Commit | Line | Data |
---|---|---|
3bd94003 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 4 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
5 | * |
6 | * This file is released under the GPL. | |
7 | */ | |
8 | ||
4cc96131 MS |
9 | #include "dm-core.h" |
10 | #include "dm-rq.h" | |
51e5b2bd | 11 | #include "dm-uevent.h" |
91ccbbac | 12 | #include "dm-ima.h" |
1da177e4 LT |
13 | |
14 | #include <linux/init.h> | |
15 | #include <linux/module.h> | |
48c9c27b | 16 | #include <linux/mutex.h> |
6958c1c6 | 17 | #include <linux/sched/mm.h> |
174cd4b1 | 18 | #include <linux/sched/signal.h> |
1da177e4 LT |
19 | #include <linux/blkpg.h> |
20 | #include <linux/bio.h> | |
1da177e4 | 21 | #include <linux/mempool.h> |
f26c5719 | 22 | #include <linux/dax.h> |
1da177e4 LT |
23 | #include <linux/slab.h> |
24 | #include <linux/idr.h> | |
7e026c8c | 25 | #include <linux/uio.h> |
3ac51e74 | 26 | #include <linux/hdreg.h> |
3f77316d | 27 | #include <linux/delay.h> |
ffcc3936 | 28 | #include <linux/wait.h> |
71cdb697 | 29 | #include <linux/pr.h> |
b0b4d7c6 | 30 | #include <linux/refcount.h> |
c6a564ff | 31 | #include <linux/part_stat.h> |
a892c8d5 | 32 | #include <linux/blk-crypto.h> |
1e8d44bd | 33 | #include <linux/blk-crypto-profile.h> |
55782138 | 34 | |
72d94861 AK |
35 | #define DM_MSG_PREFIX "core" |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
b99fdcdc ML |
44 | /* |
45 | * For REQ_POLLED fs bio, this flag is set if we link mapped underlying | |
46 | * dm_io into one list, and reuse bio->bi_private as the list head. Before | |
47 | * ending this fs bio, we will recover its ->bi_private. | |
48 | */ | |
49 | #define REQ_DM_POLL_LIST REQ_DRV | |
50 | ||
1da177e4 LT |
51 | static const char *_name = DM_NAME; |
52 | ||
2f06cd12 HM |
53 | static unsigned int major; |
54 | static unsigned int _major; | |
1da177e4 | 55 | |
d15b774c AK |
56 | static DEFINE_IDR(_minor_idr); |
57 | ||
f32c10b0 | 58 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
59 | |
60 | static void do_deferred_remove(struct work_struct *w); | |
61 | ||
62 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
63 | ||
acfe0ad7 MP |
64 | static struct workqueue_struct *deferred_remove_workqueue; |
65 | ||
93e6442c MP |
66 | atomic_t dm_global_event_nr = ATOMIC_INIT(0); |
67 | DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq); | |
68 | ||
62e08243 MP |
69 | void dm_issue_global_event(void) |
70 | { | |
71 | atomic_inc(&dm_global_event_nr); | |
72 | wake_up(&dm_global_eventq); | |
73 | } | |
74 | ||
442761fd MS |
75 | DEFINE_STATIC_KEY_FALSE(stats_enabled); |
76 | DEFINE_STATIC_KEY_FALSE(swap_bios_enabled); | |
77 | DEFINE_STATIC_KEY_FALSE(zoned_enabled); | |
78 | ||
1da177e4 | 79 | /* |
64f52b0e | 80 | * One of these is allocated (on-stack) per original bio. |
1da177e4 | 81 | */ |
64f52b0e | 82 | struct clone_info { |
64f52b0e MS |
83 | struct dm_table *map; |
84 | struct bio *bio; | |
85 | struct dm_io *io; | |
86 | sector_t sector; | |
86a3238c | 87 | unsigned int sector_count; |
4edadf6d MS |
88 | bool is_abnormal_io:1; |
89 | bool submit_as_polled:1; | |
64f52b0e MS |
90 | }; |
91 | ||
6c23f0bd CH |
92 | static inline struct dm_target_io *clone_to_tio(struct bio *clone) |
93 | { | |
94 | return container_of(clone, struct dm_target_io, clone); | |
95 | } | |
96 | ||
64f52b0e MS |
97 | void *dm_per_bio_data(struct bio *bio, size_t data_size) |
98 | { | |
655f3aad | 99 | if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO)) |
62f26317 JX |
100 | return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size; |
101 | return (char *)bio - DM_IO_BIO_OFFSET - data_size; | |
64f52b0e MS |
102 | } |
103 | EXPORT_SYMBOL_GPL(dm_per_bio_data); | |
104 | ||
105 | struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size) | |
106 | { | |
107 | struct dm_io *io = (struct dm_io *)((char *)data + data_size); | |
b30f1607 | 108 | |
64f52b0e | 109 | if (io->magic == DM_IO_MAGIC) |
62f26317 | 110 | return (struct bio *)((char *)io + DM_IO_BIO_OFFSET); |
64f52b0e | 111 | BUG_ON(io->magic != DM_TIO_MAGIC); |
62f26317 | 112 | return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET); |
64f52b0e MS |
113 | } |
114 | EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data); | |
115 | ||
86a3238c | 116 | unsigned int dm_bio_get_target_bio_nr(const struct bio *bio) |
64f52b0e MS |
117 | { |
118 | return container_of(bio, struct dm_target_io, clone)->target_bio_nr; | |
119 | } | |
120 | EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr); | |
121 | ||
ba61fdd1 JM |
122 | #define MINOR_ALLOCED ((void *)-1) |
123 | ||
115485e8 | 124 | #define DM_NUMA_NODE NUMA_NO_NODE |
115485e8 | 125 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 126 | |
a666e5c0 MP |
127 | #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE) |
128 | static int swap_bios = DEFAULT_SWAP_BIOS; | |
129 | static int get_swap_bios(void) | |
130 | { | |
131 | int latch = READ_ONCE(swap_bios); | |
b30f1607 | 132 | |
a666e5c0 MP |
133 | if (unlikely(latch <= 0)) |
134 | latch = DEFAULT_SWAP_BIOS; | |
135 | return latch; | |
136 | } | |
137 | ||
86f1152b BM |
138 | struct table_device { |
139 | struct list_head list; | |
b0b4d7c6 | 140 | refcount_t count; |
86f1152b BM |
141 | struct dm_dev dm_dev; |
142 | }; | |
143 | ||
e8603136 MS |
144 | /* |
145 | * Bio-based DM's mempools' reserved IOs set by the user. | |
146 | */ | |
4cc96131 | 147 | #define RESERVED_BIO_BASED_IOS 16 |
86a3238c | 148 | static unsigned int reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
e8603136 | 149 | |
115485e8 MS |
150 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
151 | { | |
6aa7de05 | 152 | int param = READ_ONCE(*module_param); |
115485e8 MS |
153 | int modified_param = 0; |
154 | bool modified = true; | |
155 | ||
156 | if (param < min) | |
157 | modified_param = min; | |
158 | else if (param > max) | |
159 | modified_param = max; | |
160 | else | |
161 | modified = false; | |
162 | ||
163 | if (modified) { | |
164 | (void)cmpxchg(module_param, param, modified_param); | |
165 | param = modified_param; | |
166 | } | |
167 | ||
168 | return param; | |
169 | } | |
170 | ||
86a3238c | 171 | unsigned int __dm_get_module_param(unsigned int *module_param, unsigned int def, unsigned int max) |
f4790826 | 172 | { |
86a3238c HM |
173 | unsigned int param = READ_ONCE(*module_param); |
174 | unsigned int modified_param = 0; | |
f4790826 | 175 | |
09c2d531 MS |
176 | if (!param) |
177 | modified_param = def; | |
178 | else if (param > max) | |
179 | modified_param = max; | |
f4790826 | 180 | |
09c2d531 MS |
181 | if (modified_param) { |
182 | (void)cmpxchg(module_param, param, modified_param); | |
183 | param = modified_param; | |
f4790826 MS |
184 | } |
185 | ||
09c2d531 | 186 | return param; |
f4790826 MS |
187 | } |
188 | ||
86a3238c | 189 | unsigned int dm_get_reserved_bio_based_ios(void) |
e8603136 | 190 | { |
09c2d531 | 191 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 192 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
193 | } |
194 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
195 | ||
86a3238c | 196 | static unsigned int dm_get_numa_node(void) |
115485e8 MS |
197 | { |
198 | return __dm_get_module_param_int(&dm_numa_node, | |
199 | DM_NUMA_NODE, num_online_nodes() - 1); | |
200 | } | |
201 | ||
1da177e4 LT |
202 | static int __init local_init(void) |
203 | { | |
e689fbab | 204 | int r; |
1ae49ea2 | 205 | |
51e5b2bd | 206 | r = dm_uevent_init(); |
51157b4a | 207 | if (r) |
e689fbab | 208 | return r; |
51e5b2bd | 209 | |
57bbf99c | 210 | deferred_remove_workqueue = alloc_ordered_workqueue("kdmremove", 0); |
acfe0ad7 MP |
211 | if (!deferred_remove_workqueue) { |
212 | r = -ENOMEM; | |
213 | goto out_uevent_exit; | |
214 | } | |
215 | ||
1da177e4 LT |
216 | _major = major; |
217 | r = register_blkdev(_major, _name); | |
51157b4a | 218 | if (r < 0) |
acfe0ad7 | 219 | goto out_free_workqueue; |
1da177e4 LT |
220 | |
221 | if (!_major) | |
222 | _major = r; | |
223 | ||
224 | return 0; | |
51157b4a | 225 | |
acfe0ad7 MP |
226 | out_free_workqueue: |
227 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
228 | out_uevent_exit: |
229 | dm_uevent_exit(); | |
51157b4a KU |
230 | |
231 | return r; | |
1da177e4 LT |
232 | } |
233 | ||
234 | static void local_exit(void) | |
235 | { | |
acfe0ad7 | 236 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 237 | |
00d59405 | 238 | unregister_blkdev(_major, _name); |
51e5b2bd | 239 | dm_uevent_exit(); |
1da177e4 LT |
240 | |
241 | _major = 0; | |
242 | ||
243 | DMINFO("cleaned up"); | |
244 | } | |
245 | ||
b9249e55 | 246 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
247 | local_init, |
248 | dm_target_init, | |
249 | dm_linear_init, | |
250 | dm_stripe_init, | |
952b3557 | 251 | dm_io_init, |
945fa4d2 | 252 | dm_kcopyd_init, |
1da177e4 | 253 | dm_interface_init, |
fd2ed4d2 | 254 | dm_statistics_init, |
1da177e4 LT |
255 | }; |
256 | ||
b9249e55 | 257 | static void (*_exits[])(void) = { |
1da177e4 LT |
258 | local_exit, |
259 | dm_target_exit, | |
260 | dm_linear_exit, | |
261 | dm_stripe_exit, | |
952b3557 | 262 | dm_io_exit, |
945fa4d2 | 263 | dm_kcopyd_exit, |
1da177e4 | 264 | dm_interface_exit, |
fd2ed4d2 | 265 | dm_statistics_exit, |
1da177e4 LT |
266 | }; |
267 | ||
268 | static int __init dm_init(void) | |
269 | { | |
270 | const int count = ARRAY_SIZE(_inits); | |
1da177e4 LT |
271 | int r, i; |
272 | ||
f1cd6cb2 TS |
273 | #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE)) |
274 | DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled." | |
275 | " Duplicate IMA measurements will not be recorded in the IMA log."); | |
276 | #endif | |
277 | ||
1da177e4 LT |
278 | for (i = 0; i < count; i++) { |
279 | r = _inits[i](); | |
280 | if (r) | |
281 | goto bad; | |
282 | } | |
283 | ||
284 | return 0; | |
f1cd6cb2 | 285 | bad: |
1da177e4 LT |
286 | while (i--) |
287 | _exits[i](); | |
288 | ||
289 | return r; | |
290 | } | |
291 | ||
292 | static void __exit dm_exit(void) | |
293 | { | |
294 | int i = ARRAY_SIZE(_exits); | |
295 | ||
296 | while (i--) | |
297 | _exits[i](); | |
d15b774c AK |
298 | |
299 | /* | |
300 | * Should be empty by this point. | |
301 | */ | |
d15b774c | 302 | idr_destroy(&_minor_idr); |
1da177e4 LT |
303 | } |
304 | ||
305 | /* | |
306 | * Block device functions | |
307 | */ | |
432a212c MA |
308 | int dm_deleting_md(struct mapped_device *md) |
309 | { | |
310 | return test_bit(DMF_DELETING, &md->flags); | |
311 | } | |
312 | ||
05bdb996 | 313 | static int dm_blk_open(struct gendisk *disk, blk_mode_t mode) |
1da177e4 LT |
314 | { |
315 | struct mapped_device *md; | |
316 | ||
fba9f90e JM |
317 | spin_lock(&_minor_lock); |
318 | ||
d32e2bf8 | 319 | md = disk->private_data; |
fba9f90e JM |
320 | if (!md) |
321 | goto out; | |
322 | ||
5c6bd75d | 323 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 324 | dm_deleting_md(md)) { |
fba9f90e JM |
325 | md = NULL; |
326 | goto out; | |
327 | } | |
328 | ||
1da177e4 | 329 | dm_get(md); |
5c6bd75d | 330 | atomic_inc(&md->open_count); |
fba9f90e JM |
331 | out: |
332 | spin_unlock(&_minor_lock); | |
333 | ||
334 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
335 | } |
336 | ||
ae220766 | 337 | static void dm_blk_close(struct gendisk *disk) |
1da177e4 | 338 | { |
63a4f065 | 339 | struct mapped_device *md; |
6e9624b8 | 340 | |
4a1aeb98 MB |
341 | spin_lock(&_minor_lock); |
342 | ||
63a4f065 MS |
343 | md = disk->private_data; |
344 | if (WARN_ON(!md)) | |
345 | goto out; | |
346 | ||
2c140a24 MP |
347 | if (atomic_dec_and_test(&md->open_count) && |
348 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 349 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 350 | |
1da177e4 | 351 | dm_put(md); |
63a4f065 | 352 | out: |
4a1aeb98 | 353 | spin_unlock(&_minor_lock); |
1da177e4 LT |
354 | } |
355 | ||
5c6bd75d AK |
356 | int dm_open_count(struct mapped_device *md) |
357 | { | |
358 | return atomic_read(&md->open_count); | |
359 | } | |
360 | ||
361 | /* | |
362 | * Guarantees nothing is using the device before it's deleted. | |
363 | */ | |
2c140a24 | 364 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
365 | { |
366 | int r = 0; | |
367 | ||
368 | spin_lock(&_minor_lock); | |
369 | ||
2c140a24 | 370 | if (dm_open_count(md)) { |
5c6bd75d | 371 | r = -EBUSY; |
2c140a24 MP |
372 | if (mark_deferred) |
373 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
374 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
375 | r = -EEXIST; | |
5c6bd75d AK |
376 | else |
377 | set_bit(DMF_DELETING, &md->flags); | |
378 | ||
379 | spin_unlock(&_minor_lock); | |
380 | ||
381 | return r; | |
382 | } | |
383 | ||
2c140a24 MP |
384 | int dm_cancel_deferred_remove(struct mapped_device *md) |
385 | { | |
386 | int r = 0; | |
387 | ||
388 | spin_lock(&_minor_lock); | |
389 | ||
390 | if (test_bit(DMF_DELETING, &md->flags)) | |
391 | r = -EBUSY; | |
392 | else | |
393 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
394 | ||
395 | spin_unlock(&_minor_lock); | |
396 | ||
397 | return r; | |
398 | } | |
399 | ||
400 | static void do_deferred_remove(struct work_struct *w) | |
401 | { | |
402 | dm_deferred_remove(); | |
403 | } | |
404 | ||
3ac51e74 DW |
405 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
406 | { | |
407 | struct mapped_device *md = bdev->bd_disk->private_data; | |
408 | ||
409 | return dm_get_geometry(md, geo); | |
410 | } | |
411 | ||
971888c4 | 412 | static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, |
5bd5e8d8 | 413 | struct block_device **bdev) |
aa129a22 | 414 | { |
564b5c54 | 415 | struct dm_target *ti; |
6c182cd8 | 416 | struct dm_table *map; |
971888c4 | 417 | int r; |
aa129a22 | 418 | |
6c182cd8 | 419 | retry: |
e56f81e0 | 420 | r = -ENOTTY; |
971888c4 | 421 | map = dm_get_live_table(md, srcu_idx); |
aa129a22 | 422 | if (!map || !dm_table_get_size(map)) |
971888c4 | 423 | return r; |
aa129a22 MB |
424 | |
425 | /* We only support devices that have a single target */ | |
2aec377a | 426 | if (map->num_targets != 1) |
971888c4 | 427 | return r; |
aa129a22 | 428 | |
564b5c54 MS |
429 | ti = dm_table_get_target(map, 0); |
430 | if (!ti->type->prepare_ioctl) | |
971888c4 | 431 | return r; |
519049af | 432 | |
971888c4 MS |
433 | if (dm_suspended_md(md)) |
434 | return -EAGAIN; | |
aa129a22 | 435 | |
564b5c54 | 436 | r = ti->type->prepare_ioctl(ti, bdev); |
5bbbfdf6 | 437 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
971888c4 | 438 | dm_put_live_table(md, *srcu_idx); |
238d991f | 439 | fsleep(10000); |
6c182cd8 HR |
440 | goto retry; |
441 | } | |
971888c4 | 442 | |
e56f81e0 CH |
443 | return r; |
444 | } | |
445 | ||
971888c4 | 446 | static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) |
971888c4 MS |
447 | { |
448 | dm_put_live_table(md, srcu_idx); | |
449 | } | |
450 | ||
05bdb996 | 451 | static int dm_blk_ioctl(struct block_device *bdev, blk_mode_t mode, |
e56f81e0 CH |
452 | unsigned int cmd, unsigned long arg) |
453 | { | |
454 | struct mapped_device *md = bdev->bd_disk->private_data; | |
971888c4 | 455 | int r, srcu_idx; |
e56f81e0 | 456 | |
5bd5e8d8 | 457 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
e56f81e0 | 458 | if (r < 0) |
971888c4 | 459 | goto out; |
6c182cd8 | 460 | |
e56f81e0 CH |
461 | if (r > 0) { |
462 | /* | |
e980f623 CH |
463 | * Target determined this ioctl is being issued against a |
464 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 465 | */ |
e980f623 | 466 | if (!capable(CAP_SYS_RAWIO)) { |
0378c625 | 467 | DMDEBUG_LIMIT( |
e980f623 CH |
468 | "%s: sending ioctl %x to DM device without required privilege.", |
469 | current->comm, cmd); | |
470 | r = -ENOIOCTLCMD; | |
e56f81e0 | 471 | goto out; |
e980f623 | 472 | } |
e56f81e0 | 473 | } |
6c182cd8 | 474 | |
a7cb3d2f CH |
475 | if (!bdev->bd_disk->fops->ioctl) |
476 | r = -ENOTTY; | |
477 | else | |
478 | r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg); | |
e56f81e0 | 479 | out: |
971888c4 | 480 | dm_unprepare_ioctl(md, srcu_idx); |
aa129a22 MB |
481 | return r; |
482 | } | |
483 | ||
7465d7ac MS |
