65797 | 1582 | |
1c3b13e6 | 1583 | return 0; |
0ce65797 MS |
1584 | } |
1585 | ||
978e51ba MS |
1586 | static void init_clone_info(struct clone_info *ci, struct mapped_device *md, |
1587 | struct dm_table *map, struct bio *bio) | |
1588 | { | |
1589 | ci->map = map; | |
1590 | ci->io = alloc_io(md, bio); | |
1591 | ci->sector = bio->bi_iter.bi_sector; | |
1592 | } | |
1593 | ||
a1e1cb72 MS |
1594 | #define __dm_part_stat_sub(part, field, subnd) \ |
1595 | (part_stat_get(part, field) -= (subnd)) | |
1596 | ||
1da177e4 | 1597 | /* |
14fe594d | 1598 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1599 | */ |
978e51ba MS |
1600 | static blk_qc_t __split_and_process_bio(struct mapped_device *md, |
1601 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1602 | { |
1da177e4 | 1603 | struct clone_info ci; |
978e51ba | 1604 | blk_qc_t ret = BLK_QC_T_NONE; |
512875bd | 1605 | int error = 0; |
1da177e4 | 1606 | |
978e51ba | 1607 | init_clone_info(&ci, md, map, bio); |
0ce65797 | 1608 | |
1eff9d32 | 1609 | if (bio->bi_opf & REQ_PREFLUSH) { |
dbe3ece1 JA |
1610 | struct bio flush_bio; |
1611 | ||
1612 | /* | |
1613 | * Use an on-stack bio for this, it's safe since we don't | |
1614 | * need to reference it after submit. It's just used as | |
1615 | * the basis for the clone(s). | |
1616 | */ | |
1617 | bio_init(&flush_bio, NULL, 0); | |
1618 | flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; | |
1619 | ci.bio = &flush_bio; | |
b372d360 | 1620 | ci.sector_count = 0; |
14fe594d | 1621 | error = __send_empty_flush(&ci); |
b372d360 | 1622 | /* dec_pending submits any data associated with flush */ |
a4aa5e56 DLM |
1623 | } else if (bio_op(bio) == REQ_OP_ZONE_RESET) { |
1624 | ci.bio = bio; | |
1625 | ci.sector_count = 0; | |
1626 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1627 | } else { |
6a8736d1 | 1628 | ci.bio = bio; |
d87f4c14 | 1629 | ci.sector_count = bio_sectors(bio); |
18a25da8 | 1630 | while (ci.sector_count && !error) { |
14fe594d | 1631 | error = __split_and_process_non_flush(&ci); |
18a25da8 N |
1632 | if (current->bio_list && ci.sector_count && !error) { |
1633 | /* | |
1634 | * Remainder must be passed to generic_make_request() | |
1635 | * so that it gets handled *after* bios already submitted | |
1636 | * have been completely processed. | |
1637 | * We take a clone of the original to store in | |
745dc570 | 1638 | * ci.io->orig_bio to be used by end_io_acct() and |
18a25da8 | 1639 | * for dec_pending to use for completion handling. |
18a25da8 | 1640 | */ |
f21c601a MS |
1641 | struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count, |
1642 | GFP_NOIO, &md->queue->bio_split); | |
745dc570 | 1643 | ci.io->orig_bio = b; |
a1e1cb72 MS |
1644 | |
1645 | /* | |
1646 | * Adjust IO stats for each split, otherwise upon queue | |
1647 | * reentry there will be redundant IO accounting. | |
1648 | * NOTE: this is a stop-gap fix, a proper fix involves | |
1649 | * significant refactoring of DM core's bio splitting | |
1650 | * (by eliminating DM's splitting and just using bio_split) | |
1651 | */ | |
1652 | part_stat_lock(); | |
1653 | __dm_part_stat_sub(&dm_disk(md)->part0, | |
1654 | sectors[op_stat_group(bio_op(bio))], ci.sector_count); | |
1655 | part_stat_unlock(); | |
1656 | ||
18a25da8 | 1657 | bio_chain(b, bio); |
075c18c3 | 1658 | trace_block_split(md->queue, b, bio->bi_iter.bi_sector); |
978e51ba | 1659 | ret = generic_make_request(bio); |
18a25da8 N |
1660 | break; |
1661 | } | |
1662 | } | |
d87f4c14 | 1663 | } |
0ce65797 | 1664 | |
1da177e4 | 1665 | /* drop the extra reference count */ |
54385bf7 | 1666 | dec_pending(ci.io, errno_to_blk_status(error)); |
978e51ba | 1667 | return ret; |
0ce65797 MS |
1668 | } |
1669 | ||
cec47e3d | 1670 | /* |
978e51ba MS |
1671 | * Optimized variant of __split_and_process_bio that leverages the |
1672 | * fact that targets that use it do _not_ have a need to split bios. | |
cec47e3d | 1673 | */ |
568c73a3 MS |
1674 | static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map, |
1675 | struct bio *bio, struct dm_target *ti) | |
978e51ba MS |
1676 | { |
1677 | struct clone_info ci; | |
1678 | blk_qc_t ret = BLK_QC_T_NONE; | |
1679 | int error = 0; | |
1680 | ||
978e51ba MS |
1681 | init_clone_info(&ci, md, map, bio); |
1682 | ||
1683 | if (bio->bi_opf & REQ_PREFLUSH) { | |
dbe3ece1 JA |
1684 | struct bio flush_bio; |
1685 | ||
1686 | /* | |
1687 | * Use an on-stack bio for this, it's safe since we don't | |
1688 | * need to reference it after submit. It's just used as | |
1689 | * the basis for the clone(s). | |
1690 | */ | |
1691 | bio_init(&flush_bio, NULL, 0); | |
1692 | flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; | |
1693 | ci.bio = &flush_bio; | |
978e51ba MS |
1694 | ci.sector_count = 0; |
1695 | error = __send_empty_flush(&ci); | |
1696 | /* dec_pending submits any data associated with flush */ | |
1697 | } else { | |
978e51ba MS |
1698 | struct dm_target_io *tio; |
1699 | ||
978e51ba MS |
1700 | ci.bio = bio; |
1701 | ci.sector_count = bio_sectors(bio); | |
568c73a3 | 1702 | if (__process_abnormal_io(&ci, ti, &error)) |
0519c71e MS |
1703 | goto out; |
1704 | ||
1705 | tio = alloc_tio(&ci, ti, 0, GFP_NOIO); | |
978e51ba MS |
1706 | ret = __clone_and_map_simple_bio(&ci, tio, NULL); |
1707 | } | |
1708 | out: | |
1709 | /* drop the extra reference count */ | |
1710 | dec_pending(ci.io, errno_to_blk_status(error)); | |
1711 | return ret; | |
1712 | } | |
1713 | ||
568c73a3 MS |
1714 | static void dm_queue_split(struct mapped_device *md, struct dm_target *ti, struct bio **bio) |
1715 | { | |
1716 | unsigned len, sector_count; | |
1717 | ||
1718 | sector_count = bio_sectors(*bio); | |
1719 | len = min_t(sector_t, max_io_len((*bio)->bi_iter.bi_sector, ti), sector_count); | |
1720 | ||
1721 | if (sector_count > len) { | |
1722 | struct bio *split = bio_split(*bio, len, GFP_NOIO, &md->queue->bio_split); | |
1723 | ||
1724 | bio_chain(split, *bio); | |
1725 | trace_block_split(md->queue, split, (*bio)->bi_iter.bi_sector); | |
1726 | generic_make_request(*bio); | |
1727 | *bio = split; | |
1728 | } | |
1729 | } | |
1730 | ||
6548c7c5 MS |
1731 | static blk_qc_t dm_process_bio(struct mapped_device *md, |
1732 | struct dm_table *map, struct bio *bio) | |
1733 | { | |
568c73a3 MS |
1734 | blk_qc_t ret = BLK_QC_T_NONE; |
1735 | struct dm_target *ti = md->immutable_target; | |
1736 | ||
1737 | if (unlikely(!map)) { | |
1738 | bio_io_error(bio); | |
1739 | return ret; | |
1740 | } | |
1741 | ||
1742 | if (!ti) { | |
1743 | ti = dm_table_find_target(map, bio->bi_iter.bi_sector); | |
1744 | if (unlikely(!ti || !dm_target_is_valid(ti))) { | |
1745 | bio_io_error(bio); | |
1746 | return ret; | |
1747 | } | |
1748 | } | |
1749 | ||
1750 | /* | |
1751 | * If in ->make_request_fn we need to use blk_queue_split(), otherwise | |
1752 | * queue_limits for abnormal requests (e.g. discard, writesame, etc) | |
1753 | * won't be imposed. | |
1754 | */ | |
1755 | if (current->bio_list) { | |
1756 | if (is_abnormal_io(bio)) | |
1757 | blk_queue_split(md->queue, &bio); | |
1758 | else | |
1759 | dm_queue_split(md, ti, &bio); | |
1760 | } | |
1761 | ||
6548c7c5 | 1762 | if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED) |
568c73a3 | 1763 | return __process_bio(md, map, bio, ti); |
6548c7c5 MS |
1764 | else |
1765 | return __split_and_process_bio(md, map, bio); | |
1766 | } | |
1767 | ||
24113d48 | 1768 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1769 | { |
