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 | ||
e8603136 MS |
161 | /* |
162 | * Bio-based DM's mempools' reserved IOs set by the user. | |
163 | */ | |
4cc96131 | 164 | #define RESERVED_BIO_BASED_IOS 16 |
e8603136 MS |
165 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
166 | ||
115485e8 MS |
167 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
168 | { | |
6aa7de05 | 169 | int param = READ_ONCE(*module_param); |
115485e8 MS |
170 | int modified_param = 0; |
171 | bool modified = true; | |
172 | ||
173 | if (param < min) | |
174 | modified_param = min; | |
175 | else if (param > max) | |
176 | modified_param = max; | |
177 | else | |
178 | modified = false; | |
179 | ||
180 | if (modified) { | |
181 | (void)cmpxchg(module_param, param, modified_param); | |
182 | param = modified_param; | |
183 | } | |
184 | ||
185 | return param; | |
186 | } | |
187 | ||
4cc96131 MS |
188 | unsigned __dm_get_module_param(unsigned *module_param, |
189 | unsigned def, unsigned max) | |
f4790826 | 190 | { |
6aa7de05 | 191 | unsigned param = READ_ONCE(*module_param); |
09c2d531 | 192 | unsigned modified_param = 0; |
f4790826 | 193 | |
09c2d531 MS |
194 | if (!param) |
195 | modified_param = def; | |
196 | else if (param > max) | |
197 | modified_param = max; | |
f4790826 | 198 | |
09c2d531 MS |
199 | if (modified_param) { |
200 | (void)cmpxchg(module_param, param, modified_param); | |
201 | param = modified_param; | |
f4790826 MS |
202 | } |
203 | ||
09c2d531 | 204 | return param; |
f4790826 MS |
205 | } |
206 | ||
e8603136 MS |
207 | unsigned dm_get_reserved_bio_based_ios(void) |
208 | { | |
09c2d531 | 209 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 210 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
211 | } |
212 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
213 | ||
115485e8 MS |
214 | static unsigned dm_get_numa_node(void) |
215 | { | |
216 | return __dm_get_module_param_int(&dm_numa_node, | |
217 | DM_NUMA_NODE, num_online_nodes() - 1); | |
218 | } | |
219 | ||
1da177e4 LT |
220 | static int __init local_init(void) |
221 | { | |
e689fbab | 222 | int r; |
1ae49ea2 | 223 | |
51e5b2bd | 224 | r = dm_uevent_init(); |
51157b4a | 225 | if (r) |
e689fbab | 226 | return r; |
51e5b2bd | 227 | |
acfe0ad7 MP |
228 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
229 | if (!deferred_remove_workqueue) { | |
230 | r = -ENOMEM; | |
231 | goto out_uevent_exit; | |
232 | } | |
233 | ||
1da177e4 LT |
234 | _major = major; |
235 | r = register_blkdev(_major, _name); | |
51157b4a | 236 | if (r < 0) |
acfe0ad7 | 237 | goto out_free_workqueue; |
1da177e4 LT |
238 | |
239 | if (!_major) | |
240 | _major = r; | |
241 | ||
242 | return 0; | |
51157b4a | 243 | |
acfe0ad7 MP |
244 | out_free_workqueue: |
245 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
246 | out_uevent_exit: |
247 | dm_uevent_exit(); | |
51157b4a KU |
248 | |
249 | return r; | |
1da177e4 LT |
250 | } |
251 | ||
252 | static void local_exit(void) | |
253 | { | |
2c140a24 | 254 | flush_scheduled_work(); |
acfe0ad7 | 255 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 256 | |
00d59405 | 257 | unregister_blkdev(_major, _name); |
51e5b2bd | 258 | dm_uevent_exit(); |
1da177e4 LT |
259 | |
260 | _major = 0; | |
261 | ||
262 | DMINFO("cleaned up"); | |
263 | } | |
264 | ||
b9249e55 | 265 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
266 | local_init, |
267 | dm_target_init, | |
268 | dm_linear_init, | |
269 | dm_stripe_init, | |
952b3557 | 270 | dm_io_init, |
945fa4d2 | 271 | dm_kcopyd_init, |
1da177e4 | 272 | dm_interface_init, |
fd2ed4d2 | 273 | dm_statistics_init, |
1da177e4 LT |
274 | }; |
275 | ||
b9249e55 | 276 | static void (*_exits[])(void) = { |
1da177e4 LT |
277 | local_exit, |
278 | dm_target_exit, | |
279 | dm_linear_exit, | |
280 | dm_stripe_exit, | |
952b3557 | 281 | dm_io_exit, |
945fa4d2 | 282 | dm_kcopyd_exit, |
1da177e4 | 283 | dm_interface_exit, |
fd2ed4d2 | 284 | dm_statistics_exit, |
1da177e4 LT |
285 | }; |
286 | ||
287 | static int __init dm_init(void) | |
288 | { | |
289 | const int count = ARRAY_SIZE(_inits); | |
290 | ||
291 | int r, i; | |
292 | ||
293 | for (i = 0; i < count; i++) { | |
294 | r = _inits[i](); | |
295 | if (r) | |
296 | goto bad; | |
297 | } | |
298 | ||
299 | return 0; | |
300 | ||
301 | bad: | |
302 | while (i--) | |
303 | _exits[i](); | |
304 | ||
305 | return r; | |
306 | } | |
307 | ||
308 | static void __exit dm_exit(void) | |
309 | { | |
310 | int i = ARRAY_SIZE(_exits); | |
311 | ||
312 | while (i--) | |
313 | _exits[i](); | |
d15b774c AK |
314 | |
315 | /* | |
316 | * Should be empty by this point. | |
317 | */ | |
d15b774c | 318 | idr_destroy(&_minor_idr); |
1da177e4 LT |
319 | } |
320 | ||
321 | /* | |
322 | * Block device functions | |
323 | */ | |
432a212c MA |
324 | int dm_deleting_md(struct mapped_device *md) |
325 | { | |
326 | return test_bit(DMF_DELETING, &md->flags); | |
327 | } | |
328 | ||
fe5f9f2c | 329 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
330 | { |
331 | struct mapped_device *md; | |
332 | ||
fba9f90e JM |
333 | spin_lock(&_minor_lock); |
334 | ||
fe5f9f2c | 335 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
336 | if (!md) |
337 | goto out; | |
338 | ||
5c6bd75d | 339 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 340 | dm_deleting_md(md)) { |
fba9f90e JM |
341 | md = NULL; |
342 | goto out; | |
343 | } | |
344 | ||
1da177e4 | 345 | dm_get(md); |
5c6bd75d | 346 | atomic_inc(&md->open_count); |
fba9f90e JM |
347 | out: |
348 | spin_unlock(&_minor_lock); | |
349 | ||
350 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
351 | } |
352 | ||
db2a144b | 353 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 354 | { |
63a4f065 | 355 | struct mapped_device *md; |
6e9624b8 | 356 | |
4a1aeb98 MB |
357 | spin_lock(&_minor_lock); |
358 | ||
63a4f065 MS |
359 | md = disk->private_data; |
360 | if (WARN_ON(!md)) | |
361 | goto out; | |
362 | ||
2c140a24 MP |
363 | if (atomic_dec_and_test(&md->open_count) && |
364 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 365 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 366 | |
1da177e4 | 367 | dm_put(md); |
63a4f065 | 368 | out: |
4a1aeb98 | 369 | spin_unlock(&_minor_lock); |
1da177e4 LT |
370 | } |
371 | ||
5c6bd75d AK |
372 | int dm_open_count(struct mapped_device *md) |
373 | { | |
374 | return atomic_read(&md->open_count); | |
375 | } | |
376 | ||
377 | /* | |
378 | * Guarantees nothing is using the device before it's deleted. | |
379 | */ | |
2c140a24 | 380 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
381 | { |
382 | int r = 0; | |
383 | ||
384 | spin_lock(&_minor_lock); | |
385 | ||
2c140a24 | 386 | if (dm_open_count(md)) { |
5c6bd75d | 387 | r = -EBUSY; |
2c140a24 MP |
388 | if (mark_deferred) |
389 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
390 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
391 | r = -EEXIST; | |
5c6bd75d AK |
392 | else |
393 | set_bit(DMF_DELETING, &md->flags); | |
394 | ||
395 | spin_unlock(&_minor_lock); | |
396 | ||
397 | return r; | |
398 | } | |
399 | ||
2c140a24 MP |
400 | int dm_cancel_deferred_remove(struct mapped_device *md) |
401 | { | |
402 | int r = 0; | |
403 | ||
404 | spin_lock(&_minor_lock); | |
405 | ||
406 | if (test_bit(DMF_DELETING, &md->flags)) | |
407 | r = -EBUSY; | |
408 | else | |
409 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
410 | ||
411 | spin_unlock(&_minor_lock); | |
412 | ||
413 | return r; | |
414 | } | |
415 | ||
416 | static void do_deferred_remove(struct work_struct *w) | |
417 | { | |
418 | dm_deferred_remove(); | |
419 | } | |
420 | ||
fd2ed4d2 MP |
421 | sector_t dm_get_size(struct mapped_device *md) |
422 | { | |
423 | return get_capacity(md->disk); | |
424 | } | |
425 | ||
9974fa2c MS |
426 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
427 | { | |
428 | return md->queue; | |
429 | } | |
430 | ||
fd2ed4d2 MP |
431 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
432 | { | |
433 | return &md->stats; | |
434 | } | |
435 | ||
3ac51e74 DW |
436 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
437 | { | |
438 | struct mapped_device *md = bdev->bd_disk->private_data; | |
439 | ||
440 | return dm_get_geometry(md, geo); | |
441 | } | |
442 | ||
e76239a3 | 443 | static int dm_blk_report_zones(struct gendisk *disk, sector_t sector, |
bd976e52 | 444 | struct blk_zone *zones, unsigned int *nr_zones) |
e76239a3 CH |
445 | { |
446 | #ifdef CONFIG_BLK_DEV_ZONED | |
447 | struct mapped_device *md = disk->private_data; | |
448 | struct dm_target *tgt; | |
449 | struct dm_table *map; | |
450 | int srcu_idx, ret; | |
451 | ||
452 | if (dm_suspended_md(md)) | |
453 | return -EAGAIN; | |
454 | ||
455 | map = dm_get_live_table(md, &srcu_idx); | |
456 | if (!map) | |
457 | return -EIO; | |
458 | ||
459 | tgt = dm_table_find_target(map, sector); | |
123d87d5 | 460 | if (!tgt) { |
e76239a3 CH |
461 | ret = -EIO; |
462 | goto out; | |
463 | } | |
464 | ||
465 | /* | |
466 | * If we are executing this, we already know that the block device | |
467 | * is a zoned device and so each target should have support for that | |
468 | * type of drive. A missing report_zones method means that the target | |
469 | * driver has a problem. | |
470 | */ | |
471 | if (WARN_ON(!tgt->type->report_zones)) { | |
472 | ret = -EIO; | |
473 | goto out; | |
474 | } | |
475 | ||
476 | /* | |
477 | * blkdev_report_zones() will loop and call this again to cover all the | |
478 | * zones of the target, eventually moving on to the next target. | |
479 | * So there is no need to loop here trying to fill the entire array | |
480 | * of zones. | |
481 | */ | |
bd976e52 | 482 | ret = tgt->type->report_zones(tgt, sector, zones, nr_zones); |
e76239a3 CH |
483 | |
484 | out: | |
485 | dm_put_live_table(md, srcu_idx); | |
486 | return ret; | |
487 | #else | |
488 | return -ENOTSUPP; | |
489 | #endif | |
490 | } | |
491 | ||
971888c4 | 492 | static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, |
5bd5e8d8 | 493 | struct block_device **bdev) |
971888c4 | 494 | __acquires(md->io_barrier) |
aa129a22 | 495 | { |
66482026 | 496 | struct dm_target *tgt; |
6c182cd8 | 497 | struct dm_table *map; |
971888c4 | 498 | int r; |
aa129a22 | 499 | |
6c182cd8 | 500 | retry: |
e56f81e0 | 501 | r = -ENOTTY; |
971888c4 | 502 | map = dm_get_live_table(md, srcu_idx); |
aa129a22 | 503 | if (!map || !dm_table_get_size(map)) |
971888c4 | 504 | return r; |
aa129a22 MB |
505 | |
506 | /* We only support devices that have a single target */ | |
507 | if (dm_table_get_num_targets(map) != 1) | |
971888c4 | 508 | return r; |
aa129a22 | 509 | |
66482026 MS |
510 | tgt = dm_table_get_target(map, 0); |
511 | if (!tgt->type->prepare_ioctl) | |
971888c4 | 512 | return r; |
519049af | 513 | |
971888c4 MS |
514 | if (dm_suspended_md(md)) |
515 | return -EAGAIN; | |
aa129a22 | 516 | |
5bd5e8d8 | 517 | r = tgt->type->prepare_ioctl(tgt, bdev); |
5bbbfdf6 | 518 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
971888c4 | 519 | dm_put_live_table(md, *srcu_idx); |
6c182cd8 HR |
520 | msleep(10); |
521 | goto retry; | |
522 | } | |
971888c4 | 523 | |
e56f81e0 CH |
524 | return r; |
525 | } | |
526 | ||
971888c4 MS |
527 | static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) |
528 | __releases(md->io_barrier) | |
529 | { | |
530 | dm_put_live_table(md, srcu_idx); | |
531 | } | |
532 | ||
e56f81e0 CH |
533 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
534 | unsigned int cmd, unsigned long arg) | |
535 | { | |
536 | struct mapped_device *md = bdev->bd_disk->private_data; | |
971888c4 | 537 | int r, srcu_idx; |
e56f81e0 | 538 | |
5bd5e8d8 | 539 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
e56f81e0 | 540 | if (r < 0) |
971888c4 | 541 | goto out; |
6c182cd8 | 542 | |
e56f81e0 CH |
543 | if (r > 0) { |
544 | /* | |
e980f623 CH |
545 | * Target determined this ioctl is being issued against a |
546 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 547 | */ |
e980f623 CH |
548 | if (!capable(CAP_SYS_RAWIO)) { |
549 | DMWARN_LIMIT( | |
550 | "%s: sending ioctl %x to DM device without required privilege.", | |
551 | current->comm, cmd); | |
552 | r = -ENOIOCTLCMD; | |
e56f81e0 | 553 | goto out; |
e980f623 | 554 | } |
e56f81e0 | 555 | } |
6c182cd8 | 556 | |
66482026 | 557 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 558 | out: |
971888c4 | 559 | dm_unprepare_ioctl(md, srcu_idx); |
aa129a22 MB |
560 | return r; |
561 | } | |
562 | ||
978e51ba MS |
563 | static void start_io_acct(struct dm_io *io); |
564 | ||
565 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) | |
1da177e4 | 566 | { |
64f52b0e MS |
567 | struct dm_io *io; |
568 | struct dm_target_io *tio; | |
569 | struct bio *clone; | |
570 | ||
6f1c819c | 571 | clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs); |
64f52b0e MS |
572 | if (!clone) |
573 | return NULL; | |
574 | ||
575 | tio = container_of(clone, struct dm_target_io, clone); | |
576 | tio->inside_dm_io = true; | |
577 | tio->io = NULL; | |
578 | ||
579 | io = container_of(tio, struct dm_io, tio); | |
580 | io->magic = DM_IO_MAGIC; | |
978e51ba MS |
581 | io->status = 0; |
582 | atomic_set(&io->io_count, 1); | |
583 | io->orig_bio = bio; | |
584 | io->md = md; | |
585 | spin_lock_init(&io->endio_lock); | |
586 | ||
587 | start_io_acct(io); | |
64f52b0e MS |
588 | |
589 | return io; | |
1da177e4 LT |
590 | } |
591 | ||
028867ac | 592 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 | 593 | { |
64f52b0e MS |
594 | bio_put(&io->tio.clone); |
595 | } | |
596 | ||
597 | static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, | |
598 | unsigned target_bio_nr, gfp_t gfp_mask) | |