484 | u64 dm_start_time_ns_from_clone(struct bio *bio) |
485 | { | |
6c23f0bd | 486 | return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time); |
7465d7ac MS |
487 | } |
488 | EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone); | |
489 | ||
06eed768 | 490 | static inline bool bio_is_flush_with_data(struct bio *bio) |
7465d7ac | 491 | { |
8d394bc4 MS |
492 | return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size); |
493 | } | |
494 | ||
06eed768 | 495 | static inline unsigned int dm_io_sectors(struct dm_io *io, struct bio *bio) |
8d394bc4 | 496 | { |
d3de6d12 ML |
497 | /* |
498 | * If REQ_PREFLUSH set, don't account payload, it will be | |
499 | * submitted (and accounted) after this flush completes. | |
500 | */ | |
501 | if (bio_is_flush_with_data(bio)) | |
06eed768 MS |
502 | return 0; |
503 | if (unlikely(dm_io_flagged(io, DM_IO_WAS_SPLIT))) | |
504 | return io->sectors; | |
505 | return bio_sectors(bio); | |
506 | } | |
8d394bc4 | 507 | |
06eed768 MS |
508 | static void dm_io_acct(struct dm_io *io, bool end) |
509 | { | |
510 | struct bio *bio = io->orig_bio; | |
8d394bc4 | 511 | |
526d1006 LN |
512 | if (dm_io_flagged(io, DM_IO_BLK_STAT)) { |
513 | if (!end) | |
514 | bdev_start_io_acct(bio->bi_bdev, bio_op(bio), | |
06eed768 | 515 | io->start_time); |
526d1006 LN |
516 | else |
517 | bdev_end_io_acct(bio->bi_bdev, bio_op(bio), | |
06eed768 MS |
518 | dm_io_sectors(io, bio), |
519 | io->start_time); | |
526d1006 | 520 | } |
7465d7ac | 521 | |
442761fd | 522 | if (static_branch_unlikely(&stats_enabled) && |
06eed768 | 523 | unlikely(dm_stats_used(&io->md->stats))) { |
7dd76d1f ML |
524 | sector_t sector; |
525 | ||
06eed768 | 526 | if (unlikely(dm_io_flagged(io, DM_IO_WAS_SPLIT))) |
7dd76d1f | 527 | sector = bio_end_sector(bio) - io->sector_offset; |
06eed768 MS |
528 | else |
529 | sector = bio->bi_iter.bi_sector; | |
7dd76d1f | 530 | |
06eed768 MS |
531 | dm_stats_account_io(&io->md->stats, bio_data_dir(bio), |
532 | sector, dm_io_sectors(io, bio), | |
533 | end, io->start_time, &io->stats_aux); | |
7dd76d1f | 534 | } |
8d394bc4 MS |
535 | } |
536 | ||
b992b40d | 537 | static void __dm_start_io_acct(struct dm_io *io) |
8d394bc4 | 538 | { |
e6926ad0 | 539 | dm_io_acct(io, false); |
7465d7ac MS |
540 | } |
541 | ||
0fbb4d93 | 542 | static void dm_start_io_acct(struct dm_io *io, struct bio *clone) |
7465d7ac | 543 | { |
0fbb4d93 MS |
544 | /* |
545 | * Ensure IO accounting is only ever started once. | |
0fbb4d93 | 546 | */ |
3b03f7c1 MS |
547 | if (dm_io_flagged(io, DM_IO_ACCOUNTED)) |
548 | return; | |
549 | ||
550 | /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */ | |
551 | if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) { | |
82f6cdcc MS |
552 | dm_io_set_flag(io, DM_IO_ACCOUNTED); |
553 | } else { | |
554 | unsigned long flags; | |
655f3aad | 555 | /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */ |
4d7bca13 | 556 | spin_lock_irqsave(&io->lock, flags); |
10eb3a0d BM |
557 | if (dm_io_flagged(io, DM_IO_ACCOUNTED)) { |
558 | spin_unlock_irqrestore(&io->lock, flags); | |
559 | return; | |
560 | } | |
82f6cdcc | 561 | dm_io_set_flag(io, DM_IO_ACCOUNTED); |
4d7bca13 | 562 | spin_unlock_irqrestore(&io->lock, flags); |
82f6cdcc | 563 | } |
7465d7ac | 564 | |
b992b40d | 565 | __dm_start_io_acct(io); |
0fbb4d93 | 566 | } |
7465d7ac | 567 | |
b992b40d | 568 | static void dm_end_io_acct(struct dm_io *io) |
0fbb4d93 | 569 | { |
e6926ad0 | 570 | dm_io_acct(io, true); |
7465d7ac | 571 | } |
978e51ba | 572 | |
6f25dd1c | 573 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio, gfp_t gfp_mask) |
1da177e4 | 574 | { |
64f52b0e MS |
575 | struct dm_io *io; |
576 | struct dm_target_io *tio; | |
577 | struct bio *clone; | |
578 | ||
6f25dd1c MS |
579 | clone = bio_alloc_clone(NULL, bio, gfp_mask, &md->mempools->io_bs); |
580 | if (unlikely(!clone)) | |
581 | return NULL; | |
6c23f0bd | 582 | tio = clone_to_tio(clone); |
655f3aad MS |
583 | tio->flags = 0; |
584 | dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO); | |
64f52b0e MS |
585 | tio->io = NULL; |
586 | ||
587 | io = container_of(tio, struct dm_io, tio); | |
588 | io->magic = DM_IO_MAGIC; | |
84b98f4c | 589 | io->status = BLK_STS_OK; |
0f14d60a ML |
590 | |
591 | /* one ref is for submission, the other is for completion */ | |
592 | atomic_set(&io->io_count, 2); | |
9f6dc633 | 593 | this_cpu_inc(*md->pending_io); |
7dd76d1f | 594 | io->orig_bio = bio; |
978e51ba | 595 | io->md = md; |
4d7bca13 | 596 | spin_lock_init(&io->lock); |
b879f915 | 597 | io->start_time = jiffies; |
82f6cdcc | 598 | io->flags = 0; |
526d1006 LN |
599 | if (blk_queue_io_stat(md->queue)) |
600 | dm_io_set_flag(io, DM_IO_BLK_STAT); | |
64f52b0e | 601 | |
c4f512d2 MS |
602 | if (static_branch_unlikely(&stats_enabled) && |
603 | unlikely(dm_stats_used(&md->stats))) | |
442761fd | 604 | dm_stats_record_start(&md->stats, &io->stats_aux); |
64f52b0e MS |
605 | |
606 | return io; | |
1da177e4 LT |
607 | } |
608 | ||
0119ab14 | 609 | static void free_io(struct dm_io *io) |
1da177e4 | 610 | { |
64f52b0e MS |
611 | bio_put(&io->tio.clone); |
612 | } | |
613 | ||
1d1068ce | 614 | static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti, |
86a3238c | 615 | unsigned int target_bio_nr, unsigned int *len, gfp_t gfp_mask) |
64f52b0e | 616 | { |
9dd1cd32 | 617 | struct mapped_device *md = ci->io->md; |
64f52b0e | 618 | struct dm_target_io *tio; |
018b05eb | 619 | struct bio *clone; |
64f52b0e MS |
620 | |
621 | if (!ci->io->tio.io) { | |
622 | /* the dm_target_io embedded in ci->io is available */ | |
623 | tio = &ci->io->tio; | |
018b05eb MS |
624 | /* alloc_io() already initialized embedded clone */ |
625 | clone = &tio->clone; | |
64f52b0e | 626 | } else { |
29dec90a CH |
627 | clone = bio_alloc_clone(NULL, ci->bio, gfp_mask, |
628 | &md->mempools->bs); | |
64f52b0e MS |
629 | if (!clone) |
630 | return NULL; | |
631 | ||
b99fdcdc ML |
632 | /* REQ_DM_POLL_LIST shouldn't be inherited */ |
633 | clone->bi_opf &= ~REQ_DM_POLL_LIST; | |
634 | ||
6c23f0bd | 635 | tio = clone_to_tio(clone); |
655f3aad | 636 | tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */ |
64f52b0e MS |
637 | } |
638 | ||
639 | tio->magic = DM_TIO_MAGIC; | |
640 | tio->io = ci->io; | |
641 | tio->ti = ti; | |
642 | tio->target_bio_nr = target_bio_nr; | |
dc8e2021 | 643 | tio->len_ptr = len; |
743598f0 | 644 | tio->old_sector = 0; |
64f52b0e | 645 | |
9dd1cd32 MS |
646 | /* Set default bdev, but target must bio_set_dev() before issuing IO */ |
647 | clone->bi_bdev = md->disk->part0; | |
648 | if (unlikely(ti->needs_bio_set_dev)) | |
649 | bio_set_dev(clone, md->disk->part0); | |
650 | ||
018b05eb MS |
651 | if (len) { |
652 | clone->bi_iter.bi_size = to_bytes(*len); | |
653 | if (bio_integrity(clone)) | |
654 | bio_integrity_trim(clone); | |
655 | } | |
64f52b0e | 656 | |
018b05eb | 657 | return clone; |
1da177e4 LT |
658 | } |
659 | ||
1d1068ce | 660 | static void free_tio(struct bio *clone) |
1da177e4 | 661 | { |
655f3aad | 662 | if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO)) |
64f52b0e | 663 | return; |
1d1068ce | 664 | bio_put(clone); |
1da177e4 LT |
665 | } |
666 | ||
667 | /* | |
668 | * Add the bio to the list of deferred io. | |
669 | */ | |
92c63902 | 670 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 671 | { |
05447420 | 672 | unsigned long flags; |
1da177e4 | 673 | |
05447420 | 674 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 675 | bio_list_add(&md->deferred, bio); |
05447420 | 676 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 677 | queue_work(md->wq, &md->work); |
1da177e4 LT |
678 | } |
679 | ||
680 | /* | |
681 | * Everyone (including functions in this file), should use this | |
682 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 683 | * dm_put_live_table() when finished. |
1da177e4 | 684 | */ |
563a225c MS |
685 | struct dm_table *dm_get_live_table(struct mapped_device *md, |
686 | int *srcu_idx) __acquires(md->io_barrier) | |
1da177e4 | 687 | { |
83d5e5b0 MP |
688 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
689 | ||
690 | return srcu_dereference(md->map, &md->io_barrier); | |
691 | } | |
1da177e4 | 692 | |
563a225c MS |
693 | void dm_put_live_table(struct mapped_device *md, |
694 | int srcu_idx) __releases(md->io_barrier) | |
83d5e5b0 MP |
695 | { |
696 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
697 | } | |
698 | ||
699 | void dm_sync_table(struct mapped_device *md) | |
700 | { | |
701 | synchronize_srcu(&md->io_barrier); | |
702 | synchronize_rcu_expedited(); | |
703 | } | |
704 | ||
705 | /* | |
706 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
707 | * The caller must not block between these two functions. | |
708 | */ | |
709 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
710 | { | |
711 | rcu_read_lock(); | |
712 | return rcu_dereference(md->map); | |
713 | } | |
1da177e4 | 714 | |
83d5e5b0 MP |
715 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
716 | { | |
717 | rcu_read_unlock(); | |
1da177e4 LT |
718 | } |
719 | ||
971888c4 MS |
720 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
721 | ||
86f1152b BM |
722 | /* |
723 | * Open a table device so we can use it as a map destination. | |
724 | */ | |
b9a785d2 | 725 | static struct table_device *open_table_device(struct mapped_device *md, |
05bdb996 | 726 | dev_t dev, blk_mode_t mode) |
86f1152b | 727 | { |
b9a785d2 | 728 | struct table_device *td; |
c2fce61f | 729 | struct bdev_handle *bdev_handle; |
cd913c76 | 730 | u64 part_off; |
86f1152b BM |
731 | int r; |
732 | ||
b9a785d2 CH |
733 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
734 | if (!td) | |
735 | return ERR_PTR(-ENOMEM); | |
736 | refcount_set(&td->count, 1); | |
86f1152b | 737 | |
c2fce61f JK |
738 | bdev_handle = bdev_open_by_dev(dev, mode, _dm_claim_ptr, NULL); |
739 | if (IS_ERR(bdev_handle)) { | |
740 | r = PTR_ERR(bdev_handle); | |
b9a785d2 CH |
741 | goto out_free_td; |
742 | } | |
86f1152b | 743 | |
1a581b72 CH |
744 | /* |
745 | * We can be called before the dm disk is added. In that case we can't | |
746 | * register the holder relation here. It will be done once add_disk was | |
747 | * called. | |
748 | */ | |
749 | if (md->disk->slave_dir) { | |
c2fce61f | 750 | r = bd_link_disk_holder(bdev_handle->bdev, md->disk); |
1a581b72 CH |
751 | if (r) |
752 | goto out_blkdev_put; | |
753 | } | |
86f1152b | 754 | |
b9a785d2 | 755 | td->dm_dev.mode = mode; |
c2fce61f JK |
756 | td->dm_dev.bdev = bdev_handle->bdev; |
757 | td->dm_dev.bdev_handle = bdev_handle; | |
758 | td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev_handle->bdev, &part_off, | |
759 | NULL, NULL); | |
b9a785d2 CH |
760 | format_dev_t(td->dm_dev.name, dev); |
761 | list_add(&td->list, &md->table_devices); | |
762 | return td; | |
763 | ||
764 | out_blkdev_put: | |
c2fce61f | 765 | bdev_release(bdev_handle); |
b9a785d2 CH |
766 | out_free_td: |
767 | kfree(td); | |
768 | return ERR_PTR(r); | |
86f1152b BM |
769 | } |
770 | ||
771 | /* | |
772 | * Close a table device that we've been using. | |
773 | */ | |
774 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
775 | { | |
1a581b72 CH |
776 | if (md->disk->slave_dir) |
777 | bd_unlink_disk_holder(td->dm_dev.bdev, md->disk); | |
c2fce61f | 778 | bdev_release(td->dm_dev.bdev_handle); |
817bf402 | 779 | put_dax(td->dm_dev.dax_dev); |
7b586583 CH |
780 | list_del(&td->list); |
781 | kfree(td); | |
86f1152b BM |
782 | } |
783 | ||
784 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
05bdb996 | 785 | blk_mode_t mode) |
8454fca4 | 786 | { |
86f1152b BM |
787 | struct table_device *td; |
788 | ||
789 | list_for_each_entry(td, l, list) | |
790 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
791 | return td; | |
792 | ||
793 | return NULL; | |
794 | } | |
795 | ||
05bdb996 | 796 | int dm_get_table_device(struct mapped_device *md, dev_t dev, blk_mode_t mode, |
8454fca4 SS |
797 | struct dm_dev **result) |
798 | { | |
86f1152b BM |
799 | struct table_device *td; |
800 | ||
801 | mutex_lock(&md->table_devices_lock); | |
802 | td = find_table_device(&md->table_devices, dev, mode); | |
803 | if (!td) { | |
b9a785d2 CH |
804 | td = open_table_device(md, dev, mode); |
805 | if (IS_ERR(td)) { | |
86f1152b | 806 | mutex_unlock(&md->table_devices_lock); |
b9a785d2 | 807 | return PTR_ERR(td); |
86f1152b | 808 | } |
b0b4d7c6 ER |
809 | } else { |
810 | refcount_inc(&td->count); | |
86f1152b | 811 | } |
86f1152b BM |
812 | mutex_unlock(&md->table_devices_lock); |
813 | ||
814 | *result = &td->dm_dev; | |
815 | return 0; | |
816 | } | |
86f1152b BM |
817 | |
818 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
819 | { | |
820 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
821 | ||
822 | mutex_lock(&md->table_devices_lock); | |
7b586583 | 823 | if (refcount_dec_and_test(&td->count)) |
86f1152b | 824 | close_table_device(td, md); |
86f1152b BM |
825 | mutex_unlock(&md->table_devices_lock); |
826 | } | |
86f1152b | 827 | |
3ac51e74 DW |
828 | /* |
829 | * Get the geometry associated with a dm device | |
830 | */ | |
831 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
832 | { | |
833 | *geo = md->geometry; | |
834 | ||
835 | return 0; | |
836 | } | |
837 | ||
838 | /* | |
839 | * Set the geometry of a device. | |
840 | */ | |
841 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
842 | { | |
843 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
844 | ||
845 | if (geo->start > sz) { | |
43e6c111 | 846 | DMERR("Start sector is beyond the geometry limits."); |
3ac51e74 DW |
847 | return -EINVAL; |
848 | } | |
849 | ||
850 | md->geometry = *geo; | |
851 | ||
852 | return 0; | |
853 | } | |
854 | ||
2e93ccc1 KU |
855 | static int __noflush_suspending(struct mapped_device *md) |
856 | { | |
857 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
858 | } | |
859 | ||
8b211aac | 860 | static void dm_requeue_add_io(struct dm_io *io, bool first_stage) |
1da177e4 | 861 | { |
b35f8caa | 862 | struct mapped_device *md = io->md; |
1da177e4 | 863 | |
8b211aac ML |
864 | if (first_stage) { |
865 | struct dm_io *next = md->requeue_list; | |
866 | ||
867 | md->requeue_list = io; | |
868 | io->next = next; | |
869 | } else { | |
870 | bio_list_add_head(&md->deferred, io->orig_bio); | |
871 | } | |
872 | } | |
873 | ||
874 | static void dm_kick_requeue(struct mapped_device *md, bool first_stage) | |
875 | { | |
876 | if (first_stage) | |
877 | queue_work(md->wq, &md->requeue_work); | |
878 | else | |
879 | queue_work(md->wq, &md->work); | |
880 | } | |
881 | ||
444fe04f ML |
882 | /* |
883 | * Return true if the dm_io's original bio is requeued. | |
884 | * io->status is updated with error if requeue disallowed. | |
885 | */ | |
8b211aac | 886 | static bool dm_handle_requeue(struct dm_io *io, bool first_stage) |
1da177e4 | 887 | { |
8b211aac | 888 | struct bio *bio = io->orig_bio; |
444fe04f ML |
889 | bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE); |
890 | bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) && | |
891 | (bio->bi_opf & REQ_POLLED)); | |
892 | struct mapped_device *md = io->md; | |
893 | bool requeued = false; | |
1da177e4 | 894 | |
444fe04f | 895 | if (handle_requeue || handle_polled_eagain) { |
e2736347 | 896 | unsigned long flags; |
444fe04f ML |
897 | |
898 | if (bio->bi_opf & REQ_POLLED) { | |
899 | /* | |
900 | * Upper layer won't help us poll split bio | |
901 | * (io->orig_bio may only reflect a subset of the | |
902 | * pre-split original) so clear REQ_POLLED. | |
903 | */ | |
904 | bio_clear_polled(bio); | |
905 | } | |
906 | ||
e2736347 | 907 | /* |
444fe04f ML |
908 | * Target requested pushing back the I/O or |
909 | * polled IO hit BLK_STS_AGAIN. | |
e2736347 MS |
910 | */ |
911 | spin_lock_irqsave(&md->deferred_lock, flags); | |
444fe04f ML |
912 | if ((__noflush_suspending(md) && |
913 | !WARN_ON_ONCE(dm_is_zone_write(md, bio))) || | |
8b211aac ML |
914 | handle_polled_eagain || first_stage) { |
915 | dm_requeue_add_io(io, first_stage); | |
444fe04f | 916 | requeued = true; |
e2736347 | 917 | } else { |
2e93ccc1 | 918 | /* |