1770 | struct mapped_device *md = q->queuedata; | |
978e51ba | 1771 | blk_qc_t ret = BLK_QC_T_NONE; |
83d5e5b0 MP |
1772 | int srcu_idx; |
1773 | struct dm_table *map; | |
cec47e3d | 1774 | |
83d5e5b0 | 1775 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1776 | |
6a8736d1 TH |
1777 | /* if we're suspended, we have to queue this io for later */ |
1778 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1779 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1780 | |
1eff9d32 | 1781 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1782 | queue_io(md, bio); |
1783 | else | |
54d9a1b4 | 1784 | bio_io_error(bio); |
978e51ba | 1785 | return ret; |
cec47e3d | 1786 | } |
1da177e4 | 1787 | |
6548c7c5 | 1788 | ret = dm_process_bio(md, map, bio); |
978e51ba | 1789 | |
83d5e5b0 | 1790 | dm_put_live_table(md, srcu_idx); |
978e51ba MS |
1791 | return ret; |
1792 | } | |
1793 | ||
1da177e4 LT |
1794 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1795 | { | |
8a57dfc6 CS |
1796 | int r = bdi_bits; |
1797 | struct mapped_device *md = congested_data; | |
1798 | struct dm_table *map; | |
1da177e4 | 1799 | |
1eb787ec | 1800 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1801 | if (dm_request_based(md)) { |
cec47e3d | 1802 | /* |
e522c039 MS |
1803 | * With request-based DM we only need to check the |
1804 | * top-level queue for congestion. | |
cec47e3d | 1805 | */ |
dc3b17cc | 1806 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1807 | } else { |
1808 | map = dm_get_live_table_fast(md); | |
1809 | if (map) | |
cec47e3d | 1810 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1811 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1812 | } |
1813 | } | |
1814 | ||
1da177e4 LT |
1815 | return r; |
1816 | } | |
1817 | ||
1818 | /*----------------------------------------------------------------- | |
1819 | * An IDR is used to keep track of allocated minor numbers. | |
1820 | *---------------------------------------------------------------*/ | |
2b06cfff | 1821 | static void free_minor(int minor) |
1da177e4 | 1822 | { |
f32c10b0 | 1823 | spin_lock(&_minor_lock); |
1da177e4 | 1824 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1825 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1826 | } |
1827 | ||
1828 | /* | |
1829 | * See if the device with a specific minor # is free. | |
1830 | */ | |
cf13ab8e | 1831 | static int specific_minor(int minor) |
1da177e4 | 1832 | { |
c9d76be6 | 1833 | int r; |
1da177e4 LT |
1834 | |
1835 | if (minor >= (1 << MINORBITS)) | |
1836 | return -EINVAL; | |
1837 | ||
c9d76be6 | 1838 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1839 | spin_lock(&_minor_lock); |
1da177e4 | 1840 | |
c9d76be6 | 1841 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1842 | |
f32c10b0 | 1843 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1844 | idr_preload_end(); |
1845 | if (r < 0) | |
1846 | return r == -ENOSPC ? -EBUSY : r; | |
1847 | return 0; | |
1da177e4 LT |
1848 | } |
1849 | ||
cf13ab8e | 1850 | static int next_free_minor(int *minor) |
1da177e4 | 1851 | { |
c9d76be6 | 1852 | int r; |
62f75c2f | 1853 | |
c9d76be6 | 1854 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1855 | spin_lock(&_minor_lock); |
1da177e4 | 1856 | |
c9d76be6 | 1857 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1858 | |
f32c10b0 | 1859 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1860 | idr_preload_end(); |
1861 | if (r < 0) | |
1862 | return r; | |
1863 | *minor = r; | |
1864 | return 0; | |
1da177e4 LT |
1865 | } |
1866 | ||
83d5cde4 | 1867 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1868 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1869 | |
53d5914f MP |
1870 | static void dm_wq_work(struct work_struct *work); |
1871 | ||
c12c9a3c | 1872 | static void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1873 | { |
bfebd1cd MS |
1874 | /* |
1875 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1876 | */ | |
dc3b17cc | 1877 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf MS |
1878 | } |
1879 | ||
0f20972f MS |
1880 | static void cleanup_mapped_device(struct mapped_device *md) |
1881 | { | |
0f20972f MS |
1882 | if (md->wq) |
1883 | destroy_workqueue(md->wq); | |
6f1c819c KO |
1884 | bioset_exit(&md->bs); |
1885 | bioset_exit(&md->io_bs); | |
0f20972f | 1886 | |
f26c5719 DW |
1887 | if (md->dax_dev) { |
1888 | kill_dax(md->dax_dev); | |
1889 | put_dax(md->dax_dev); | |
1890 | md->dax_dev = NULL; | |
1891 | } | |
1892 | ||
0f20972f MS |
1893 | if (md->disk) { |
1894 | spin_lock(&_minor_lock); | |
1895 | md->disk->private_data = NULL; | |
1896 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1897 | del_gendisk(md->disk); |
1898 | put_disk(md->disk); | |
1899 | } | |
1900 | ||
1901 | if (md->queue) | |
1902 | blk_cleanup_queue(md->queue); | |
1903 | ||
d09960b0 TE |
1904 | cleanup_srcu_struct(&md->io_barrier); |
1905 | ||
0f20972f MS |
1906 | if (md->bdev) { |
1907 | bdput(md->bdev); | |
1908 | md->bdev = NULL; | |
1909 | } | |
4cc96131 | 1910 | |
d5ffebdd MS |
1911 | mutex_destroy(&md->suspend_lock); |
1912 | mutex_destroy(&md->type_lock); | |
1913 | mutex_destroy(&md->table_devices_lock); | |
1914 | ||
4cc96131 | 1915 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
1916 | } |
1917 | ||
1da177e4 LT |
1918 | /* |
1919 | * Allocate and initialise a blank device with a given minor. | |
1920 | */ | |
2b06cfff | 1921 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1922 | { |
115485e8 | 1923 | int r, numa_node_id = dm_get_numa_node(); |
976431b0 | 1924 | struct dax_device *dax_dev = NULL; |
115485e8 | 1925 | struct mapped_device *md; |
ba61fdd1 | 1926 | void *old_md; |
1da177e4 | 1927 | |
856eb091 | 1928 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1929 | if (!md) { |
1930 | DMWARN("unable to allocate device, out of memory."); | |
1931 | return NULL; | |
1932 | } | |
1933 | ||
10da4f79 | 1934 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1935 | goto bad_module_get; |
10da4f79 | 1936 | |
1da177e4 | 1937 | /* get a minor number for the dev */ |
2b06cfff | 1938 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1939 | r = next_free_minor(&minor); |
2b06cfff | 1940 | else |
cf13ab8e | 1941 | r = specific_minor(minor); |
1da177e4 | 1942 | if (r < 0) |
6ed7ade8 | 1943 | goto bad_minor; |
1da177e4 | 1944 | |
83d5e5b0 MP |
1945 | r = init_srcu_struct(&md->io_barrier); |
1946 | if (r < 0) | |
1947 | goto bad_io_barrier; | |
1948 | ||
115485e8 | 1949 | md->numa_node_id = numa_node_id; |
591ddcfc | 1950 | md->init_tio_pdu = false; |
a5664dad | 1951 | md->type = DM_TYPE_NONE; |
e61290a4 | 1952 | mutex_init(&md->suspend_lock); |
a5664dad | 1953 | mutex_init(&md->type_lock); |
86f1152b | 1954 | mutex_init(&md->table_devices_lock); |
022c2611 | 1955 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1956 | atomic_set(&md->holders, 1); |
5c6bd75d | 1957 | atomic_set(&md->open_count, 0); |
1da177e4 | 1958 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1959 | atomic_set(&md->uevent_seq, 0); |
1960 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1961 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1962 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1963 | |
6d469642 | 1964 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1965 | if (!md->queue) |
0f20972f | 1966 | goto bad; |
c12c9a3c MS |
1967 | md->queue->queuedata = md; |
1968 | md->queue->backing_dev_info->congested_data = md; | |
1da177e4 | 1969 | |
c12c9a3c | 1970 | md->disk = alloc_disk_node(1, md->numa_node_id); |
