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 | |
acfe0ad7 MP |
210 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
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 | ||
fe5f9f2c | 313 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
314 | { |
315 | struct mapped_device *md; | |
316 | ||
fba9f90e JM |
317 | spin_lock(&_minor_lock); |
318 | ||
fe5f9f2c | 319 | md = bdev->bd_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 | ||
db2a144b | 337 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
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 | ||
e56f81e0 CH |
451 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
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 | ||
8d394bc4 | 490 | static 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 | ||
e6926ad0 | 495 | static void dm_io_acct(struct dm_io *io, bool end) |
8d394bc4 | 496 | { |
e6926ad0 ML |
497 | struct dm_stats_aux *stats_aux = &io->stats_aux; |
498 | unsigned long start_time = io->start_time; | |
499 | struct mapped_device *md = io->md; | |
500 | struct bio *bio = io->orig_bio; | |
d3de6d12 ML |
501 | unsigned int sectors; |
502 | ||
503 | /* | |
504 | * If REQ_PREFLUSH set, don't account payload, it will be | |
505 | * submitted (and accounted) after this flush completes. | |
506 | */ | |
507 | if (bio_is_flush_with_data(bio)) | |
508 | sectors = 0; | |
7dd76d1f | 509 | else if (likely(!(dm_io_flagged(io, DM_IO_WAS_SPLIT)))) |
d3de6d12 | 510 | sectors = bio_sectors(bio); |
7dd76d1f ML |
511 | else |
512 | sectors = io->sectors; | |
8d394bc4 MS |
513 | |
514 | if (!end) | |
5f275713 | 515 | bdev_start_io_acct(bio->bi_bdev, bio_op(bio), start_time); |
8d394bc4 | 516 | else |
5f275713 YK |
517 | bdev_end_io_acct(bio->bi_bdev, bio_op(bio), sectors, |
518 | start_time); | |
7465d7ac | 519 | |
442761fd | 520 | if (static_branch_unlikely(&stats_enabled) && |
7dd76d1f ML |
521 | unlikely(dm_stats_used(&md->stats))) { |
522 | sector_t sector; | |
523 | ||
524 | if (likely(!dm_io_flagged(io, DM_IO_WAS_SPLIT))) | |
525 | sector = bio->bi_iter.bi_sector; | |
526 | else | |
527 | sector = bio_end_sector(bio) - io->sector_offset; | |
528 | ||
7465d7ac | 529 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
7dd76d1f | 530 | sector, sectors, |
8d394bc4 | 531 | end, start_time, stats_aux); |
7dd76d1f | 532 | } |
8d394bc4 MS |
533 | } |
534 | ||
b992b40d | 535 | static void __dm_start_io_acct(struct dm_io *io) |
8d394bc4 | 536 | { |
e6926ad0 | 537 | dm_io_acct(io, false); |
7465d7ac MS |
538 | } |
539 | ||
0fbb4d93 | 540 | static void dm_start_io_acct(struct dm_io *io, struct bio *clone) |
7465d7ac | 541 | { |
0fbb4d93 MS |
542 | /* |
543 | * Ensure IO accounting is only ever started once. | |
0fbb4d93 | 544 | */ |
3b03f7c1 MS |
545 | if (dm_io_flagged(io, DM_IO_ACCOUNTED)) |
546 | return; | |
547 | ||
548 | /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */ | |
549 | if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) { | |
82f6cdcc MS |
550 | dm_io_set_flag(io, DM_IO_ACCOUNTED); |
551 | } else { | |
552 | unsigned long flags; | |
655f3aad | 553 | /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */ |
4d7bca13 | 554 | spin_lock_irqsave(&io->lock, flags); |
10eb3a0d BM |
555 | if (dm_io_flagged(io, DM_IO_ACCOUNTED)) { |
556 | spin_unlock_irqrestore(&io->lock, flags); | |
557 | return; | |
558 | } | |
82f6cdcc | 559 | dm_io_set_flag(io, DM_IO_ACCOUNTED); |
4d7bca13 | 560 | spin_unlock_irqrestore(&io->lock, flags); |
82f6cdcc | 561 | } |
7465d7ac | 562 | |
b992b40d | 563 | __dm_start_io_acct(io); |
0fbb4d93 | 564 | } |
7465d7ac | 565 | |
b992b40d | 566 | static void dm_end_io_acct(struct dm_io *io) |
0fbb4d93 | 567 | { |
e6926ad0 | 568 | dm_io_acct(io, true); |
7465d7ac | 569 | } |
978e51ba MS |
570 | |
571 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) | |
1da177e4 | 572 | { |
64f52b0e MS |
573 | struct dm_io *io; |
574 | struct dm_target_io *tio; | |
575 | struct bio *clone; | |
576 | ||
29dec90a | 577 | clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs); |
6c23f0bd | 578 | tio = clone_to_tio(clone); |
655f3aad MS |