599 | { | |
600 | struct dm_target_io *tio; | |
601 | ||
602 | if (!ci->io->tio.io) { | |
603 | /* the dm_target_io embedded in ci->io is available */ | |
604 | tio = &ci->io->tio; | |
605 | } else { | |
6f1c819c | 606 | struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs); |
64f52b0e MS |
607 | if (!clone) |
608 | return NULL; | |
609 | ||
610 | tio = container_of(clone, struct dm_target_io, clone); | |
611 | tio->inside_dm_io = false; | |
612 | } | |
613 | ||
614 | tio->magic = DM_TIO_MAGIC; | |
615 | tio->io = ci->io; | |
616 | tio->ti = ti; | |
617 | tio->target_bio_nr = target_bio_nr; | |
618 | ||
619 | return tio; | |
1da177e4 LT |
620 | } |
621 | ||
cfae7529 | 622 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 623 | { |
64f52b0e MS |
624 | if (tio->inside_dm_io) |
625 | return; | |
dba14160 | 626 | bio_put(&tio->clone); |
1da177e4 LT |
627 | } |
628 | ||
c4576aed | 629 | static bool md_in_flight_bios(struct mapped_device *md) |
90abb8c4 | 630 | { |
6f757231 MP |
631 | int cpu; |
632 | struct hd_struct *part = &dm_disk(md)->part0; | |
b7934ba4 | 633 | long sum = 0; |
6f757231 MP |
634 | |
635 | for_each_possible_cpu(cpu) { | |
b7934ba4 JA |
636 | sum += part_stat_local_read_cpu(part, in_flight[0], cpu); |
637 | sum += part_stat_local_read_cpu(part, in_flight[1], cpu); | |
6f757231 MP |
638 | } |
639 | ||
b7934ba4 | 640 | return sum != 0; |
90abb8c4 KU |
641 | } |
642 | ||
c4576aed MS |
643 | static bool md_in_flight(struct mapped_device *md) |
644 | { | |
645 | if (queue_is_mq(md->queue)) | |
3c94d83c | 646 | return blk_mq_queue_inflight(md->queue); |
c4576aed MS |
647 | else |
648 | return md_in_flight_bios(md); | |
90abb8c4 KU |
649 | } |
650 | ||
3eaf840e JNN |
651 | static void start_io_acct(struct dm_io *io) |
652 | { | |
653 | struct mapped_device *md = io->md; | |
745dc570 | 654 | struct bio *bio = io->orig_bio; |
3eaf840e JNN |
655 | |
656 | io->start_time = jiffies; | |
657 | ||
ddcf35d3 MC |
658 | generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio), |
659 | &dm_disk(md)->part0); | |
f3986374 | 660 | |
fd2ed4d2 | 661 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
662 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
663 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
664 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
665 | } |
666 | ||
d221d2e7 | 667 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
668 | { |
669 | struct mapped_device *md = io->md; | |
745dc570 | 670 | struct bio *bio = io->orig_bio; |
3eaf840e | 671 | unsigned long duration = jiffies - io->start_time; |
3eaf840e | 672 | |
ddcf35d3 MC |
673 | generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0, |
674 | io->start_time); | |
3eaf840e | 675 | |
fd2ed4d2 | 676 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
677 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
678 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
679 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 680 | |
d221d2e7 | 681 | /* nudge anyone waiting on suspend queue */ |
645efa84 | 682 | if (unlikely(wq_has_sleeper(&md->wait))) |
d221d2e7 | 683 | wake_up(&md->wait); |
3eaf840e JNN |
684 | } |
685 | ||
1da177e4 LT |
686 | /* |
687 | * Add the bio to the list of deferred io. | |
688 | */ | |
92c63902 | 689 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 690 | { |
05447420 | 691 | unsigned long flags; |
1da177e4 | 692 | |
05447420 | 693 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 694 | bio_list_add(&md->deferred, bio); |
05447420 | 695 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 696 | queue_work(md->wq, &md->work); |
1da177e4 LT |
697 | } |
698 | ||
699 | /* | |
700 | * Everyone (including functions in this file), should use this | |
701 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 702 | * dm_put_live_table() when finished. |
1da177e4 | 703 | */ |
83d5e5b0 | 704 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 705 | { |
83d5e5b0 MP |
706 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
707 | ||
708 | return srcu_dereference(md->map, &md->io_barrier); | |
709 | } | |
1da177e4 | 710 | |
83d5e5b0 MP |
711 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
712 | { | |
713 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
714 | } | |
715 | ||
716 | void dm_sync_table(struct mapped_device *md) | |
717 | { | |
718 | synchronize_srcu(&md->io_barrier); | |
719 | synchronize_rcu_expedited(); | |
720 | } | |
721 | ||
722 | /* | |
723 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
724 | * The caller must not block between these two functions. | |
725 | */ | |
726 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
727 | { | |
728 | rcu_read_lock(); | |
729 | return rcu_dereference(md->map); | |
730 | } | |
1da177e4 | 731 | |
83d5e5b0 MP |
732 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
733 | { | |
734 | rcu_read_unlock(); | |
1da177e4 LT |
735 | } |
736 | ||
971888c4 MS |
737 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
738 | ||
86f1152b BM |
739 | /* |
740 | * Open a table device so we can use it as a map destination. | |
741 | */ | |
742 | static int open_table_device(struct table_device *td, dev_t dev, | |
743 | struct mapped_device *md) | |
744 | { | |
86f1152b BM |
745 | struct block_device *bdev; |
746 | ||
747 | int r; | |
748 | ||
749 | BUG_ON(td->dm_dev.bdev); | |
750 | ||
519049af | 751 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr); |
86f1152b BM |
752 | if (IS_ERR(bdev)) |
753 | return PTR_ERR(bdev); | |
754 | ||
755 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
756 | if (r) { | |
757 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
758 | return r; | |
759 | } | |
760 | ||
761 | td->dm_dev.bdev = bdev; | |
817bf402 | 762 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
763 | return 0; |
764 | } | |
765 | ||
766 | /* | |
767 | * Close a table device that we've been using. | |
768 | */ | |
769 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
770 | { | |
771 | if (!td->dm_dev.bdev) | |
772 | return; | |
773 | ||
774 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
775 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 776 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 777 | td->dm_dev.bdev = NULL; |
817bf402 | 778 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
779 | } |
780 | ||
781 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
8454fca4 SS |
782 | fmode_t mode) |
783 | { | |
86f1152b BM |
784 | struct table_device *td; |
785 | ||
786 | list_for_each_entry(td, l, list) | |
787 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
788 | return td; | |
789 | ||
790 | return NULL; | |
791 | } | |
792 | ||
793 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
8454fca4 SS |
794 | struct dm_dev **result) |
795 | { | |
86f1152b BM |
796 | int r; |
797 | struct table_device *td; | |
798 | ||
799 | mutex_lock(&md->table_devices_lock); | |
800 | td = find_table_device(&md->table_devices, dev, mode); | |
801 | if (!td) { | |
115485e8 | 802 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
803 | if (!td) { |
804 | mutex_unlock(&md->table_devices_lock); | |
805 | return -ENOMEM; | |
806 | } | |
807 | ||
808 | td->dm_dev.mode = mode; | |
809 | td->dm_dev.bdev = NULL; | |
810 | ||
811 | if ((r = open_table_device(td, dev, md))) { | |
812 | mutex_unlock(&md->table_devices_lock); | |
813 | kfree(td); | |
814 | return r; | |
815 | } | |
816 | ||
817 | format_dev_t(td->dm_dev.name, dev); | |
818 | ||
b0b4d7c6 | 819 | refcount_set(&td->count, 1); |
86f1152b | 820 | list_add(&td->list, &md->table_devices); |
b0b4d7c6 ER |
821 | } else { |
822 | refcount_inc(&td->count); | |
86f1152b | 823 | } |
86f1152b BM |
824 | mutex_unlock(&md->table_devices_lock); |
825 | ||
826 | *result = &td->dm_dev; | |
827 | return 0; | |
828 | } | |
829 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
830 | ||
831 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
832 | { | |
833 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
834 | ||
835 | mutex_lock(&md->table_devices_lock); | |
b0b4d7c6 | 836 | if (refcount_dec_and_test(&td->count)) { |
86f1152b BM |
837 | close_table_device(td, md); |
838 | list_del(&td->list); | |
839 | kfree(td); | |
840 | } | |
841 | mutex_unlock(&md->table_devices_lock); | |
842 | } | |
843 | EXPORT_SYMBOL(dm_put_table_device); | |
844 | ||
845 | static void free_table_devices(struct list_head *devices) | |
846 | { | |
847 | struct list_head *tmp, *next; | |
848 | ||
849 | list_for_each_safe(tmp, next, devices) { | |
850 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
851 | ||
852 | DMWARN("dm_destroy: %s still exists with %d references", | |
b0b4d7c6 | 853 | td->dm_dev.name, refcount_read(&td->count)); |
86f1152b BM |
854 | kfree(td); |
855 | } | |
856 | } | |
857 | ||
3ac51e74 DW |
858 | /* |
859 | * Get the geometry associated with a dm device | |
860 | */ | |
861 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
862 | { | |
863 | *geo = md->geometry; | |
864 | ||
865 | return 0; | |
866 | } | |
867 | ||
868 | /* | |
869 | * Set the geometry of a device. | |
870 | */ | |
871 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
872 | { | |
873 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
874 | ||
875 | if (geo->start > sz) { | |
876 | DMWARN("Start sector is beyond the geometry limits."); | |
877 | return -EINVAL; | |
878 | } | |
879 | ||
880 | md->geometry = *geo; | |
881 | ||
882 | return 0; | |
883 | } | |
884 | ||
2e93ccc1 KU |
885 | static int __noflush_suspending(struct mapped_device *md) |
886 | { | |
887 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
888 | } | |
889 | ||
1da177e4 LT |
890 | /* |
891 | * Decrements the number of outstanding ios that a bio has been | |
892 | * cloned into, completing the original io if necc. | |
893 | */ | |
4e4cbee9 | 894 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 895 | { |
2e93ccc1 | 896 | unsigned long flags; |
4e4cbee9 | 897 | blk_status_t io_error; |
b35f8caa MB |
898 | struct bio *bio; |
899 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
900 | |
901 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
902 | if (unlikely(error)) { |
903 | spin_lock_irqsave(&io->endio_lock, flags); | |
745dc570 | 904 | if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) |
4e4cbee9 | 905 | io->status = error; |
f88fb981 KU |
906 | spin_unlock_irqrestore(&io->endio_lock, flags); |
907 | } | |
1da177e4 LT |
908 | |
909 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 910 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
911 | /* |
912 | * Target requested pushing back the I/O. | |
2e93ccc1 | 913 | */ |
022c2611 | 914 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 | 915 | if (__noflush_suspending(md)) |
745dc570 MS |
916 | /* NOTE early return due to BLK_STS_DM_REQUEUE below */ |
917 | bio_list_add_head(&md->deferred, io->orig_bio); | |
6a8736d1 | 918 | else |
2e93ccc1 | 919 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 920 | io->status = BLK_STS_IOERR; |
022c2611 | 921 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
922 | } |
923 | ||
4e4cbee9 | 924 | io_error = io->status; |
745dc570 | 925 | bio = io->orig_bio; |
6a8736d1 TH |
926 | end_io_acct(io); |
927 | free_io(md, io); | |
928 | ||
4e4cbee9 | 929 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 930 | return; |
2e93ccc1 | 931 | |
1eff9d32 | 932 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 933 | /* |
6a8736d1 | 934 | * Preflush done for flush with data, reissue |
28a8f0d3 | 935 | * without REQ_PREFLUSH. |
af7e466a | 936 | */ |
1eff9d32 | 937 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 938 | queue_io(md, bio); |
af7e466a | 939 | } else { |
b372d360 | 940 | /* done with normal IO or empty flush */ |
8dd601fa N |
941 | if (io_error) |
942 | bio->bi_status = io_error; | |
4246a0b6 | 943 | bio_endio(bio); |
b35f8caa | 944 | } |
1da177e4 LT |
945 | } |
946 | } | |
947 | ||
bcb44433 MS |
948 | void disable_discard(struct mapped_device *md) |
949 | { | |
950 | struct queue_limits *limits = dm_get_queue_limits(md); | |
951 | ||
952 | /* device doesn't really support DISCARD, disable it */ | |
953 | limits->max_discard_sectors = 0; | |
954 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue); | |
955 | } | |
956 | ||
4cc96131 | 957 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
958 | { |
959 | struct queue_limits *limits = dm_get_queue_limits(md); | |
960 | ||
961 | /* device doesn't really support WRITE SAME, disable it */ | |
962 | limits->max_write_same_sectors = 0; | |
963 | } | |
964 | ||
ac62d620 CH |
965 | void disable_write_zeroes(struct mapped_device *md) |
966 | { | |
967 | struct queue_limits *limits = dm_get_queue_limits(md); | |
968 | ||
969 | /* device doesn't really support WRITE ZEROES, disable it */ | |
970 | limits->max_write_zeroes_sectors = 0; | |
971 | } | |
972 | ||
4246a0b6 | 973 | static void clone_endio(struct bio *bio) |
1da177e4 | 974 | { |
4e4cbee9 | 975 | blk_status_t error = bio->bi_status; |
bfc6d41c | 976 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 977 | struct dm_io *io = tio->io; |
9faf400f | 978 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
979 | dm_endio_fn endio = tio->ti->type->end_io; |
980 | ||
978e51ba | 981 | if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) { |
bcb44433 MS |
982 | if (bio_op(bio) == REQ_OP_DISCARD && |
983 | !bio->bi_disk->queue->limits.max_discard_sectors) | |
984 | disable_discard(md); | |
985 | else if (bio_op(bio) == REQ_OP_WRITE_SAME && | |
986 | !bio->bi_disk->queue->limits.max_write_same_sectors) | |
ac62d620 | 987 | disable_write_same(md); |
bcb44433 MS |