1da177e4 | 1971 | if (!md->disk) |
0f20972f | 1972 | goto bad; |
1da177e4 | 1973 | |
f0b04115 | 1974 | init_waitqueue_head(&md->wait); |
53d5914f | 1975 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1976 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1977 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 1978 | |
1da177e4 LT |
1979 | md->disk->major = _major; |
1980 | md->disk->first_minor = minor; | |
1981 | md->disk->fops = &dm_blk_dops; | |
1982 | md->disk->queue = md->queue; | |
1983 | md->disk->private_data = md; | |
1984 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 | 1985 | |
976431b0 DW |
1986 | if (IS_ENABLED(CONFIG_DAX_DRIVER)) { |
1987 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); | |
1988 | if (!dax_dev) | |
1989 | goto bad; | |
1990 | } | |
f26c5719 DW |
1991 | md->dax_dev = dax_dev; |
1992 | ||
c100ec49 | 1993 | add_disk_no_queue_reg(md->disk); |
7e51f257 | 1994 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1995 | |
670368a8 | 1996 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1997 | if (!md->wq) |
0f20972f | 1998 | goto bad; |
304f3f6a | 1999 | |
32a926da MP |
2000 | md->bdev = bdget_disk(md->disk, 0); |
2001 | if (!md->bdev) | |
0f20972f | 2002 | goto bad; |
32a926da | 2003 | |
fd2ed4d2 MP |
2004 | dm_stats_init(&md->stats); |
2005 | ||
ba61fdd1 | 2006 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2007 | spin_lock(&_minor_lock); |
ba61fdd1 | 2008 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2009 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2010 | |
2011 | BUG_ON(old_md != MINOR_ALLOCED); | |
2012 | ||
1da177e4 LT |
2013 | return md; |
2014 | ||
0f20972f MS |
2015 | bad: |
2016 | cleanup_mapped_device(md); | |
83d5e5b0 | 2017 | bad_io_barrier: |
1da177e4 | 2018 | free_minor(minor); |
6ed7ade8 | 2019 | bad_minor: |
10da4f79 | 2020 | module_put(THIS_MODULE); |
6ed7ade8 | 2021 | bad_module_get: |
856eb091 | 2022 | kvfree(md); |
1da177e4 LT |
2023 | return NULL; |
2024 | } | |
2025 | ||
ae9da83f JN |
2026 | static void unlock_fs(struct mapped_device *md); |
2027 | ||
1da177e4 LT |
2028 | static void free_dev(struct mapped_device *md) |
2029 | { | |
f331c029 | 2030 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2031 | |
32a926da | 2032 | unlock_fs(md); |
2eb6e1e3 | 2033 | |
0f20972f | 2034 | cleanup_mapped_device(md); |
63a4f065 | 2035 | |
86f1152b | 2036 | free_table_devices(&md->table_devices); |
63a4f065 | 2037 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2038 | free_minor(minor); |
2039 | ||
10da4f79 | 2040 | module_put(THIS_MODULE); |
856eb091 | 2041 | kvfree(md); |
1da177e4 LT |
2042 | } |
2043 | ||
2a2a4c51 | 2044 | static int __bind_mempools(struct mapped_device *md, struct dm_table *t) |
e6ee8c0b | 2045 | { |
c0820cf5 | 2046 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
2a2a4c51 | 2047 | int ret = 0; |
e6ee8c0b | 2048 | |
0776aa0e | 2049 | if (dm_table_bio_based(t)) { |
64f52b0e MS |
2050 | /* |
2051 | * The md may already have mempools that need changing. | |
2052 | * If so, reload bioset because front_pad may have changed | |
2053 | * because a different table was loaded. | |
2054 | */ | |
6f1c819c KO |
2055 | bioset_exit(&md->bs); |
2056 | bioset_exit(&md->io_bs); | |
0776aa0e | 2057 | |
6f1c819c | 2058 | } else if (bioset_initialized(&md->bs)) { |
4e6e36c3 MS |
2059 | /* |
2060 | * There's no need to reload with request-based dm | |
2061 | * because the size of front_pad doesn't change. | |
2062 | * Note for future: If you are to reload bioset, | |
2063 | * prep-ed requests in the queue may refer | |
2064 | * to bio from the old bioset, so you must walk | |
2065 | * through the queue to unprep. | |
2066 | */ | |
2067 | goto out; | |
c0820cf5 | 2068 | } |
e6ee8c0b | 2069 | |
6f1c819c KO |
2070 | BUG_ON(!p || |
2071 | bioset_initialized(&md->bs) || | |
2072 | bioset_initialized(&md->io_bs)); | |
cbc4e3c1 | 2073 | |
2a2a4c51 JA |
2074 | ret = bioset_init_from_src(&md->bs, &p->bs); |
2075 | if (ret) | |
2076 | goto out; | |
2077 | ret = bioset_init_from_src(&md->io_bs, &p->io_bs); | |
2078 | if (ret) | |
2079 | bioset_exit(&md->bs); | |
e6ee8c0b | 2080 | out: |
02233342 | 2081 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b | 2082 | dm_table_free_md_mempools(t); |
2a2a4c51 | 2083 | return ret; |
e6ee8c0b KU |
2084 | } |
2085 | ||
1da177e4 LT |
2086 | /* |
2087 | * Bind a table to the device. | |
2088 | */ | |
2089 | static void event_callback(void *context) | |
2090 | { | |
7a8c3d3b MA |
2091 | unsigned long flags; |
2092 | LIST_HEAD(uevents); | |
1da177e4 LT |
2093 | struct mapped_device *md = (struct mapped_device *) context; |
2094 | ||
7a8c3d3b MA |
2095 | spin_lock_irqsave(&md->uevent_lock, flags); |
2096 | list_splice_init(&md->uevent_list, &uevents); | |
2097 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2098 | ||
ed9e1982 | 2099 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2100 | |
1da177e4 LT |
2101 | atomic_inc(&md->event_nr); |
2102 | wake_up(&md->eventq); | |
62e08243 | 2103 | dm_issue_global_event(); |
1da177e4 LT |
2104 | } |
2105 | ||
c217649b MS |
2106 | /* |
2107 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2108 | */ | |
4e90188b | 2109 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2110 | { |
1ea0654e BVA |
2111 | lockdep_assert_held(&md->suspend_lock); |
2112 | ||
4e90188b | 2113 | set_capacity(md->disk, size); |
1da177e4 | 2114 | |
db8fef4f | 2115 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2116 | } |
2117 | ||
042d2a9b AK |
2118 | /* |
2119 | * Returns old map, which caller must destroy. | |
2120 | */ | |
2121 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2122 | struct queue_limits *limits) | |
1da177e4 | 2123 | { |
042d2a9b | 2124 | struct dm_table *old_map; |
165125e1 | 2125 | struct request_queue *q = md->queue; |
978e51ba | 2126 | bool request_based = dm_table_request_based(t); |
1da177e4 | 2127 | sector_t size; |
2a2a4c51 | 2128 | int ret; |
1da177e4 | 2129 | |
5a8f1f80 BVA |
2130 | lockdep_assert_held(&md->suspend_lock); |
2131 | ||
1da177e4 | 2132 | size = dm_table_get_size(t); |
3ac51e74 DW |
2133 | |
2134 | /* | |
2135 | * Wipe any geometry if the size of the table changed. | |
2136 | */ | |
fd2ed4d2 | 2137 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2138 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2139 | ||
32a926da | 2140 | __set_size(md, size); |
d5816876 | 2141 | |
2ca3310e AK |
2142 | dm_table_event_callback(t, event_callback, md); |
2143 | ||
e6ee8c0b KU |
2144 | /* |
2145 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2146 | * for request-based during suspension. So stop it to prevent | |
2147 | * I/O mapping before resume. | |
2148 | * This must be done before setting the queue restrictions, | |
2149 | * because request-based dm may be run just after the setting. | |
2150 | */ | |
978e51ba | 2151 | if (request_based) |
eca7ee6d | 2152 | dm_stop_queue(q); |
978e51ba MS |
2153 | |
2154 | if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) { | |
16f12266 | 2155 | /* |
978e51ba MS |
2156 | * Leverage the fact that request-based DM targets and |
2157 | * NVMe bio based targets are immutable singletons | |
2158 | * - used to optimize both dm_request_fn and dm_mq_queue_rq; | |
2159 | * and __process_bio. | |
16f12266 MS |
2160 | */ |
2161 | md->immutable_target = dm_table_get_immutable_target(t); | |
2162 | } | |
e6ee8c0b | 2163 | |
2a2a4c51 JA |
2164 | ret = __bind_mempools(md, t); |
2165 | if (ret) { | |