e2736347 MS |
919 | * noflush suspend was interrupted or this is |
920 | * a write to a zoned target. | |
2e93ccc1 | 921 | */ |
e2736347 | 922 | io->status = BLK_STS_IOERR; |
2e93ccc1 | 923 | } |
e2736347 MS |
924 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
925 | } | |
2e93ccc1 | 926 | |
444fe04f | 927 | if (requeued) |
8b211aac | 928 | dm_kick_requeue(md, first_stage); |
444fe04f ML |
929 | |
930 | return requeued; | |
931 | } | |
932 | ||
8b211aac | 933 | static void __dm_io_complete(struct dm_io *io, bool first_stage) |
444fe04f | 934 | { |
8b211aac | 935 | struct bio *bio = io->orig_bio; |
444fe04f ML |
936 | struct mapped_device *md = io->md; |
937 | blk_status_t io_error; | |
938 | bool requeued; | |
939 | ||
8b211aac ML |
940 | requeued = dm_handle_requeue(io, first_stage); |
941 | if (requeued && first_stage) | |
942 | return; | |
444fe04f | 943 | |
e2736347 | 944 | io_error = io->status; |
82f6cdcc | 945 | if (dm_io_flagged(io, DM_IO_ACCOUNTED)) |
b992b40d | 946 | dm_end_io_acct(io); |
e2736347 MS |
947 | else if (!io_error) { |
948 | /* | |
949 | * Must handle target that DM_MAPIO_SUBMITTED only to | |
950 | * then bio_endio() rather than dm_submit_bio_remap() | |
951 | */ | |
b992b40d ML |
952 | __dm_start_io_acct(io); |
953 | dm_end_io_acct(io); | |
e2736347 MS |
954 | } |
955 | free_io(io); | |
956 | smp_wmb(); | |
957 | this_cpu_dec(*md->pending_io); | |
6a8736d1 | 958 | |
e2736347 MS |
959 | /* nudge anyone waiting on suspend queue */ |
960 | if (unlikely(wq_has_sleeper(&md->wait))) | |
961 | wake_up(&md->wait); | |
2e93ccc1 | 962 | |
444fe04f ML |
963 | /* Return early if the original bio was requeued */ |
964 | if (requeued) | |
965 | return; | |
e2736347 MS |
966 | |
967 | if (bio_is_flush_with_data(bio)) { | |
968 | /* | |
969 | * Preflush done for flush with data, reissue | |
970 | * without REQ_PREFLUSH. | |
971 | */ | |
972 | bio->bi_opf &= ~REQ_PREFLUSH; | |
973 | queue_io(md, bio); | |
974 | } else { | |
975 | /* done with normal IO or empty flush */ | |
976 | if (io_error) | |
977 | bio->bi_status = io_error; | |
978 | bio_endio(bio); | |
979 | } | |
980 | } | |
981 | ||
8b211aac ML |
982 | static void dm_wq_requeue_work(struct work_struct *work) |
983 | { | |
984 | struct mapped_device *md = container_of(work, struct mapped_device, | |
985 | requeue_work); | |
986 | unsigned long flags; | |
987 | struct dm_io *io; | |
988 | ||
989 | /* reuse deferred lock to simplify dm_handle_requeue */ | |
990 | spin_lock_irqsave(&md->deferred_lock, flags); | |
991 | io = md->requeue_list; | |
992 | md->requeue_list = NULL; | |
993 | spin_unlock_irqrestore(&md->deferred_lock, flags); | |
994 | ||
995 | while (io) { | |
996 | struct dm_io *next = io->next; | |
997 | ||
46754bd0 | 998 | dm_io_rewind(io, &md->disk->bio_split); |
8b211aac ML |
999 | |
1000 | io->next = NULL; | |
1001 | __dm_io_complete(io, false); | |
1002 | io = next; | |
f77692d6 | 1003 | cond_resched(); |
8b211aac ML |
1004 | } |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * Two staged requeue: | |
1009 | * | |
1010 | * 1) io->orig_bio points to the real original bio, and the part mapped to | |
1011 | * this io must be requeued, instead of other parts of the original bio. | |
1012 | * | |
1013 | * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io. | |
1014 | */ | |
1015 | static void dm_io_complete(struct dm_io *io) | |
1016 | { | |
1017 | bool first_requeue; | |
1018 | ||
1019 | /* | |
1020 | * Only dm_io that has been split needs two stage requeue, otherwise | |
1021 | * we may run into long bio clone chain during suspend and OOM could | |
1022 | * be triggered. | |
1023 | * | |
1024 | * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they | |
1025 | * also aren't handled via the first stage requeue. | |
1026 | */ | |
1027 | if (dm_io_flagged(io, DM_IO_WAS_SPLIT)) | |
1028 | first_requeue = true; | |
1029 | else | |
1030 | first_requeue = false; | |
1031 | ||
1032 | __dm_io_complete(io, first_requeue); | |
1033 | } | |
1034 | ||
1da177e4 LT |
1035 | /* |
1036 | * Decrements the number of outstanding ios that a bio has been | |
1037 | * cloned into, completing the original io if necc. | |
1038 | */ | |
84b98f4c MS |
1039 | static inline void __dm_io_dec_pending(struct dm_io *io) |
1040 | { | |
1041 | if (atomic_dec_and_test(&io->io_count)) | |
1042 | dm_io_complete(io); | |
1043 | } | |
1044 | ||
1045 | static void dm_io_set_error(struct dm_io *io, blk_status_t error) | |
1da177e4 | 1046 | { |
84b98f4c MS |
1047 | unsigned long flags; |
1048 | ||
2e93ccc1 | 1049 | /* Push-back supersedes any I/O errors */ |
84b98f4c MS |
1050 | spin_lock_irqsave(&io->lock, flags); |
1051 | if (!(io->status == BLK_STS_DM_REQUEUE && | |
1052 | __noflush_suspending(io->md))) { | |
1053 | io->status = error; | |
1da177e4 | 1054 | } |
84b98f4c MS |
1055 | spin_unlock_irqrestore(&io->lock, flags); |
1056 | } | |
1da177e4 | 1057 | |
2e803cd9 | 1058 | static void dm_io_dec_pending(struct dm_io *io, blk_status_t error) |
84b98f4c MS |
1059 | { |
1060 | if (unlikely(error)) | |
1061 | dm_io_set_error(io, error); | |
1062 | ||
1063 | __dm_io_dec_pending(io); | |
1da177e4 LT |
1064 | } |
1065 | ||
f7995089 MS |
1066 | /* |
1067 | * The queue_limits are only valid as long as you have a reference | |
1068 | * count on 'md'. But _not_ imposing verification to avoid atomic_read(), | |
1069 | */ | |
1070 | static inline struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
1071 | { | |
1072 | return &md->queue->limits; | |
1073 | } | |
1074 | ||
bcb44433 MS |
1075 | void disable_discard(struct mapped_device *md) |
1076 | { | |
1077 | struct queue_limits *limits = dm_get_queue_limits(md); | |
1078 | ||
1079 | /* device doesn't really support DISCARD, disable it */ | |
1080 | limits->max_discard_sectors = 0; | |
bcb44433 MS |
1081 | } |
1082 | ||
ac62d620 CH |
1083 | void disable_write_zeroes(struct mapped_device *md) |
1084 | { | |
1085 | struct queue_limits *limits = dm_get_queue_limits(md); | |
1086 | ||
1087 | /* device doesn't really support WRITE ZEROES, disable it */ | |
1088 | limits->max_write_zeroes_sectors = 0; | |
1089 | } | |
1090 | ||
a666e5c0 MP |
1091 | static bool swap_bios_limit(struct dm_target *ti, struct bio *bio) |
1092 | { | |
1093 | return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios); | |
1094 | } | |
1095 | ||
4246a0b6 | 1096 | static void clone_endio(struct bio *bio) |
1da177e4 | 1097 | { |
4e4cbee9 | 1098 | blk_status_t error = bio->bi_status; |
6c23f0bd | 1099 | struct dm_target_io *tio = clone_to_tio(bio); |
6cbce280 MS |
1100 | struct dm_target *ti = tio->ti; |
1101 | dm_endio_fn endio = ti->type->end_io; | |
b35f8caa | 1102 | struct dm_io *io = tio->io; |
6cbce280 | 1103 | struct mapped_device *md = io->md; |
1da177e4 | 1104 | |
dddf3056 MS |
1105 | if (unlikely(error == BLK_STS_TARGET)) { |
1106 | if (bio_op(bio) == REQ_OP_DISCARD && | |
1107 | !bdev_max_discard_sectors(bio->bi_bdev)) | |
1108 | disable_discard(md); | |
1109 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES && | |
1110 | !bdev_write_zeroes_sectors(bio->bi_bdev)) | |
1111 | disable_write_zeroes(md); | |
ca522482 | 1112 | } |
415c79e1 | 1113 | |
dddf3056 | 1114 | if (static_branch_unlikely(&zoned_enabled) && |
edd1dbc8 | 1115 | unlikely(bdev_is_zoned(bio->bi_bdev))) |
dddf3056 MS |
1116 | dm_zone_endio(io, bio); |
1117 | ||
1be56909 | 1118 | if (endio) { |
6cbce280 | 1119 | int r = endio(ti, bio, &error); |
b30f1607 | 1120 | |
1be56909 CH |
1121 | switch (r) { |
1122 | case DM_ENDIO_REQUEUE: | |
442761fd MS |
1123 | if (static_branch_unlikely(&zoned_enabled)) { |
1124 | /* | |
1125 | * Requeuing writes to a sequential zone of a zoned | |
1126 | * target will break the sequential write pattern: | |
1127 | * fail such IO. | |
1128 | */ | |
1129 | if (WARN_ON_ONCE(dm_is_zone_write(md, bio))) | |
1130 | error = BLK_STS_IOERR; | |
1131 | else | |
1132 | error = BLK_STS_DM_REQUEUE; | |
1133 | } else | |
bf14e2b2 | 1134 | error = BLK_STS_DM_REQUEUE; |
df561f66 | 1135 | fallthrough; |
1be56909 CH |
1136 | case DM_ENDIO_DONE: |
1137 | break; | |
1138 | case DM_ENDIO_INCOMPLETE: | |
1139 | /* The target will handle the io */ | |
1140 | return; | |
1141 | default: | |
43e6c111 | 1142 | DMCRIT("unimplemented target endio return value: %d", r); |
1be56909 CH |
1143 | BUG(); |
1144 | } | |
1145 | } | |
1146 | ||
442761fd MS |
1147 | if (static_branch_unlikely(&swap_bios_enabled) && |
1148 | unlikely(swap_bios_limit(ti, bio))) | |
a666e5c0 | 1149 | up(&md->swap_bios_semaphore); |
a666e5c0 | 1150 | |
1d1068ce | 1151 | free_tio(bio); |
e2118b3c | 1152 | dm_io_dec_pending(io, error); |
1da177e4 LT |
1153 | } |
1154 | ||
56a67df7 MS |
1155 | /* |
1156 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1157 | * target boundary. | |
1158 | */ | |
3720281d MS |
1159 | static inline sector_t max_io_len_target_boundary(struct dm_target *ti, |
1160 | sector_t target_offset) | |
56a67df7 | 1161 | { |
56a67df7 MS |
1162 | return ti->len - target_offset; |
1163 | } | |
1164 | ||
06961c48 | 1165 | static sector_t __max_io_len(struct dm_target *ti, sector_t sector, |
be04c14a MS |
1166 | unsigned int max_granularity, |
1167 | unsigned int max_sectors) | |
1da177e4 | 1168 | { |
3720281d MS |
1169 | sector_t target_offset = dm_target_offset(ti, sector); |
1170 | sector_t len = max_io_len_target_boundary(ti, target_offset); | |
1da177e4 LT |
1171 | |
1172 | /* | |
3ee16db3 MS |
1173 | * Does the target need to split IO even further? |
1174 | * - varied (per target) IO splitting is a tenet of DM; this | |
1175 | * explains why stacked chunk_sectors based splitting via | |
5a97806f | 1176 | * bio_split_to_limits() isn't possible here. |
1da177e4 | 1177 | */ |
06961c48 | 1178 | if (!max_granularity) |
c3949322 CH |
1179 | return len; |
1180 | return min_t(sector_t, len, | |
be04c14a | 1181 | min(max_sectors ? : queue_max_sectors(ti->table->md->queue), |
06961c48 MS |
1182 | blk_chunk_sectors_left(target_offset, max_granularity))); |
1183 | } | |
1184 | ||
1185 | static inline sector_t max_io_len(struct dm_target *ti, sector_t sector) | |
1186 | { | |
be04c14a | 1187 | return __max_io_len(ti, sector, ti->max_io_len, 0); |
1da177e4 LT |
1188 | } |
1189 | ||
542f9038 MS |
1190 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1191 | { | |
1192 | if (len > UINT_MAX) { | |
1193 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1194 | (unsigned long long)len, UINT_MAX); | |
1195 | ti->error = "Maximum size of target IO is too large"; | |
1196 | return -EINVAL; | |
1197 | } | |
1198 | ||
75ae1936 | 1199 | ti->max_io_len = (uint32_t) len; |
542f9038 MS |
1200 | |
1201 | return 0; | |
1202 | } | |
1203 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1204 | ||
f26c5719 | 1205 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
3d97c829 MS |
1206 | sector_t sector, int *srcu_idx) |
1207 | __acquires(md->io_barrier) | |
545ed20e | 1208 | { |
545ed20e TK |
1209 | struct dm_table *map; |
1210 | struct dm_target *ti; | |
545ed20e | 1211 | |
f26c5719 | 1212 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1213 | if (!map) |
f26c5719 | 1214 | return NULL; |
545ed20e TK |
1215 | |
1216 | ti = dm_table_find_target(map, sector); | |
123d87d5 | 1217 | if (!ti) |
f26c5719 | 1218 | return NULL; |
545ed20e | 1219 | |
f26c5719 DW |
1220 | return ti; |
1221 | } | |
545ed20e | 1222 | |
f26c5719 | 1223 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
e511c4a3 JC |
1224 | long nr_pages, enum dax_access_mode mode, void **kaddr, |
1225 | pfn_t *pfn) | |
f26c5719 DW |
1226 | { |
1227 | struct mapped_device *md = dax_get_private(dax_dev); | |
1228 | sector_t sector = pgoff * PAGE_SECTORS; | |
1229 | struct dm_target *ti; | |
1230 | long len, ret = -EIO; | |
1231 | int srcu_idx; | |
545ed20e | 1232 | |
f26c5719 | 1233 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1234 | |
f26c5719 DW |
1235 | if (!ti) |
1236 | goto out; | |
1237 | if (!ti->type->direct_access) | |
1238 | goto out; | |
3720281d | 1239 | len = max_io_len(ti, sector) / PAGE_SECTORS; |
f26c5719 DW |
1240 | if (len < 1) |
1241 | goto out; | |
1242 | nr_pages = min(len, nr_pages); | |
e511c4a3 | 1243 | ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn); |
817bf402 | 1244 | |
f26c5719 | 1245 | out: |
545ed20e | 1246 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1247 | |
1248 | return ret; | |
545ed20e TK |
1249 | } |
1250 | ||
cdf6cdcd VG |
1251 | static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, |
1252 | size_t nr_pages) | |
1253 | { | |
1254 | struct mapped_device *md = dax_get_private(dax_dev); | |
1255 | sector_t sector = pgoff * PAGE_SECTORS; | |
1256 | struct dm_target *ti; | |
1257 | int ret = -EIO; | |
1258 | int srcu_idx; | |
1259 | ||
1260 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1261 | ||
1262 | if (!ti) | |
1263 | goto out; | |
1264 | if (WARN_ON(!ti->type->dax_zero_page_range)) { | |
1265 | /* | |
1266 | * ->zero_page_range() is mandatory dax operation. If we are | |
1267 | * here, something is wrong. | |
1268 | */ | |
cdf6cdcd VG |
1269 | goto out; |
1270 | } | |
1271 | ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages); | |
cdf6cdcd VG |
1272 | out: |
1273 | dm_put_live_table(md, srcu_idx); | |
1274 | ||
1275 | return ret; | |
1276 | } | |
1277 | ||
047218ec JC |
1278 | static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff, |
1279 | void *addr, size_t bytes, struct iov_iter *i) | |
1280 | { | |
1281 | struct mapped_device *md = dax_get_private(dax_dev); | |
1282 | sector_t sector = pgoff * PAGE_SECTORS; | |
1283 | struct dm_target *ti; | |
1284 | int srcu_idx; | |
1285 | long ret = 0; | |
1286 | ||
1287 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1288 | if (!ti || !ti->type->dax_recovery_write) | |
1289 | goto out; | |
1290 | ||
1291 | ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i); | |
1292 | out: | |
1293 | dm_put_live_table(md, srcu_idx); | |
1294 | return ret; | |
1295 | } | |
1296 | ||
1dd40c3e MP |
1297 | /* |
1298 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
6842d264 | 1299 | * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management |
e6fc9f62 MS |
1300 | * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by |
1301 | * __send_duplicate_bios(). | |
1dd40c3e MP |
1302 | * |
1303 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1304 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1305 | * sent in a next bio. | |
1306 | * | |
1307 | * A diagram that explains the arithmetics: | |
1308 | * +--------------------+---------------+-------+ | |
1309 | * | 1 | 2 | 3 | | |
1310 | * +--------------------+---------------+-------+ | |
1311 | * | |
1312 | * <-------------- *tio->len_ptr ---------------> | |
bdb34759 | 1313 | * <----- bio_sectors -----> |
1dd40c3e MP |
1314 | * <-- n_sectors --> |
1315 | * | |
1316 | * Region 1 was already iterated over with bio_advance or similar function. | |
1317 | * (it may be empty if the target doesn't use bio_advance) | |
1318 | * Region 2 is the remaining bio size that the target wants to process. | |
1319 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1320 | * to make it empty) | |
1321 | * The target requires that region 3 is to be sent in the next bio. | |
1322 | * | |
1323 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1324 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1325 | * copies of the bio. | |
1326 | */ | |
86a3238c | 1327 | void dm_accept_partial_bio(struct bio *bio, unsigned int n_sectors) |
1dd40c3e | 1328 | { |
6c23f0bd | 1329 | struct dm_target_io *tio = clone_to_tio(bio); |
8b211aac | 1330 | struct dm_io *io = tio->io; |
86a3238c | 1331 | unsigned int bio_sectors = bio_sectors(bio); |
6842d264 | 1332 | |
655f3aad | 1333 | BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO)); |
6842d264 DLM |
1334 | BUG_ON(op_is_zone_mgmt(bio_op(bio))); |
1335 | BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND); | |
bdb34759 MS |
1336 | BUG_ON(bio_sectors > *tio->len_ptr); |
1337 | BUG_ON(n_sectors > bio_sectors); | |
6842d264 | 1338 | |