988 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES && |
989 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) | |
ac62d620 CH |
990 | disable_write_zeroes(md); |
991 | } | |
7eee4ae2 | 992 | |
1be56909 | 993 | if (endio) { |
4e4cbee9 | 994 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
995 | switch (r) { |
996 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 997 | error = BLK_STS_DM_REQUEUE; |
1be56909 CH |
998 | /*FALLTHRU*/ |
999 | case DM_ENDIO_DONE: | |
1000 | break; | |
1001 | case DM_ENDIO_INCOMPLETE: | |
1002 | /* The target will handle the io */ | |
1003 | return; | |
1004 | default: | |
1005 | DMWARN("unimplemented target endio return value: %d", r); | |
1006 | BUG(); | |
1007 | } | |
1008 | } | |
1009 | ||
cfae7529 | 1010 | free_tio(tio); |
b35f8caa | 1011 | dec_pending(io, error); |
1da177e4 LT |
1012 | } |
1013 | ||
56a67df7 MS |
1014 | /* |
1015 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1016 | * target boundary. | |
1017 | */ | |
1018 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1019 | { | |
1020 | sector_t target_offset = dm_target_offset(ti, sector); | |
1021 | ||
1022 | return ti->len - target_offset; | |
1023 | } | |
1024 | ||
1025 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1026 | { |
56a67df7 | 1027 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1028 | sector_t offset, max_len; |
1da177e4 LT |
1029 | |
1030 | /* | |
542f9038 | 1031 | * Does the target need to split even further? |
1da177e4 | 1032 | */ |
542f9038 MS |
1033 | if (ti->max_io_len) { |
1034 | offset = dm_target_offset(ti, sector); | |
1035 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1036 | max_len = sector_div(offset, ti->max_io_len); | |
1037 | else | |
1038 | max_len = offset & (ti->max_io_len - 1); | |
1039 | max_len = ti->max_io_len - max_len; | |
1040 | ||
1041 | if (len > max_len) | |
1042 | len = max_len; | |
1da177e4 LT |
1043 | } |
1044 | ||
1045 | return len; | |
1046 | } | |
1047 | ||
542f9038 MS |
1048 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1049 | { | |
1050 | if (len > UINT_MAX) { | |
1051 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1052 | (unsigned long long)len, UINT_MAX); | |
1053 | ti->error = "Maximum size of target IO is too large"; | |
1054 | return -EINVAL; | |
1055 | } | |
1056 | ||
75ae1936 | 1057 | ti->max_io_len = (uint32_t) len; |
542f9038 MS |
1058 | |
1059 | return 0; | |
1060 | } | |
1061 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1062 | ||
f26c5719 | 1063 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
3d97c829 MS |
1064 | sector_t sector, int *srcu_idx) |
1065 | __acquires(md->io_barrier) | |
545ed20e | 1066 | { |
545ed20e TK |
1067 | struct dm_table *map; |
1068 | struct dm_target *ti; | |
545ed20e | 1069 | |
f26c5719 | 1070 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1071 | if (!map) |
f26c5719 | 1072 | return NULL; |
545ed20e TK |
1073 | |
1074 | ti = dm_table_find_target(map, sector); | |
123d87d5 | 1075 | if (!ti) |
f26c5719 | 1076 | return NULL; |
545ed20e | 1077 | |
f26c5719 DW |
1078 | return ti; |
1079 | } | |
545ed20e | 1080 | |
f26c5719 | 1081 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1082 | long nr_pages, void **kaddr, pfn_t *pfn) |
f26c5719 DW |
1083 | { |
1084 | struct mapped_device *md = dax_get_private(dax_dev); | |
1085 | sector_t sector = pgoff * PAGE_SECTORS; | |
1086 | struct dm_target *ti; | |
1087 | long len, ret = -EIO; | |
1088 | int srcu_idx; | |
545ed20e | 1089 | |
f26c5719 | 1090 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1091 | |
f26c5719 DW |
1092 | if (!ti) |
1093 | goto out; | |
1094 | if (!ti->type->direct_access) | |
1095 | goto out; | |
1096 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
1097 | if (len < 1) | |
1098 | goto out; | |
1099 | nr_pages = min(len, nr_pages); | |
dbc62659 | 1100 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
817bf402 | 1101 | |
f26c5719 | 1102 | out: |
545ed20e | 1103 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1104 | |
1105 | return ret; | |
545ed20e TK |
1106 | } |
1107 | ||
7bf7eac8 DW |
1108 | static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev, |
1109 | int blocksize, sector_t start, sector_t len) | |
1110 | { | |
1111 | struct mapped_device *md = dax_get_private(dax_dev); | |
1112 | struct dm_table *map; | |
1113 | int srcu_idx; | |
1114 | bool ret; | |
1115 | ||
1116 | map = dm_get_live_table(md, &srcu_idx); | |
1117 | if (!map) | |
1118 | return false; | |
1119 | ||
2e9ee095 | 1120 | ret = dm_table_supports_dax(map, device_supports_dax, &blocksize); |
7bf7eac8 DW |
1121 | |
1122 | dm_put_live_table(md, srcu_idx); | |
1123 | ||
1124 | return ret; | |
1125 | } | |
1126 | ||
7e026c8c | 1127 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1128 | void *addr, size_t bytes, struct iov_iter *i) |
7e026c8c DW |
1129 | { |
1130 | struct mapped_device *md = dax_get_private(dax_dev); | |
1131 | sector_t sector = pgoff * PAGE_SECTORS; | |
1132 | struct dm_target *ti; | |
1133 | long ret = 0; | |
1134 | int srcu_idx; | |
1135 | ||
1136 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1137 | ||
1138 | if (!ti) | |
1139 | goto out; | |
1140 | if (!ti->type->dax_copy_from_iter) { | |
1141 | ret = copy_from_iter(addr, bytes, i); | |
1142 | goto out; | |
1143 | } | |
1144 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
1145 | out: | |
1146 | dm_put_live_table(md, srcu_idx); | |
1147 | ||
1148 | return ret; | |
1149 | } | |
1150 | ||
b3a9a0c3 DW |
1151 | static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
1152 | void *addr, size_t bytes, struct iov_iter *i) | |
1153 | { | |
1154 | struct mapped_device *md = dax_get_private(dax_dev); | |
1155 | sector_t sector = pgoff * PAGE_SECTORS; | |
1156 | struct dm_target *ti; | |
1157 | long ret = 0; | |
1158 | int srcu_idx; | |
1159 | ||
1160 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1161 | ||
1162 | if (!ti) | |
1163 | goto out; | |
1164 | if (!ti->type->dax_copy_to_iter) { | |
1165 | ret = copy_to_iter(addr, bytes, i); | |
1166 | goto out; | |
1167 | } | |
1168 | ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i); | |
1169 | out: | |
1170 | dm_put_live_table(md, srcu_idx); | |
1171 | ||
1172 | return ret; | |
1173 | } | |
1174 | ||
1dd40c3e MP |
1175 | /* |
1176 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
c06b3e58 | 1177 | * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET. |
1dd40c3e MP |
1178 | * |
1179 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1180 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1181 | * sent in a next bio. | |
1182 | * | |
1183 | * A diagram that explains the arithmetics: | |
1184 | * +--------------------+---------------+-------+ | |
1185 | * | 1 | 2 | 3 | | |
1186 | * +--------------------+---------------+-------+ | |
1187 | * | |
1188 | * <-------------- *tio->len_ptr ---------------> | |
1189 | * <------- bi_size -------> | |
1190 | * <-- n_sectors --> | |
1191 | * | |
1192 | * Region 1 was already iterated over with bio_advance or similar function. | |
1193 | * (it may be empty if the target doesn't use bio_advance) | |
1194 | * Region 2 is the remaining bio size that the target wants to process. | |
1195 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1196 | * to make it empty) | |
1197 | * The target requires that region 3 is to be sent in the next bio. | |
1198 | * | |
1199 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1200 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1201 | * copies of the bio. | |
1202 | */ | |
1203 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1204 | { | |
1205 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1206 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1207 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1208 | BUG_ON(bi_size > *tio->len_ptr); |
1209 | BUG_ON(n_sectors > bi_size); | |
1210 | *tio->len_ptr -= bi_size - n_sectors; | |
1211 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1212 | } | |
1213 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1214 | ||
10999307 | 1215 | /* |
e76239a3 CH |
1216 | * The zone descriptors obtained with a zone report indicate |
1217 | * zone positions within the underlying device of the target. The zone | |
1218 | * descriptors must be remapped to match their position within the dm device. | |
1219 | * The caller target should obtain the zones information using | |
1220 | * blkdev_report_zones() to ensure that remapping for partition offset is | |
1221 | * already handled. | |
10999307 | 1222 | */ |
e76239a3 CH |
1223 | void dm_remap_zone_report(struct dm_target *ti, sector_t start, |
1224 | struct blk_zone *zones, unsigned int *nr_zones) | |
10999307 DLM |
1225 | { |
1226 | #ifdef CONFIG_BLK_DEV_ZONED | |
10999307 | 1227 | struct blk_zone *zone; |
e76239a3 CH |
1228 | unsigned int nrz = *nr_zones; |
1229 | int i; | |
9864cd5d | 1230 | |
10999307 | 1231 | /* |
e76239a3 CH |
1232 | * Remap the start sector and write pointer position of the zones in |
1233 | * the array. Since we may have obtained from the target underlying | |
1234 | * device more zones that the target size, also adjust the number | |
1235 | * of zones. | |
10999307 | 1236 | */ |
e76239a3 CH |
1237 | for (i = 0; i < nrz; i++) { |
1238 | zone = zones + i; | |
1239 | if (zone->start >= start + ti->len) { | |
1240 | memset(zone, 0, sizeof(struct blk_zone) * (nrz - i)); | |
1241 | break; | |
10999307 DLM |
1242 | } |
1243 | ||
e76239a3 CH |
1244 | zone->start = zone->start + ti->begin - start; |
1245 | if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL) | |
1246 | continue; | |
10999307 | 1247 | |
e76239a3 CH |
1248 | if (zone->cond == BLK_ZONE_COND_FULL) |
1249 | zone->wp = zone->start + zone->len; | |
1250 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
1251 | zone->wp = zone->start; | |
1252 | else | |
1253 | zone->wp = zone->wp + ti->begin - start; | |
10999307 DLM |
1254 | } |
1255 | ||
e76239a3 | 1256 | *nr_zones = i; |
10999307 | 1257 | #else /* !CONFIG_BLK_DEV_ZONED */ |
e76239a3 | 1258 | *nr_zones = 0; |
10999307 DLM |
1259 | #endif |
1260 | } | |
1261 | EXPORT_SYMBOL_GPL(dm_remap_zone_report); | |
1262 | ||
978e51ba | 1263 | static blk_qc_t __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1264 | { |
1265 | int r; | |
2056a782 | 1266 | sector_t sector; |
dba14160 | 1267 | struct bio *clone = &tio->clone; |
64f52b0e | 1268 | struct dm_io *io = tio->io; |
978e51ba | 1269 | struct mapped_device *md = io->md; |
bd2a49b8 | 1270 | struct dm_target *ti = tio->ti; |
978e51ba | 1271 | blk_qc_t ret = BLK_QC_T_NONE; |
1da177e4 | 1272 | |
1da177e4 | 1273 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1274 | |
1275 | /* | |
1276 | * Map the clone. If r == 0 we don't need to do | |
1277 | * anything, the target has assumed ownership of | |
1278 | * this io. | |
1279 | */ | |
64f52b0e | 1280 | atomic_inc(&io->io_count); |
4f024f37 | 1281 | sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1282 | |
7de3ee57 | 1283 | r = ti->type->map(ti, clone); |
846785e6 CH |
1284 | switch (r) { |
1285 | case DM_MAPIO_SUBMITTED: | |
1286 | break; | |
1287 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1288 | /* the bio has been remapped so dispatch it */ |
74d46992 | 1289 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
64f52b0e | 1290 | bio_dev(io->orig_bio), sector); |
978e51ba MS |
1291 | if (md->type == DM_TYPE_NVME_BIO_BASED) |
1292 | ret = direct_make_request(clone); | |
1293 | else | |
1294 | ret = generic_make_request(clone); | |
846785e6 CH |
1295 | break; |
1296 | case DM_MAPIO_KILL: | |
4e4cbee9 | 1297 | free_tio(tio); |
64f52b0e | 1298 | dec_pending(io, BLK_STS_IOERR); |
4e4cbee9 | 1299 | break; |
846785e6 | 1300 | case DM_MAPIO_REQUEUE: |
cfae7529 | 1301 | free_tio(tio); |
64f52b0e | 1302 | dec_pending(io, BLK_STS_DM_REQUEUE); |
846785e6 CH |
1303 | break; |
1304 | default: | |
45cbcd79 KU |
1305 | DMWARN("unimplemented target map return value: %d", r); |
1306 | BUG(); | |
1da177e4 | 1307 | } |
1da177e4 | 1308 | |
978e51ba | 1309 | return ret; |
1da177e4 | 1310 | } |
1da177e4 | 1311 | |
e0d6609a | 1312 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1313 | { |
4f024f37 KO |
1314 | bio->bi_iter.bi_sector = sector; |
1315 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1316 | } |
1317 | ||
1318 | /* | |
1319 | * Creates a bio that consists of range of complete bvecs. | |
1320 | */ | |
c80914e8 MS |
1321 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1322 | sector_t sector, unsigned len) | |
1da177e4 | 1323 | { |
dba14160 | 1324 | struct bio *clone = &tio->clone; |
1da177e4 | 1325 | |
1c3b13e6 KO |
1326 | __bio_clone_fast(clone, bio); |
1327 | ||
57c36519 | 1328 | if (bio_integrity(bio)) { |
e2460f2a MP |
1329 | int r; |
1330 | ||
1331 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1332 | !dm_target_passes_integrity(tio->ti->type))) { | |
1333 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1334 | dm_device_name(tio->io->md), | |
1335 | tio->ti->type->name); | |
1336 | return -EIO; | |
1337 | } | |
1338 | ||
1339 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1340 | if (r < 0) |
1341 | return r; | |
1342 | } | |
bd2a49b8 | 1343 | |
fa8db494 MS |
1344 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1345 | clone->bi_iter.bi_size = to_bytes(len); | |
1346 | ||
1347 | if (bio_integrity(bio)) | |
1348 | bio_integrity_trim(clone); | |
c80914e8 MS |
1349 | |
1350 | return 0; | |
1da177e4 LT |
1351 | } |
1352 | ||
318716dd MS |
1353 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
1354 | struct dm_target *ti, unsigned num_bios) | |
f9ab94ce | 1355 | { |
dba14160 | 1356 | struct dm_target_io *tio; |