579 | tio->flags = 0; |
580 | dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO); | |
64f52b0e MS |
581 | tio->io = NULL; |
582 | ||
583 | io = container_of(tio, struct dm_io, tio); | |
584 | io->magic = DM_IO_MAGIC; | |
84b98f4c | 585 | io->status = BLK_STS_OK; |
0f14d60a ML |
586 | |
587 | /* one ref is for submission, the other is for completion */ | |
588 | atomic_set(&io->io_count, 2); | |
9f6dc633 | 589 | this_cpu_inc(*md->pending_io); |
7dd76d1f | 590 | io->orig_bio = bio; |
978e51ba | 591 | io->md = md; |
4d7bca13 | 592 | spin_lock_init(&io->lock); |
b879f915 | 593 | io->start_time = jiffies; |
82f6cdcc | 594 | io->flags = 0; |
64f52b0e | 595 | |
442761fd MS |
596 | if (static_branch_unlikely(&stats_enabled)) |
597 | dm_stats_record_start(&md->stats, &io->stats_aux); | |
64f52b0e MS |
598 | |
599 | return io; | |
1da177e4 LT |
600 | } |
601 | ||
0119ab14 | 602 | static void free_io(struct dm_io *io) |
1da177e4 | 603 | { |
64f52b0e MS |
604 | bio_put(&io->tio.clone); |
605 | } | |
606 | ||
1d1068ce | 607 | static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti, |
86a3238c | 608 | unsigned int target_bio_nr, unsigned int *len, gfp_t gfp_mask) |
64f52b0e | 609 | { |
9dd1cd32 | 610 | struct mapped_device *md = ci->io->md; |
64f52b0e | 611 | struct dm_target_io *tio; |
018b05eb | 612 | struct bio *clone; |
64f52b0e MS |
613 | |
614 | if (!ci->io->tio.io) { | |
615 | /* the dm_target_io embedded in ci->io is available */ | |
616 | tio = &ci->io->tio; | |
018b05eb MS |
617 | /* alloc_io() already initialized embedded clone */ |
618 | clone = &tio->clone; | |
64f52b0e | 619 | } else { |
29dec90a CH |
620 | clone = bio_alloc_clone(NULL, ci->bio, gfp_mask, |
621 | &md->mempools->bs); | |
64f52b0e MS |
622 | if (!clone) |
623 | return NULL; | |
624 | ||
b99fdcdc ML |
625 | /* REQ_DM_POLL_LIST shouldn't be inherited */ |
626 | clone->bi_opf &= ~REQ_DM_POLL_LIST; | |
627 | ||
6c23f0bd | 628 | tio = clone_to_tio(clone); |
655f3aad | 629 | tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */ |
64f52b0e MS |
630 | } |
631 | ||
632 | tio->magic = DM_TIO_MAGIC; | |
633 | tio->io = ci->io; | |
634 | tio->ti = ti; | |
635 | tio->target_bio_nr = target_bio_nr; | |
dc8e2021 | 636 | tio->len_ptr = len; |
743598f0 | 637 | tio->old_sector = 0; |
64f52b0e | 638 | |
9dd1cd32 MS |
639 | /* Set default bdev, but target must bio_set_dev() before issuing IO */ |
640 | clone->bi_bdev = md->disk->part0; | |
641 | if (unlikely(ti->needs_bio_set_dev)) | |
642 | bio_set_dev(clone, md->disk->part0); | |
643 | ||
018b05eb MS |
644 | if (len) { |
645 | clone->bi_iter.bi_size = to_bytes(*len); | |
646 | if (bio_integrity(clone)) | |
647 | bio_integrity_trim(clone); | |
648 | } | |
64f52b0e | 649 | |
018b05eb | 650 | return clone; |
1da177e4 LT |
651 | } |
652 | ||
1d1068ce | 653 | static void free_tio(struct bio *clone) |
1da177e4 | 654 | { |
655f3aad | 655 | if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO)) |
64f52b0e | 656 | return; |
1d1068ce | 657 | bio_put(clone); |
1da177e4 LT |
658 | } |
659 | ||
660 | /* | |
661 | * Add the bio to the list of deferred io. | |
662 | */ | |
92c63902 | 663 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 664 | { |
05447420 | 665 | unsigned long flags; |
1da177e4 | 666 | |
05447420 | 667 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 668 | bio_list_add(&md->deferred, bio); |
05447420 | 669 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 670 | queue_work(md->wq, &md->work); |
1da177e4 LT |
671 | } |
672 | ||
673 | /* | |
674 | * Everyone (including functions in this file), should use this | |
675 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 676 | * dm_put_live_table() when finished. |
1da177e4 | 677 | */ |
563a225c MS |
678 | struct dm_table *dm_get_live_table(struct mapped_device *md, |
679 | int *srcu_idx) __acquires(md->io_barrier) | |
1da177e4 | 680 | { |
83d5e5b0 MP |
681 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
682 | ||
683 | return srcu_dereference(md->map, &md->io_barrier); | |
684 | } | |
1da177e4 | 685 | |
563a225c MS |
686 | void dm_put_live_table(struct mapped_device *md, |