bdb34759 | 1339 | *tio->len_ptr -= bio_sectors - n_sectors; |
1dd40c3e | 1340 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; |
7dd76d1f ML |
1341 | |
1342 | /* | |
1343 | * __split_and_process_bio() may have already saved mapped part | |
1344 | * for accounting but it is being reduced so update accordingly. | |
1345 | */ | |
8b211aac ML |
1346 | dm_io_set_flag(io, DM_IO_WAS_SPLIT); |
1347 | io->sectors = n_sectors; | |
1348 | io->sector_offset = bio_sectors(io->orig_bio); | |
1dd40c3e MP |
1349 | } |
1350 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1351 | ||
0fbb4d93 MS |
1352 | /* |
1353 | * @clone: clone bio that DM core passed to target's .map function | |
1354 | * @tgt_clone: clone of @clone bio that target needs submitted | |
0fbb4d93 MS |
1355 | * |
1356 | * Targets should use this interface to submit bios they take | |
1357 | * ownership of when returning DM_MAPIO_SUBMITTED. | |
1358 | * | |
1359 | * Target should also enable ti->accounts_remapped_io | |
1360 | */ | |
b7f8dff0 | 1361 | void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone) |
0fbb4d93 MS |
1362 | { |
1363 | struct dm_target_io *tio = clone_to_tio(clone); | |
1364 | struct dm_io *io = tio->io; | |
1365 | ||
1366 | /* establish bio that will get submitted */ | |
1367 | if (!tgt_clone) | |
1368 | tgt_clone = clone; | |
1369 | ||
1370 | /* | |
1371 | * Account io->origin_bio to DM dev on behalf of target | |
1372 | * that took ownership of IO with DM_MAPIO_SUBMITTED. | |
1373 | */ | |
9d20653f | 1374 | dm_start_io_acct(io, clone); |
0fbb4d93 | 1375 | |
9d20653f | 1376 | trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk), |
0fbb4d93 | 1377 | tio->old_sector); |
9d20653f | 1378 | submit_bio_noacct(tgt_clone); |
0fbb4d93 MS |
1379 | } |
1380 | EXPORT_SYMBOL_GPL(dm_submit_bio_remap); | |
1381 | ||
a666e5c0 MP |
1382 | static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch) |
1383 | { | |
1384 | mutex_lock(&md->swap_bios_lock); | |
1385 | while (latch < md->swap_bios) { | |
1386 | cond_resched(); | |
1387 | down(&md->swap_bios_semaphore); | |
1388 | md->swap_bios--; | |
1389 | } | |
1390 | while (latch > md->swap_bios) { | |
1391 | cond_resched(); | |
1392 | up(&md->swap_bios_semaphore); | |
1393 | md->swap_bios++; | |
1394 | } | |
1395 | mutex_unlock(&md->swap_bios_lock); | |
1396 | } | |
1397 | ||
1561b396 | 1398 | static void __map_bio(struct bio *clone) |
1da177e4 | 1399 | { |
1561b396 | 1400 | struct dm_target_io *tio = clone_to_tio(clone); |
bd2a49b8 | 1401 | struct dm_target *ti = tio->ti; |
6cbce280 MS |
1402 | struct dm_io *io = tio->io; |
1403 | struct mapped_device *md = io->md; | |
1404 | int r; | |
1da177e4 | 1405 | |
1da177e4 | 1406 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1407 | |
1408 | /* | |
0fbb4d93 | 1409 | * Map the clone. |
1da177e4 | 1410 | */ |
743598f0 | 1411 | tio->old_sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1412 | |
442761fd MS |
1413 | if (static_branch_unlikely(&swap_bios_enabled) && |
1414 | unlikely(swap_bios_limit(ti, clone))) { | |
a666e5c0 | 1415 | int latch = get_swap_bios(); |
b30f1607 | 1416 | |
a666e5c0 MP |
1417 | if (unlikely(latch != md->swap_bios)) |
1418 | __set_swap_bios_limit(md, latch); | |
1419 | down(&md->swap_bios_semaphore); | |
1420 | } | |
1421 | ||
442761fd MS |
1422 | if (static_branch_unlikely(&zoned_enabled)) { |
1423 | /* | |
1424 | * Check if the IO needs a special mapping due to zone append | |
1425 | * emulation on zoned target. In this case, dm_zone_map_bio() | |
1426 | * calls the target map operation. | |
1427 | */ | |
1428 | if (unlikely(dm_emulate_zone_append(md))) | |
1429 | r = dm_zone_map_bio(tio); | |
f1445032 MP |
1430 | else |
1431 | goto do_map; | |
1432 | } else { | |
1433 | do_map: | |
1434 | if (likely(ti->type->map == linear_map)) | |
1435 | r = linear_map(ti, clone); | |
1436 | else if (ti->type->map == stripe_map) | |
1437 | r = stripe_map(ti, clone); | |
442761fd MS |
1438 | else |
1439 | r = ti->type->map(ti, clone); | |
f1445032 | 1440 | } |
bb37d772 | 1441 | |
846785e6 CH |
1442 | switch (r) { |
1443 | case DM_MAPIO_SUBMITTED: | |
0fbb4d93 MS |
1444 | /* target has assumed ownership of this io */ |
1445 | if (!ti->accounts_remapped_io) | |
9d20653f | 1446 | dm_start_io_acct(io, clone); |
846785e6 CH |
1447 | break; |
1448 | case DM_MAPIO_REMAPPED: | |
9d20653f | 1449 | dm_submit_bio_remap(clone, NULL); |
846785e6 CH |
1450 | break; |
1451 | case DM_MAPIO_KILL: | |
846785e6 | 1452 | case DM_MAPIO_REQUEUE: |
442761fd MS |
1453 | if (static_branch_unlikely(&swap_bios_enabled) && |
1454 | unlikely(swap_bios_limit(ti, clone))) | |
6cbce280 | 1455 | up(&md->swap_bios_semaphore); |
1d1068ce | 1456 | free_tio(clone); |
90a2326e MS |
1457 | if (r == DM_MAPIO_KILL) |
1458 | dm_io_dec_pending(io, BLK_STS_IOERR); | |
1459 | else | |
1460 | dm_io_dec_pending(io, BLK_STS_DM_REQUEUE); | |
846785e6 CH |
1461 | break; |
1462 | default: | |
43e6c111 | 1463 | DMCRIT("unimplemented target map return value: %d", r); |
45cbcd79 | 1464 | BUG(); |
1da177e4 LT |
1465 | } |
1466 | } | |
1da177e4 | 1467 | |
86a3238c | 1468 | static void setup_split_accounting(struct clone_info *ci, unsigned int len) |
7dd76d1f ML |
1469 | { |
1470 | struct dm_io *io = ci->io; | |
1471 | ||
1472 | if (ci->sector_count > len) { | |
1473 | /* | |
1474 | * Split needed, save the mapped part for accounting. | |
1475 | * NOTE: dm_accept_partial_bio() will update accordingly. | |
1476 | */ | |
1477 | dm_io_set_flag(io, DM_IO_WAS_SPLIT); | |
1478 | io->sectors = len; | |
8b211aac | 1479 | io->sector_offset = bio_sectors(ci->bio); |
7dd76d1f ML |
1480 | } |
1481 | } | |
1482 | ||
318716dd | 1483 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
f7b58a69 | 1484 | struct dm_target *ti, unsigned int num_bios, |
4a2fe296 | 1485 | unsigned *len, gfp_t gfp_flag) |
f9ab94ce | 1486 | { |
1d1068ce | 1487 | struct bio *bio; |
4a2fe296 | 1488 | int try = (gfp_flag & GFP_NOWAIT) ? 0 : 1; |
dba14160 | 1489 | |
4a2fe296 | 1490 | for (; try < 2; try++) { |
318716dd | 1491 | int bio_nr; |
318716dd | 1492 | |
4a2fe296 | 1493 | if (try && num_bios > 1) |
bc02cdbe | 1494 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd | 1495 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
f7b58a69 | 1496 | bio = alloc_tio(ci, ti, bio_nr, len, |
dc8e2021 | 1497 | try ? GFP_NOIO : GFP_NOWAIT); |
1d1068ce | 1498 | if (!bio) |
318716dd MS |
1499 | break; |
1500 | ||
1d1068ce | 1501 | bio_list_add(blist, bio); |
318716dd | 1502 | } |
4a2fe296 | 1503 | if (try && num_bios > 1) |
bc02cdbe | 1504 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1505 | if (bio_nr == num_bios) |
1506 | return; | |
1507 | ||
6c23f0bd | 1508 | while ((bio = bio_list_pop(blist))) |
1d1068ce | 1509 | free_tio(bio); |
318716dd | 1510 | } |
9015df24 AK |
1511 | } |
1512 | ||
34dbaa88 | 1513 | static unsigned int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
4a2fe296 MS |
1514 | unsigned int num_bios, unsigned int *len, |
1515 | gfp_t gfp_flag) | |
06a426ce | 1516 | { |
318716dd | 1517 | struct bio_list blist = BIO_EMPTY_LIST; |
8eabf5d0 | 1518 | struct bio *clone; |
564b5c54 | 1519 | unsigned int ret = 0; |
06a426ce | 1520 | |
4a2fe296 MS |
1521 | if (WARN_ON_ONCE(num_bios == 0)) /* num_bios = 0 is a bug in caller */ |
1522 | return 0; | |
1523 | ||
1524 | /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */ | |
1525 | if (len) | |
1526 | setup_split_accounting(ci, *len); | |
1527 | ||
1528 | /* | |
1529 | * Using alloc_multiple_bios(), even if num_bios is 1, to consistently | |
1530 | * support allocating using GFP_NOWAIT with GFP_NOIO fallback. | |
1531 | */ | |
1532 | alloc_multiple_bios(&blist, ci, ti, num_bios, len, gfp_flag); | |
1533 | while ((clone = bio_list_pop(&blist))) { | |
1534 | if (num_bios > 1) | |
655f3aad | 1535 | dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO); |
4a2fe296 MS |
1536 | __map_bio(clone); |
1537 | ret += 1; | |
318716dd | 1538 | } |
0f14d60a ML |
1539 | |
1540 | return ret; | |
06a426ce MS |
1541 | } |
1542 | ||
332f2b1e | 1543 | static void __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1544 | { |
564b5c54 | 1545 | struct dm_table *t = ci->map; |
828678b8 MS |
1546 | struct bio flush_bio; |
1547 | ||
1548 | /* | |
1549 | * Use an on-stack bio for this, it's safe since we don't | |
1550 | * need to reference it after submit. It's just used as | |
1551 | * the basis for the clone(s). | |
1552 | */ | |
49add496 CH |
1553 | bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0, |
1554 | REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC); | |
47d95102 | 1555 | |
828678b8 MS |
1556 | ci->bio = &flush_bio; |
1557 | ci->sector_count = 0; | |
92b914e2 | 1558 | ci->io->tio.clone.bi_iter.bi_size = 0; |
f9ab94ce | 1559 | |
564b5c54 MS |
1560 | for (unsigned int i = 0; i < t->num_targets; i++) { |
1561 | unsigned int bios; | |
1562 | struct dm_target *ti = dm_table_get_target(t, i); | |
0f14d60a | 1563 | |
4a2fe296 MS |
1564 | if (unlikely(ti->num_flush_bios == 0)) |
1565 | continue; | |
1566 | ||
0f14d60a | 1567 | atomic_add(ti->num_flush_bios, &ci->io->io_count); |
4a2fe296 MS |
1568 | bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, |
1569 | NULL, GFP_NOWAIT); | |
0f14d60a ML |
1570 | atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count); |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * alloc_io() takes one extra reference for submission, so the | |
1575 | * reference won't reach 0 without the following subtraction | |
1576 | */ | |
1577 | atomic_sub(1, &ci->io->io_count); | |
828678b8 MS |
1578 | |
1579 | bio_uninit(ci->bio); | |
f9ab94ce MP |
1580 | } |
1581 | ||
4a2fe296 MS |
1582 | static void __send_abnormal_io(struct clone_info *ci, struct dm_target *ti, |
1583 | unsigned int num_bios, unsigned int max_granularity, | |
1584 | unsigned int max_sectors) | |
ba1cbad9 | 1585 | { |
86a3238c | 1586 | unsigned int len, bios; |
ba1cbad9 | 1587 | |
3720281d | 1588 | len = min_t(sector_t, ci->sector_count, |
be04c14a | 1589 | __max_io_len(ti, ci->sector, max_granularity, max_sectors)); |
51b86f9a | 1590 | |
0f14d60a | 1591 | atomic_add(num_bios, &ci->io->io_count); |
4a2fe296 | 1592 | bios = __send_duplicate_bios(ci, ti, num_bios, &len, GFP_NOIO); |
0f14d60a ML |
1593 | /* |
1594 | * alloc_io() takes one extra reference for submission, so the | |
1595 | * reference won't reach 0 without the following (+1) subtraction | |
1596 | */ | |
1597 | atomic_sub(num_bios - bios + 1, &ci->io->io_count); | |
7dd06a25 | 1598 | |
3d7f4562 MS |
1599 | ci->sector += len; |
1600 | ci->sector_count -= len; | |
ba1cbad9 MS |
1601 | } |
1602 | ||
568c73a3 MS |
1603 | static bool is_abnormal_io(struct bio *bio) |
1604 | { | |
a3282b43 | 1605 | enum req_op op = bio_op(bio); |
568c73a3 | 1606 | |
4edadf6d MS |
1607 | if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) { |
1608 | switch (op) { | |
1609 | case REQ_OP_DISCARD: | |
1610 | case REQ_OP_SECURE_ERASE: | |
1611 | case REQ_OP_WRITE_ZEROES: | |
1612 | return true; | |
1613 | default: | |
1614 | break; | |
1615 | } | |
568c73a3 MS |
1616 | } |
1617 | ||
4edadf6d | 1618 | return false; |
568c73a3 MS |
1619 | } |
1620 | ||
4edadf6d MS |
1621 | static blk_status_t __process_abnormal_io(struct clone_info *ci, |
1622 | struct dm_target *ti) | |
0519c71e | 1623 | { |
86a3238c | 1624 | unsigned int num_bios = 0; |
06961c48 | 1625 | unsigned int max_granularity = 0; |
be04c14a | 1626 | unsigned int max_sectors = 0; |
13f6facf | 1627 | struct queue_limits *limits = dm_get_queue_limits(ti->table->md); |
0519c71e | 1628 | |
e6fc9f62 | 1629 | switch (bio_op(ci->bio)) { |
9679b5a7 MS |
1630 | case REQ_OP_DISCARD: |
1631 | num_bios = ti->num_discard_bios; | |
be04c14a | 1632 | max_sectors = limits->max_discard_sectors; |
13f6facf | 1633 | if (ti->max_discard_granularity) |
be04c14a | 1634 | max_granularity = max_sectors; |
9679b5a7 MS |
1635 | break; |
1636 | case REQ_OP_SECURE_ERASE: | |
1637 | num_bios = ti->num_secure_erase_bios; | |
be04c14a | 1638 | max_sectors = limits->max_secure_erase_sectors; |
13f6facf | 1639 | if (ti->max_secure_erase_granularity) |
be04c14a | 1640 | max_granularity = max_sectors; |
9679b5a7 | 1641 | break; |
9679b5a7 MS |
1642 | case REQ_OP_WRITE_ZEROES: |
1643 | num_bios = ti->num_write_zeroes_bios; | |
be04c14a | 1644 | max_sectors = limits->max_write_zeroes_sectors; |
13f6facf | 1645 | if (ti->max_write_zeroes_granularity) |
be04c14a | 1646 | max_granularity = max_sectors; |
9679b5a7 | 1647 | break; |
2d9b02be BVA |
1648 | default: |
1649 | break; | |
9679b5a7 | 1650 | } |
0519c71e | 1651 | |
e6fc9f62 MS |
1652 | /* |
1653 | * Even though the device advertised support for this type of | |
1654 | * request, that does not mean every target supports it, and | |
1655 | * reconfiguration might also have changed that since the | |
1656 | * check was performed. | |
1657 | */ | |
84b98f4c | 1658 | if (unlikely(!num_bios)) |
4edadf6d MS |
1659 | return BLK_STS_NOTSUPP; |
1660 | ||
4a2fe296 MS |
1661 | __send_abnormal_io(ci, ti, num_bios, max_granularity, max_sectors); |
1662 | ||
4edadf6d | 1663 | return BLK_STS_OK; |
0519c71e MS |
1664 | } |
1665 | ||
b99fdcdc | 1666 | /* |
ec211631 | 1667 | * Reuse ->bi_private as dm_io list head for storing all dm_io instances |
b99fdcdc ML |
1668 | * associated with this bio, and this bio's bi_private needs to be |
1669 | * stored in dm_io->data before the reuse. | |
1670 | * | |
1671 | * bio->bi_private is owned by fs or upper layer, so block layer won't | |
1672 | * touch it after splitting. Meantime it won't be changed by anyone after | |
1673 | * bio is submitted. So this reuse is safe. | |
1674 | */ | |
ec211631 | 1675 | static inline struct dm_io **dm_poll_list_head(struct bio *bio) |
b99fdcdc | 1676 | { |
ec211631 | 1677 | return (struct dm_io **)&bio->bi_private; |
b99fdcdc ML |
1678 | } |
1679 | ||
1680 | static void dm_queue_poll_io(struct bio *bio, struct dm_io *io) | |
1681 | { | |
ec211631 | 1682 | struct dm_io **head = dm_poll_list_head(bio); |
b99fdcdc ML |
1683 | |
1684 | if (!(bio->bi_opf & REQ_DM_POLL_LIST)) { | |
1685 | bio->bi_opf |= REQ_DM_POLL_LIST; | |
1686 | /* | |
1687 | * Save .bi_private into dm_io, so that we can reuse | |
ec211631 | 1688 | * .bi_private as dm_io list head for storing dm_io list |
b99fdcdc ML |
1689 | */ |
1690 | io->data = bio->bi_private; | |
1691 | ||
b99fdcdc ML |
1692 | /* tell block layer to poll for completion */ |
1693 | bio->bi_cookie = ~BLK_QC_T_NONE; | |
ec211631 ML |
1694 | |
1695 | io->next = NULL; | |
b99fdcdc ML |
1696 | } else { |
1697 | /* | |
1698 | * bio recursed due to split, reuse original poll list, | |
1699 | * and save bio->bi_private too. | |
1700 | */ | |
ec211631 ML |
1701 | io->data = (*head)->data; |
1702 | io->next = *head; | |
b99fdcdc ML |
1703 | } |
1704 | ||
ec211631 | 1705 | *head = io; |
b99fdcdc ML |
1706 | } |
1707 | ||
e4c93811 AK |
1708 | /* |
1709 | * Select the correct strategy for processing a non-flush bio. | |
1710 | */ | |
84b98f4c | 1711 | static blk_status_t __split_and_process_bio(struct clone_info *ci) |
0ce65797 | 1712 | { |
66bdaa43 | 1713 | struct bio *clone; |
512875bd | 1714 | struct dm_target *ti; |
86a3238c | 1715 | unsigned int len; |
0ce65797 | 1716 | |
512875bd | 1717 | ti = dm_table_find_target(ci->map, ci->sector); |
4edadf6d MS |
1718 | if (unlikely(!ti)) |
1719 | return BLK_STS_IOERR; | |
1ee88de3 | 1720 | |
1ee88de3 | 1721 | if (unlikely(ci->is_abnormal_io)) |
4edadf6d | 1722 | return __process_abnormal_io(ci, ti); |
3d7f4562 | 1723 | |
b99fdcdc ML |
1724 | /* |
1725 | * Only support bio polling for normal IO, and the target io is | |
1726 | * exactly inside the dm_io instance (verified in dm_poll_dm_io) | |
1727 | */ | |
a3282b43 | 1728 | ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED); |
0ce65797 | 1729 | |
3720281d | 1730 | len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count); |
7dd76d1f | 1731 | setup_split_accounting(ci, len); |
6f25dd1c MS |
1732 | |
1733 | if (unlikely(ci->bio->bi_opf & REQ_NOWAIT)) { | |
1734 | if (unlikely(!dm_target_supports_nowait(ti->type))) | |