318716dd | 1357 | int try; |
dba14160 | 1358 | |
318716dd MS |
1359 | if (!num_bios) |
1360 | return; | |
f9ab94ce | 1361 | |
318716dd MS |
1362 | if (num_bios == 1) { |
1363 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); | |
1364 | bio_list_add(blist, &tio->clone); | |
1365 | return; | |
1366 | } | |
9015df24 | 1367 | |
318716dd MS |
1368 | for (try = 0; try < 2; try++) { |
1369 | int bio_nr; | |
1370 | struct bio *bio; | |
1371 | ||
1372 | if (try) | |
bc02cdbe | 1373 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd MS |
1374 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
1375 | tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); | |
1376 | if (!tio) | |
1377 | break; | |
1378 | ||
1379 | bio_list_add(blist, &tio->clone); | |
1380 | } | |
1381 | if (try) | |
bc02cdbe | 1382 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1383 | if (bio_nr == num_bios) |
1384 | return; | |
1385 | ||
1386 | while ((bio = bio_list_pop(blist))) { | |
1387 | tio = container_of(bio, struct dm_target_io, clone); | |
1388 | free_tio(tio); | |
1389 | } | |
1390 | } | |
9015df24 AK |
1391 | } |
1392 | ||
978e51ba MS |
1393 | static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci, |
1394 | struct dm_target_io *tio, unsigned *len) | |
9015df24 | 1395 | { |
dba14160 | 1396 | struct bio *clone = &tio->clone; |
9015df24 | 1397 | |
1dd40c3e MP |
1398 | tio->len_ptr = len; |
1399 | ||
99778273 | 1400 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1401 | if (len) |
1dd40c3e | 1402 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1403 | |
978e51ba | 1404 | return __map_bio(tio); |
f9ab94ce MP |
1405 | } |
1406 | ||
14fe594d | 1407 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1408 | unsigned num_bios, unsigned *len) |
06a426ce | 1409 | { |
318716dd MS |
1410 | struct bio_list blist = BIO_EMPTY_LIST; |
1411 | struct bio *bio; | |
1412 | struct dm_target_io *tio; | |
1413 | ||
1414 | alloc_multiple_bios(&blist, ci, ti, num_bios); | |
06a426ce | 1415 | |
318716dd MS |
1416 | while ((bio = bio_list_pop(&blist))) { |
1417 | tio = container_of(bio, struct dm_target_io, clone); | |
978e51ba | 1418 | (void) __clone_and_map_simple_bio(ci, tio, len); |
318716dd | 1419 | } |
06a426ce MS |
1420 | } |
1421 | ||
14fe594d | 1422 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1423 | { |
06a426ce | 1424 | unsigned target_nr = 0; |
f9ab94ce MP |
1425 | struct dm_target *ti; |
1426 | ||
892ad71f | 1427 | /* |
dbe3ece1 JA |
1428 | * Empty flush uses a statically initialized bio, as the base for |
1429 | * cloning. However, blkg association requires that a bdev is | |
1430 | * associated with a gendisk, which doesn't happen until the bdev is | |
1431 | * opened. So, blkg association is done at issue time of the flush | |
1432 | * rather than when the device is created in alloc_dev(). | |
892ad71f DZ |
1433 | */ |
1434 | bio_set_dev(ci->bio, ci->io->md->bdev); | |
1435 | ||
b372d360 | 1436 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1437 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1438 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1439 | |
892ad71f DZ |
1440 | bio_disassociate_blkg(ci->bio); |
1441 | ||
f9ab94ce MP |
1442 | return 0; |
1443 | } | |
1444 | ||
c80914e8 | 1445 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
f31c21e4 | 1446 | sector_t sector, unsigned *len) |
5ae89a87 | 1447 | { |
dba14160 | 1448 | struct bio *bio = ci->bio; |
5ae89a87 | 1449 | struct dm_target_io *tio; |
f31c21e4 | 1450 | int r; |
5ae89a87 | 1451 | |
318716dd | 1452 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); |
f31c21e4 N |
1453 | tio->len_ptr = len; |
1454 | r = clone_bio(tio, bio, sector, *len); | |
1455 | if (r < 0) { | |
1456 | free_tio(tio); | |
1457 | return r; | |
b0d8ed4d | 1458 | } |
978e51ba | 1459 | (void) __map_bio(tio); |
c80914e8 | 1460 | |
f31c21e4 | 1461 | return 0; |
5ae89a87 MS |
1462 | } |
1463 | ||
55a62eef | 1464 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1465 | |
55a62eef | 1466 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1467 | { |
55a62eef | 1468 | return ti->num_discard_bios; |
23508a96 MS |
1469 | } |
1470 | ||
00716545 DS |
1471 | static unsigned get_num_secure_erase_bios(struct dm_target *ti) |
1472 | { | |
1473 | return ti->num_secure_erase_bios; | |
1474 | } | |
1475 | ||
55a62eef | 1476 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1477 | { |
55a62eef | 1478 | return ti->num_write_same_bios; |
23508a96 MS |
1479 | } |
1480 | ||
ac62d620 CH |
1481 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1482 | { | |
1483 | return ti->num_write_zeroes_bios; | |
1484 | } | |
1485 | ||
3d7f4562 | 1486 | static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
61697a6a | 1487 | unsigned num_bios) |
ba1cbad9 | 1488 | { |
51b86f9a | 1489 | unsigned len; |
ba1cbad9 | 1490 | |
3d7f4562 MS |
1491 | /* |
1492 | * Even though the device advertised support for this type of | |
1493 | * request, that does not mean every target supports it, and | |
1494 | * reconfiguration might also have changed that since the | |
1495 | * check was performed. | |
1496 | */ | |
3d7f4562 MS |
1497 | if (!num_bios) |
1498 | return -EOPNOTSUPP; | |
ba1cbad9 | 1499 | |
51b86f9a ML |
1500 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
1501 | ||
3d7f4562 | 1502 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1503 | |
3d7f4562 MS |
1504 | ci->sector += len; |
1505 | ci->sector_count -= len; | |
5ae89a87 MS |
1506 | |
1507 | return 0; | |
ba1cbad9 MS |
1508 | } |
1509 | ||
3d7f4562 | 1510 | static int __send_discard(struct clone_info *ci, struct dm_target *ti) |
23508a96 | 1511 | { |
61697a6a | 1512 | return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti)); |
23508a96 | 1513 | } |
0ce65797 | 1514 | |
00716545 DS |
1515 | static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti) |
1516 | { | |
61697a6a | 1517 | return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti)); |
00716545 DS |
1518 | } |
1519 | ||
3d7f4562 | 1520 | static int __send_write_same(struct clone_info *ci, struct dm_target *ti) |
0ce65797 | 1521 | { |
61697a6a | 1522 | return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti)); |
0ce65797 MS |
1523 | } |
1524 | ||
3d7f4562 | 1525 | static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti) |
ac62d620 | 1526 | { |
61697a6a | 1527 | return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti)); |
ac62d620 CH |
1528 | } |
1529 | ||
568c73a3 MS |
1530 | static bool is_abnormal_io(struct bio *bio) |
1531 | { | |
1532 | bool r = false; | |
1533 | ||
1534 | switch (bio_op(bio)) { | |
1535 | case REQ_OP_DISCARD: | |
1536 | case REQ_OP_SECURE_ERASE: | |
1537 | case REQ_OP_WRITE_SAME: | |
1538 | case REQ_OP_WRITE_ZEROES: | |
1539 | r = true; | |
1540 | break; | |
1541 | } | |
1542 | ||
1543 | return r; | |
1544 | } | |
1545 | ||
0519c71e MS |
1546 | static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti, |
1547 | int *result) | |
1548 | { | |
1549 | struct bio *bio = ci->bio; | |
1550 | ||
1551 | if (bio_op(bio) == REQ_OP_DISCARD) | |
1552 | *result = __send_discard(ci, ti); | |
00716545 DS |
1553 | else if (bio_op(bio) == REQ_OP_SECURE_ERASE) |
1554 | *result = __send_secure_erase(ci, ti); | |
0519c71e MS |
1555 | else if (bio_op(bio) == REQ_OP_WRITE_SAME) |
1556 | *result = __send_write_same(ci, ti); | |
1557 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES) | |
1558 | *result = __send_write_zeroes(ci, ti); | |
1559 | else | |
1560 | return false; | |
1561 | ||
1562 | return true; | |
1563 | } | |
1564 | ||
e4c93811 AK |
1565 | /* |
1566 | * Select the correct strategy for processing a non-flush bio. | |
1567 | */ | |
14fe594d | 1568 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1569 | { |
512875bd | 1570 | struct dm_target *ti; |
1c3b13e6 | 1571 | unsigned len; |
c80914e8 | 1572 | int r; |
0ce65797 | 1573 | |
512875bd | 1574 | ti = dm_table_find_target(ci->map, ci->sector); |
123d87d5 | 1575 | if (!ti) |
512875bd JN |
1576 | return -EIO; |
1577 | ||
568c73a3 | 1578 | if (__process_abnormal_io(ci, ti, &r)) |
0519c71e | 1579 | return r; |
3d7f4562 | 1580 | |
e76239a3 | 1581 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
0ce65797 | 1582 | |
c80914e8 MS |
1583 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1584 | if (r < 0) | |
1585 | return r; | |
0ce65797 | 1586 | |
1c3b13e6 KO |
1587 | ci->sector += len; |
1588 | ci->sector_count -= len; | |
0ce65797 | 1589 | |
1c3b13e6 | 1590 | return 0; |
0ce65797 MS |
1591 | } |
1592 | ||
978e51ba MS |
1593 | static void init_clone_info(struct clone_info *ci, struct mapped_device *md, |
1594 | struct dm_table *map, struct bio *bio) | |
1595 | { | |
1596 | ci->map = map; | |
1597 | ci->io = alloc_io(md, bio); | |
1598 | ci->sector = bio->bi_iter.bi_sector; | |
1599 | } | |
1600 | ||
a1e1cb72 MS |
1601 | #define __dm_part_stat_sub(part, field, subnd) \ |
1602 | (part_stat_get(part, field) -= (subnd)) | |
1603 | ||
1da177e4 | 1604 | /* |
14fe594d | 1605 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1606 | */ |
978e51ba MS |
1607 | static blk_qc_t __split_and_process_bio(struct mapped_device *md, |
1608 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1609 | { |
1da177e4 | 1610 | struct clone_info ci; |
978e51ba | 1611 | blk_qc_t ret = BLK_QC_T_NONE; |
512875bd | 1612 | int error = 0; |
1da177e4 | 1613 | |
978e51ba | 1614 | init_clone_info(&ci, md, map, bio); |
0ce65797 | 1615 | |
1eff9d32 | 1616 | if (bio->bi_opf & REQ_PREFLUSH) { |
dbe3ece1 JA |
1617 | struct bio flush_bio; |
1618 | ||
1619 | /* | |
1620 | * Use an on-stack bio for this, it's safe since we don't | |
1621 | * need to reference it after submit. It's just used as | |
1622 | * the basis for the clone(s). | |
1623 | */ | |
1624 | bio_init(&flush_bio, NULL, 0); | |
1625 | flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; | |
1626 | ci.bio = &flush_bio; | |
b372d360 | 1627 | ci.sector_count = 0; |
14fe594d | 1628 | error = __send_empty_flush(&ci); |
b372d360 | 1629 | /* dec_pending submits any data associated with flush */ |
a4aa5e56 DLM |
1630 | } else if (bio_op(bio) == REQ_OP_ZONE_RESET) { |
1631 | ci.bio = bio; | |
1632 | ci.sector_count = 0; | |
1633 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1634 | } else { |
6a8736d1 | 1635 | ci.bio = bio; |
d87f4c14 | 1636 | ci.sector_count = bio_sectors(bio); |
18a25da8 | 1637 | while (ci.sector_count && !error) { |
14fe594d | 1638 | error = __split_and_process_non_flush(&ci); |
18a25da8 N |
1639 | if (current->bio_list && ci.sector_count && !error) { |
1640 | /* | |
1641 | * Remainder must be passed to generic_make_request() | |
1642 | * so that it gets handled *after* bios already submitted | |
1643 | * have been completely processed. | |
1644 | * We take a clone of the original to store in | |
745dc570 | 1645 | * ci.io->orig_bio to be used by end_io_acct() and |
18a25da8 | 1646 | * for dec_pending to use for completion handling. |
18a25da8 | 1647 | */ |
f21c601a MS |
1648 | struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count, |
1649 | GFP_NOIO, &md->queue->bio_split); | |
745dc570 | 1650 | ci.io->orig_bio = b; |
a1e1cb72 MS |
1651 | |
1652 | /* | |
1653 | * Adjust IO stats for each split, otherwise upon queue | |
1654 | * reentry there will be redundant IO accounting. | |
1655 | * NOTE: this is a stop-gap fix, a proper fix involves | |
1656 | * significant refactoring of DM core's bio splitting | |
1657 | * (by eliminating DM's splitting and just using bio_split) | |
1658 | */ | |
1659 | part_stat_lock(); | |
1660 | __dm_part_stat_sub(&dm_disk(md)->part0, | |
1661 | sectors[op_stat_group(bio_op(bio))], ci.sector_count); | |
1662 | part_stat_unlock(); | |
1663 | ||
18a25da8 | 1664 | bio_chain(b, bio); |
075c18c3 | 1665 | trace_block_split(md->queue, b, bio->bi_iter.bi_sector); |
978e51ba | 1666 | ret = generic_make_request(bio); |
18a25da8 N |
1667 | break; |
1668 | } | |
1669 | } | |
d87f4c14 | 1670 | } |
0ce65797 | 1671 | |
1da177e4 | 1672 | /* drop the extra reference count */ |
54385bf7 | 1673 | dec_pending(ci.io, errno_to_blk_status(error)); |
978e51ba | 1674 | return ret; |
0ce65797 MS |
1675 | } |
1676 | ||
cec47e3d | 1677 | /* |
978e51ba MS |
1678 | * Optimized variant of __split_and_process_bio that leverages the |
1679 | * fact that targets that use it do _not_ have a need to split bios. | |
cec47e3d | 1680 | */ |
568c73a3 MS |
1681 | static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map, |
1682 | struct bio *bio, struct dm_target *ti) | |
978e51ba MS |
1683 | { |
1684 | struct clone_info ci; | |
1685 | blk_qc_t ret = BLK_QC_T_NONE; | |
1686 | int error = 0; | |
1687 | ||
978e51ba MS |
1688 | init_clone_info(&ci, md, map, bio); |
1689 | ||
1690 | if (bio->bi_opf & REQ_PREFLUSH) { | |
dbe3ece1 JA |
1691 | struct bio flush_bio; |
1692 | ||
1693 | /* | |
1694 | * Use an on-stack bio for this, it's safe since we don't | |
1695 | * need to reference it after submit. It's just used as | |
1696 | * the basis for the clone(s). | |
1697 | */ | |
1698 | bio_init(&flush_bio, NULL, 0); | |
1699 | flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; | |
1700 | ci.bio = &flush_bio; | |
978e51ba MS |
1701 | ci.sector_count = 0; |
1702 | error = __send_empty_flush(&ci); | |
1703 | /* dec_pending submits any data associated with flush */ | |
1704 | } else { | |
978e51ba MS |
1705 | struct dm_target_io *tio; |
1706 | ||
978e51ba MS |
1707 | ci.bio = bio; |
1708 | ci.sector_count = bio_sectors(bio); | |
568c73a3 | 1709 | if (__process_abnormal_io(&ci, ti, &error)) |
0519c71e MS |
1710 | goto out; |
1711 | ||
1712 | tio = alloc_tio(&ci, ti, 0, GFP_NOIO); | |
978e51ba MS |