687 | int srcu_idx) __releases(md->io_barrier) | |
83d5e5b0 MP |
688 | { |
689 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
690 | } | |
691 | ||
692 | void dm_sync_table(struct mapped_device *md) | |
693 | { | |
694 | synchronize_srcu(&md->io_barrier); | |
695 | synchronize_rcu_expedited(); | |
696 | } | |
697 | ||
698 | /* | |
699 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
700 | * The caller must not block between these two functions. | |
701 | */ | |
702 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
703 | { | |
704 | rcu_read_lock(); | |
705 | return rcu_dereference(md->map); | |
706 | } | |
1da177e4 | 707 | |
83d5e5b0 MP |
708 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
709 | { | |
710 | rcu_read_unlock(); | |
1da177e4 LT |
711 | } |
712 | ||
563a225c | 713 | static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md, |
a3282b43 | 714 | int *srcu_idx, blk_opf_t bio_opf) |
563a225c | 715 | { |
5d7362d0 | 716 | if (bio_opf & REQ_NOWAIT) |
563a225c MS |
717 | return dm_get_live_table_fast(md); |
718 | else | |
719 | return dm_get_live_table(md, srcu_idx); | |
720 | } | |
721 | ||
722 | static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx, | |
a3282b43 | 723 | blk_opf_t bio_opf) |
563a225c | 724 | { |
5d7362d0 | 725 | if (bio_opf & REQ_NOWAIT) |
563a225c MS |
726 | dm_put_live_table_fast(md); |
727 | else | |
728 | dm_put_live_table(md, srcu_idx); | |
729 | } | |
730 | ||
971888c4 MS |
731 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
732 | ||
86f1152b BM |
733 | /* |
734 | * Open a table device so we can use it as a map destination. | |
735 | */ | |
b9a785d2 CH |
736 | static struct table_device *open_table_device(struct mapped_device *md, |
737 | dev_t dev, fmode_t mode) | |
86f1152b | 738 | { |
b9a785d2 | 739 | struct table_device *td; |
86f1152b | 740 | struct block_device *bdev; |
cd913c76 | 741 | u64 part_off; |
86f1152b BM |
742 | int r; |
743 | ||
b9a785d2 CH |
744 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
745 | if (!td) | |
746 | return ERR_PTR(-ENOMEM); | |
747 | refcount_set(&td->count, 1); | |
86f1152b | 748 | |
b9a785d2 CH |
749 | bdev = blkdev_get_by_dev(dev, mode | FMODE_EXCL, _dm_claim_ptr); |
750 | if (IS_ERR(bdev)) { | |
751 | r = PTR_ERR(bdev); | |
752 | goto out_free_td; | |
753 | } | |
86f1152b | 754 | |
1a581b72 CH |
755 | /* |
756 | * We can be called before the dm disk is added. In that case we can't | |
757 | * register the holder relation here. It will be done once add_disk was | |
758 | * called. | |
759 | */ | |
760 | if (md->disk->slave_dir) { | |
761 | r = bd_link_disk_holder(bdev, md->disk); | |
762 | if (r) | |
763 | goto out_blkdev_put; | |
764 | } | |
86f1152b | 765 | |
b9a785d2 | 766 | td->dm_dev.mode = mode; |
86f1152b | 767 | td->dm_dev.bdev = bdev; |
8012b866 | 768 | td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off, NULL, NULL); |
b9a785d2 CH |
769 | format_dev_t(td->dm_dev.name, dev); |
770 | list_add(&td->list, &md->table_devices); | |
771 | return td; | |
772 | ||
773 | out_blkdev_put: | |
774 | blkdev_put(bdev, mode | FMODE_EXCL); | |
775 | out_free_td: | |
776 | kfree(td); | |
777 | return ERR_PTR(r); | |
86f1152b BM |
778 | } |
779 | ||
780 | /* | |
781 | * Close a table device that we've been using. | |
782 | */ | |
783 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
784 | { | |
1a581b72 CH |
785 | if (md->disk->slave_dir) |
786 | bd_unlink_disk_holder(td->dm_dev.bdev, md->disk); | |
86f1152b | 787 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); |
817bf402 | 788 | put_dax(td->dm_dev.dax_dev); |
7b586583 CH |
789 | list_del(&td->list); |
790 | kfree(td); | |
86f1152b BM |
791 | } |
792 | ||
793 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
8454fca4 SS |
794 | fmode_t mode) |
795 | { | |
86f1152b BM |
796 | struct table_device *td; |
797 | ||
798 | list_for_each_entry(td, l, list) | |
799 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
800 | return td; | |
801 | ||
802 | return NULL; | |
803 | } | |
804 | ||
805 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
8454fca4 SS |