2166 | old_map = ERR_PTR(ret); | |
2167 | goto out; | |
2168 | } | |
e6ee8c0b | 2169 | |
a12f5d48 | 2170 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2171 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2172 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2173 | ||
754c5fc7 | 2174 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2175 | if (old_map) |
2176 | dm_sync_table(md); | |
1da177e4 | 2177 | |
2a2a4c51 | 2178 | out: |
042d2a9b | 2179 | return old_map; |
1da177e4 LT |
2180 | } |
2181 | ||
a7940155 AK |
2182 | /* |
2183 | * Returns unbound table for the caller to free. | |
2184 | */ | |
2185 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2186 | { |
a12f5d48 | 2187 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2188 | |
2189 | if (!map) | |
a7940155 | 2190 | return NULL; |
1da177e4 LT |
2191 | |
2192 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2193 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2194 | dm_sync_table(md); |
a7940155 AK |
2195 | |
2196 | return map; | |
1da177e4 LT |
2197 | } |
2198 | ||
2199 | /* | |
2200 | * Constructor for a new device. | |
2201 | */ | |
2b06cfff | 2202 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 | 2203 | { |
c12c9a3c | 2204 | int r; |
1da177e4 LT |
2205 | struct mapped_device *md; |
2206 | ||
2b06cfff | 2207 | md = alloc_dev(minor); |
1da177e4 LT |
2208 | if (!md) |
2209 | return -ENXIO; | |
2210 | ||
c12c9a3c MS |
2211 | r = dm_sysfs_init(md); |
2212 | if (r) { | |
2213 | free_dev(md); | |
2214 | return r; | |
2215 | } | |
784aae73 | 2216 | |
1da177e4 LT |
2217 | *result = md; |
2218 | return 0; | |
2219 | } | |
2220 | ||
a5664dad MS |
2221 | /* |
2222 | * Functions to manage md->type. | |
2223 | * All are required to hold md->type_lock. | |
2224 | */ | |
2225 | void dm_lock_md_type(struct mapped_device *md) | |
2226 | { | |
2227 | mutex_lock(&md->type_lock); | |
2228 | } | |
2229 | ||
2230 | void dm_unlock_md_type(struct mapped_device *md) | |
2231 | { | |
2232 | mutex_unlock(&md->type_lock); | |
2233 | } | |
2234 | ||
7e0d574f | 2235 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2236 | { |
00c4fc3b | 2237 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2238 | md->type = type; |
2239 | } | |
2240 | ||
7e0d574f | 2241 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2242 | { |
2243 | return md->type; | |
2244 | } | |
2245 | ||
36a0456f AK |
2246 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2247 | { | |
2248 | return md->immutable_target_type; | |
2249 | } | |
2250 | ||
f84cb8a4 MS |
2251 | /* |
2252 | * The queue_limits are only valid as long as you have a reference | |
2253 | * count on 'md'. | |
2254 | */ | |
2255 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2256 | { | |
2257 | BUG_ON(!atomic_read(&md->holders)); | |
2258 | return &md->queue->limits; | |
2259 | } | |
2260 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2261 | ||
4a0b4ddf MS |
2262 | /* |
2263 | * Setup the DM device's queue based on md's type | |
2264 | */ | |
591ddcfc | 2265 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2266 | { |
bfebd1cd | 2267 | int r; |
c100ec49 | 2268 | struct queue_limits limits; |
7e0d574f | 2269 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2270 | |
545ed20e | 2271 | switch (type) { |
bfebd1cd | 2272 | case DM_TYPE_REQUEST_BASED: |
e83068a5 | 2273 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2274 | if (r) { |
eca7ee6d | 2275 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2276 | return r; |
2277 | } | |
2278 | break; | |
2279 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2280 | case DM_TYPE_DAX_BIO_BASED: |
978e51ba MS |
2281 | case DM_TYPE_NVME_BIO_BASED: |
2282 | dm_init_normal_md_queue(md); | |
24113d48 | 2283 | blk_queue_make_request(md->queue, dm_make_request); |
bfebd1cd | 2284 | break; |
7e0d574f BVA |
2285 | case DM_TYPE_NONE: |
2286 | WARN_ON_ONCE(true); | |
2287 | break; | |
4a0b4ddf MS |
2288 | } |
2289 | ||
c100ec49 MS |
2290 | r = dm_calculate_queue_limits(t, &limits); |
2291 | if (r) { | |
2292 | DMERR("Cannot calculate initial queue limits"); | |
2293 | return r; | |
2294 | } | |
2295 | dm_table_set_restrictions(t, md->queue, &limits); | |
2296 | blk_register_queue(md->disk); | |
2297 | ||
4a0b4ddf MS |
2298 | return 0; |
2299 | } | |
2300 | ||
2bec1f4a | 2301 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2302 | { |
2303 | struct mapped_device *md; | |
1da177e4 LT |
2304 | unsigned minor = MINOR(dev); |
2305 | ||
2306 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2307 | return NULL; | |
2308 | ||
f32c10b0 | 2309 | spin_lock(&_minor_lock); |
1da177e4 LT |
2310 | |
2311 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2312 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2313 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2314 | md = NULL; | |
2315 | goto out; | |
fba9f90e | 2316 | } |
49de5769 | 2317 | dm_get(md); |
fba9f90e | 2318 | out: |
f32c10b0 | 2319 | spin_unlock(&_minor_lock); |
1da177e4 | 2320 | |
637842cf DT |
2321 | return md; |
2322 | } | |
3cf2e4ba | 2323 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2324 | |
9ade92a9 | 2325 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2326 | { |
9ade92a9 | 2327 | return md->interface_ptr; |
1da177e4 LT |
2328 | } |
2329 | ||
2330 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2331 | { | |
2332 | md->interface_ptr = ptr; | |
2333 | } | |
2334 | ||
2335 | void dm_get(struct mapped_device *md) | |
2336 | { | |
2337 | atomic_inc(&md->holders); | |
3f77316d | 2338 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2339 | } |
2340 | ||
09ee96b2 MP |
2341 | int dm_hold(struct mapped_device *md) |
2342 | { | |
2343 | spin_lock(&_minor_lock); | |
2344 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2345 | spin_unlock(&_minor_lock); | |
2346 | return -EBUSY; | |
2347 | } | |
2348 | dm_get(md); | |
2349 | spin_unlock(&_minor_lock); | |
2350 | return 0; | |
2351 | } | |
2352 | EXPORT_SYMBOL_GPL(dm_hold); | |
2353 | ||
72d94861 AK |
2354 | const char *dm_device_name(struct mapped_device *md) |
2355 | { | |
2356 | return md->name; | |
2357 | } | |
2358 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2359 | ||
3f77316d | 2360 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2361 | { |
1134e5ae | 2362 | struct dm_table *map; |
83d5e5b0 | 2363 | int srcu_idx; |
1da177e4 | 2364 | |
3f77316d | 2365 | might_sleep(); |
fba9f90e | 2366 | |
63a4f065 | 2367 | spin_lock(&_minor_lock); |
3f77316d KU |
2368 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2369 | set_bit(DMF_FREEING, &md->flags); | |
2370 | spin_unlock(&_minor_lock); | |
3b785fbc | 2371 | |
c12c9a3c | 2372 | blk_set_queue_dying(md->queue); |
3f77316d | 2373 | |
ab7c7bb6 MP |
2374 | /* |
2375 | * Take suspend_lock so that presuspend and postsuspend methods | |
2376 | * do not race with internal suspend. | |
2377 | */ | |
2378 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2379 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2380 | if (!dm_suspended_md(md)) { |
2381 | dm_table_presuspend_targets(map); | |
2382 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2383 | } |
83d5e5b0 MP |
2384 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2385 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2386 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2387 | |
3f77316d KU |
2388 | /* |
2389 | * Rare, but there may be I/O requests still going to complete, | |