1713 | ret = __clone_and_map_simple_bio(&ci, tio, NULL); |
1714 | } | |
1715 | out: | |
1716 | /* drop the extra reference count */ | |
1717 | dec_pending(ci.io, errno_to_blk_status(error)); | |
1718 | return ret; | |
1719 | } | |
1720 | ||
568c73a3 MS |
1721 | static void dm_queue_split(struct mapped_device *md, struct dm_target *ti, struct bio **bio) |
1722 | { | |
1723 | unsigned len, sector_count; | |
1724 | ||
1725 | sector_count = bio_sectors(*bio); | |
1726 | len = min_t(sector_t, max_io_len((*bio)->bi_iter.bi_sector, ti), sector_count); | |
1727 | ||
1728 | if (sector_count > len) { | |
1729 | struct bio *split = bio_split(*bio, len, GFP_NOIO, &md->queue->bio_split); | |
1730 | ||
1731 | bio_chain(split, *bio); | |
1732 | trace_block_split(md->queue, split, (*bio)->bi_iter.bi_sector); | |
1733 | generic_make_request(*bio); | |
1734 | *bio = split; | |
1735 | } | |
1736 | } | |
1737 | ||
6548c7c5 MS |
1738 | static blk_qc_t dm_process_bio(struct mapped_device *md, |
1739 | struct dm_table *map, struct bio *bio) | |
1740 | { | |
568c73a3 MS |
1741 | blk_qc_t ret = BLK_QC_T_NONE; |
1742 | struct dm_target *ti = md->immutable_target; | |
1743 | ||
1744 | if (unlikely(!map)) { | |
1745 | bio_io_error(bio); | |
1746 | return ret; | |
1747 | } | |
1748 | ||
1749 | if (!ti) { | |
1750 | ti = dm_table_find_target(map, bio->bi_iter.bi_sector); | |
123d87d5 | 1751 | if (unlikely(!ti)) { |
568c73a3 MS |
1752 | bio_io_error(bio); |
1753 | return ret; | |
1754 | } | |
1755 | } | |
1756 | ||
1757 | /* | |
1758 | * If in ->make_request_fn we need to use blk_queue_split(), otherwise | |
1759 | * queue_limits for abnormal requests (e.g. discard, writesame, etc) | |
1760 | * won't be imposed. | |
1761 | */ | |
1762 | if (current->bio_list) { | |
effd58c9 MS |
1763 | blk_queue_split(md->queue, &bio); |
1764 | if (!is_abnormal_io(bio)) | |
568c73a3 MS |
1765 | dm_queue_split(md, ti, &bio); |
1766 | } | |
1767 | ||
6548c7c5 | 1768 | if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED) |
568c73a3 | 1769 | return __process_bio(md, map, bio, ti); |
6548c7c5 MS |
1770 | else |
1771 | return __split_and_process_bio(md, map, bio); | |
1772 | } | |
1773 | ||
24113d48 | 1774 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1775 | { |
1776 | struct mapped_device *md = q->queuedata; | |
978e51ba | 1777 | blk_qc_t ret = BLK_QC_T_NONE; |
83d5e5b0 MP |
1778 | int srcu_idx; |
1779 | struct dm_table *map; | |
cec47e3d | 1780 | |
83d5e5b0 | 1781 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1782 | |
6a8736d1 TH |
1783 | /* if we're suspended, we have to queue this io for later */ |
1784 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1785 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1786 | |
1eff9d32 | 1787 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1788 | queue_io(md, bio); |
1789 | else | |
54d9a1b4 | 1790 | bio_io_error(bio); |
978e51ba | 1791 | return ret; |
cec47e3d | 1792 | } |
1da177e4 | 1793 | |
6548c7c5 | 1794 | ret = dm_process_bio(md, map, bio); |
978e51ba | 1795 | |
83d5e5b0 | 1796 | dm_put_live_table(md, srcu_idx); |
978e51ba MS |
1797 | return ret; |
1798 | } | |
1799 | ||
1da177e4 LT |
1800 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1801 | { | |
8a57dfc6 CS |
1802 | int r = bdi_bits; |
1803 | struct mapped_device *md = congested_data; | |
1804 | struct dm_table *map; | |
1da177e4 | 1805 | |
1eb787ec | 1806 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1807 | if (dm_request_based(md)) { |
cec47e3d | 1808 | /* |
e522c039 MS |
1809 | * With request-based DM we only need to check the |
1810 | * top-level queue for congestion. | |
cec47e3d | 1811 | */ |
dc3b17cc | 1812 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1813 | } else { |
1814 | map = dm_get_live_table_fast(md); | |
1815 | if (map) | |
cec47e3d | 1816 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1817 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1818 | } |
1819 | } | |
1820 | ||
1da177e4 LT |
1821 | return r; |
1822 | } | |
1823 | ||
1824 | /*----------------------------------------------------------------- | |
1825 | * An IDR is used to keep track of allocated minor numbers. | |
1826 | *---------------------------------------------------------------*/ | |
2b06cfff | 1827 | static void free_minor(int minor) |
1da177e4 | 1828 | { |
f32c10b0 | 1829 | spin_lock(&_minor_lock); |
1da177e4 | 1830 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1831 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1832 | } |
1833 | ||
1834 | /* | |
1835 | * See if the device with a specific minor # is free. | |
1836 | */ | |
cf13ab8e | 1837 | static int specific_minor(int minor) |
1da177e4 | 1838 | { |
c9d76be6 | 1839 | int r; |
1da177e4 LT |
1840 | |
1841 | if (minor >= (1 << MINORBITS)) | |
1842 | return -EINVAL; | |
1843 | ||
c9d76be6 | 1844 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1845 | spin_lock(&_minor_lock); |
1da177e4 | 1846 | |
c9d76be6 | 1847 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1848 | |
f32c10b0 | 1849 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1850 | idr_preload_end(); |
1851 | if (r < 0) | |
1852 | return r == -ENOSPC ? -EBUSY : r; | |
1853 | return 0; | |
1da177e4 LT |
1854 | } |
1855 | ||
cf13ab8e | 1856 | static int next_free_minor(int *minor) |
1da177e4 | 1857 | { |
c9d76be6 | 1858 | int r; |
62f75c2f | 1859 | |
c9d76be6 | 1860 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1861 | spin_lock(&_minor_lock); |
1da177e4 | 1862 | |
c9d76be6 | 1863 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1864 | |
f32c10b0 | 1865 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1866 | idr_preload_end(); |
1867 | if (r < 0) | |
1868 | return r; | |
1869 | *minor = r; | |
1870 | return 0; | |
1da177e4 LT |
1871 | } |
1872 | ||
83d5cde4 | 1873 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1874 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1875 | |
53d5914f MP |
1876 | static void dm_wq_work(struct work_struct *work); |
1877 | ||
c12c9a3c | 1878 | static void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1879 | { |
bfebd1cd MS |
1880 | /* |
1881 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1882 | */ | |
dc3b17cc | 1883 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf MS |
1884 | } |
1885 | ||
0f20972f MS |
1886 | static void cleanup_mapped_device(struct mapped_device *md) |
1887 | { | |
0f20972f MS |
1888 | if (md->wq) |
1889 | destroy_workqueue(md->wq); | |
6f1c819c KO |
1890 | bioset_exit(&md->bs); |
1891 | bioset_exit(&md->io_bs); | |
0f20972f | 1892 | |
f26c5719 DW |
1893 | if (md->dax_dev) { |
1894 | kill_dax(md->dax_dev); | |
1895 | put_dax(md->dax_dev); | |
1896 | md->dax_dev = NULL; | |
1897 | } | |
1898 | ||
0f20972f MS |
1899 | if (md->disk) { |
1900 | spin_lock(&_minor_lock); | |
1901 | md->disk->private_data = NULL; | |
1902 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1903 | del_gendisk(md->disk); |
1904 | put_disk(md->disk); | |
1905 | } | |
1906 | ||
1907 | if (md->queue) | |
1908 | blk_cleanup_queue(md->queue); | |
1909 | ||
d09960b0 TE |
1910 | cleanup_srcu_struct(&md->io_barrier); |
1911 | ||
0f20972f MS |
1912 | if (md->bdev) { |
1913 | bdput(md->bdev); | |
1914 | md->bdev = NULL; | |
1915 | } | |
4cc96131 | 1916 | |
d5ffebdd MS |
1917 | mutex_destroy(&md->suspend_lock); |
1918 | mutex_destroy(&md->type_lock); | |
1919 | mutex_destroy(&md->table_devices_lock); | |
1920 | ||
4cc96131 | 1921 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
1922 | } |
1923 | ||
1da177e4 LT |
1924 | /* |
1925 | * Allocate and initialise a blank device with a given minor. | |
1926 | */ | |
2b06cfff | 1927 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1928 | { |
115485e8 MS |
1929 | int r, numa_node_id = dm_get_numa_node(); |
1930 | struct mapped_device *md; | |
ba61fdd1 | 1931 | void *old_md; |
1da177e4 | 1932 | |
856eb091 | 1933 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1934 | if (!md) { |
1935 | DMWARN("unable to allocate device, out of memory."); | |
1936 | return NULL; | |
1937 | } | |
1938 | ||
10da4f79 | 1939 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1940 | goto bad_module_get; |
10da4f79 | 1941 | |
1da177e4 | 1942 | /* get a minor number for the dev */ |
2b06cfff | 1943 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1944 | r = next_free_minor(&minor); |
2b06cfff | 1945 | else |
cf13ab8e | 1946 | r = specific_minor(minor); |
1da177e4 | 1947 | if (r < 0) |
6ed7ade8 | 1948 | goto bad_minor; |
1da177e4 | 1949 | |
83d5e5b0 MP |
1950 | r = init_srcu_struct(&md->io_barrier); |
1951 | if (r < 0) | |
1952 | goto bad_io_barrier; | |
1953 | ||
115485e8 | 1954 | md->numa_node_id = numa_node_id; |
591ddcfc | 1955 | md->init_tio_pdu = false; |
a5664dad | 1956 | md->type = DM_TYPE_NONE; |
e61290a4 | 1957 | mutex_init(&md->suspend_lock); |
a5664dad | 1958 | mutex_init(&md->type_lock); |
86f1152b | 1959 | mutex_init(&md->table_devices_lock); |
022c2611 | 1960 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1961 | atomic_set(&md->holders, 1); |
5c6bd75d | 1962 | atomic_set(&md->open_count, 0); |
1da177e4 | 1963 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1964 | atomic_set(&md->uevent_seq, 0); |
1965 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1966 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1967 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1968 | |
6d469642 | 1969 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1970 | if (!md->queue) |
0f20972f | 1971 | goto bad; |
c12c9a3c MS |
1972 | md->queue->queuedata = md; |
1973 | md->queue->backing_dev_info->congested_data = md; | |
1da177e4 | 1974 | |
c12c9a3c | 1975 | md->disk = alloc_disk_node(1, md->numa_node_id); |
1da177e4 | 1976 | if (!md->disk) |
0f20972f | 1977 | goto bad; |
1da177e4 | 1978 | |
f0b04115 | 1979 | init_waitqueue_head(&md->wait); |
53d5914f | 1980 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1981 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1982 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 1983 | |
1da177e4 LT |
1984 | md->disk->major = _major; |
1985 | md->disk->first_minor = minor; | |
1986 | md->disk->fops = &dm_blk_dops; | |
1987 | md->disk->queue = md->queue; | |
1988 | md->disk->private_data = md; | |
1989 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 | 1990 | |
976431b0 | 1991 | if (IS_ENABLED(CONFIG_DAX_DRIVER)) { |
fefc1d97 PG |
1992 | md->dax_dev = alloc_dax(md, md->disk->disk_name, |
1993 | &dm_dax_ops, 0); | |
514cf4f8 | 1994 | if (!md->dax_dev) |
976431b0 DW |
1995 | goto bad; |
1996 | } | |
f26c5719 | 1997 | |
c100ec49 | 1998 | add_disk_no_queue_reg(md->disk); |
7e51f257 | 1999 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2000 | |
670368a8 | 2001 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 2002 | if (!md->wq) |
0f20972f | 2003 | goto bad; |
304f3f6a | 2004 | |
32a926da MP |
2005 | md->bdev = bdget_disk(md->disk, 0); |
2006 | if (!md->bdev) | |
0f20972f | 2007 | goto bad; |
32a926da | 2008 | |
fd2ed4d2 MP |
2009 | dm_stats_init(&md->stats); |
2010 | ||
ba61fdd1 | 2011 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2012 | spin_lock(&_minor_lock); |
ba61fdd1 | 2013 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2014 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2015 | |
2016 | BUG_ON(old_md != MINOR_ALLOCED); | |
2017 | ||
1da177e4 LT |
2018 | return md; |
2019 | ||
0f20972f MS |
2020 | bad: |
2021 | cleanup_mapped_device(md); | |
83d5e5b0 | 2022 | bad_io_barrier: |
1da177e4 | 2023 | free_minor(minor); |
6ed7ade8 | 2024 | bad_minor: |
10da4f79 | 2025 | module_put(THIS_MODULE); |
6ed7ade8 | 2026 | bad_module_get: |
856eb091 | 2027 | kvfree(md); |
1da177e4 LT |
2028 | return NULL; |
2029 | } | |
2030 | ||
ae9da83f JN |
2031 | static void unlock_fs(struct mapped_device *md); |
2032 | ||
1da177e4 LT |
2033 | static void free_dev(struct mapped_device *md) |
2034 | { | |
f331c029 | 2035 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2036 | |
32a926da | 2037 | unlock_fs(md); |
2eb6e1e3 | 2038 | |
0f20972f | 2039 | cleanup_mapped_device(md); |
63a4f065 | 2040 | |
86f1152b | 2041 | free_table_devices(&md->table_devices); |
63a4f065 | 2042 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2043 | free_minor(minor); |
2044 | ||
10da4f79 | 2045 | module_put(THIS_MODULE); |
856eb091 | 2046 | kvfree(md); |
1da177e4 LT |
2047 | } |
2048 | ||
2a2a4c51 | 2049 | static int __bind_mempools(struct mapped_device *md, struct dm_table *t) |
e6ee8c0b | 2050 | { |
c0820cf5 | 2051 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
2a2a4c51 | 2052 | int ret = 0; |
e6ee8c0b | 2053 | |
0776aa0e | 2054 | if (dm_table_bio_based(t)) { |
64f52b0e MS |
2055 | /* |
2056 | * The md may already have mempools that need changing. | |
2057 | * If so, reload bioset because front_pad may have changed | |
2058 | * because a different table was loaded. | |
2059 | */ | |
6f1c819c KO |
2060 | bioset_exit(&md->bs); |
2061 | bioset_exit(&md->io_bs); | |
0776aa0e | 2062 | |
6f1c819c | 2063 | } else if (bioset_initialized(&md->bs)) { |
4e6e36c3 MS |
2064 | /* |
2065 | * There's no need to reload with request-based dm | |
2066 | * because the size of front_pad doesn't change. | |
2067 | * Note for future: If you are to reload bioset, | |
2068 | * prep-ed requests in the queue may refer | |
2069 | * to bio from the old bioset, so you must walk | |
2070 | * through the queue to unprep. | |
2071 | */ | |
2072 | goto out; | |
c0820cf5 | 2073 | } |
e6ee8c0b | 2074 | |
6f1c819c KO |
2075 | BUG_ON(!p || |
2076 | bioset_initialized(&md->bs) || | |
2077 | bioset_initialized(&md->io_bs)); | |
cbc4e3c1 | 2078 | |
2a2a4c51 JA |