806 | struct dm_dev **result) |
807 | { | |
86f1152b BM |
808 | struct table_device *td; |
809 | ||
810 | mutex_lock(&md->table_devices_lock); | |
811 | td = find_table_device(&md->table_devices, dev, mode); | |
812 | if (!td) { | |
b9a785d2 CH |
813 | td = open_table_device(md, dev, mode); |
814 | if (IS_ERR(td)) { | |
86f1152b | 815 | mutex_unlock(&md->table_devices_lock); |
b9a785d2 | 816 | return PTR_ERR(td); |
86f1152b | 817 | } |
b0b4d7c6 ER |
818 | } else { |
819 | refcount_inc(&td->count); | |
86f1152b | 820 | } |
86f1152b BM |
821 | mutex_unlock(&md->table_devices_lock); |
822 | ||
823 | *result = &td->dm_dev; | |
824 | return 0; | |
825 | } | |
86f1152b BM |
826 | |
827 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
828 | { | |
829 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
830 | ||
831 | mutex_lock(&md->table_devices_lock); | |
7b586583 | 832 | if (refcount_dec_and_test(&td->count)) |
86f1152b | 833 | close_table_device(td, md); |
86f1152b BM |
834 | mutex_unlock(&md->table_devices_lock); |
835 | } | |
86f1152b | 836 | |
3ac51e74 DW |
837 | /* |
838 | * Get the geometry associated with a dm device | |
839 | */ | |
840 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
841 | { | |
842 | *geo = md->geometry; | |
843 | ||
844 | return 0; | |
845 | } | |
846 | ||
847 | /* | |
848 | * Set the geometry of a device. | |
849 | */ | |
850 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
851 | { | |
852 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
853 | ||
854 | if (geo->start > sz) { | |
43e6c111 | 855 | DMERR("Start sector is beyond the geometry limits."); |
3ac51e74 DW |
856 | return -EINVAL; |
857 | } | |
858 | ||
859 | md->geometry = *geo; | |
860 | ||
861 | return 0; | |
862 | } | |
863 | ||
2e93ccc1 KU |
864 | static int __noflush_suspending(struct mapped_device *md) |
865 | { | |
866 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
867 | } | |
868 | ||
8b211aac | 869 | static void dm_requeue_add_io(struct dm_io *io, bool first_stage) |
1da177e4 | 870 | { |
b35f8caa | 871 | struct mapped_device *md = io->md; |
1da177e4 | 872 | |
8b211aac ML |
873 | if (first_stage) { |
874 | struct dm_io *next = md->requeue_list; | |
875 | ||
876 | md->requeue_list = io; | |
877 | io->next = next; | |
878 | } else { | |
879 | bio_list_add_head(&md->deferred, io->orig_bio); | |
880 | } | |
881 | } | |
882 | ||
883 | static void dm_kick_requeue(struct mapped_device *md, bool first_stage) | |
884 | { | |
885 | if (first_stage) | |
886 | queue_work(md->wq, &md->requeue_work); | |
887 | else | |
888 | queue_work(md->wq, &md->work); | |
889 | } | |
890 | ||
444fe04f ML |
891 | /* |
892 | * Return true if the dm_io's original bio is requeued. | |
893 | * io->status is updated with error if requeue disallowed. | |
894 | */ | |
8b211aac | 895 | static bool dm_handle_requeue(struct dm_io *io, bool first_stage) |
1da177e4 | 896 | { |
8b211aac | 897 | struct bio *bio = io->orig_bio; |
444fe04f ML |
898 | bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE); |
899 | bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) && | |
900 | (bio->bi_opf & REQ_POLLED)); | |
901 | struct mapped_device *md = io->md; | |
902 | bool requeued = false; | |
1da177e4 | 903 | |
444fe04f | 904 | if (handle_requeue || handle_polled_eagain) { |
e2736347 | 905 | unsigned long flags; |
444fe04f ML |
906 | |
907 | if (bio->bi_opf & REQ_POLLED) { | |
908 | /* | |
909 | * Upper layer won't help us poll split bio | |
910 | * (io->orig_bio may only reflect a subset of the | |
911 | * pre-split original) so clear REQ_POLLED. | |
912 | */ | |
913 | bio_clear_polled(bio); | |
914 | } | |
915 | ||
e2736347 | 916 | /* |
444fe04f ML |
917 | * Target requested pushing back the I/O or |
918 | * polled IO hit BLK_STS_AGAIN. | |
e2736347 MS |
919 | */ |
920 | spin_lock_irqsave(&md->deferred_lock, flags); | |
444fe04f ML |
921 | if ((__noflush_suspending(md) && |
922 | !WARN_ON_ONCE(dm_is_zone_write(md, bio))) || | |
8b211aac ML |
923 | handle_polled_eagain || first_stage) { |
924 | dm_requeue_add_io(io, first_stage); | |
444fe04f | 925 | requeued = true; |