2390 | * for example. Wait for all references to disappear. | |
2391 | * No one should increment the reference count of the mapped_device, | |
2392 | * after the mapped_device state becomes DMF_FREEING. | |
2393 | */ | |
2394 | if (wait) | |
2395 | while (atomic_read(&md->holders)) | |
2396 | msleep(1); | |
2397 | else if (atomic_read(&md->holders)) | |
2398 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2399 | dm_device_name(md), atomic_read(&md->holders)); | |
2400 | ||
2401 | dm_sysfs_exit(md); | |
3f77316d KU |
2402 | dm_table_destroy(__unbind(md)); |
2403 | free_dev(md); | |
2404 | } | |
2405 | ||
2406 | void dm_destroy(struct mapped_device *md) | |
2407 | { | |
2408 | __dm_destroy(md, true); | |
2409 | } | |
2410 | ||
2411 | void dm_destroy_immediate(struct mapped_device *md) | |
2412 | { | |
2413 | __dm_destroy(md, false); | |
2414 | } | |
2415 | ||
2416 | void dm_put(struct mapped_device *md) | |
2417 | { | |
2418 | atomic_dec(&md->holders); | |
1da177e4 | 2419 | } |
79eb885c | 2420 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2421 | |
b48633f8 | 2422 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2423 | { |
2424 | int r = 0; | |
9f4c3f87 | 2425 | DEFINE_WAIT(wait); |
46125c1c MB |
2426 | |
2427 | while (1) { | |
9f4c3f87 | 2428 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2429 | |
b4324fee | 2430 | if (!md_in_flight(md)) |
46125c1c MB |
2431 | break; |
2432 | ||
e3fabdfd | 2433 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2434 | r = -EINTR; |
2435 | break; | |
2436 | } | |
2437 | ||
2438 | io_schedule(); | |
2439 | } | |
9f4c3f87 | 2440 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2441 | |
46125c1c MB |
2442 | return r; |
2443 | } | |
2444 | ||
1da177e4 LT |
2445 | /* |
2446 | * Process the deferred bios | |
2447 | */ | |
ef208587 | 2448 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2449 | { |
ef208587 MP |
2450 | struct mapped_device *md = container_of(work, struct mapped_device, |
2451 | work); | |
6d6f10df | 2452 | struct bio *c; |
83d5e5b0 MP |
2453 | int srcu_idx; |
2454 | struct dm_table *map; | |
1da177e4 | 2455 | |
83d5e5b0 | 2456 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2457 | |
3b00b203 | 2458 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2459 | spin_lock_irq(&md->deferred_lock); |
2460 | c = bio_list_pop(&md->deferred); | |
2461 | spin_unlock_irq(&md->deferred_lock); | |
2462 | ||
6a8736d1 | 2463 | if (!c) |
df12ee99 | 2464 | break; |
022c2611 | 2465 | |
e6ee8c0b | 2466 | if (dm_request_based(md)) |
6548c7c5 | 2467 | (void) generic_make_request(c); |
6a8736d1 | 2468 | else |
6548c7c5 | 2469 | (void) dm_process_bio(md, map, c); |
022c2611 | 2470 | } |
73d410c0 | 2471 | |
83d5e5b0 | 2472 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2473 | } |
2474 | ||
9a1fb464 | 2475 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2476 | { |
3b00b203 | 2477 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2478 | smp_mb__after_atomic(); |
53d5914f | 2479 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2480 | } |
2481 | ||
1da177e4 | 2482 | /* |
042d2a9b | 2483 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2484 | */ |
042d2a9b | 2485 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2486 | { |
87eb5b21 | 2487 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2488 | struct queue_limits limits; |
042d2a9b | 2489 | int r; |
1da177e4 | 2490 | |
e61290a4 | 2491 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2492 | |
2493 | /* device must be suspended */ | |
4f186f8b | 2494 | if (!dm_suspended_md(md)) |
93c534ae | 2495 | goto out; |
1da177e4 | 2496 | |
3ae70656 MS |
2497 | /* |
2498 | * If the new table has no data devices, retain the existing limits. | |
2499 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2500 | * then new I/O is queued based on these limits, and then some paths | |
2501 | * reappear. | |
2502 | */ | |
2503 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2504 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2505 | if (live_map) |
2506 | limits = md->queue->limits; | |
83d5e5b0 | 2507 | dm_put_live_table_fast(md); |
3ae70656 MS |
2508 | } |
2509 | ||
87eb5b21 MC |
2510 | if (!live_map) { |
2511 | r = dm_calculate_queue_limits(table, &limits); | |
2512 | if (r) { | |
2513 | map = ERR_PTR(r); | |
2514 | goto out; | |
2515 | } | |
042d2a9b | 2516 | } |
754c5fc7 | 2517 | |
042d2a9b | 2518 | map = __bind(md, table, &limits); |
62e08243 | 2519 | dm_issue_global_event(); |
1da177e4 | 2520 | |
93c534ae | 2521 | out: |
e61290a4 | 2522 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2523 | return map; |
1da177e4 LT |
2524 | } |
2525 | ||
2526 | /* | |
2527 | * Functions to lock and unlock any filesystem running on the | |
2528 | * device. | |
2529 | */ | |
2ca3310e | 2530 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2531 | { |
e39e2e95 | 2532 | int r; |
1da177e4 LT |
2533 | |
2534 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2535 | |
db8fef4f | 2536 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2537 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2538 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2539 | md->frozen_sb = NULL; |
2540 | return r; | |
dfbe03f6 AK |
2541 | } |
2542 | ||
aa8d7c2f AK |
2543 | set_bit(DMF_FROZEN, &md->flags); |
2544 | ||
1da177e4 LT |
2545 | return 0; |
2546 | } | |
2547 | ||
2ca3310e | 2548 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2549 | { |
aa8d7c2f AK |
2550 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2551 | return; | |
2552 | ||
db8fef4f | 2553 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2554 | md->frozen_sb = NULL; |
aa8d7c2f | 2555 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2556 | } |
2557 | ||
2558 | /* | |
b48633f8 BVA |
2559 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2560 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2561 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2562 | * | |
ffcc3936 MS |
2563 | * If __dm_suspend returns 0, the device is completely quiescent |
2564 | * now. There is no request-processing activity. All new requests | |
2565 | * are being added to md->deferred list. | |
cec47e3d | 2566 | */ |
ffcc3936 | 2567 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2568 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2569 | int dmf_suspended_flag) |
1da177e4 | 2570 | { |
ffcc3936 MS |
2571 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2572 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2573 | int r; | |
1da177e4 | 2574 | |
5a8f1f80 BVA |
2575 | lockdep_assert_held(&md->suspend_lock); |
2576 | ||
2e93ccc1 KU |
2577 | /* |
2578 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2579 | * This flag is cleared before dm_suspend returns. | |
2580 | */ | |
2581 | if (noflush) | |
2582 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 BVA |
2583 | else |
2584 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); | |
2e93ccc1 | 2585 | |
d67ee213 MS |
2586 | /* |
2587 | * This gets reverted if there's an error later and the targets | |
2588 | * provide the .presuspend_undo hook. | |
2589 | */ | |
cf222b37 AK |
2590 | dm_table_presuspend_targets(map); |
2591 | ||
32a926da | 2592 | /* |
9f518b27 KU |
2593 | * Flush I/O to the device. |
2594 | * Any I/O submitted after lock_fs() may not be flushed. | |
2595 | * noflush takes precedence over do_lockfs. | |
2596 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2597 | */ |
2598 | if (!noflush && do_lockfs) { | |