2079 | ret = bioset_init_from_src(&md->bs, &p->bs); |
2080 | if (ret) | |
2081 | goto out; | |
2082 | ret = bioset_init_from_src(&md->io_bs, &p->io_bs); | |
2083 | if (ret) | |
2084 | bioset_exit(&md->bs); | |
e6ee8c0b | 2085 | out: |
02233342 | 2086 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b | 2087 | dm_table_free_md_mempools(t); |
2a2a4c51 | 2088 | return ret; |
e6ee8c0b KU |
2089 | } |
2090 | ||
1da177e4 LT |
2091 | /* |
2092 | * Bind a table to the device. | |
2093 | */ | |
2094 | static void event_callback(void *context) | |
2095 | { | |
7a8c3d3b MA |
2096 | unsigned long flags; |
2097 | LIST_HEAD(uevents); | |
1da177e4 LT |
2098 | struct mapped_device *md = (struct mapped_device *) context; |
2099 | ||
7a8c3d3b MA |
2100 | spin_lock_irqsave(&md->uevent_lock, flags); |
2101 | list_splice_init(&md->uevent_list, &uevents); | |
2102 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2103 | ||
ed9e1982 | 2104 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2105 | |
1da177e4 LT |
2106 | atomic_inc(&md->event_nr); |
2107 | wake_up(&md->eventq); | |
62e08243 | 2108 | dm_issue_global_event(); |
1da177e4 LT |
2109 | } |
2110 | ||
c217649b MS |
2111 | /* |
2112 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2113 | */ | |
4e90188b | 2114 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2115 | { |
1ea0654e BVA |
2116 | lockdep_assert_held(&md->suspend_lock); |
2117 | ||
4e90188b | 2118 | set_capacity(md->disk, size); |
1da177e4 | 2119 | |
db8fef4f | 2120 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2121 | } |
2122 | ||
042d2a9b AK |
2123 | /* |
2124 | * Returns old map, which caller must destroy. | |
2125 | */ | |
2126 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2127 | struct queue_limits *limits) | |
1da177e4 | 2128 | { |
042d2a9b | 2129 | struct dm_table *old_map; |
165125e1 | 2130 | struct request_queue *q = md->queue; |
978e51ba | 2131 | bool request_based = dm_table_request_based(t); |
1da177e4 | 2132 | sector_t size; |
2a2a4c51 | 2133 | int ret; |
1da177e4 | 2134 | |
5a8f1f80 BVA |
2135 | lockdep_assert_held(&md->suspend_lock); |
2136 | ||
1da177e4 | 2137 | size = dm_table_get_size(t); |
3ac51e74 DW |
2138 | |
2139 | /* | |
2140 | * Wipe any geometry if the size of the table changed. | |
2141 | */ | |
fd2ed4d2 | 2142 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2143 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2144 | ||
32a926da | 2145 | __set_size(md, size); |
d5816876 | 2146 | |
2ca3310e AK |
2147 | dm_table_event_callback(t, event_callback, md); |
2148 | ||
e6ee8c0b KU |
2149 | /* |
2150 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2151 | * for request-based during suspension. So stop it to prevent | |
2152 | * I/O mapping before resume. | |
2153 | * This must be done before setting the queue restrictions, | |
2154 | * because request-based dm may be run just after the setting. | |
2155 | */ | |
978e51ba | 2156 | if (request_based) |
eca7ee6d | 2157 | dm_stop_queue(q); |
978e51ba MS |
2158 | |
2159 | if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) { | |
16f12266 | 2160 | /* |
978e51ba MS |
2161 | * Leverage the fact that request-based DM targets and |
2162 | * NVMe bio based targets are immutable singletons | |
2163 | * - used to optimize both dm_request_fn and dm_mq_queue_rq; | |
2164 | * and __process_bio. | |
16f12266 MS |
2165 | */ |
2166 | md->immutable_target = dm_table_get_immutable_target(t); | |
2167 | } | |
e6ee8c0b | 2168 | |
2a2a4c51 JA |
2169 | ret = __bind_mempools(md, t); |
2170 | if (ret) { | |
2171 | old_map = ERR_PTR(ret); | |
2172 | goto out; | |
2173 | } | |
e6ee8c0b | 2174 | |
a12f5d48 | 2175 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2176 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2177 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2178 | ||
754c5fc7 | 2179 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2180 | if (old_map) |
2181 | dm_sync_table(md); | |
1da177e4 | 2182 | |
2a2a4c51 | 2183 | out: |
042d2a9b | 2184 | return old_map; |
1da177e4 LT |
2185 | } |
2186 | ||
a7940155 AK |
2187 | /* |
2188 | * Returns unbound table for the caller to free. | |
2189 | */ | |
2190 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2191 | { |
a12f5d48 | 2192 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2193 | |
2194 | if (!map) | |
a7940155 | 2195 | return NULL; |
1da177e4 LT |
2196 | |
2197 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2198 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2199 | dm_sync_table(md); |
a7940155 AK |
2200 | |
2201 | return map; | |
1da177e4 LT |
2202 | } |
2203 | ||
2204 | /* | |
2205 | * Constructor for a new device. | |
2206 | */ | |
2b06cfff | 2207 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 | 2208 | { |
c12c9a3c | 2209 | int r; |
1da177e4 LT |
2210 | struct mapped_device *md; |
2211 | ||
2b06cfff | 2212 | md = alloc_dev(minor); |
1da177e4 LT |
2213 | if (!md) |
2214 | return -ENXIO; | |
2215 | ||
c12c9a3c MS |
2216 | r = dm_sysfs_init(md); |
2217 | if (r) { | |
2218 | free_dev(md); | |
2219 | return r; | |
2220 | } | |
784aae73 | 2221 | |
1da177e4 LT |
2222 | *result = md; |
2223 | return 0; | |
2224 | } | |
2225 | ||
a5664dad MS |
2226 | /* |
2227 | * Functions to manage md->type. | |
2228 | * All are required to hold md->type_lock. | |
2229 | */ | |
2230 | void dm_lock_md_type(struct mapped_device *md) | |
2231 | { | |
2232 | mutex_lock(&md->type_lock); | |
2233 | } | |
2234 | ||
2235 | void dm_unlock_md_type(struct mapped_device *md) | |
2236 | { | |
2237 | mutex_unlock(&md->type_lock); | |
2238 | } | |
2239 | ||
7e0d574f | 2240 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2241 | { |
00c4fc3b | 2242 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2243 | md->type = type; |
2244 | } | |
2245 | ||
7e0d574f | 2246 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2247 | { |
2248 | return md->type; | |
2249 | } | |
2250 | ||
36a0456f AK |
2251 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2252 | { | |
2253 | return md->immutable_target_type; | |
2254 | } | |
2255 | ||
f84cb8a4 MS |
2256 | /* |
2257 | * The queue_limits are only valid as long as you have a reference | |
2258 | * count on 'md'. | |
2259 | */ | |
2260 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2261 | { | |
2262 | BUG_ON(!atomic_read(&md->holders)); | |
2263 | return &md->queue->limits; | |
2264 | } | |
2265 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2266 | ||
4a0b4ddf MS |
2267 | /* |
2268 | * Setup the DM device's queue based on md's type | |
2269 | */ | |
591ddcfc | 2270 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2271 | { |
bfebd1cd | 2272 | int r; |
c100ec49 | 2273 | struct queue_limits limits; |
7e0d574f | 2274 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2275 | |
545ed20e | 2276 | switch (type) { |
bfebd1cd | 2277 | case DM_TYPE_REQUEST_BASED: |
e83068a5 | 2278 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2279 | if (r) { |
eca7ee6d | 2280 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2281 | return r; |
2282 | } | |
2283 | break; | |
2284 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2285 | case DM_TYPE_DAX_BIO_BASED: |
978e51ba MS |
2286 | case DM_TYPE_NVME_BIO_BASED: |
2287 | dm_init_normal_md_queue(md); | |
24113d48 | 2288 | blk_queue_make_request(md->queue, dm_make_request); |
bfebd1cd | 2289 | break; |
7e0d574f BVA |
2290 | case DM_TYPE_NONE: |
2291 | WARN_ON_ONCE(true); | |
2292 | break; | |
4a0b4ddf MS |
2293 | } |
2294 | ||
c100ec49 MS |
2295 | r = dm_calculate_queue_limits(t, &limits); |
2296 | if (r) { | |
2297 | DMERR("Cannot calculate initial queue limits"); | |
2298 | return r; | |
2299 | } | |
2300 | dm_table_set_restrictions(t, md->queue, &limits); | |
2301 | blk_register_queue(md->disk); | |
2302 | ||
4a0b4ddf MS |
2303 | return 0; |
2304 | } | |
2305 | ||
2bec1f4a | 2306 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2307 | { |
2308 | struct mapped_device *md; | |
1da177e4 LT |
2309 | unsigned minor = MINOR(dev); |
2310 | ||
2311 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2312 | return NULL; | |
2313 | ||
f32c10b0 | 2314 | spin_lock(&_minor_lock); |
1da177e4 LT |
2315 | |
2316 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2317 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2318 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2319 | md = NULL; | |
2320 | goto out; | |
fba9f90e | 2321 | } |
49de5769 | 2322 | dm_get(md); |
fba9f90e | 2323 | out: |
f32c10b0 | 2324 | spin_unlock(&_minor_lock); |
1da177e4 | 2325 | |
637842cf DT |
2326 | return md; |
2327 | } | |
3cf2e4ba | 2328 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2329 | |
9ade92a9 | 2330 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2331 | { |
9ade92a9 | 2332 | return md->interface_ptr; |
1da177e4 LT |
2333 | } |
2334 | ||
2335 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2336 | { | |
2337 | md->interface_ptr = ptr; | |
2338 | } | |
2339 | ||
2340 | void dm_get(struct mapped_device *md) | |
2341 | { | |
2342 | atomic_inc(&md->holders); | |
3f77316d | 2343 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2344 | } |
2345 | ||
09ee96b2 MP |
2346 | int dm_hold(struct mapped_device *md) |
2347 | { | |
2348 | spin_lock(&_minor_lock); | |
2349 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2350 | spin_unlock(&_minor_lock); | |
2351 | return -EBUSY; | |
2352 | } | |
2353 | dm_get(md); | |
2354 | spin_unlock(&_minor_lock); | |
2355 | return 0; | |
2356 | } | |
2357 | EXPORT_SYMBOL_GPL(dm_hold); | |
2358 | ||
72d94861 AK |
2359 | const char *dm_device_name(struct mapped_device *md) |
2360 | { | |
2361 | return md->name; | |
2362 | } | |
2363 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2364 | ||
3f77316d | 2365 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2366 | { |
1134e5ae | 2367 | struct dm_table *map; |
83d5e5b0 | 2368 | int srcu_idx; |
1da177e4 | 2369 | |
3f77316d | 2370 | might_sleep(); |
fba9f90e | 2371 | |
63a4f065 | 2372 | spin_lock(&_minor_lock); |
3f77316d KU |
2373 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2374 | set_bit(DMF_FREEING, &md->flags); | |
2375 | spin_unlock(&_minor_lock); | |
3b785fbc | 2376 | |
c12c9a3c | 2377 | blk_set_queue_dying(md->queue); |
3f77316d | 2378 | |
ab7c7bb6 MP |
2379 | /* |
2380 | * Take suspend_lock so that presuspend and postsuspend methods | |
2381 | * do not race with internal suspend. | |
2382 | */ | |
2383 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2384 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2385 | if (!dm_suspended_md(md)) { |
2386 | dm_table_presuspend_targets(map); | |
2387 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2388 | } |
83d5e5b0 MP |
2389 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2390 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2391 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2392 | |
3f77316d KU |
2393 | /* |
2394 | * Rare, but there may be I/O requests still going to complete, | |
2395 | * for example. Wait for all references to disappear. | |
2396 | * No one should increment the reference count of the mapped_device, | |
2397 | * after the mapped_device state becomes DMF_FREEING. | |
2398 | */ | |
2399 | if (wait) | |
2400 | while (atomic_read(&md->holders)) | |
2401 | msleep(1); | |
2402 | else if (atomic_read(&md->holders)) | |
2403 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2404 | dm_device_name(md), atomic_read(&md->holders)); | |
2405 | ||
2406 | dm_sysfs_exit(md); | |
3f77316d KU |
2407 | dm_table_destroy(__unbind(md)); |
2408 | free_dev(md); | |
2409 | } | |
2410 | ||
2411 | void dm_destroy(struct mapped_device *md) | |
2412 | { | |
2413 | __dm_destroy(md, true); | |
2414 | } | |
2415 | ||
2416 | void dm_destroy_immediate(struct mapped_device *md) | |
2417 | { | |
2418 | __dm_destroy(md, false); | |
2419 | } | |
2420 | ||
2421 | void dm_put(struct mapped_device *md) | |
2422 | { | |
2423 | atomic_dec(&md->holders); | |
1da177e4 | 2424 | } |
79eb885c | 2425 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2426 | |
b48633f8 | 2427 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2428 | { |
2429 | int r = 0; | |
9f4c3f87 | 2430 | DEFINE_WAIT(wait); |
46125c1c MB |
2431 | |
2432 | while (1) { | |
9f4c3f87 | 2433 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2434 | |
b4324fee | 2435 | if (!md_in_flight(md)) |
46125c1c MB |
2436 | break; |
2437 | ||
e3fabdfd | 2438 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2439 | r = -EINTR; |
2440 | break; | |
2441 | } | |
2442 | ||
2443 | io_schedule(); | |
2444 | } | |
9f4c3f87 | 2445 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2446 | |
46125c1c MB |
2447 | return r; |
2448 | } | |
2449 | ||
1da177e4 LT |
2450 | /* |
2451 | * Process the deferred bios | |
2452 | */ | |
ef208587 | 2453 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2454 | { |
ef208587 MP |
2455 | struct mapped_device *md = container_of(work, struct mapped_device, |
2456 | work); | |
6d6f10df | 2457 | struct bio *c; |
83d5e5b0 MP |
2458 | int srcu_idx; |
2459 | struct dm_table *map; | |
1da177e4 | 2460 | |
83d5e5b0 | 2461 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2462 | |
3b00b203 | 2463 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2464 | spin_lock_irq(&md->deferred_lock); |
2465 | c = bio_list_pop(&md->deferred); | |
2466 | spin_unlock_irq(&md->deferred_lock); | |
2467 | ||
6a8736d1 | 2468 | if (!c) |
df12ee99 | 2469 | break; |
022c2611 | 2470 | |
e6ee8c0b | 2471 | if (dm_request_based(md)) |
6548c7c5 | 2472 | (void) generic_make_request(c); |
6a8736d1 | 2473 | else |
6548c7c5 | 2474 | (void) dm_process_bio(md, map, c); |
022c2611 | 2475 | } |