1735 | return BLK_STS_NOTSUPP; | |
1736 | ||
1737 | clone = alloc_tio(ci, ti, 0, &len, GFP_NOWAIT); | |
1738 | if (unlikely(!clone)) | |
1739 | return BLK_STS_AGAIN; | |
1740 | } else { | |
1741 | clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO); | |
1742 | } | |
66bdaa43 | 1743 | __map_bio(clone); |
0ce65797 | 1744 | |
1c3b13e6 KO |
1745 | ci->sector += len; |
1746 | ci->sector_count -= len; | |
0ce65797 | 1747 | |
84b98f4c | 1748 | return BLK_STS_OK; |
0ce65797 MS |
1749 | } |
1750 | ||
6f25dd1c | 1751 | static void init_clone_info(struct clone_info *ci, struct dm_io *io, |
4edadf6d | 1752 | struct dm_table *map, struct bio *bio, bool is_abnormal) |
978e51ba MS |
1753 | { |
1754 | ci->map = map; | |
6f25dd1c | 1755 | ci->io = io; |
d41e077a | 1756 | ci->bio = bio; |
4edadf6d | 1757 | ci->is_abnormal_io = is_abnormal; |
b99fdcdc | 1758 | ci->submit_as_polled = false; |
978e51ba | 1759 | ci->sector = bio->bi_iter.bi_sector; |
d41e077a MS |
1760 | ci->sector_count = bio_sectors(bio); |
1761 | ||
1762 | /* Shouldn't happen but sector_count was being set to 0 so... */ | |
442761fd MS |
1763 | if (static_branch_unlikely(&zoned_enabled) && |
1764 | WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count)) | |
d41e077a | 1765 | ci->sector_count = 0; |
978e51ba MS |
1766 | } |
1767 | ||
1da177e4 | 1768 | /* |
14fe594d | 1769 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1770 | */ |
96c9865c MS |
1771 | static void dm_split_and_process_bio(struct mapped_device *md, |
1772 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1773 | { |
1da177e4 | 1774 | struct clone_info ci; |
4857abf6 | 1775 | struct dm_io *io; |
84b98f4c | 1776 | blk_status_t error = BLK_STS_OK; |
4edadf6d MS |
1777 | bool is_abnormal; |
1778 | ||
1779 | is_abnormal = is_abnormal_io(bio); | |
1780 | if (unlikely(is_abnormal)) { | |
1781 | /* | |
5a97806f | 1782 | * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc) |
4edadf6d MS |
1783 | * otherwise associated queue_limits won't be imposed. |
1784 | */ | |
5a97806f | 1785 | bio = bio_split_to_limits(bio); |
613b1488 JA |
1786 | if (!bio) |
1787 | return; | |
4edadf6d | 1788 | } |
1da177e4 | 1789 | |
6f25dd1c MS |
1790 | /* Only support nowait for normal IO */ |
1791 | if (unlikely(bio->bi_opf & REQ_NOWAIT) && !is_abnormal) { | |
1792 | io = alloc_io(md, bio, GFP_NOWAIT); | |
1793 | if (unlikely(!io)) { | |
1794 | /* Unable to do anything without dm_io. */ | |
1795 | bio_wouldblock_error(bio); | |
1796 | return; | |
1797 | } | |
1798 | } else { | |
1799 | io = alloc_io(md, bio, GFP_NOIO); | |
1800 | } | |
1801 | init_clone_info(&ci, io, map, bio, is_abnormal); | |
0ce65797 | 1802 | |
1eff9d32 | 1803 | if (bio->bi_opf & REQ_PREFLUSH) { |
332f2b1e | 1804 | __send_empty_flush(&ci); |
e2736347 | 1805 | /* dm_io_complete submits any data associated with flush */ |
d41e077a | 1806 | goto out; |
d87f4c14 | 1807 | } |
0ce65797 | 1808 | |
d41e077a MS |
1809 | error = __split_and_process_bio(&ci); |
1810 | if (error || !ci.sector_count) | |
1811 | goto out; | |
d41e077a MS |
1812 | /* |
1813 | * Remainder must be passed to submit_bio_noacct() so it gets handled | |
1814 | * *after* bios already submitted have been completely processed. | |
d41e077a | 1815 | */ |
8b211aac ML |
1816 | bio_trim(bio, io->sectors, ci.sector_count); |
1817 | trace_block_split(bio, bio->bi_iter.bi_sector); | |
1818 | bio_inc_remaining(bio); | |
d41e077a MS |
1819 | submit_bio_noacct(bio); |
1820 | out: | |
b99fdcdc ML |
1821 | /* |
1822 | * Drop the extra reference count for non-POLLED bio, and hold one | |
1823 | * reference for POLLED bio, which will be released in dm_poll_bio | |
1824 | * | |
ec211631 ML |
1825 | * Add every dm_io instance into the dm_io list head which is stored |
1826 | * in bio->bi_private, so that dm_poll_bio can poll them all. | |
b99fdcdc | 1827 | */ |
0f14d60a ML |
1828 | if (error || !ci.submit_as_polled) { |
1829 | /* | |
1830 | * In case of submission failure, the extra reference for | |
1831 | * submitting io isn't consumed yet | |
1832 | */ | |
1833 | if (error) | |
1834 | atomic_dec(&io->io_count); | |
1835 | dm_io_dec_pending(io, error); | |
1836 | } else | |
4857abf6 | 1837 | dm_queue_poll_io(bio, io); |
0ce65797 MS |
1838 | } |
1839 | ||
3e08773c | 1840 | static void dm_submit_bio(struct bio *bio) |
cec47e3d | 1841 | { |
309dca30 | 1842 | struct mapped_device *md = bio->bi_bdev->bd_disk->private_data; |
83d5e5b0 MP |
1843 | int srcu_idx; |
1844 | struct dm_table *map; | |
cec47e3d | 1845 | |
a9ce3853 | 1846 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1847 | |
fa247089 MS |
1848 | /* If suspended, or map not yet available, queue this IO for later */ |
1849 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) || | |
1850 | unlikely(!map)) { | |
6abc4946 KK |
1851 | if (bio->bi_opf & REQ_NOWAIT) |
1852 | bio_wouldblock_error(bio); | |
b2abdb1b | 1853 | else if (bio->bi_opf & REQ_RAHEAD) |
54d9a1b4 | 1854 | bio_io_error(bio); |
b2abdb1b MS |
1855 | else |
1856 | queue_io(md, bio); | |
1857 | goto out; | |
cec47e3d | 1858 | } |
1da177e4 | 1859 | |
96c9865c | 1860 | dm_split_and_process_bio(md, map, bio); |
b2abdb1b | 1861 | out: |
a9ce3853 | 1862 | dm_put_live_table(md, srcu_idx); |
978e51ba MS |
1863 | } |
1864 | ||
b99fdcdc ML |
1865 | static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob, |
1866 | unsigned int flags) | |
1867 | { | |
655f3aad | 1868 | WARN_ON_ONCE(!dm_tio_is_normal(&io->tio)); |
b99fdcdc ML |
1869 | |
1870 | /* don't poll if the mapped io is done */ | |
1871 | if (atomic_read(&io->io_count) > 1) | |
1872 | bio_poll(&io->tio.clone, iob, flags); | |
1873 | ||
1874 | /* bio_poll holds the last reference */ | |
1875 | return atomic_read(&io->io_count) == 1; | |
1876 | } | |
1877 | ||
1878 | static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob, | |
1879 | unsigned int flags) | |
1880 | { | |
ec211631 ML |
1881 | struct dm_io **head = dm_poll_list_head(bio); |
1882 | struct dm_io *list = *head; | |
1883 | struct dm_io *tmp = NULL; | |
1884 | struct dm_io *curr, *next; | |
b99fdcdc ML |
1885 | |
1886 | /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */ | |
1887 | if (!(bio->bi_opf & REQ_DM_POLL_LIST)) | |
1888 | return 0; | |
1889 | ||
ec211631 | 1890 | WARN_ON_ONCE(!list); |
b99fdcdc ML |
1891 | |
1892 | /* | |
1893 | * Restore .bi_private before possibly completing dm_io. | |
1894 | * | |
1895 | * bio_poll() is only possible once @bio has been completely | |
1896 | * submitted via submit_bio_noacct()'s depth-first submission. | |
1897 | * So there is no dm_queue_poll_io() race associated with | |
1898 | * clearing REQ_DM_POLL_LIST here. | |
1899 | */ | |
1900 | bio->bi_opf &= ~REQ_DM_POLL_LIST; | |
ec211631 | 1901 | bio->bi_private = list->data; |
b99fdcdc | 1902 | |
ec211631 ML |
1903 | for (curr = list, next = curr->next; curr; curr = next, next = |
1904 | curr ? curr->next : NULL) { | |
1905 | if (dm_poll_dm_io(curr, iob, flags)) { | |
b99fdcdc | 1906 | /* |
84b98f4c MS |
1907 | * clone_endio() has already occurred, so no |
1908 | * error handling is needed here. | |
b99fdcdc | 1909 | */ |
ec211631 ML |
1910 | __dm_io_dec_pending(curr); |
1911 | } else { | |
1912 | curr->next = tmp; | |
1913 | tmp = curr; | |
b99fdcdc ML |
1914 | } |
1915 | } | |
1916 | ||
1917 | /* Not done? */ | |
ec211631 | 1918 | if (tmp) { |
b99fdcdc ML |
1919 | bio->bi_opf |= REQ_DM_POLL_LIST; |
1920 | /* Reset bio->bi_private to dm_io list head */ | |
ec211631 | 1921 | *head = tmp; |
b99fdcdc ML |
1922 | return 0; |
1923 | } | |
1924 | return 1; | |
1925 | } | |
1926 | ||
a4a82ce3 HM |
1927 | /* |
1928 | *--------------------------------------------------------------- | |
1da177e4 | 1929 | * An IDR is used to keep track of allocated minor numbers. |
a4a82ce3 HM |
1930 | *--------------------------------------------------------------- |
1931 | */ | |
2b06cfff | 1932 | static void free_minor(int minor) |
1da177e4 | 1933 | { |
f32c10b0 | 1934 | spin_lock(&_minor_lock); |
1da177e4 | 1935 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1936 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1937 | } |
1938 | ||
1939 | /* | |
1940 | * See if the device with a specific minor # is free. | |
1941 | */ | |
cf13ab8e | 1942 | static int specific_minor(int minor) |
1da177e4 | 1943 | { |
c9d76be6 | 1944 | int r; |
1da177e4 LT |
1945 | |
1946 | if (minor >= (1 << MINORBITS)) | |
1947 | return -EINVAL; | |
1948 | ||
c9d76be6 | 1949 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1950 | spin_lock(&_minor_lock); |
1da177e4 | 1951 | |
c9d76be6 | 1952 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1953 | |
f32c10b0 | 1954 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1955 | idr_preload_end(); |
1956 | if (r < 0) | |
1957 | return r == -ENOSPC ? -EBUSY : r; | |
1958 | return 0; | |
1da177e4 LT |
1959 | } |
1960 | ||
cf13ab8e | 1961 | static int next_free_minor(int *minor) |
1da177e4 | 1962 | { |
c9d76be6 | 1963 | int r; |
62f75c2f | 1964 | |
c9d76be6 | 1965 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1966 | spin_lock(&_minor_lock); |
1da177e4 | 1967 | |
c9d76be6 | 1968 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1969 | |
f32c10b0 | 1970 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1971 | idr_preload_end(); |
1972 | if (r < 0) | |
1973 | return r; | |
1974 | *minor = r; | |
1975 | return 0; | |
1da177e4 LT |
1976 | } |
1977 | ||
83d5cde4 | 1978 | static const struct block_device_operations dm_blk_dops; |
681cc5e8 | 1979 | static const struct block_device_operations dm_rq_blk_dops; |
f26c5719 | 1980 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1981 | |
53d5914f MP |
1982 | static void dm_wq_work(struct work_struct *work); |
1983 | ||
aa6ce87a | 1984 | #ifdef CONFIG_BLK_INLINE_ENCRYPTION |
cb77cb5a | 1985 | static void dm_queue_destroy_crypto_profile(struct request_queue *q) |
aa6ce87a | 1986 | { |
cb77cb5a | 1987 | dm_destroy_crypto_profile(q->crypto_profile); |
aa6ce87a ST |
1988 | } |
1989 | ||
1990 | #else /* CONFIG_BLK_INLINE_ENCRYPTION */ | |
1991 | ||
cb77cb5a | 1992 | static inline void dm_queue_destroy_crypto_profile(struct request_queue *q) |
aa6ce87a ST |
1993 | { |
1994 | } | |
1995 | #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */ | |
1996 | ||
0f20972f MS |
1997 | static void cleanup_mapped_device(struct mapped_device *md) |
1998 | { | |
0f20972f MS |
1999 | if (md->wq) |
2000 | destroy_workqueue(md->wq); | |
29dec90a | 2001 | dm_free_md_mempools(md->mempools); |
0f20972f | 2002 | |
f26c5719 | 2003 | if (md->dax_dev) { |
fb08a190 | 2004 | dax_remove_host(md->disk); |
f26c5719 DW |
2005 | kill_dax(md->dax_dev); |
2006 | put_dax(md->dax_dev); | |
2007 | md->dax_dev = NULL; | |
2008 | } | |
2009 | ||
588b7f5d | 2010 | dm_cleanup_zoned_dev(md); |
0f20972f MS |
2011 | if (md->disk) { |
2012 | spin_lock(&_minor_lock); | |
2013 | md->disk->private_data = NULL; | |
2014 | spin_unlock(&_minor_lock); | |
89f871af | 2015 | if (dm_get_md_type(md) != DM_TYPE_NONE) { |
1a581b72 CH |
2016 | struct table_device *td; |
2017 | ||
89f871af | 2018 | dm_sysfs_exit(md); |
1a581b72 CH |
2019 | list_for_each_entry(td, &md->table_devices, list) { |
2020 | bd_unlink_disk_holder(td->dm_dev.bdev, | |
2021 | md->disk); | |
2022 | } | |
d563792c YK |
2023 | |
2024 | /* | |
2025 | * Hold lock to make sure del_gendisk() won't concurrent | |
2026 | * with open/close_table_device(). | |
2027 | */ | |
2028 | mutex_lock(&md->table_devices_lock); | |
89f871af | 2029 | del_gendisk(md->disk); |
d563792c | 2030 | mutex_unlock(&md->table_devices_lock); |
89f871af | 2031 | } |
cb77cb5a | 2032 | dm_queue_destroy_crypto_profile(md->queue); |
8b9ab626 | 2033 | put_disk(md->disk); |
74a2b6ec | 2034 | } |
0f20972f | 2035 | |
9f6dc633 MS |
2036 | if (md->pending_io) { |
2037 | free_percpu(md->pending_io); | |
2038 | md->pending_io = NULL; | |
2039 | } | |
2040 | ||
d09960b0 TE |
2041 | cleanup_srcu_struct(&md->io_barrier); |
2042 | ||
d5ffebdd MS |
2043 | mutex_destroy(&md->suspend_lock); |
2044 | mutex_destroy(&md->type_lock); | |
2045 | mutex_destroy(&md->table_devices_lock); | |
a666e5c0 | 2046 | mutex_destroy(&md->swap_bios_lock); |
d5ffebdd | 2047 | |
4cc96131 | 2048 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
2049 | } |
2050 | ||
1da177e4 LT |
2051 | /* |
2052 | * Allocate and initialise a blank device with a given minor. | |
2053 | */ | |
2b06cfff | 2054 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 2055 | { |
115485e8 MS |
2056 | int r, numa_node_id = dm_get_numa_node(); |
2057 | struct mapped_device *md; | |
ba61fdd1 | 2058 | void *old_md; |
1da177e4 | 2059 | |
856eb091 | 2060 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 | 2061 | if (!md) { |
43e6c111 | 2062 | DMERR("unable to allocate device, out of memory."); |
1da177e4 LT |
2063 | return NULL; |
2064 | } | |
2065 | ||
10da4f79 | 2066 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2067 | goto bad_module_get; |
10da4f79 | 2068 | |
1da177e4 | 2069 | /* get a minor number for the dev */ |
2b06cfff | 2070 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2071 | r = next_free_minor(&minor); |
2b06cfff | 2072 | else |
cf13ab8e | 2073 | r = specific_minor(minor); |
1da177e4 | 2074 | if (r < 0) |
6ed7ade8 | 2075 | goto bad_minor; |
1da177e4 | 2076 | |
83d5e5b0 MP |
2077 | r = init_srcu_struct(&md->io_barrier); |
2078 | if (r < 0) | |
2079 | goto bad_io_barrier; | |
2080 | ||
115485e8 | 2081 | md->numa_node_id = numa_node_id; |
591ddcfc | 2082 | md->init_tio_pdu = false; |
a5664dad | 2083 | md->type = DM_TYPE_NONE; |
e61290a4 | 2084 | mutex_init(&md->suspend_lock); |
a5664dad | 2085 | mutex_init(&md->type_lock); |
86f1152b | 2086 | mutex_init(&md->table_devices_lock); |
022c2611 | 2087 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2088 | atomic_set(&md->holders, 1); |
5c6bd75d | 2089 | atomic_set(&md->open_count, 0); |
1da177e4 | 2090 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2091 | atomic_set(&md->uevent_seq, 0); |
2092 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2093 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2094 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2095 | |
47ace7e0 | 2096 | /* |
c62b37d9 CH |
2097 | * default to bio-based until DM table is loaded and md->type |
2098 | * established. If request-based table is loaded: blk-mq will | |
2099 | * override accordingly. | |
47ace7e0 | 2100 | */ |
74fe6ba9 | 2101 | md->disk = blk_alloc_disk(md->numa_node_id); |
1da177e4 | 2102 | if (!md->disk) |
0f20972f | 2103 | goto bad; |
74fe6ba9 | 2104 | md->queue = md->disk->queue; |
1da177e4 | 2105 | |
f0b04115 | 2106 | init_waitqueue_head(&md->wait); |
53d5914f | 2107 | INIT_WORK(&md->work, dm_wq_work); |
8b211aac | 2108 | INIT_WORK(&md->requeue_work, dm_wq_requeue_work); |
f0b04115 | 2109 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2110 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 2111 | |
8b211aac | 2112 | md->requeue_list = NULL; |
a666e5c0 MP |
2113 | md->swap_bios = get_swap_bios(); |
2114 | sema_init(&md->swap_bios_semaphore, md->swap_bios); | |
2115 | mutex_init(&md->swap_bios_lock); | |
2116 | ||
1da177e4 LT |
2117 | md->disk->major = _major; |
2118 | md->disk->first_minor = minor; | |
74fe6ba9 | 2119 | md->disk->minors = 1; |
1ebe2e5f | 2120 | md->disk->flags |= GENHD_FL_NO_PART; |
1da177e4 | 2121 | md->disk->fops = &dm_blk_dops; |
1da177e4 LT |
2122 | md->disk->private_data = md; |
2123 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 | 2124 | |
5d2a228b | 2125 | if (IS_ENABLED(CONFIG_FS_DAX)) { |
30c6828a | 2126 | md->dax_dev = alloc_dax(md, &dm_dax_ops); |
d7519392 CH |
2127 | if (IS_ERR(md->dax_dev)) { |
2128 | md->dax_dev = NULL; | |
976431b0 | 2129 | goto bad; |
d7519392 | 2130 | } |
7ac5360c CH |
2131 | set_dax_nocache(md->dax_dev); |
2132 | set_dax_nomc(md->dax_dev); | |
fb08a190 | 2133 | if (dax_add_host(md->dax_dev, md->disk)) |
976431b0 DW |
2134 | goto bad; |
2135 | } | |
f26c5719 | 2136 | |
7e51f257 | 2137 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2138 | |