2599 | r = lock_fs(md); | |
d67ee213 MS |
2600 | if (r) { |
2601 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2602 | return r; |
d67ee213 | 2603 | } |
aa8d7c2f | 2604 | } |
1da177e4 LT |
2605 | |
2606 | /* | |
3b00b203 MP |
2607 | * Here we must make sure that no processes are submitting requests |
2608 | * to target drivers i.e. no one may be executing | |
2609 | * __split_and_process_bio. This is called from dm_request and | |
2610 | * dm_wq_work. | |
2611 | * | |
2612 | * To get all processes out of __split_and_process_bio in dm_request, | |
2613 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2614 | * __split_and_process_bio from dm_request and quiesce the thread |
2615 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2616 | * flush_workqueue(md->wq). | |
1da177e4 | 2617 | */ |
1eb787ec | 2618 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2619 | if (map) |
2620 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2621 | |
d0bcb878 | 2622 | /* |
29e4013d TH |
2623 | * Stop md->queue before flushing md->wq in case request-based |
2624 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2625 | */ |
6a23e05c | 2626 | if (dm_request_based(md)) |
eca7ee6d | 2627 | dm_stop_queue(md->queue); |
cec47e3d | 2628 | |
d0bcb878 KU |
2629 | flush_workqueue(md->wq); |
2630 | ||
1da177e4 | 2631 | /* |
3b00b203 MP |
2632 | * At this point no more requests are entering target request routines. |
2633 | * We call dm_wait_for_completion to wait for all existing requests | |
2634 | * to finish. | |
1da177e4 | 2635 | */ |
b48633f8 | 2636 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2637 | if (!r) |
2638 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2639 | |
6d6f10df | 2640 | if (noflush) |
022c2611 | 2641 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2642 | if (map) |
2643 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2644 | |
1da177e4 | 2645 | /* were we interrupted ? */ |
46125c1c | 2646 | if (r < 0) { |
9a1fb464 | 2647 | dm_queue_flush(md); |
73d410c0 | 2648 | |
cec47e3d | 2649 | if (dm_request_based(md)) |
eca7ee6d | 2650 | dm_start_queue(md->queue); |
cec47e3d | 2651 | |
2ca3310e | 2652 | unlock_fs(md); |
d67ee213 | 2653 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2654 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2655 | } |
1da177e4 | 2656 | |
ffcc3936 MS |
2657 | return r; |
2658 | } | |
2659 | ||
2660 | /* | |
2661 | * We need to be able to change a mapping table under a mounted | |
2662 | * filesystem. For example we might want to move some data in | |
2663 | * the background. Before the table can be swapped with | |
2664 | * dm_bind_table, dm_suspend must be called to flush any in | |
2665 | * flight bios and ensure that any further io gets deferred. | |
2666 | */ | |
2667 | /* | |
2668 | * Suspend mechanism in request-based dm. | |
2669 | * | |
2670 | * 1. Flush all I/Os by lock_fs() if needed. | |
2671 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2672 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2673 | * | |
2674 | * To abort suspend, start the request_queue. | |
2675 | */ | |
2676 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2677 | { | |
2678 | struct dm_table *map = NULL; | |
2679 | int r = 0; | |
2680 | ||
2681 | retry: | |
2682 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2683 | ||
2684 | if (dm_suspended_md(md)) { | |
2685 | r = -EINVAL; | |
2686 | goto out_unlock; | |
2687 | } | |
2688 | ||
2689 | if (dm_suspended_internally_md(md)) { | |
2690 | /* already internally suspended, wait for internal resume */ | |
2691 | mutex_unlock(&md->suspend_lock); | |
2692 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2693 | if (r) | |
2694 | return r; | |
2695 | goto retry; | |
2696 | } | |
2697 | ||
a12f5d48 | 2698 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2699 | |
eaf9a736 | 2700 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2701 | if (r) |
2702 | goto out_unlock; | |
3b00b203 | 2703 | |
4d4471cb KU |
2704 | dm_table_postsuspend_targets(map); |
2705 | ||
d287483d | 2706 | out_unlock: |
e61290a4 | 2707 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2708 | return r; |
1da177e4 LT |
2709 | } |
2710 | ||
ffcc3936 MS |
2711 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2712 | { | |
2713 | if (map) { | |
2714 | int r = dm_table_resume_targets(map); | |
2715 | if (r) | |
2716 | return r; | |
2717 | } | |
2718 | ||
2719 | dm_queue_flush(md); | |
2720 | ||
2721 | /* | |
2722 | * Flushing deferred I/Os must be done after targets are resumed | |
2723 | * so that mapping of targets can work correctly. | |
2724 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2725 | */ | |
2726 | if (dm_request_based(md)) | |
eca7ee6d | 2727 | dm_start_queue(md->queue); |
ffcc3936 MS |
2728 | |
2729 | unlock_fs(md); | |
2730 | ||
2731 | return 0; | |
2732 | } | |
2733 | ||
1da177e4 LT |
2734 | int dm_resume(struct mapped_device *md) |
2735 | { | |
8dc23658 | 2736 | int r; |
cf222b37 | 2737 | struct dm_table *map = NULL; |
1da177e4 | 2738 | |
ffcc3936 | 2739 | retry: |
8dc23658 | 2740 | r = -EINVAL; |
ffcc3936 MS |
2741 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2742 | ||
4f186f8b | 2743 | if (!dm_suspended_md(md)) |
cf222b37 | 2744 | goto out; |
cf222b37 | 2745 | |
ffcc3936 MS |
2746 | if (dm_suspended_internally_md(md)) { |
2747 | /* already internally suspended, wait for internal resume */ | |
2748 | mutex_unlock(&md->suspend_lock); | |
2749 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2750 | if (r) | |
2751 | return r; | |
2752 | goto retry; | |
2753 | } | |
2754 | ||
a12f5d48 | 2755 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2756 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2757 | goto out; |
1da177e4 | 2758 | |
ffcc3936 | 2759 | r = __dm_resume(md, map); |
8757b776 MB |
2760 | if (r) |
2761 | goto out; | |
2ca3310e | 2762 | |
2ca3310e | 2763 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2764 | out: |
e61290a4 | 2765 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2766 | |
cf222b37 | 2767 | return r; |
1da177e4 LT |
2768 | } |
2769 | ||
fd2ed4d2 MP |
2770 | /* |
2771 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2772 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2773 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2774 | */ |
2775 | ||
ffcc3936 | 2776 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2777 | { |
ffcc3936 MS |
2778 | struct dm_table *map = NULL; |
2779 | ||
1ea0654e BVA |
2780 | lockdep_assert_held(&md->suspend_lock); |
2781 | ||
96b26c8c | 2782 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2783 | return; /* nested internal suspend */ |
2784 | ||
2785 | if (dm_suspended_md(md)) { | |
2786 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2787 | return; /* nest suspend */ | |
2788 | } | |
2789 | ||
a12f5d48 | 2790 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2791 | |
2792 | /* | |
2793 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2794 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2795 | * would require changing .presuspend to return an error -- avoid this | |
2796 | * until there is a need for more elaborate variants of internal suspend. | |
2797 | */ | |
eaf9a736 MS |
2798 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2799 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2800 | |
2801 | dm_table_postsuspend_targets(map); | |
2802 | } | |
2803 | ||
2804 | static void __dm_internal_resume(struct mapped_device *md) | |