73d410c0 | 2476 | |
83d5e5b0 | 2477 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2478 | } |
2479 | ||
9a1fb464 | 2480 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2481 | { |
3b00b203 | 2482 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2483 | smp_mb__after_atomic(); |
53d5914f | 2484 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2485 | } |
2486 | ||
1da177e4 | 2487 | /* |
042d2a9b | 2488 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2489 | */ |
042d2a9b | 2490 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2491 | { |
87eb5b21 | 2492 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2493 | struct queue_limits limits; |
042d2a9b | 2494 | int r; |
1da177e4 | 2495 | |
e61290a4 | 2496 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2497 | |
2498 | /* device must be suspended */ | |
4f186f8b | 2499 | if (!dm_suspended_md(md)) |
93c534ae | 2500 | goto out; |
1da177e4 | 2501 | |
3ae70656 MS |
2502 | /* |
2503 | * If the new table has no data devices, retain the existing limits. | |
2504 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2505 | * then new I/O is queued based on these limits, and then some paths | |
2506 | * reappear. | |
2507 | */ | |
2508 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2509 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2510 | if (live_map) |
2511 | limits = md->queue->limits; | |
83d5e5b0 | 2512 | dm_put_live_table_fast(md); |
3ae70656 MS |
2513 | } |
2514 | ||
87eb5b21 MC |
2515 | if (!live_map) { |
2516 | r = dm_calculate_queue_limits(table, &limits); | |
2517 | if (r) { | |
2518 | map = ERR_PTR(r); | |
2519 | goto out; | |
2520 | } | |
042d2a9b | 2521 | } |
754c5fc7 | 2522 | |
042d2a9b | 2523 | map = __bind(md, table, &limits); |
62e08243 | 2524 | dm_issue_global_event(); |
1da177e4 | 2525 | |
93c534ae | 2526 | out: |
e61290a4 | 2527 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2528 | return map; |
1da177e4 LT |
2529 | } |
2530 | ||
2531 | /* | |
2532 | * Functions to lock and unlock any filesystem running on the | |
2533 | * device. | |
2534 | */ | |
2ca3310e | 2535 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2536 | { |
e39e2e95 | 2537 | int r; |
1da177e4 LT |
2538 | |
2539 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2540 | |
db8fef4f | 2541 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2542 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2543 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2544 | md->frozen_sb = NULL; |
2545 | return r; | |
dfbe03f6 AK |
2546 | } |
2547 | ||
aa8d7c2f AK |
2548 | set_bit(DMF_FROZEN, &md->flags); |
2549 | ||
1da177e4 LT |
2550 | return 0; |
2551 | } | |
2552 | ||
2ca3310e | 2553 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2554 | { |
aa8d7c2f AK |
2555 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2556 | return; | |
2557 | ||
db8fef4f | 2558 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2559 | md->frozen_sb = NULL; |
aa8d7c2f | 2560 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2561 | } |
2562 | ||
2563 | /* | |
b48633f8 BVA |
2564 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2565 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2566 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2567 | * | |
ffcc3936 MS |
2568 | * If __dm_suspend returns 0, the device is completely quiescent |
2569 | * now. There is no request-processing activity. All new requests | |
2570 | * are being added to md->deferred list. | |
cec47e3d | 2571 | */ |
ffcc3936 | 2572 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2573 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2574 | int dmf_suspended_flag) |
1da177e4 | 2575 | { |
ffcc3936 MS |
2576 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2577 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2578 | int r; | |
1da177e4 | 2579 | |
5a8f1f80 BVA |
2580 | lockdep_assert_held(&md->suspend_lock); |
2581 | ||
2e93ccc1 KU |
2582 | /* |
2583 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2584 | * This flag is cleared before dm_suspend returns. | |
2585 | */ | |
2586 | if (noflush) | |
2587 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 BVA |
2588 | else |
2589 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); | |
2e93ccc1 | 2590 | |
d67ee213 MS |
2591 | /* |
2592 | * This gets reverted if there's an error later and the targets | |
2593 | * provide the .presuspend_undo hook. | |
2594 | */ | |
cf222b37 AK |
2595 | dm_table_presuspend_targets(map); |
2596 | ||
32a926da | 2597 | /* |
9f518b27 KU |
2598 | * Flush I/O to the device. |
2599 | * Any I/O submitted after lock_fs() may not be flushed. | |
2600 | * noflush takes precedence over do_lockfs. | |
2601 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2602 | */ |
2603 | if (!noflush && do_lockfs) { | |
2604 | r = lock_fs(md); | |
d67ee213 MS |
2605 | if (r) { |
2606 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2607 | return r; |
d67ee213 | 2608 | } |
aa8d7c2f | 2609 | } |
1da177e4 LT |
2610 | |
2611 | /* | |
3b00b203 MP |
2612 | * Here we must make sure that no processes are submitting requests |
2613 | * to target drivers i.e. no one may be executing | |
2614 | * __split_and_process_bio. This is called from dm_request and | |
2615 | * dm_wq_work. | |
2616 | * | |
2617 | * To get all processes out of __split_and_process_bio in dm_request, | |
2618 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2619 | * __split_and_process_bio from dm_request and quiesce the thread |
2620 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2621 | * flush_workqueue(md->wq). | |
1da177e4 | 2622 | */ |
1eb787ec | 2623 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2624 | if (map) |
2625 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2626 | |
d0bcb878 | 2627 | /* |
29e4013d TH |
2628 | * Stop md->queue before flushing md->wq in case request-based |
2629 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2630 | */ |
6a23e05c | 2631 | if (dm_request_based(md)) |
eca7ee6d | 2632 | dm_stop_queue(md->queue); |
cec47e3d | 2633 | |
d0bcb878 KU |
2634 | flush_workqueue(md->wq); |
2635 | ||
1da177e4 | 2636 | /* |
3b00b203 MP |
2637 | * At this point no more requests are entering target request routines. |
2638 | * We call dm_wait_for_completion to wait for all existing requests | |
2639 | * to finish. | |
1da177e4 | 2640 | */ |
b48633f8 | 2641 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2642 | if (!r) |
2643 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2644 | |
6d6f10df | 2645 | if (noflush) |
022c2611 | 2646 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2647 | if (map) |
2648 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2649 | |
1da177e4 | 2650 | /* were we interrupted ? */ |
46125c1c | 2651 | if (r < 0) { |
9a1fb464 | 2652 | dm_queue_flush(md); |
73d410c0 | 2653 | |
cec47e3d | 2654 | if (dm_request_based(md)) |
eca7ee6d | 2655 | dm_start_queue(md->queue); |
cec47e3d | 2656 | |
2ca3310e | 2657 | unlock_fs(md); |
d67ee213 | 2658 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2659 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2660 | } |
1da177e4 | 2661 | |
ffcc3936 MS |
2662 | return r; |
2663 | } | |
2664 | ||
2665 | /* | |
2666 | * We need to be able to change a mapping table under a mounted | |
2667 | * filesystem. For example we might want to move some data in | |
2668 | * the background. Before the table can be swapped with | |
2669 | * dm_bind_table, dm_suspend must be called to flush any in | |
2670 | * flight bios and ensure that any further io gets deferred. | |
2671 | */ | |
2672 | /* | |
2673 | * Suspend mechanism in request-based dm. | |
2674 | * | |
2675 | * 1. Flush all I/Os by lock_fs() if needed. | |
2676 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2677 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2678 | * | |
2679 | * To abort suspend, start the request_queue. | |
2680 | */ | |
2681 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2682 | { | |
2683 | struct dm_table *map = NULL; | |
2684 | int r = 0; | |
2685 | ||
2686 | retry: | |
2687 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2688 | ||
2689 | if (dm_suspended_md(md)) { | |
2690 | r = -EINVAL; | |
2691 | goto out_unlock; | |
2692 | } | |
2693 | ||
2694 | if (dm_suspended_internally_md(md)) { | |
2695 | /* already internally suspended, wait for internal resume */ | |
2696 | mutex_unlock(&md->suspend_lock); | |
2697 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2698 | if (r) | |
2699 | return r; | |
2700 | goto retry; | |
2701 | } | |
2702 | ||
a12f5d48 | 2703 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2704 | |
eaf9a736 | 2705 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2706 | if (r) |
2707 | goto out_unlock; | |
3b00b203 | 2708 | |
4d4471cb KU |
2709 | dm_table_postsuspend_targets(map); |
2710 | ||
d287483d | 2711 | out_unlock: |
e61290a4 | 2712 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2713 | return r; |
1da177e4 LT |
2714 | } |
2715 | ||
ffcc3936 MS |
2716 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2717 | { | |
2718 | if (map) { | |
2719 | int r = dm_table_resume_targets(map); | |
2720 | if (r) | |
2721 | return r; | |
2722 | } | |
2723 | ||
2724 | dm_queue_flush(md); | |
2725 | ||
2726 | /* | |
2727 | * Flushing deferred I/Os must be done after targets are resumed | |
2728 | * so that mapping of targets can work correctly. | |
2729 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2730 | */ | |
2731 | if (dm_request_based(md)) | |
eca7ee6d | 2732 | dm_start_queue(md->queue); |
ffcc3936 MS |
2733 | |
2734 | unlock_fs(md); | |
2735 | ||
2736 | return 0; | |
2737 | } | |
2738 | ||
1da177e4 LT |
2739 | int dm_resume(struct mapped_device *md) |
2740 | { | |
8dc23658 | 2741 | int r; |
cf222b37 | 2742 | struct dm_table *map = NULL; |
1da177e4 | 2743 | |
ffcc3936 | 2744 | retry: |
8dc23658 | 2745 | r = -EINVAL; |
ffcc3936 MS |
2746 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2747 | ||
4f186f8b | 2748 | if (!dm_suspended_md(md)) |
cf222b37 | 2749 | goto out; |
cf222b37 | 2750 | |
ffcc3936 MS |
2751 | if (dm_suspended_internally_md(md)) { |
2752 | /* already internally suspended, wait for internal resume */ | |
2753 | mutex_unlock(&md->suspend_lock); | |
2754 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2755 | if (r) | |
2756 | return r; | |
2757 | goto retry; | |
2758 | } | |
2759 | ||
a12f5d48 | 2760 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2761 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2762 | goto out; |
1da177e4 | 2763 | |
ffcc3936 | 2764 | r = __dm_resume(md, map); |
8757b776 MB |
2765 | if (r) |
2766 | goto out; | |
2ca3310e | 2767 | |
2ca3310e | 2768 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2769 | out: |
e61290a4 | 2770 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2771 | |
cf222b37 | 2772 | return r; |
1da177e4 LT |
2773 | } |
2774 | ||
fd2ed4d2 MP |
2775 | /* |
2776 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2777 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2778 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2779 | */ |
2780 | ||
ffcc3936 | 2781 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2782 | { |
ffcc3936 MS |
2783 | struct dm_table *map = NULL; |
2784 | ||
1ea0654e BVA |
2785 | lockdep_assert_held(&md->suspend_lock); |
2786 | ||
96b26c8c | 2787 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2788 | return; /* nested internal suspend */ |
2789 | ||
2790 | if (dm_suspended_md(md)) { | |
2791 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2792 | return; /* nest suspend */ | |
2793 | } | |
2794 | ||
a12f5d48 | 2795 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2796 | |
2797 | /* | |
2798 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2799 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2800 | * would require changing .presuspend to return an error -- avoid this | |
2801 | * until there is a need for more elaborate variants of internal suspend. | |
2802 | */ | |
eaf9a736 MS |
2803 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2804 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2805 | |
2806 | dm_table_postsuspend_targets(map); | |
2807 | } | |
2808 | ||
2809 | static void __dm_internal_resume(struct mapped_device *md) | |
2810 | { | |
96b26c8c MP |
2811 | BUG_ON(!md->internal_suspend_count); |
2812 | ||
2813 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2814 | return; /* resume from nested internal suspend */ |
2815 | ||
fd2ed4d2 | 2816 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2817 | goto done; /* resume from nested suspend */ |
2818 | ||
2819 | /* | |
2820 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2821 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2822 | */ | |
2823 | (void) __dm_resume(md, NULL); | |
2824 | ||
2825 | done: | |
2826 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2827 | smp_mb__after_atomic(); | |
2828 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2829 | } | |
2830 | ||
2831 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2832 | { | |
2833 | mutex_lock(&md->suspend_lock); | |
2834 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2835 | mutex_unlock(&md->suspend_lock); | |
2836 | } | |
2837 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2838 | ||
2839 | void dm_internal_resume(struct mapped_device *md) | |
2840 | { | |
2841 | mutex_lock(&md->suspend_lock); | |
2842 | __dm_internal_resume(md); | |
2843 | mutex_unlock(&md->suspend_lock); | |
2844 | } | |
2845 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2846 | ||
2847 | /* | |