c7c879ee | 2139 | md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name); |
304f3f6a | 2140 | if (!md->wq) |
0f20972f | 2141 | goto bad; |
304f3f6a | 2142 | |
9f6dc633 MS |
2143 | md->pending_io = alloc_percpu(unsigned long); |
2144 | if (!md->pending_io) | |
2145 | goto bad; | |
2146 | ||
d3aa3e06 JJ |
2147 | r = dm_stats_init(&md->stats); |
2148 | if (r < 0) | |
2149 | goto bad; | |
fd2ed4d2 | 2150 | |
ba61fdd1 | 2151 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2152 | spin_lock(&_minor_lock); |
ba61fdd1 | 2153 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2154 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2155 | |
2156 | BUG_ON(old_md != MINOR_ALLOCED); | |
2157 | ||
1da177e4 LT |
2158 | return md; |
2159 | ||
0f20972f MS |
2160 | bad: |
2161 | cleanup_mapped_device(md); | |
83d5e5b0 | 2162 | bad_io_barrier: |
1da177e4 | 2163 | free_minor(minor); |
6ed7ade8 | 2164 | bad_minor: |
10da4f79 | 2165 | module_put(THIS_MODULE); |
6ed7ade8 | 2166 | bad_module_get: |
856eb091 | 2167 | kvfree(md); |
1da177e4 LT |
2168 | return NULL; |
2169 | } | |
2170 | ||
ae9da83f JN |
2171 | static void unlock_fs(struct mapped_device *md); |
2172 | ||
1da177e4 LT |
2173 | static void free_dev(struct mapped_device *md) |
2174 | { | |
f331c029 | 2175 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2176 | |
32a926da | 2177 | unlock_fs(md); |
2eb6e1e3 | 2178 | |
0f20972f | 2179 | cleanup_mapped_device(md); |
63a4f065 | 2180 | |
992ec6a9 | 2181 | WARN_ON_ONCE(!list_empty(&md->table_devices)); |
63a4f065 | 2182 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2183 | free_minor(minor); |
2184 | ||
10da4f79 | 2185 | module_put(THIS_MODULE); |
856eb091 | 2186 | kvfree(md); |
1da177e4 LT |
2187 | } |
2188 | ||
2189 | /* | |
2190 | * Bind a table to the device. | |
2191 | */ | |
2192 | static void event_callback(void *context) | |
2193 | { | |
7a8c3d3b MA |
2194 | unsigned long flags; |
2195 | LIST_HEAD(uevents); | |
d695e441 | 2196 | struct mapped_device *md = context; |
1da177e4 | 2197 | |
7a8c3d3b MA |
2198 | spin_lock_irqsave(&md->uevent_lock, flags); |
2199 | list_splice_init(&md->uevent_list, &uevents); | |
2200 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2201 | ||
ed9e1982 | 2202 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2203 | |
1da177e4 LT |
2204 | atomic_inc(&md->event_nr); |
2205 | wake_up(&md->eventq); | |
62e08243 | 2206 | dm_issue_global_event(); |
1da177e4 LT |
2207 | } |
2208 | ||
042d2a9b AK |
2209 | /* |
2210 | * Returns old map, which caller must destroy. | |
2211 | */ | |
2212 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2213 | struct queue_limits *limits) | |
1da177e4 | 2214 | { |
042d2a9b | 2215 | struct dm_table *old_map; |
1da177e4 | 2216 | sector_t size; |
2a2a4c51 | 2217 | int ret; |
1da177e4 | 2218 | |
5a8f1f80 BVA |
2219 | lockdep_assert_held(&md->suspend_lock); |
2220 | ||
1da177e4 | 2221 | size = dm_table_get_size(t); |
3ac51e74 DW |
2222 | |
2223 | /* | |
2224 | * Wipe any geometry if the size of the table changed. | |
2225 | */ | |
fd2ed4d2 | 2226 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2227 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2228 | ||
7533afa1 | 2229 | set_capacity(md->disk, size); |
d5816876 | 2230 | |
2ca3310e AK |
2231 | dm_table_event_callback(t, event_callback, md); |
2232 | ||
f5b4aee1 | 2233 | if (dm_table_request_based(t)) { |
16f12266 | 2234 | /* |
9c37de29 MS |
2235 | * Leverage the fact that request-based DM targets are |
2236 | * immutable singletons - used to optimize dm_mq_queue_rq. | |
16f12266 MS |
2237 | */ |
2238 | md->immutable_target = dm_table_get_immutable_target(t); | |
e6ee8c0b | 2239 | |
29dec90a CH |
2240 | /* |
2241 | * There is no need to reload with request-based dm because the | |
2242 | * size of front_pad doesn't change. | |
2243 | * | |
2244 | * Note for future: If you are to reload bioset, prep-ed | |
2245 | * requests in the queue may refer to bio from the old bioset, | |
2246 | * so you must walk through the queue to unprep. | |
2247 | */ | |
2248 | if (!md->mempools) { | |
2249 | md->mempools = t->mempools; | |
2250 | t->mempools = NULL; | |
2251 | } | |
2252 | } else { | |
2253 | /* | |
2254 | * The md may already have mempools that need changing. | |
2255 | * If so, reload bioset because front_pad may have changed | |
2256 | * because a different table was loaded. | |
2257 | */ | |
2258 | dm_free_md_mempools(md->mempools); | |
2259 | md->mempools = t->mempools; | |
2260 | t->mempools = NULL; | |
2a2a4c51 | 2261 | } |
e6ee8c0b | 2262 | |
f5b4aee1 | 2263 | ret = dm_table_set_restrictions(t, md->queue, limits); |
bb37d772 DLM |
2264 | if (ret) { |
2265 | old_map = ERR_PTR(ret); | |
2266 | goto out; | |
2267 | } | |
2268 | ||
a12f5d48 | 2269 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2270 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2271 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2272 | ||
41abc4e1 HR |
2273 | if (old_map) |
2274 | dm_sync_table(md); | |
2a2a4c51 | 2275 | out: |
042d2a9b | 2276 | return old_map; |
1da177e4 LT |
2277 | } |
2278 | ||
a7940155 AK |
2279 | /* |
2280 | * Returns unbound table for the caller to free. | |
2281 | */ | |
2282 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2283 | { |
a12f5d48 | 2284 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2285 | |
2286 | if (!map) | |
a7940155 | 2287 | return NULL; |
1da177e4 LT |
2288 | |
2289 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2290 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2291 | dm_sync_table(md); |
a7940155 AK |
2292 | |
2293 | return map; | |
1da177e4 LT |
2294 | } |
2295 | ||
2296 | /* | |
2297 | * Constructor for a new device. | |
2298 | */ | |
2b06cfff | 2299 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2300 | { |
2301 | struct mapped_device *md; | |
2302 | ||
2b06cfff | 2303 | md = alloc_dev(minor); |
1da177e4 LT |
2304 | if (!md) |
2305 | return -ENXIO; | |
2306 | ||
91ccbbac TS |
2307 | dm_ima_reset_data(md); |
2308 | ||
1da177e4 LT |
2309 | *result = md; |
2310 | return 0; | |
2311 | } | |
2312 | ||
a5664dad MS |
2313 | /* |
2314 | * Functions to manage md->type. | |
2315 | * All are required to hold md->type_lock. | |
2316 | */ | |
2317 | void dm_lock_md_type(struct mapped_device *md) | |
2318 | { | |
2319 | mutex_lock(&md->type_lock); | |
2320 | } | |
2321 | ||
2322 | void dm_unlock_md_type(struct mapped_device *md) | |
2323 | { | |
2324 | mutex_unlock(&md->type_lock); | |
2325 | } | |
2326 | ||
7e0d574f | 2327 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2328 | { |
00c4fc3b | 2329 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2330 | md->type = type; |
2331 | } | |
2332 | ||
7e0d574f | 2333 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2334 | { |
2335 | return md->type; | |
2336 | } | |
2337 | ||
36a0456f AK |
2338 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2339 | { | |
2340 | return md->immutable_target_type; | |
2341 | } | |
2342 | ||
4a0b4ddf MS |
2343 | /* |
2344 | * Setup the DM device's queue based on md's type | |
2345 | */ | |
591ddcfc | 2346 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2347 | { |
ba305859 | 2348 | enum dm_queue_mode type = dm_table_get_type(t); |
c100ec49 | 2349 | struct queue_limits limits; |
1a581b72 | 2350 | struct table_device *td; |
ba305859 | 2351 | int r; |
bfebd1cd | 2352 | |
545ed20e | 2353 | switch (type) { |
bfebd1cd | 2354 | case DM_TYPE_REQUEST_BASED: |
681cc5e8 | 2355 | md->disk->fops = &dm_rq_blk_dops; |
e83068a5 | 2356 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2357 | if (r) { |
681cc5e8 | 2358 | DMERR("Cannot initialize queue for request-based dm mapped device"); |
bfebd1cd MS |
2359 | return r; |
2360 | } | |
2361 | break; | |
2362 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2363 | case DM_TYPE_DAX_BIO_BASED: |
526d1006 | 2364 | blk_queue_flag_set(QUEUE_FLAG_IO_STAT, md->queue); |
bfebd1cd | 2365 | break; |
7e0d574f BVA |
2366 | case DM_TYPE_NONE: |
2367 | WARN_ON_ONCE(true); | |
2368 | break; | |
4a0b4ddf MS |
2369 | } |
2370 | ||
c100ec49 MS |
2371 | r = dm_calculate_queue_limits(t, &limits); |
2372 | if (r) { | |
2373 | DMERR("Cannot calculate initial queue limits"); | |
2374 | return r; | |
2375 | } | |
bb37d772 DLM |
2376 | r = dm_table_set_restrictions(t, md->queue, &limits); |
2377 | if (r) | |
2378 | return r; | |
2379 | ||
d563792c YK |
2380 | /* |
2381 | * Hold lock to make sure add_disk() and del_gendisk() won't concurrent | |
2382 | * with open_table_device() and close_table_device(). | |
2383 | */ | |
2384 | mutex_lock(&md->table_devices_lock); | |
e7089f65 | 2385 | r = add_disk(md->disk); |
d563792c | 2386 | mutex_unlock(&md->table_devices_lock); |
e7089f65 LC |
2387 | if (r) |
2388 | return r; | |
c100ec49 | 2389 | |
1a581b72 CH |
2390 | /* |
2391 | * Register the holder relationship for devices added before the disk | |
2392 | * was live. | |
2393 | */ | |
2394 | list_for_each_entry(td, &md->table_devices, list) { | |
2395 | r = bd_link_disk_holder(td->dm_dev.bdev, md->disk); | |
2396 | if (r) | |
2397 | goto out_undo_holders; | |
89f871af | 2398 | } |
d563792c | 2399 | |
1a581b72 CH |
2400 | r = dm_sysfs_init(md); |
2401 | if (r) | |
2402 | goto out_undo_holders; | |
2403 | ||
89f871af | 2404 | md->type = type; |
4a0b4ddf | 2405 | return 0; |
1a581b72 CH |
2406 | |
2407 | out_undo_holders: | |
2408 | list_for_each_entry_continue_reverse(td, &md->table_devices, list) | |
2409 | bd_unlink_disk_holder(td->dm_dev.bdev, md->disk); | |
2410 | mutex_lock(&md->table_devices_lock); | |
2411 | del_gendisk(md->disk); | |
2412 | mutex_unlock(&md->table_devices_lock); | |
2413 | return r; | |
4a0b4ddf MS |
2414 | } |
2415 | ||
2bec1f4a | 2416 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2417 | { |
2418 | struct mapped_device *md; | |
86a3238c | 2419 | unsigned int minor = MINOR(dev); |
1da177e4 LT |
2420 | |
2421 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2422 | return NULL; | |
2423 | ||
f32c10b0 | 2424 | spin_lock(&_minor_lock); |
1da177e4 LT |
2425 | |
2426 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2427 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2428 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2429 | md = NULL; | |
2430 | goto out; | |
fba9f90e | 2431 | } |
49de5769 | 2432 | dm_get(md); |
fba9f90e | 2433 | out: |
f32c10b0 | 2434 | spin_unlock(&_minor_lock); |
1da177e4 | 2435 | |
637842cf DT |
2436 | return md; |
2437 | } | |
3cf2e4ba | 2438 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2439 | |
9ade92a9 | 2440 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2441 | { |
9ade92a9 | 2442 | return md->interface_ptr; |
1da177e4 LT |
2443 | } |
2444 | ||
2445 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2446 | { | |
2447 | md->interface_ptr = ptr; | |
2448 | } | |
2449 | ||
2450 | void dm_get(struct mapped_device *md) | |
2451 | { | |
2452 | atomic_inc(&md->holders); | |
3f77316d | 2453 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2454 | } |
2455 | ||
09ee96b2 MP |
2456 | int dm_hold(struct mapped_device *md) |
2457 | { | |
2458 | spin_lock(&_minor_lock); | |
2459 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2460 | spin_unlock(&_minor_lock); | |
2461 | return -EBUSY; | |
2462 | } | |
2463 | dm_get(md); | |
2464 | spin_unlock(&_minor_lock); | |
2465 | return 0; | |
2466 | } | |
2467 | EXPORT_SYMBOL_GPL(dm_hold); | |
2468 | ||
72d94861 AK |
2469 | const char *dm_device_name(struct mapped_device *md) |
2470 | { | |
2471 | return md->name; | |
2472 | } | |
2473 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2474 | ||
3f77316d | 2475 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2476 | { |
1134e5ae | 2477 | struct dm_table *map; |
83d5e5b0 | 2478 | int srcu_idx; |
1da177e4 | 2479 | |
3f77316d | 2480 | might_sleep(); |
fba9f90e | 2481 | |
63a4f065 | 2482 | spin_lock(&_minor_lock); |
3f77316d KU |
2483 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2484 | set_bit(DMF_FREEING, &md->flags); | |
2485 | spin_unlock(&_minor_lock); | |
3b785fbc | 2486 | |
7a5428dc | 2487 | blk_mark_disk_dead(md->disk); |
3f77316d | 2488 | |
ab7c7bb6 MP |
2489 | /* |
2490 | * Take suspend_lock so that presuspend and postsuspend methods | |
2491 | * do not race with internal suspend. | |
2492 | */ | |
2493 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2494 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2495 | if (!dm_suspended_md(md)) { |
2496 | dm_table_presuspend_targets(map); | |
adc0daad | 2497 | set_bit(DMF_SUSPENDED, &md->flags); |
5df96f2b | 2498 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
3f77316d | 2499 | dm_table_postsuspend_targets(map); |
1da177e4 | 2500 | } |
238d991f | 2501 | /* dm_put_live_table must be before fsleep, otherwise deadlock is possible */ |
83d5e5b0 | 2502 | dm_put_live_table(md, srcu_idx); |
2a708cff | 2503 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2504 | |
3f77316d KU |
2505 | /* |
2506 | * Rare, but there may be I/O requests still going to complete, | |
2507 | * for example. Wait for all references to disappear. | |
2508 | * No one should increment the reference count of the mapped_device, | |
2509 | * after the mapped_device state becomes DMF_FREEING. | |
2510 | */ | |
2511 | if (wait) | |
2512 | while (atomic_read(&md->holders)) | |
238d991f | 2513 | fsleep(1000); |
3f77316d KU |
2514 | else if (atomic_read(&md->holders)) |
2515 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2516 | dm_device_name(md), atomic_read(&md->holders)); | |
2517 | ||
3f77316d KU |
2518 | dm_table_destroy(__unbind(md)); |
2519 | free_dev(md); | |
2520 | } | |
2521 | ||
2522 | void dm_destroy(struct mapped_device *md) | |
2523 | { | |
2524 | __dm_destroy(md, true); | |
2525 | } | |
2526 | ||
2527 | void dm_destroy_immediate(struct mapped_device *md) | |
2528 | { | |
2529 | __dm_destroy(md, false); | |
2530 | } | |
2531 | ||
2532 | void dm_put(struct mapped_device *md) | |
2533 | { | |
2534 | atomic_dec(&md->holders); | |
1da177e4 | 2535 | } |
79eb885c | 2536 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2537 | |
9f6dc633 | 2538 | static bool dm_in_flight_bios(struct mapped_device *md) |
85067747 ML |
2539 | { |
2540 | int cpu; | |
9f6dc633 | 2541 | unsigned long sum = 0; |
85067747 | 2542 | |
9f6dc633 MS |
2543 | for_each_possible_cpu(cpu) |
2544 | sum += *per_cpu_ptr(md->pending_io, cpu); | |
85067747 ML |
2545 | |
2546 | return sum != 0; | |
2547 | } | |
2548 | ||
2f064a59 | 2549 | static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state) |
46125c1c MB |
2550 | { |
2551 | int r = 0; | |
9f4c3f87 | 2552 | DEFINE_WAIT(wait); |
46125c1c | 2553 | |
85067747 | 2554 | while (true) { |
9f4c3f87 | 2555 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2556 | |
9f6dc633 | 2557 | if (!dm_in_flight_bios(md)) |
46125c1c MB |
2558 | break; |
2559 | ||
e3fabdfd | 2560 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2561 | r = -EINTR; |
2562 | break; | |
2563 | } | |
2564 | ||
2565 | io_schedule(); | |
2566 | } | |
9f4c3f87 | 2567 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2568 | |
9f6dc633 MS |
2569 | smp_rmb(); |
2570 | ||
46125c1c MB |
2571 | return r; |
2572 | } | |
2573 | ||
2f064a59 | 2574 | static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state) |
85067747 ML |
2575 | { |
2576 | int r = 0; | |
2577 | ||
2578 | if (!queue_is_mq(md->queue)) | |
2579 | return dm_wait_for_bios_completion(md, task_state); | |
2580 | ||
2581 | while (true) { | |
2582 | if (!blk_mq_queue_inflight(md->queue)) | |
2583 | break; | |
2584 | ||
2585 | if (signal_pending_state(task_state, current)) { | |
2586 | r = -EINTR; | |
2587 | break; | |
2588 | } | |
2589 | ||
238d991f | 2590 | fsleep(5000); |
85067747 ML |
2591 | } |
2592 | ||
2593 | return r; | |
2594 | } | |
2595 | ||
1da177e4 LT |
2596 | /* |
2597 | * Process the deferred bios | |
2598 | */ | |
ef208587 | 2599 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2600 | { |
0c2915b8 MS |
2601 | struct mapped_device *md = container_of(work, struct mapped_device, work); |
2602 | struct bio *bio; | |