e2736347 | 926 | } else { |
2e93ccc1 | 927 | /* |
e2736347 MS |
928 | * noflush suspend was interrupted or this is |
929 | * a write to a zoned target. | |
2e93ccc1 | 930 | */ |
e2736347 | 931 | io->status = BLK_STS_IOERR; |
2e93ccc1 | 932 | } |
e2736347 MS |
933 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
934 | } | |
2e93ccc1 | 935 | |
444fe04f | 936 | if (requeued) |
8b211aac | 937 | dm_kick_requeue(md, first_stage); |
444fe04f ML |
938 | |
939 | return requeued; | |
940 | } | |
941 | ||
8b211aac | 942 | static void __dm_io_complete(struct dm_io *io, bool first_stage) |
444fe04f | 943 | { |
8b211aac | 944 | struct bio *bio = io->orig_bio; |
444fe04f ML |
945 | struct mapped_device *md = io->md; |
946 | blk_status_t io_error; | |
947 | bool requeued; | |
948 | ||
8b211aac ML |
949 | requeued = dm_handle_requeue(io, first_stage); |
950 | if (requeued && first_stage) | |
951 | return; | |
444fe04f | 952 | |
e2736347 | 953 | io_error = io->status; |
82f6cdcc | 954 | if (dm_io_flagged(io, DM_IO_ACCOUNTED)) |
b992b40d | 955 | dm_end_io_acct(io); |
e2736347 MS |
956 | else if (!io_error) { |
957 | /* | |
958 | * Must handle target that DM_MAPIO_SUBMITTED only to | |
959 | * then bio_endio() rather than dm_submit_bio_remap() | |
960 | */ | |
b992b40d ML |
961 | __dm_start_io_acct(io); |
962 | dm_end_io_acct(io); | |
e2736347 MS |
963 | } |
964 | free_io(io); | |
965 | smp_wmb(); | |
966 | this_cpu_dec(*md->pending_io); | |
6a8736d1 | 967 | |
e2736347 MS |
968 | /* nudge anyone waiting on suspend queue */ |
969 | if (unlikely(wq_has_sleeper(&md->wait))) | |
970 | wake_up(&md->wait); | |
2e93ccc1 | 971 | |
444fe04f ML |
972 | /* Return early if the original bio was requeued */ |
973 | if (requeued) | |
974 | return; | |
e2736347 MS |
975 | |
976 | if (bio_is_flush_with_data(bio)) { | |
977 | /* | |
978 | * Preflush done for flush with data, reissue | |
979 | * without REQ_PREFLUSH. | |
980 | */ | |
981 | bio->bi_opf &= ~REQ_PREFLUSH; | |
982 | queue_io(md, bio); | |
983 | } else { | |
984 | /* done with normal IO or empty flush */ | |
985 | if (io_error) | |
986 | bio->bi_status = io_error; | |
987 | bio_endio(bio); | |
988 | } | |
989 | } | |
990 | ||
8b211aac ML |
991 | static void dm_wq_requeue_work(struct work_struct *work) |
992 | { | |
993 | struct mapped_device *md = container_of(work, struct mapped_device, | |
994 | requeue_work); | |
995 | unsigned long flags; | |
996 | struct dm_io *io; | |
997 | ||
998 | /* reuse deferred lock to simplify dm_handle_requeue */ | |
999 | spin_lock_irqsave(&md->deferred_lock, flags); | |
1000 | io = md->requeue_list; | |
1001 | md->requeue_list = NULL; | |
1002 | spin_unlock_irqrestore(&md->deferred_lock, flags); | |
1003 | ||
1004 | while (io) { | |
1005 | struct dm_io *next = io->next; | |
1006 | ||
46754bd0 | 1007 | dm_io_rewind(io, &md->disk->bio_split); |
8b211aac ML |
1008 | |
1009 | io->next = NULL; | |
1010 | __dm_io_complete(io, false); | |
1011 | io = next; | |
f77692d6 | 1012 | cond_resched(); |
8b211aac ML |
1013 | } |
1014 | } | |
1015 | ||
1016 | /* | |
1017 | * Two staged requeue: | |
1018 | * | |
1019 | * 1) io->orig_bio points to the real original bio, and the part mapped to | |
1020 | * this io must be requeued, instead of other parts of the original bio. | |
1021 | * | |
1022 | * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io. | |
1023 | */ | |
1024 | static void dm_io_complete(struct dm_io *io) | |
1025 | { | |
1026 | bool first_requeue; | |
1027 | ||
1028 | /* | |
1029 | * Only dm_io that has been split needs two stage requeue, otherwise | |
1030 | * we may run into long bio clone chain during suspend and OOM could | |
1031 | * be triggered. | |
1032 | * | |
1033 | * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they | |
1034 | * also aren't handled via the first stage requeue. | |
1035 | */ | |
1036 | if (dm_io_flagged(io, DM_IO_WAS_SPLIT)) | |
1037 | first_requeue = true; | |
1038 | else | |
1039 | first_requeue = false; | |
1040 | ||
1041 | __dm_io_complete(io, first_requeue); | |
1042 | } | |
1043 | ||
1da177e4 LT |
1044 | /* |