2805 | { | |
96b26c8c MP |
2806 | BUG_ON(!md->internal_suspend_count); |
2807 | ||
2808 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2809 | return; /* resume from nested internal suspend */ |
2810 | ||
fd2ed4d2 | 2811 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2812 | goto done; /* resume from nested suspend */ |
2813 | ||
2814 | /* | |
2815 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2816 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2817 | */ | |
2818 | (void) __dm_resume(md, NULL); | |
2819 | ||
2820 | done: | |
2821 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2822 | smp_mb__after_atomic(); | |
2823 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2824 | } | |
2825 | ||
2826 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2827 | { | |
2828 | mutex_lock(&md->suspend_lock); | |
2829 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2830 | mutex_unlock(&md->suspend_lock); | |
2831 | } | |
2832 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2833 | ||
2834 | void dm_internal_resume(struct mapped_device *md) | |
2835 | { | |
2836 | mutex_lock(&md->suspend_lock); | |
2837 | __dm_internal_resume(md); | |
2838 | mutex_unlock(&md->suspend_lock); | |
2839 | } | |
2840 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2841 | ||
2842 | /* | |
2843 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2844 | * which prevents interaction with userspace-driven suspend. | |
2845 | */ | |
2846 | ||
2847 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2848 | { | |
2849 | mutex_lock(&md->suspend_lock); | |
2850 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2851 | return; |
2852 | ||
2853 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2854 | synchronize_srcu(&md->io_barrier); | |
2855 | flush_workqueue(md->wq); | |
2856 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2857 | } | |
b735fede | 2858 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2859 | |
ffcc3936 | 2860 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2861 | { |
ffcc3936 | 2862 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2863 | goto done; |
2864 | ||
2865 | dm_queue_flush(md); | |
2866 | ||
2867 | done: | |
2868 | mutex_unlock(&md->suspend_lock); | |
2869 | } | |
b735fede | 2870 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2871 | |
1da177e4 LT |
2872 | /*----------------------------------------------------------------- |
2873 | * Event notification. | |
2874 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2875 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2876 | unsigned cookie) |
69267a30 | 2877 | { |
60935eb2 MB |
2878 | char udev_cookie[DM_COOKIE_LENGTH]; |
2879 | char *envp[] = { udev_cookie, NULL }; | |
2880 | ||
2881 | if (!cookie) | |
3abf85b5 | 2882 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2883 | else { |
2884 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2885 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2886 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2887 | action, envp); | |
60935eb2 | 2888 | } |
69267a30 AK |
2889 | } |
2890 | ||
7a8c3d3b MA |
2891 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2892 | { | |
2893 | return atomic_add_return(1, &md->uevent_seq); | |
2894 | } | |
2895 | ||
1da177e4 LT |
2896 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2897 | { | |
2898 | return atomic_read(&md->event_nr); | |
2899 | } | |
2900 | ||
2901 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2902 | { | |
2903 | return wait_event_interruptible(md->eventq, | |
2904 | (event_nr != atomic_read(&md->event_nr))); | |
2905 | } | |
2906 | ||
7a8c3d3b MA |
2907 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2908 | { | |
2909 | unsigned long flags; | |
2910 | ||
2911 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2912 | list_add(elist, &md->uevent_list); | |
2913 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2914 | } | |
2915 | ||
1da177e4 LT |
2916 | /* |
2917 | * The gendisk is only valid as long as you have a reference | |
2918 | * count on 'md'. | |
2919 | */ | |
2920 | struct gendisk *dm_disk(struct mapped_device *md) | |
2921 | { | |
2922 | return md->disk; | |
2923 | } | |
65ff5b7d | 2924 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2925 | |
784aae73 MB |
2926 | struct kobject *dm_kobject(struct mapped_device *md) |
2927 | { | |
2995fa78 | 2928 | return &md->kobj_holder.kobj; |
784aae73 MB |
2929 | } |
2930 | ||
784aae73 MB |
2931 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2932 | { | |
2933 | struct mapped_device *md; | |
2934 | ||
2995fa78 | 2935 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2936 | |
b9a41d21 HT |
2937 | spin_lock(&_minor_lock); |
2938 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2939 | md = NULL; | |
2940 | goto out; | |
2941 | } | |
784aae73 | 2942 | dm_get(md); |
b9a41d21 HT |
2943 | out: |
2944 | spin_unlock(&_minor_lock); | |
2945 | ||
784aae73 MB |
2946 | return md; |
2947 | } | |
2948 | ||
4f186f8b | 2949 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2950 | { |
2951 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2952 | } | |
2953 | ||
ffcc3936 MS |
2954 | int dm_suspended_internally_md(struct mapped_device *md) |
2955 | { | |
2956 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2957 | } | |
2958 | ||
2c140a24 MP |
2959 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2960 | { | |
2961 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2962 | } | |
2963 | ||
64dbce58 KU |
2964 | int dm_suspended(struct dm_target *ti) |
2965 | { | |
ecdb2e25 | 2966 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2967 | } |
2968 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2969 | ||
2e93ccc1 KU |
2970 | int dm_noflush_suspending(struct dm_target *ti) |
2971 | { | |
ecdb2e25 | 2972 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2973 | } |
2974 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2975 | ||
7e0d574f | 2976 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
0776aa0e MS |
2977 | unsigned integrity, unsigned per_io_data_size, |
2978 | unsigned min_pool_size) | |
e6ee8c0b | 2979 | { |
115485e8 | 2980 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2981 | unsigned int pool_size = 0; |
64f52b0e | 2982 | unsigned int front_pad, io_front_pad; |
6f1c819c | 2983 | int ret; |
e6ee8c0b KU |
2984 | |
2985 | if (!pools) | |
4e6e36c3 | 2986 | return NULL; |
e6ee8c0b | 2987 | |
78d8e58a MS |
2988 | switch (type) { |
2989 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2990 | case DM_TYPE_DAX_BIO_BASED: |
22c11858 | 2991 | case DM_TYPE_NVME_BIO_BASED: |
0776aa0e | 2992 | pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); |
30187e1d | 2993 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
64f52b0e | 2994 | io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio); |
6f1c819c KO |
2995 | ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0); |
2996 | if (ret) | |
64f52b0e | 2997 | goto out; |
6f1c819c | 2998 | if (integrity && bioset_integrity_create(&pools->io_bs, pool_size)) |
eb8db831 | 2999 | goto out; |
78d8e58a MS |
3000 | break; |
3001 | case DM_TYPE_REQUEST_BASED: | |
0776aa0e | 3002 | pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); |
78d8e58a | 3003 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 3004 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
3005 | break; |
3006 | default: | |
3007 | BUG(); | |
3008 | } | |
3009 | ||
6f1c819c KO |
3010 | ret = bioset_init(&pools->bs, pool_size, front_pad, 0); |