ef208587 | 2603 | |
3b00b203 | 2604 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 | 2605 | spin_lock_irq(&md->deferred_lock); |
0c2915b8 | 2606 | bio = bio_list_pop(&md->deferred); |
df12ee99 AK |
2607 | spin_unlock_irq(&md->deferred_lock); |
2608 | ||
0c2915b8 | 2609 | if (!bio) |
df12ee99 | 2610 | break; |
022c2611 | 2611 | |
0c2915b8 | 2612 | submit_bio_noacct(bio); |
0ca44fce | 2613 | cond_resched(); |
022c2611 | 2614 | } |
1da177e4 LT |
2615 | } |
2616 | ||
9a1fb464 | 2617 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2618 | { |
3b00b203 | 2619 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2620 | smp_mb__after_atomic(); |
53d5914f | 2621 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2622 | } |
2623 | ||
1da177e4 | 2624 | /* |
042d2a9b | 2625 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2626 | */ |
042d2a9b | 2627 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2628 | { |
87eb5b21 | 2629 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2630 | struct queue_limits limits; |
042d2a9b | 2631 | int r; |
1da177e4 | 2632 | |
e61290a4 | 2633 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2634 | |
2635 | /* device must be suspended */ | |
4f186f8b | 2636 | if (!dm_suspended_md(md)) |
93c534ae | 2637 | goto out; |
1da177e4 | 2638 | |
3ae70656 MS |
2639 | /* |
2640 | * If the new table has no data devices, retain the existing limits. | |
2641 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2642 | * then new I/O is queued based on these limits, and then some paths | |
2643 | * reappear. | |
2644 | */ | |
2645 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2646 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2647 | if (live_map) |
2648 | limits = md->queue->limits; | |
83d5e5b0 | 2649 | dm_put_live_table_fast(md); |
3ae70656 MS |
2650 | } |
2651 | ||
87eb5b21 MC |
2652 | if (!live_map) { |
2653 | r = dm_calculate_queue_limits(table, &limits); | |
2654 | if (r) { | |
2655 | map = ERR_PTR(r); | |
2656 | goto out; | |
2657 | } | |
042d2a9b | 2658 | } |
754c5fc7 | 2659 | |
042d2a9b | 2660 | map = __bind(md, table, &limits); |
62e08243 | 2661 | dm_issue_global_event(); |
1da177e4 | 2662 | |
93c534ae | 2663 | out: |
e61290a4 | 2664 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2665 | return map; |
1da177e4 LT |
2666 | } |
2667 | ||
2668 | /* | |
2669 | * Functions to lock and unlock any filesystem running on the | |
2670 | * device. | |
2671 | */ | |
2ca3310e | 2672 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2673 | { |
e39e2e95 | 2674 | int r; |
1da177e4 | 2675 | |
040f04bd | 2676 | WARN_ON(test_bit(DMF_FROZEN, &md->flags)); |
aa8d7c2f | 2677 | |
982c3b30 | 2678 | r = bdev_freeze(md->disk->part0); |
040f04bd CH |
2679 | if (!r) |
2680 | set_bit(DMF_FROZEN, &md->flags); | |
2681 | return r; | |
1da177e4 LT |
2682 | } |
2683 | ||
2ca3310e | 2684 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2685 | { |
aa8d7c2f AK |
2686 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2687 | return; | |
982c3b30 | 2688 | bdev_thaw(md->disk->part0); |
aa8d7c2f | 2689 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2690 | } |
2691 | ||
2692 | /* | |
b48633f8 BVA |
2693 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2694 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2695 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2696 | * | |
ffcc3936 MS |
2697 | * If __dm_suspend returns 0, the device is completely quiescent |
2698 | * now. There is no request-processing activity. All new requests | |
2699 | * are being added to md->deferred list. | |
cec47e3d | 2700 | */ |
ffcc3936 | 2701 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
86a3238c | 2702 | unsigned int suspend_flags, unsigned int task_state, |
eaf9a736 | 2703 | int dmf_suspended_flag) |
1da177e4 | 2704 | { |
ffcc3936 MS |
2705 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2706 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2707 | int r; | |
1da177e4 | 2708 | |
5a8f1f80 BVA |
2709 | lockdep_assert_held(&md->suspend_lock); |
2710 | ||
2e93ccc1 KU |
2711 | /* |
2712 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2713 | * This flag is cleared before dm_suspend returns. | |
2714 | */ | |
2715 | if (noflush) | |
2716 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 | 2717 | else |
ac75b09f | 2718 | DMDEBUG("%s: suspending with flush", dm_device_name(md)); |
2e93ccc1 | 2719 | |
d67ee213 MS |
2720 | /* |
2721 | * This gets reverted if there's an error later and the targets | |
2722 | * provide the .presuspend_undo hook. | |
2723 | */ | |
cf222b37 AK |
2724 | dm_table_presuspend_targets(map); |
2725 | ||
32a926da | 2726 | /* |
9f518b27 KU |
2727 | * Flush I/O to the device. |
2728 | * Any I/O submitted after lock_fs() may not be flushed. | |
2729 | * noflush takes precedence over do_lockfs. | |
2730 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2731 | */ |
2732 | if (!noflush && do_lockfs) { | |
2733 | r = lock_fs(md); | |
d67ee213 MS |
2734 | if (r) { |
2735 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2736 | return r; |
d67ee213 | 2737 | } |
aa8d7c2f | 2738 | } |
1da177e4 LT |
2739 | |
2740 | /* | |
3b00b203 MP |
2741 | * Here we must make sure that no processes are submitting requests |
2742 | * to target drivers i.e. no one may be executing | |
96c9865c | 2743 | * dm_split_and_process_bio from dm_submit_bio. |
3b00b203 | 2744 | * |
96c9865c | 2745 | * To get all processes out of dm_split_and_process_bio in dm_submit_bio, |
3b00b203 | 2746 | * we take the write lock. To prevent any process from reentering |
96c9865c | 2747 | * dm_split_and_process_bio from dm_submit_bio and quiesce the thread |
0cede372 | 2748 | * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call |
6a8736d1 | 2749 | * flush_workqueue(md->wq). |
1da177e4 | 2750 | */ |
1eb787ec | 2751 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2752 | if (map) |
2753 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2754 | |
d0bcb878 | 2755 | /* |
29e4013d TH |
2756 | * Stop md->queue before flushing md->wq in case request-based |
2757 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2758 | */ |
6a23e05c | 2759 | if (dm_request_based(md)) |
eca7ee6d | 2760 | dm_stop_queue(md->queue); |
cec47e3d | 2761 | |
d0bcb878 KU |
2762 | flush_workqueue(md->wq); |
2763 | ||
1da177e4 | 2764 | /* |
3b00b203 MP |
2765 | * At this point no more requests are entering target request routines. |
2766 | * We call dm_wait_for_completion to wait for all existing requests | |
2767 | * to finish. | |
1da177e4 | 2768 | */ |
b48633f8 | 2769 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2770 | if (!r) |
2771 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2772 | |
6d6f10df | 2773 | if (noflush) |
022c2611 | 2774 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2775 | if (map) |
2776 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2777 | |
1da177e4 | 2778 | /* were we interrupted ? */ |
46125c1c | 2779 | if (r < 0) { |
9a1fb464 | 2780 | dm_queue_flush(md); |
73d410c0 | 2781 | |
cec47e3d | 2782 | if (dm_request_based(md)) |
eca7ee6d | 2783 | dm_start_queue(md->queue); |
cec47e3d | 2784 | |
2ca3310e | 2785 | unlock_fs(md); |
d67ee213 | 2786 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2787 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2788 | } |
1da177e4 | 2789 | |
ffcc3936 MS |
2790 | return r; |
2791 | } | |
2792 | ||
2793 | /* | |
2794 | * We need to be able to change a mapping table under a mounted | |
2795 | * filesystem. For example we might want to move some data in | |
2796 | * the background. Before the table can be swapped with | |
2797 | * dm_bind_table, dm_suspend must be called to flush any in | |
2798 | * flight bios and ensure that any further io gets deferred. | |
2799 | */ | |
2800 | /* | |
2801 | * Suspend mechanism in request-based dm. | |
2802 | * | |
2803 | * 1. Flush all I/Os by lock_fs() if needed. | |
2804 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2805 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2806 | * | |
2807 | * To abort suspend, start the request_queue. | |
2808 | */ | |
86a3238c | 2809 | int dm_suspend(struct mapped_device *md, unsigned int suspend_flags) |
ffcc3936 MS |
2810 | { |
2811 | struct dm_table *map = NULL; | |
2812 | int r = 0; | |
2813 | ||
2814 | retry: | |
2815 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2816 | ||
2817 | if (dm_suspended_md(md)) { | |
2818 | r = -EINVAL; | |
2819 | goto out_unlock; | |
2820 | } | |
2821 | ||
2822 | if (dm_suspended_internally_md(md)) { | |
2823 | /* already internally suspended, wait for internal resume */ | |
2824 | mutex_unlock(&md->suspend_lock); | |
2825 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2826 | if (r) | |
2827 | return r; | |
2828 | goto retry; | |
2829 | } | |
2830 | ||
a12f5d48 | 2831 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2760904d LL |
2832 | if (!map) { |
2833 | /* avoid deadlock with fs/namespace.c:do_mount() */ | |
2834 | suspend_flags &= ~DM_SUSPEND_LOCKFS_FLAG; | |
2835 | } | |
ffcc3936 | 2836 | |
eaf9a736 | 2837 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2838 | if (r) |
2839 | goto out_unlock; | |
3b00b203 | 2840 | |
5df96f2b | 2841 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
4d4471cb | 2842 | dm_table_postsuspend_targets(map); |
5df96f2b | 2843 | clear_bit(DMF_POST_SUSPENDING, &md->flags); |
4d4471cb | 2844 | |
d287483d | 2845 | out_unlock: |
e61290a4 | 2846 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2847 | return r; |
1da177e4 LT |
2848 | } |
2849 | ||
ffcc3936 MS |
2850 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2851 | { | |
2852 | if (map) { | |
2853 | int r = dm_table_resume_targets(map); | |
b30f1607 | 2854 | |
ffcc3936 MS |
2855 | if (r) |
2856 | return r; | |
2857 | } | |
2858 | ||
2859 | dm_queue_flush(md); | |
2860 | ||
2861 | /* | |
2862 | * Flushing deferred I/Os must be done after targets are resumed | |
2863 | * so that mapping of targets can work correctly. | |
2864 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2865 | */ | |
2866 | if (dm_request_based(md)) | |
eca7ee6d | 2867 | dm_start_queue(md->queue); |
ffcc3936 MS |
2868 | |
2869 | unlock_fs(md); | |
2870 | ||
2871 | return 0; | |
2872 | } | |
2873 | ||
1da177e4 LT |
2874 | int dm_resume(struct mapped_device *md) |
2875 | { | |
8dc23658 | 2876 | int r; |
cf222b37 | 2877 | struct dm_table *map = NULL; |
1da177e4 | 2878 | |
ffcc3936 | 2879 | retry: |
8dc23658 | 2880 | r = -EINVAL; |
ffcc3936 MS |
2881 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2882 | ||
4f186f8b | 2883 | if (!dm_suspended_md(md)) |
cf222b37 | 2884 | goto out; |
cf222b37 | 2885 | |
ffcc3936 MS |
2886 | if (dm_suspended_internally_md(md)) { |
2887 | /* already internally suspended, wait for internal resume */ | |
2888 | mutex_unlock(&md->suspend_lock); | |
2889 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2890 | if (r) | |
2891 | return r; | |
2892 | goto retry; | |
2893 | } | |
2894 | ||
a12f5d48 | 2895 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2896 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2897 | goto out; |
1da177e4 | 2898 | |
ffcc3936 | 2899 | r = __dm_resume(md, map); |
8757b776 MB |
2900 | if (r) |
2901 | goto out; | |
2ca3310e | 2902 | |
2ca3310e | 2903 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2904 | out: |
e61290a4 | 2905 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2906 | |
cf222b37 | 2907 | return r; |
1da177e4 LT |
2908 | } |
2909 | ||
fd2ed4d2 MP |
2910 | /* |
2911 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2912 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2913 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2914 | */ |
2915 | ||
86a3238c | 2916 | static void __dm_internal_suspend(struct mapped_device *md, unsigned int suspend_flags) |
fd2ed4d2 | 2917 | { |
ffcc3936 MS |
2918 | struct dm_table *map = NULL; |
2919 | ||
1ea0654e BVA |
2920 | lockdep_assert_held(&md->suspend_lock); |
2921 | ||
96b26c8c | 2922 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2923 | return; /* nested internal suspend */ |
2924 | ||
2925 | if (dm_suspended_md(md)) { | |
2926 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2927 | return; /* nest suspend */ | |
2928 | } | |
2929 | ||
a12f5d48 | 2930 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2931 | |
2932 | /* | |
2933 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2934 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2935 | * would require changing .presuspend to return an error -- avoid this | |
2936 | * until there is a need for more elaborate variants of internal suspend. | |
2937 | */ | |
eaf9a736 MS |
2938 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2939 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 | 2940 | |
5df96f2b | 2941 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
ffcc3936 | 2942 | dm_table_postsuspend_targets(map); |
5df96f2b | 2943 | clear_bit(DMF_POST_SUSPENDING, &md->flags); |
ffcc3936 MS |
2944 | } |
2945 | ||
2946 | static void __dm_internal_resume(struct mapped_device *md) | |
2947 | { | |
96b26c8c MP |
2948 | BUG_ON(!md->internal_suspend_count); |
2949 | ||
2950 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2951 | return; /* resume from nested internal suspend */ |
2952 | ||
fd2ed4d2 | 2953 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2954 | goto done; /* resume from nested suspend */ |
2955 | ||
2956 | /* | |
2957 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2958 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2959 | */ | |
2960 | (void) __dm_resume(md, NULL); | |
2961 | ||
2962 | done: | |
2963 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2964 | smp_mb__after_atomic(); | |
2965 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2966 | } | |
2967 | ||
2968 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2969 | { | |
2970 | mutex_lock(&md->suspend_lock); | |
2971 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2972 | mutex_unlock(&md->suspend_lock); | |
2973 | } | |
2974 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2975 | ||
2976 | void dm_internal_resume(struct mapped_device *md) | |
2977 | { | |
2978 | mutex_lock(&md->suspend_lock); | |
2979 | __dm_internal_resume(md); | |
2980 | mutex_unlock(&md->suspend_lock); | |
2981 | } | |
2982 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2983 | ||
2984 | /* | |
2985 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2986 | * which prevents interaction with userspace-driven suspend. | |
2987 | */ | |
2988 | ||
2989 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2990 | { | |
2991 | mutex_lock(&md->suspend_lock); | |
2992 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2993 | return; |
2994 | ||
2995 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2996 | synchronize_srcu(&md->io_barrier); | |
2997 | flush_workqueue(md->wq); | |
2998 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2999 | } | |
b735fede | 3000 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3001 | |
ffcc3936 | 3002 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3003 | { |
ffcc3936 | 3004 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3005 | goto done; |
3006 | ||
3007 | dm_queue_flush(md); | |
3008 | ||
3009 | done: | |
3010 | mutex_unlock(&md->suspend_lock); | |
3011 | } | |
b735fede | 3012 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3013 | |
a4a82ce3 HM |
3014 | /* |
3015 | *--------------------------------------------------------------- | |
1da177e4 | 3016 | * Event notification. |
a4a82ce3 HM |
3017 | *--------------------------------------------------------------- |
3018 | */ | |
3abf85b5 | 3019 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
86a3238c | 3020 | unsigned int cookie, bool need_resize_uevent) |
69267a30 | 3021 | { |
6958c1c6 | 3022 | int r; |
86a3238c | 3023 | unsigned int noio_flag; |
60935eb2 | 3024 | char udev_cookie[DM_COOKIE_LENGTH]; |
7533afa1 MP |
3025 | char *envp[3] = { NULL, NULL, NULL }; |
3026 | char **envpp = envp; | |