1045 | * Decrements the number of outstanding ios that a bio has been | |
1046 | * cloned into, completing the original io if necc. | |
1047 | */ | |
84b98f4c MS |
1048 | static inline void __dm_io_dec_pending(struct dm_io *io) |
1049 | { | |
1050 | if (atomic_dec_and_test(&io->io_count)) | |
1051 | dm_io_complete(io); | |
1052 | } | |
1053 | ||
1054 | static void dm_io_set_error(struct dm_io *io, blk_status_t error) | |
1da177e4 | 1055 | { |
84b98f4c MS |
1056 | unsigned long flags; |
1057 | ||
2e93ccc1 | 1058 | /* Push-back supersedes any I/O errors */ |
84b98f4c MS |
1059 | spin_lock_irqsave(&io->lock, flags); |
1060 | if (!(io->status == BLK_STS_DM_REQUEUE && | |
1061 | __noflush_suspending(io->md))) { | |
1062 | io->status = error; | |
1da177e4 | 1063 | } |
84b98f4c MS |
1064 | spin_unlock_irqrestore(&io->lock, flags); |
1065 | } | |
1da177e4 | 1066 | |
2e803cd9 | 1067 | static void dm_io_dec_pending(struct dm_io *io, blk_status_t error) |
84b98f4c MS |
1068 | { |
1069 | if (unlikely(error)) | |
1070 | dm_io_set_error(io, error); | |
1071 | ||
1072 | __dm_io_dec_pending(io); | |
1da177e4 LT |
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 | ||
3720281d | 1165 | static sector_t max_io_len(struct dm_target *ti, sector_t sector) |
1da177e4 | 1166 | { |
3720281d MS |
1167 | sector_t target_offset = dm_target_offset(ti, sector); |
1168 | sector_t len = max_io_len_target_boundary(ti, target_offset); | |
1da177e4 LT |
1169 | |
1170 | /* | |
3ee16db3 MS |
1171 | * Does the target need to split IO even further? |
1172 | * - varied (per target) IO splitting is a tenet of DM; this | |
1173 | * explains why stacked chunk_sectors based splitting via | |
5a97806f | 1174 | * bio_split_to_limits() isn't possible here. |
1da177e4 | 1175 | */ |
c3949322 CH |
1176 | if (!ti->max_io_len) |
1177 | return len; | |
1178 | return min_t(sector_t, len, | |
1179 | min(queue_max_sectors(ti->table->md->queue), | |
1180 | blk_chunk_sectors_left(target_offset, ti->max_io_len))); | |
1da177e4 LT |
1181 | } |
1182 | ||
542f9038 MS |
1183 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1184 | { | |
1185 | if (len > UINT_MAX) { | |
1186 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1187 | (unsigned long long)len, UINT_MAX); | |
1188 | ti->error = "Maximum size of target IO is too large"; | |
1189 | return -EINVAL; | |
1190 | } | |
1191 | ||
75ae1936 | 1192 | ti->max_io_len = (uint32_t) len; |
542f9038 MS |
1193 | |
1194 | return 0; | |
1195 | } | |
1196 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1197 | ||
f26c5719 | 1198 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
3d97c829 MS |
1199 | sector_t sector, int *srcu_idx) |
1200 | __acquires(md->io_barrier) | |
545ed20e | 1201 | { |
545ed20e TK |
1202 | struct dm_table *map; |
1203 | struct dm_target *ti; | |
545ed20e | 1204 | |
f26c5719 | 1205 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1206 | if (!map) |
f26c5719 | 1207 | return NULL; |
545ed20e TK |
1208 | |
1209 | ti = dm_table_find_target(map, sector); | |
123d87d5 | 1210 | if (!ti) |
f26c5719 | 1211 | return NULL; |
545ed20e | 1212 | |
f26c5719 DW |
1213 | return ti; |
1214 | } | |
545ed20e | 1215 | |
f26c5719 | 1216 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
e511c4a3 JC |
1217 | long nr_pages, enum dax_access_mode mode, void **kaddr, |
1218 | pfn_t *pfn) | |
f26c5719 DW |
1219 | { |
1220 | struct mapped_device *md = dax_get_private(dax_dev); | |
1221 | sector_t sector = pgoff * PAGE_SECTORS; | |
1222 | struct dm_target *ti; | |
1223 | long len, ret = -EIO; | |
1224 | int srcu_idx; | |
545ed20e | 1225 | |
f26c5719 | 1226 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1227 | |
f26c5719 DW |
1228 | if (!ti) |
1229 | goto out; | |
1230 | if (!ti->type->direct_access) | |
1231 | goto out; | |
3720281d | 1232 | len = max_io_len(ti, sector) / PAGE_SECTORS; |
f26c5719 DW |
1233 | if (len < 1) |
1234 | goto out; | |
1235 | nr_pages = min(len, nr_pages); | |
e511c4a3 | 1236 | ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn); |
817bf402 | 1237 | |
f26c5719 | 1238 | out: |
545ed20e | 1239 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1240 | |