3011 | if (ret) | |
5f015204 | 3012 | goto out; |
e6ee8c0b | 3013 | |
6f1c819c | 3014 | if (integrity && bioset_integrity_create(&pools->bs, pool_size)) |
5f015204 | 3015 | goto out; |
a91a2785 | 3016 | |
e6ee8c0b | 3017 | return pools; |
5f1b670d | 3018 | |
5f1b670d CH |
3019 | out: |
3020 | dm_free_md_mempools(pools); | |
78d8e58a | 3021 | |
4e6e36c3 | 3022 | return NULL; |
e6ee8c0b KU |
3023 | } |
3024 | ||
3025 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3026 | { | |
3027 | if (!pools) | |
3028 | return; | |
3029 | ||
6f1c819c KO |
3030 | bioset_exit(&pools->bs); |
3031 | bioset_exit(&pools->io_bs); | |
e6ee8c0b KU |
3032 | |
3033 | kfree(pools); | |
3034 | } | |
3035 | ||
9c72bad1 CH |
3036 | struct dm_pr { |
3037 | u64 old_key; | |
3038 | u64 new_key; | |
3039 | u32 flags; | |
3040 | bool fail_early; | |
3041 | }; | |
3042 | ||
3043 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
3044 | void *data) | |
71cdb697 CH |
3045 | { |
3046 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
3047 | struct dm_table *table; |
3048 | struct dm_target *ti; | |
3049 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 3050 | |
9c72bad1 CH |
3051 | table = dm_get_live_table(md, &srcu_idx); |
3052 | if (!table || !dm_table_get_size(table)) | |
3053 | goto out; | |
71cdb697 | 3054 | |
9c72bad1 CH |
3055 | /* We only support devices that have a single target */ |
3056 | if (dm_table_get_num_targets(table) != 1) | |
3057 | goto out; | |
3058 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 3059 | |
9c72bad1 CH |
3060 | ret = -EINVAL; |
3061 | if (!ti->type->iterate_devices) | |
3062 | goto out; | |
3063 | ||
3064 | ret = ti->type->iterate_devices(ti, fn, data); | |
3065 | out: | |
3066 | dm_put_live_table(md, srcu_idx); | |
3067 | return ret; | |
3068 | } | |
3069 | ||
3070 | /* | |
3071 | * For register / unregister we need to manually call out to every path. | |
3072 | */ | |
3073 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
3074 | sector_t start, sector_t len, void *data) | |
3075 | { | |
3076 | struct dm_pr *pr = data; | |
3077 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3078 | ||
3079 | if (!ops || !ops->pr_register) | |
3080 | return -EOPNOTSUPP; | |
3081 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
3082 | } | |
3083 | ||
3084 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
3085 | u32 flags) | |
3086 | { | |
3087 | struct dm_pr pr = { | |
3088 | .old_key = old_key, | |
3089 | .new_key = new_key, | |
3090 | .flags = flags, | |
3091 | .fail_early = true, | |
3092 | }; | |
3093 | int ret; | |
3094 | ||
3095 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
3096 | if (ret && new_key) { | |
3097 | /* unregister all paths if we failed to register any path */ | |
3098 | pr.old_key = new_key; | |
3099 | pr.new_key = 0; | |
3100 | pr.flags = 0; | |
3101 | pr.fail_early = false; | |
3102 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
3103 | } | |
3104 | ||
3105 | return ret; | |
71cdb697 CH |
3106 | } |
3107 | ||
3108 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3109 | u32 flags) |
71cdb697 CH |
3110 | { |
3111 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3112 | const struct pr_ops *ops; | |
971888c4 | 3113 | int r, srcu_idx; |
71cdb697 | 3114 | |
5bd5e8d8 | 3115 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3116 | if (r < 0) |
971888c4 | 3117 | goto out; |
71cdb697 CH |
3118 | |
3119 | ops = bdev->bd_disk->fops->pr_ops; | |
3120 | if (ops && ops->pr_reserve) | |
3121 | r = ops->pr_reserve(bdev, key, type, flags); | |
3122 | else | |
3123 | r = -EOPNOTSUPP; | |
971888c4 MS |
3124 | out: |
3125 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3126 | return r; |
3127 | } | |
3128 | ||
3129 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3130 | { | |
3131 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3132 | const struct pr_ops *ops; | |
971888c4 | 3133 | int r, srcu_idx; |
71cdb697 | 3134 | |
5bd5e8d8 | 3135 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3136 | if (r < 0) |
971888c4 | 3137 | goto out; |
71cdb697 CH |
3138 | |
3139 | ops = bdev->bd_disk->fops->pr_ops; | |
3140 | if (ops && ops->pr_release) | |
3141 | r = ops->pr_release(bdev, key, type); | |
3142 | else | |
3143 | r = -EOPNOTSUPP; | |
971888c4 MS |
3144 | out: |
3145 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3146 | return r; |
3147 | } | |
3148 | ||
3149 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3150 | enum pr_type type, bool abort) |
71cdb697 CH |
3151 | { |
3152 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3153 | const struct pr_ops *ops; | |
971888c4 | 3154 | int r, srcu_idx; |
71cdb697 | 3155 | |
5bd5e8d8 | 3156 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3157 | if (r < 0) |
971888c4 | 3158 | goto out; |
71cdb697 CH |
3159 | |
3160 | ops = bdev->bd_disk->fops->pr_ops; | |
3161 | if (ops && ops->pr_preempt) | |
3162 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3163 | else | |
3164 | r = -EOPNOTSUPP; | |
971888c4 MS |
3165 | out: |
3166 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3167 | return r; |
3168 | } | |
3169 | ||
3170 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3171 | { | |
3172 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3173 | const struct pr_ops *ops; | |
971888c4 | 3174 | int r, srcu_idx; |
71cdb697 | 3175 | |
5bd5e8d8 | 3176 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3177 | if (r < 0) |
971888c4 | 3178 | goto out; |
71cdb697 CH |
3179 | |
3180 | ops = bdev->bd_disk->fops->pr_ops; | |
3181 | if (ops && ops->pr_clear) | |
3182 | r = ops->pr_clear(bdev, key); | |
3183 | else | |
3184 | r = -EOPNOTSUPP; | |
971888c4 MS |
3185 | out: |
3186 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3187 | return r; |
3188 | } | |
3189 | ||
3190 | static const struct pr_ops dm_pr_ops = { | |
3191 | .pr_register = dm_pr_register, | |
3192 | .pr_reserve = dm_pr_reserve, | |
3193 | .pr_release = dm_pr_release, | |
3194 | .pr_preempt = dm_pr_preempt, | |
3195 | .pr_clear = dm_pr_clear, | |
3196 | }; | |
3197 | ||
83d5cde4 | 3198 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3199 | .open = dm_blk_open, |
3200 | .release = dm_blk_close, | |
aa129a22 | 3201 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3202 | .getgeo = dm_blk_getgeo, |
e76239a3 | 3203 | .report_zones = dm_blk_report_zones, |
71cdb697 | 3204 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3205 | .owner = THIS_MODULE |
3206 | }; | |
3207 | ||
f26c5719 DW |
3208 | static const struct dax_operations dm_dax_ops = { |
3209 | .direct_access = dm_dax_direct_access, | |
7e026c8c | 3210 | .copy_from_iter = dm_dax_copy_from_iter, |
b3a9a0c3 | 3211 | .copy_to_iter = dm_dax_copy_to_iter, |
f26c5719 DW |
3212 | }; |
3213 | ||
1da177e4 LT |
3214 | /* |
3215 | * module hooks | |
3216 | */ | |
3217 | module_init(dm_init); | |
3218 | module_exit(dm_exit); | |
3219 | ||
3220 | module_param(major, uint, 0); | |
3221 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3222 | |
e8603136 MS |
3223 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3224 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3225 | ||
115485e8 MS |
3226 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
3227 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
3228 | ||
1da177e4 LT |
3229 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3230 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3231 | MODULE_LICENSE("GPL"); |