3027 | if (cookie) { | |
60935eb2 MB |
3028 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", |
3029 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
7533afa1 | 3030 | *envpp++ = udev_cookie; |
60935eb2 | 3031 | } |
7533afa1 MP |
3032 | if (need_resize_uevent) { |
3033 | *envpp++ = "RESIZE=1"; | |
3034 | } | |
3035 | ||
3036 | noio_flag = memalloc_noio_save(); | |
3037 | ||
3038 | r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp); | |
6958c1c6 MP |
3039 | |
3040 | memalloc_noio_restore(noio_flag); | |
3041 | ||
3042 | return r; | |
69267a30 AK |
3043 | } |
3044 | ||
7a8c3d3b MA |
3045 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3046 | { | |
3047 | return atomic_add_return(1, &md->uevent_seq); | |
3048 | } | |
3049 | ||
1da177e4 LT |
3050 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3051 | { | |
3052 | return atomic_read(&md->event_nr); | |
3053 | } | |
3054 | ||
3055 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3056 | { | |
3057 | return wait_event_interruptible(md->eventq, | |
3058 | (event_nr != atomic_read(&md->event_nr))); | |
3059 | } | |
3060 | ||
7a8c3d3b MA |
3061 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3062 | { | |
3063 | unsigned long flags; | |
3064 | ||
3065 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3066 | list_add(elist, &md->uevent_list); | |
3067 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3068 | } | |
3069 | ||
1da177e4 LT |
3070 | /* |
3071 | * The gendisk is only valid as long as you have a reference | |
3072 | * count on 'md'. | |
3073 | */ | |
3074 | struct gendisk *dm_disk(struct mapped_device *md) | |
3075 | { | |
3076 | return md->disk; | |
3077 | } | |
65ff5b7d | 3078 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3079 | |
784aae73 MB |
3080 | struct kobject *dm_kobject(struct mapped_device *md) |
3081 | { | |
2995fa78 | 3082 | return &md->kobj_holder.kobj; |
784aae73 MB |
3083 | } |
3084 | ||
784aae73 MB |
3085 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3086 | { | |
3087 | struct mapped_device *md; | |
3088 | ||
2995fa78 | 3089 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3090 | |
b9a41d21 HT |
3091 | spin_lock(&_minor_lock); |
3092 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
3093 | md = NULL; | |
3094 | goto out; | |
3095 | } | |
784aae73 | 3096 | dm_get(md); |
b9a41d21 HT |
3097 | out: |
3098 | spin_unlock(&_minor_lock); | |
3099 | ||
784aae73 MB |
3100 | return md; |
3101 | } | |
3102 | ||
4f186f8b | 3103 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3104 | { |
3105 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3106 | } | |
3107 | ||
5df96f2b MP |
3108 | static int dm_post_suspending_md(struct mapped_device *md) |
3109 | { | |
3110 | return test_bit(DMF_POST_SUSPENDING, &md->flags); | |
3111 | } | |
3112 | ||
ffcc3936 MS |
3113 | int dm_suspended_internally_md(struct mapped_device *md) |
3114 | { | |
3115 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3116 | } | |
3117 | ||
2c140a24 MP |
3118 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3119 | { | |
3120 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3121 | } | |
3122 | ||
64dbce58 KU |
3123 | int dm_suspended(struct dm_target *ti) |
3124 | { | |
33bd6f06 | 3125 | return dm_suspended_md(ti->table->md); |
64dbce58 KU |
3126 | } |
3127 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3128 | ||
5df96f2b MP |
3129 | int dm_post_suspending(struct dm_target *ti) |
3130 | { | |
33bd6f06 | 3131 | return dm_post_suspending_md(ti->table->md); |
5df96f2b MP |
3132 | } |
3133 | EXPORT_SYMBOL_GPL(dm_post_suspending); | |
3134 | ||
2e93ccc1 KU |
3135 | int dm_noflush_suspending(struct dm_target *ti) |
3136 | { | |
33bd6f06 | 3137 | return __noflush_suspending(ti->table->md); |
2e93ccc1 KU |
3138 | } |
3139 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3140 | ||
e6ee8c0b KU |
3141 | void dm_free_md_mempools(struct dm_md_mempools *pools) |
3142 | { | |
3143 | if (!pools) | |
3144 | return; | |
3145 | ||
6f1c819c KO |
3146 | bioset_exit(&pools->bs); |
3147 | bioset_exit(&pools->io_bs); | |
e6ee8c0b KU |
3148 | |
3149 | kfree(pools); | |
3150 | } | |
3151 | ||
9c72bad1 CH |
3152 | struct dm_pr { |
3153 | u64 old_key; | |
3154 | u64 new_key; | |
3155 | u32 flags; | |
c6adada5 | 3156 | bool abort; |
9c72bad1 | 3157 | bool fail_early; |
8dd87f3c | 3158 | int ret; |
70151087 | 3159 | enum pr_type type; |
8a8da082 MC |
3160 | struct pr_keys *read_keys; |
3161 | struct pr_held_reservation *rsv; | |
9c72bad1 CH |
3162 | }; |
3163 | ||
3164 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
8dd87f3c | 3165 | struct dm_pr *pr) |
71cdb697 CH |
3166 | { |
3167 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
3168 | struct dm_table *table; |
3169 | struct dm_target *ti; | |
3170 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 3171 | |
9c72bad1 CH |
3172 | table = dm_get_live_table(md, &srcu_idx); |
3173 | if (!table || !dm_table_get_size(table)) | |
3174 | goto out; | |
71cdb697 | 3175 | |
9c72bad1 | 3176 | /* We only support devices that have a single target */ |
2aec377a | 3177 | if (table->num_targets != 1) |
9c72bad1 CH |
3178 | goto out; |
3179 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 3180 | |
e120a5f1 MS |
3181 | if (dm_suspended_md(md)) { |
3182 | ret = -EAGAIN; | |
3183 | goto out; | |
3184 | } | |
3185 | ||
9c72bad1 CH |
3186 | ret = -EINVAL; |
3187 | if (!ti->type->iterate_devices) | |
3188 | goto out; | |
3189 | ||
8dd87f3c MC |
3190 | ti->type->iterate_devices(ti, fn, pr); |
3191 | ret = 0; | |
9c72bad1 CH |
3192 | out: |
3193 | dm_put_live_table(md, srcu_idx); | |
3194 | return ret; | |
3195 | } | |
3196 | ||
3197 | /* | |
3198 | * For register / unregister we need to manually call out to every path. | |
3199 | */ | |
3200 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
3201 | sector_t start, sector_t len, void *data) | |
3202 | { | |
3203 | struct dm_pr *pr = data; | |
3204 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
8dd87f3c MC |
3205 | int ret; |
3206 | ||
3207 | if (!ops || !ops->pr_register) { | |
3208 | pr->ret = -EOPNOTSUPP; | |
3209 | return -1; | |
3210 | } | |
9c72bad1 | 3211 | |
8dd87f3c MC |
3212 | ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); |
3213 | if (!ret) | |
3214 | return 0; | |
3215 | ||
3216 | if (!pr->ret) | |
3217 | pr->ret = ret; | |
9c72bad1 | 3218 | |
8dd87f3c MC |
3219 | if (pr->fail_early) |
3220 | return -1; | |
3221 | ||
3222 | return 0; | |
9c72bad1 CH |
3223 | } |
3224 | ||
3225 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
3226 | u32 flags) | |
3227 | { | |
3228 | struct dm_pr pr = { | |
3229 | .old_key = old_key, | |
3230 | .new_key = new_key, | |
3231 | .flags = flags, | |
3232 | .fail_early = true, | |
8dd87f3c | 3233 | .ret = 0, |
9c72bad1 CH |
3234 | }; |
3235 | int ret; | |
3236 | ||
3237 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
8dd87f3c MC |
3238 | if (ret) { |
3239 | /* Didn't even get to register a path */ | |
3240 | return ret; | |
9c72bad1 CH |
3241 | } |
3242 | ||
8dd87f3c MC |
3243 | if (!pr.ret) |
3244 | return 0; | |
3245 | ret = pr.ret; | |
3246 | ||
3247 | if (!new_key) | |
3248 | return ret; | |
3249 | ||
3250 | /* unregister all paths if we failed to register any path */ | |
3251 | pr.old_key = new_key; | |
3252 | pr.new_key = 0; | |
3253 | pr.flags = 0; | |
3254 | pr.fail_early = false; | |
3255 | (void) dm_call_pr(bdev, __dm_pr_register, &pr); | |
9c72bad1 | 3256 | return ret; |
71cdb697 CH |
3257 | } |
3258 | ||
70151087 MC |
3259 | |
3260 | static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev, | |
3261 | sector_t start, sector_t len, void *data) | |
3262 | { | |
3263 | struct dm_pr *pr = data; | |
3264 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3265 | ||
3266 | if (!ops || !ops->pr_reserve) { | |
3267 | pr->ret = -EOPNOTSUPP; | |
3268 | return -1; | |
3269 | } | |
3270 | ||
3271 | pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags); | |
3272 | if (!pr->ret) | |
3273 | return -1; | |
3274 | ||
3275 | return 0; | |
3276 | } | |
3277 | ||
71cdb697 | 3278 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, |
956a4025 | 3279 | u32 flags) |
71cdb697 | 3280 | { |
70151087 MC |
3281 | struct dm_pr pr = { |
3282 | .old_key = key, | |
3283 | .flags = flags, | |
3284 | .type = type, | |
3285 | .fail_early = false, | |
3286 | .ret = 0, | |
3287 | }; | |
3288 | int ret; | |
71cdb697 | 3289 | |
70151087 MC |
3290 | ret = dm_call_pr(bdev, __dm_pr_reserve, &pr); |
3291 | if (ret) | |
3292 | return ret; | |
71cdb697 | 3293 | |
70151087 | 3294 | return pr.ret; |
71cdb697 CH |
3295 | } |
3296 | ||
08a3c338 MC |
3297 | /* |
3298 | * If there is a non-All Registrants type of reservation, the release must be | |
3299 | * sent down the holding path. For the cases where there is no reservation or | |
3300 | * the path is not the holder the device will also return success, so we must | |
3301 | * try each path to make sure we got the correct path. | |
3302 | */ | |
3303 | static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev, | |
3304 | sector_t start, sector_t len, void *data) | |
3305 | { | |
3306 | struct dm_pr *pr = data; | |
3307 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3308 | ||
3309 | if (!ops || !ops->pr_release) { | |
3310 | pr->ret = -EOPNOTSUPP; | |
3311 | return -1; | |
3312 | } | |
3313 | ||
3314 | pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type); | |
3315 | if (pr->ret) | |
3316 | return -1; | |
3317 | ||
3318 | return 0; | |
71cdb697 CH |
3319 | } |
3320 | ||
3321 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3322 | { | |
08a3c338 MC |
3323 | struct dm_pr pr = { |
3324 | .old_key = key, | |
3325 | .type = type, | |
3326 | .fail_early = false, | |
3327 | }; | |
3328 | int ret; | |
71cdb697 | 3329 | |
08a3c338 MC |
3330 | ret = dm_call_pr(bdev, __dm_pr_release, &pr); |
3331 | if (ret) | |
3332 | return ret; | |
71cdb697 | 3333 | |
08a3c338 | 3334 | return pr.ret; |
71cdb697 CH |
3335 | } |
3336 | ||
c6adada5 MC |
3337 | static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev, |
3338 | sector_t start, sector_t len, void *data) | |
3339 | { | |
3340 | struct dm_pr *pr = data; | |
3341 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3342 | ||
3343 | if (!ops || !ops->pr_preempt) { | |
3344 | pr->ret = -EOPNOTSUPP; | |
3345 | return -1; | |
3346 | } | |
3347 | ||
3348 | pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type, | |
3349 | pr->abort); | |
3350 | if (!pr->ret) | |
3351 | return -1; | |
3352 | ||
3353 | return 0; | |
71cdb697 CH |
3354 | } |
3355 | ||
3356 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3357 | enum pr_type type, bool abort) |
71cdb697 | 3358 | { |
c6adada5 MC |
3359 | struct dm_pr pr = { |
3360 | .new_key = new_key, | |
3361 | .old_key = old_key, | |
3362 | .type = type, | |
3363 | .fail_early = false, | |
3364 | }; | |
3365 | int ret; | |
71cdb697 | 3366 | |
c6adada5 MC |
3367 | ret = dm_call_pr(bdev, __dm_pr_preempt, &pr); |
3368 | if (ret) | |
3369 | return ret; | |
71cdb697 | 3370 | |
c6adada5 | 3371 | return pr.ret; |
71cdb697 CH |
3372 | } |
3373 | ||
3374 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3375 | { | |
3376 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3377 | const struct pr_ops *ops; | |
971888c4 | 3378 | int r, srcu_idx; |
71cdb697 | 3379 | |
5bd5e8d8 | 3380 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3381 | if (r < 0) |
971888c4 | 3382 | goto out; |
71cdb697 CH |
3383 | |
3384 | ops = bdev->bd_disk->fops->pr_ops; | |
3385 | if (ops && ops->pr_clear) | |
3386 | r = ops->pr_clear(bdev, key); | |
3387 | else | |
3388 | r = -EOPNOTSUPP; | |
971888c4 MS |
3389 | out: |
3390 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3391 | return r; |
3392 | } | |
3393 | ||
8a8da082 MC |
3394 | static int __dm_pr_read_keys(struct dm_target *ti, struct dm_dev *dev, |
3395 | sector_t start, sector_t len, void *data) | |
3396 | { | |
3397 | struct dm_pr *pr = data; | |
3398 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3399 | ||
3400 | if (!ops || !ops->pr_read_keys) { | |
3401 | pr->ret = -EOPNOTSUPP; | |
3402 | return -1; | |
3403 | } | |
3404 | ||
3405 | pr->ret = ops->pr_read_keys(dev->bdev, pr->read_keys); | |
3406 | if (!pr->ret) | |
3407 | return -1; | |
3408 | ||
3409 | return 0; | |
3410 | } | |
3411 | ||
3412 | static int dm_pr_read_keys(struct block_device *bdev, struct pr_keys *keys) | |
3413 | { | |
3414 | struct dm_pr pr = { | |
3415 | .read_keys = keys, | |
3416 | }; | |
3417 | int ret; | |
3418 | ||
3419 | ret = dm_call_pr(bdev, __dm_pr_read_keys, &pr); | |
3420 | if (ret) | |
3421 | return ret; | |
3422 | ||
3423 | return pr.ret; | |
3424 | } | |
3425 | ||
3426 | static int __dm_pr_read_reservation(struct dm_target *ti, struct dm_dev *dev, | |
3427 | sector_t start, sector_t len, void *data) | |
3428 | { | |
3429 | struct dm_pr *pr = data; | |
3430 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3431 | ||
3432 | if (!ops || !ops->pr_read_reservation) { | |
3433 | pr->ret = -EOPNOTSUPP; | |
3434 | return -1; | |
3435 | } | |
3436 | ||
3437 | pr->ret = ops->pr_read_reservation(dev->bdev, pr->rsv); | |
3438 | if (!pr->ret) | |
3439 | return -1; | |
3440 | ||
3441 | return 0; | |
3442 | } | |
3443 | ||
3444 | static int dm_pr_read_reservation(struct block_device *bdev, | |
3445 | struct pr_held_reservation *rsv) | |
3446 | { | |
3447 | struct dm_pr pr = { | |
3448 | .rsv = rsv, | |
3449 | }; | |
3450 | int ret; | |
3451 | ||
3452 | ret = dm_call_pr(bdev, __dm_pr_read_reservation, &pr); | |
3453 | if (ret) | |
3454 | return ret; | |
3455 | ||
3456 | return pr.ret; | |
3457 | } | |
3458 | ||
71cdb697 CH |
3459 | static const struct pr_ops dm_pr_ops = { |
3460 | .pr_register = dm_pr_register, | |
3461 | .pr_reserve = dm_pr_reserve, | |
3462 | .pr_release = dm_pr_release, | |
3463 | .pr_preempt = dm_pr_preempt, | |
3464 | .pr_clear = dm_pr_clear, | |
8a8da082 MC |
3465 | .pr_read_keys = dm_pr_read_keys, |
3466 | .pr_read_reservation = dm_pr_read_reservation, | |
71cdb697 CH |
3467 | }; |
3468 | ||
83d5cde4 | 3469 | static const struct block_device_operations dm_blk_dops = { |
c62b37d9 | 3470 | .submit_bio = dm_submit_bio, |
b99fdcdc | 3471 | .poll_bio = dm_poll_bio, |
1da177e4 LT |
3472 | .open = dm_blk_open, |
3473 | .release = dm_blk_close, | |
aa129a22 | 3474 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3475 | .getgeo = dm_blk_getgeo, |
e76239a3 | 3476 | .report_zones = dm_blk_report_zones, |
71cdb697 | 3477 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3478 | .owner = THIS_MODULE |
3479 | }; | |
3480 | ||
681cc5e8 MS |
3481 | static const struct block_device_operations dm_rq_blk_dops = { |
3482 | .open = dm_blk_open, | |
3483 | .release = dm_blk_close, | |
3484 | .ioctl = dm_blk_ioctl, | |
3485 | .getgeo = dm_blk_getgeo, | |
3486 | .pr_ops = &dm_pr_ops, | |
3487 | .owner = THIS_MODULE | |
3488 | }; | |
3489 | ||
f26c5719 DW |
3490 | static const struct dax_operations dm_dax_ops = { |
3491 | .direct_access = dm_dax_direct_access, | |
cdf6cdcd | 3492 | .zero_page_range = dm_dax_zero_page_range, |
047218ec | 3493 | .recovery_write = dm_dax_recovery_write, |
f26c5719 DW |
3494 | }; |
3495 | ||
1da177e4 LT |
3496 | /* |
3497 | * module hooks | |
3498 | */ | |
3499 | module_init(dm_init); | |
3500 | module_exit(dm_exit); | |
3501 | ||
3502 | module_param(major, uint, 0); | |
3503 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3504 | |
6a808034 | 3505 | module_param(reserved_bio_based_ios, uint, 0644); |
e8603136 MS |
3506 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); |
3507 | ||
6a808034 | 3508 | module_param(dm_numa_node, int, 0644); |
115485e8 MS |
3509 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); |
3510 | ||
6a808034 | 3511 | module_param(swap_bios, int, 0644); |
a666e5c0 MP |
3512 | MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs"); |
3513 | ||
1da177e4 LT |
3514 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3515 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3516 | MODULE_LICENSE("GPL"); |