1241 | return ret; | |
545ed20e TK |
1242 | } |
1243 | ||
cdf6cdcd VG |
1244 | static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, |
1245 | size_t nr_pages) | |
1246 | { | |
1247 | struct mapped_device *md = dax_get_private(dax_dev); | |
1248 | sector_t sector = pgoff * PAGE_SECTORS; | |
1249 | struct dm_target *ti; | |
1250 | int ret = -EIO; | |
1251 | int srcu_idx; | |
1252 | ||
1253 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1254 | ||
1255 | if (!ti) | |
1256 | goto out; | |
1257 | if (WARN_ON(!ti->type->dax_zero_page_range)) { | |
1258 | /* | |
1259 | * ->zero_page_range() is mandatory dax operation. If we are | |
1260 | * here, something is wrong. | |
1261 | */ | |
cdf6cdcd VG |
1262 | goto out; |
1263 | } | |
1264 | ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages); | |
cdf6cdcd VG |
1265 | out: |
1266 | dm_put_live_table(md, srcu_idx); | |
1267 | ||
1268 | return ret; | |
1269 | } | |
1270 | ||
047218ec JC |
1271 | static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff, |
1272 | void *addr, size_t bytes, struct iov_iter *i) | |
1273 | { | |
1274 | struct mapped_device *md = dax_get_private(dax_dev); | |
1275 | sector_t sector = pgoff * PAGE_SECTORS; | |
1276 | struct dm_target *ti; | |
1277 | int srcu_idx; | |
1278 | long ret = 0; | |
1279 | ||
1280 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1281 | if (!ti || !ti->type->dax_recovery_write) | |
1282 | goto out; | |
1283 | ||
1284 | ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i); | |
1285 | out: | |
1286 | dm_put_live_table(md, srcu_idx); | |
1287 | return ret; | |
1288 | } | |
1289 | ||
1dd40c3e MP |
1290 | /* |
1291 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
6842d264 | 1292 | * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management |
e6fc9f62 MS |
1293 | * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by |
1294 | * __send_duplicate_bios(). | |
1dd40c3e MP |
1295 | * |
1296 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1297 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1298 | * sent in a next bio. | |
1299 | * | |
1300 | * A diagram that explains the arithmetics: | |
1301 | * +--------------------+---------------+-------+ | |
1302 | * | 1 | 2 | 3 | | |
1303 | * +--------------------+---------------+-------+ | |
1304 | * | |
1305 | * <-------------- *tio->len_ptr ---------------> | |
bdb34759 | 1306 | * <----- bio_sectors -----> |
1dd40c3e MP |
1307 | * <-- n_sectors --> |
1308 | * | |
1309 | * Region 1 was already iterated over with bio_advance or similar function. | |
1310 | * (it may be empty if the target doesn't use bio_advance) | |
1311 | * Region 2 is the remaining bio size that the target wants to process. | |
1312 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1313 | * to make it empty) | |
1314 | * The target requires that region 3 is to be sent in the next bio. | |
1315 | * | |
1316 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1317 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1318 | * copies of the bio. | |
1319 | */ | |
86a3238c | 1320 | void dm_accept_partial_bio(struct bio *bio, unsigned int n_sectors) |
1dd40c3e | 1321 | { |
6c23f0bd | 1322 | struct dm_target_io *tio = clone_to_tio(bio); |
8b211aac | 1323 | struct dm_io *io = tio->io; |
86a3238c | 1324 | unsigned int bio_sectors = bio_sectors(bio); |
6842d264 | 1325 | |
655f3aad | 1326 | BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO)); |
6842d264 DLM |
1327 | BUG_ON(op_is_zone_mgmt(bio_op(bio))); |
1328 | BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND); | |
bdb34759 MS |
1329 | BUG_ON(bio_sectors > *tio->len_ptr); |
1330 | BUG_ON(n_sectors > bio_sectors); | |
6842d264 | 1331 | |
bdb34759 | 1332 | *tio->len_ptr -= bio_sectors - n_sectors; |
1dd40c3e | 1333 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; |
7dd76d1f ML |
1334 | |
1335 | /* | |
1336 | * __split_and_process_bio() may have already saved mapped part | |
1337 | * for accounting but it is being reduced so update accordingly. | |
1338 | */ | |
8b211aac ML |
1339 | dm_io_set_flag(io, DM_IO_WAS_SPLIT); |