2848 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2849 | * which prevents interaction with userspace-driven suspend. | |
2850 | */ | |
2851 | ||
2852 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2853 | { | |
2854 | mutex_lock(&md->suspend_lock); | |
2855 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2856 | return; |
2857 | ||
2858 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2859 | synchronize_srcu(&md->io_barrier); | |
2860 | flush_workqueue(md->wq); | |
2861 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2862 | } | |
b735fede | 2863 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2864 | |
ffcc3936 | 2865 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2866 | { |
ffcc3936 | 2867 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2868 | goto done; |
2869 | ||
2870 | dm_queue_flush(md); | |
2871 | ||
2872 | done: | |
2873 | mutex_unlock(&md->suspend_lock); | |
2874 | } | |
b735fede | 2875 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2876 | |
1da177e4 LT |
2877 | /*----------------------------------------------------------------- |
2878 | * Event notification. | |
2879 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2880 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2881 | unsigned cookie) |
69267a30 | 2882 | { |
60935eb2 MB |
2883 | char udev_cookie[DM_COOKIE_LENGTH]; |
2884 | char *envp[] = { udev_cookie, NULL }; | |
2885 | ||
2886 | if (!cookie) | |
3abf85b5 | 2887 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2888 | else { |
2889 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2890 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2891 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2892 | action, envp); | |
60935eb2 | 2893 | } |
69267a30 AK |
2894 | } |
2895 | ||
7a8c3d3b MA |
2896 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2897 | { | |
2898 | return atomic_add_return(1, &md->uevent_seq); | |
2899 | } | |
2900 | ||
1da177e4 LT |
2901 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2902 | { | |
2903 | return atomic_read(&md->event_nr); | |
2904 | } | |
2905 | ||
2906 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2907 | { | |
2908 | return wait_event_interruptible(md->eventq, | |
2909 | (event_nr != atomic_read(&md->event_nr))); | |
2910 | } | |
2911 | ||
7a8c3d3b MA |
2912 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2913 | { | |
2914 | unsigned long flags; | |
2915 | ||
2916 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2917 | list_add(elist, &md->uevent_list); | |
2918 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2919 | } | |
2920 | ||
1da177e4 LT |
2921 | /* |
2922 | * The gendisk is only valid as long as you have a reference | |
2923 | * count on 'md'. | |
2924 | */ | |
2925 | struct gendisk *dm_disk(struct mapped_device *md) | |
2926 | { | |
2927 | return md->disk; | |
2928 | } | |
65ff5b7d | 2929 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2930 | |
784aae73 MB |
2931 | struct kobject *dm_kobject(struct mapped_device *md) |
2932 | { | |
2995fa78 | 2933 | return &md->kobj_holder.kobj; |
784aae73 MB |
2934 | } |
2935 | ||
784aae73 MB |
2936 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2937 | { | |
2938 | struct mapped_device *md; | |
2939 | ||
2995fa78 | 2940 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2941 | |
b9a41d21 HT |
2942 | spin_lock(&_minor_lock); |
2943 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2944 | md = NULL; | |
2945 | goto out; | |
2946 | } | |
784aae73 | 2947 | dm_get(md); |
b9a41d21 HT |
2948 | out: |
2949 | spin_unlock(&_minor_lock); | |
2950 | ||
784aae73 MB |
2951 | return md; |
2952 | } | |
2953 | ||
4f186f8b | 2954 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2955 | { |
2956 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2957 | } | |
2958 | ||
ffcc3936 MS |
2959 | int dm_suspended_internally_md(struct mapped_device *md) |
2960 | { | |
2961 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2962 | } | |
2963 | ||
2c140a24 MP |
2964 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2965 | { | |
2966 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2967 | } | |
2968 | ||
64dbce58 KU |
2969 | int dm_suspended(struct dm_target *ti) |
2970 | { | |
ecdb2e25 | 2971 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2972 | } |
2973 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2974 | ||
2e93ccc1 KU |
2975 | int dm_noflush_suspending(struct dm_target *ti) |
2976 | { | |
ecdb2e25 | 2977 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2978 | } |
2979 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2980 | ||
7e0d574f | 2981 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
0776aa0e MS |
2982 | unsigned integrity, unsigned per_io_data_size, |
2983 | unsigned min_pool_size) | |
e6ee8c0b | 2984 | { |
115485e8 | 2985 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2986 | unsigned int pool_size = 0; |
64f52b0e | 2987 | unsigned int front_pad, io_front_pad; |
6f1c819c | 2988 | int ret; |
e6ee8c0b KU |
2989 | |
2990 | if (!pools) | |
4e6e36c3 | 2991 | return NULL; |
e6ee8c0b | 2992 | |
78d8e58a MS |
2993 | switch (type) { |
2994 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2995 | case DM_TYPE_DAX_BIO_BASED: |
22c11858 | 2996 | case DM_TYPE_NVME_BIO_BASED: |
0776aa0e | 2997 | pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); |
30187e1d | 2998 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
64f52b0e | 2999 | io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio); |
6f1c819c KO |
3000 | ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0); |
3001 | if (ret) | |
64f52b0e | 3002 | goto out; |
6f1c819c | 3003 | if (integrity && bioset_integrity_create(&pools->io_bs, pool_size)) |
eb8db831 | 3004 | goto out; |
78d8e58a MS |
3005 | break; |
3006 | case DM_TYPE_REQUEST_BASED: | |
0776aa0e | 3007 | pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); |
78d8e58a | 3008 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 3009 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
3010 | break; |
3011 | default: | |
3012 | BUG(); | |
3013 | } | |
3014 | ||
6f1c819c KO |
3015 | ret = bioset_init(&pools->bs, pool_size, front_pad, 0); |
3016 | if (ret) | |
5f015204 | 3017 | goto out; |
e6ee8c0b | 3018 | |
6f1c819c | 3019 | if (integrity && bioset_integrity_create(&pools->bs, pool_size)) |
5f015204 | 3020 | goto out; |
a91a2785 | 3021 | |
e6ee8c0b | 3022 | return pools; |
5f1b670d | 3023 | |
5f1b670d CH |
3024 | out: |
3025 | dm_free_md_mempools(pools); | |
78d8e58a | 3026 | |
4e6e36c3 | 3027 | return NULL; |
e6ee8c0b KU |
3028 | } |
3029 | ||
3030 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3031 | { | |
3032 | if (!pools) | |
3033 | return; | |
3034 | ||
6f1c819c KO |
3035 | bioset_exit(&pools->bs); |
3036 | bioset_exit(&pools->io_bs); | |
e6ee8c0b KU |
3037 | |
3038 | kfree(pools); | |
3039 | } | |
3040 | ||
9c72bad1 CH |
3041 | struct dm_pr { |
3042 | u64 old_key; | |
3043 | u64 new_key; | |
3044 | u32 flags; | |
3045 | bool fail_early; | |
3046 | }; | |
3047 | ||
3048 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
3049 | void *data) | |
71cdb697 CH |
3050 | { |
3051 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
3052 | struct dm_table *table; |
3053 | struct dm_target *ti; | |
3054 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 3055 | |
9c72bad1 CH |
3056 | table = dm_get_live_table(md, &srcu_idx); |
3057 | if (!table || !dm_table_get_size(table)) | |
3058 | goto out; | |
71cdb697 | 3059 | |
9c72bad1 CH |
3060 | /* We only support devices that have a single target */ |
3061 | if (dm_table_get_num_targets(table) != 1) | |
3062 | goto out; | |
3063 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 3064 | |
9c72bad1 CH |
3065 | ret = -EINVAL; |
3066 | if (!ti->type->iterate_devices) | |
3067 | goto out; | |
3068 | ||
3069 | ret = ti->type->iterate_devices(ti, fn, data); | |
3070 | out: | |
3071 | dm_put_live_table(md, srcu_idx); | |
3072 | return ret; | |
3073 | } | |
3074 | ||
3075 | /* | |
3076 | * For register / unregister we need to manually call out to every path. | |
3077 | */ | |
3078 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
3079 | sector_t start, sector_t len, void *data) | |
3080 | { | |
3081 | struct dm_pr *pr = data; | |
3082 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3083 | ||
3084 | if (!ops || !ops->pr_register) | |
3085 | return -EOPNOTSUPP; | |
3086 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
3087 | } | |
3088 | ||
3089 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
3090 | u32 flags) | |
3091 | { | |
3092 | struct dm_pr pr = { | |
3093 | .old_key = old_key, | |
3094 | .new_key = new_key, | |
3095 | .flags = flags, | |
3096 | .fail_early = true, | |
3097 | }; | |
3098 | int ret; | |
3099 | ||
3100 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
3101 | if (ret && new_key) { | |
3102 | /* unregister all paths if we failed to register any path */ | |
3103 | pr.old_key = new_key; | |
3104 | pr.new_key = 0; | |
3105 | pr.flags = 0; | |
3106 | pr.fail_early = false; | |
3107 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
3108 | } | |
3109 | ||
3110 | return ret; | |
71cdb697 CH |
3111 | } |
3112 | ||
3113 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3114 | u32 flags) |
71cdb697 CH |
3115 | { |
3116 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3117 | const struct pr_ops *ops; | |
971888c4 | 3118 | int r, srcu_idx; |
71cdb697 | 3119 | |
5bd5e8d8 | 3120 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3121 | if (r < 0) |
971888c4 | 3122 | goto out; |
71cdb697 CH |
3123 | |
3124 | ops = bdev->bd_disk->fops->pr_ops; | |
3125 | if (ops && ops->pr_reserve) | |
3126 | r = ops->pr_reserve(bdev, key, type, flags); | |
3127 | else | |
3128 | r = -EOPNOTSUPP; | |
971888c4 MS |
3129 | out: |
3130 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3131 | return r; |
3132 | } | |
3133 | ||
3134 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3135 | { | |
3136 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3137 | const struct pr_ops *ops; | |
971888c4 | 3138 | int r, srcu_idx; |
71cdb697 | 3139 | |
5bd5e8d8 | 3140 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3141 | if (r < 0) |
971888c4 | 3142 | goto out; |
71cdb697 CH |
3143 | |
3144 | ops = bdev->bd_disk->fops->pr_ops; | |
3145 | if (ops && ops->pr_release) | |
3146 | r = ops->pr_release(bdev, key, type); | |
3147 | else | |
3148 | r = -EOPNOTSUPP; | |
971888c4 MS |
3149 | out: |
3150 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3151 | return r; |
3152 | } | |
3153 | ||
3154 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3155 | enum pr_type type, bool abort) |
71cdb697 CH |
3156 | { |
3157 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3158 | const struct pr_ops *ops; | |
971888c4 | 3159 | int r, srcu_idx; |
71cdb697 | 3160 | |
5bd5e8d8 | 3161 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3162 | if (r < 0) |
971888c4 | 3163 | goto out; |
71cdb697 CH |
3164 | |
3165 | ops = bdev->bd_disk->fops->pr_ops; | |
3166 | if (ops && ops->pr_preempt) | |
3167 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3168 | else | |
3169 | r = -EOPNOTSUPP; | |
971888c4 MS |
3170 | out: |
3171 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3172 | return r; |
3173 | } | |
3174 | ||
3175 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3176 | { | |
3177 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3178 | const struct pr_ops *ops; | |
971888c4 | 3179 | int r, srcu_idx; |
71cdb697 | 3180 | |
5bd5e8d8 | 3181 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3182 | if (r < 0) |
971888c4 | 3183 | goto out; |
71cdb697 CH |
3184 | |
3185 | ops = bdev->bd_disk->fops->pr_ops; | |
3186 | if (ops && ops->pr_clear) | |
3187 | r = ops->pr_clear(bdev, key); | |
3188 | else | |
3189 | r = -EOPNOTSUPP; | |
971888c4 MS |
3190 | out: |
3191 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3192 | return r; |
3193 | } | |
3194 | ||
3195 | static const struct pr_ops dm_pr_ops = { | |
3196 | .pr_register = dm_pr_register, | |
3197 | .pr_reserve = dm_pr_reserve, | |
3198 | .pr_release = dm_pr_release, | |
3199 | .pr_preempt = dm_pr_preempt, | |
3200 | .pr_clear = dm_pr_clear, | |
3201 | }; | |
3202 | ||
83d5cde4 | 3203 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3204 | .open = dm_blk_open, |
3205 | .release = dm_blk_close, | |
aa129a22 | 3206 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3207 | .getgeo = dm_blk_getgeo, |
e76239a3 | 3208 | .report_zones = dm_blk_report_zones, |
71cdb697 | 3209 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3210 | .owner = THIS_MODULE |
3211 | }; | |
3212 | ||
f26c5719 DW |
3213 | static const struct dax_operations dm_dax_ops = { |
3214 | .direct_access = dm_dax_direct_access, | |
7bf7eac8 | 3215 | .dax_supported = dm_dax_supported, |
7e026c8c | 3216 | .copy_from_iter = dm_dax_copy_from_iter, |
b3a9a0c3 | 3217 | .copy_to_iter = dm_dax_copy_to_iter, |
f26c5719 DW |
3218 | }; |
3219 | ||
1da177e4 LT |
3220 | /* |
3221 | * module hooks | |
3222 | */ | |
3223 | module_init(dm_init); | |
3224 | module_exit(dm_exit); | |
3225 | ||
3226 | module_param(major, uint, 0); | |
3227 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3228 | |
e8603136 MS |
3229 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3230 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3231 | ||
115485e8 MS |
3232 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
3233 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
3234 | ||
1da177e4 LT |
3235 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3236 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3237 | MODULE_LICENSE("GPL"); |