1340 | io->sectors = n_sectors; | |
1341 | io->sector_offset = bio_sectors(io->orig_bio); | |
1dd40c3e MP |
1342 | } |
1343 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1344 | ||
0fbb4d93 MS |
1345 | /* |
1346 | * @clone: clone bio that DM core passed to target's .map function | |
1347 | * @tgt_clone: clone of @clone bio that target needs submitted | |
0fbb4d93 MS |
1348 | * |
1349 | * Targets should use this interface to submit bios they take | |
1350 | * ownership of when returning DM_MAPIO_SUBMITTED. | |
1351 | * | |
1352 | * Target should also enable ti->accounts_remapped_io | |
1353 | */ | |
b7f8dff0 | 1354 | void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone) |
0fbb4d93 MS |
1355 | { |
1356 | struct dm_target_io *tio = clone_to_tio(clone); | |
1357 | struct dm_io *io = tio->io; | |
1358 | ||
1359 | /* establish bio that will get submitted */ | |
1360 | if (!tgt_clone) | |
1361 | tgt_clone = clone; | |
1362 | ||
1363 | /* | |
1364 | * Account io->origin_bio to DM dev on behalf of target | |
1365 | * that took ownership of IO with DM_MAPIO_SUBMITTED. | |
1366 | */ | |
9d20653f | 1367 | dm_start_io_acct(io, clone); |
0fbb4d93 | 1368 | |
9d20653f | 1369 | trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk), |
0fbb4d93 | 1370 | tio->old_sector); |
9d20653f | 1371 | submit_bio_noacct(tgt_clone); |
0fbb4d93 MS |
1372 | } |
1373 | EXPORT_SYMBOL_GPL(dm_submit_bio_remap); | |
1374 | ||
a666e5c0 MP |
1375 | static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch) |
1376 | { | |
1377 | mutex_lock(&md->swap_bios_lock); | |
1378 | while (latch < md->swap_bios) { | |
1379 | cond_resched(); | |
1380 | down(&md->swap_bios_semaphore); | |
1381 | md->swap_bios--; | |
1382 | } | |
1383 | while (latch > md->swap_bios) { | |
1384 | cond_resched(); | |
1385 | up(&md->swap_bios_semaphore); | |
1386 | md->swap_bios++; | |
1387 | } | |
1388 | mutex_unlock(&md->swap_bios_lock); | |
1389 | } | |
1390 | ||
1561b396 | 1391 | static void __map_bio(struct bio *clone) |
1da177e4 | 1392 | { |
1561b396 | 1393 | struct dm_target_io *tio = clone_to_tio(clone); |
bd2a49b8 | 1394 | struct dm_target *ti = tio->ti; |
6cbce280 MS |
1395 | struct dm_io *io = tio->io; |
1396 | struct mapped_device *md = io->md; | |
1397 | int r; | |
1da177e4 | 1398 | |
1da177e4 | 1399 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1400 | |
1401 | /* | |
0fbb4d93 | 1402 | * Map the clone. |
1da177e4 | 1403 | */ |
743598f0 | 1404 | tio->old_sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1405 | |
442761fd MS |
1406 | if (static_branch_unlikely(&swap_bios_enabled) && |
1407 | unlikely(swap_bios_limit(ti, clone))) { | |
a666e5c0 | 1408 | int latch = get_swap_bios(); |
b30f1607 | 1409 | |
a666e5c0 MP |
1410 | if (unlikely(latch != md->swap_bios)) |
1411 | __set_swap_bios_limit(md, latch); | |
1412 | down(&md->swap_bios_semaphore); | |
1413 | } | |
1414 | ||
442761fd MS |
1415 | if (static_branch_unlikely(&zoned_enabled)) { |
1416 | /* | |
1417 | * Check if the IO needs a special mapping due to zone append | |
1418 | * emulation on zoned target. In this case, dm_zone_map_bio() | |
1419 | * calls the target map operation. | |
1420 | */ | |
1421 | if (unlikely(dm_emulate_zone_append(md))) | |
1422 | r = dm_zone_map_bio(tio); | |
1423 | else | |
1424 | r = ti->type->map(ti, clone); | |
1425 | } else | |
bb37d772 DLM |
1426 | r = ti->type->map(ti, clone); |
1427 | ||
846785e6 CH |
1428 | switch (r) { |
1429 | case DM_MAPIO_SUBMITTED: | |
0fbb4d93 MS |
1430 | /* target has assumed ownership of this io */ |
1431 | if (!ti->accounts_remapped_io) | |
9d20653f | 1432 | dm_start_io_acct(io, clone); |
846785e6 CH |
1433 | break; |
1434 | case DM_MAPIO_REMAPPED: | |
9d20653f | 1435 | dm_submit_bio_remap(clone, NULL); |
846785e6 CH |
1436 | break; |
1437 | case DM_MAPIO_KILL: | |
846785e6 | 1438 | case DM_MAPIO_REQUEUE: |
442761fd MS |
1439 | if (static_branch_unlikely(&swap_bios_enabled) && |
1440 | unlikely(swap_bios_limit(ti, clone))) | |
6cbce280 | 1441 | up(&md->swap_bios_semaphore); |
1d1068ce | 1442 | free_tio(clone); |
90a2326e MS |
1443 | if (r == DM_MAPIO_KILL) |
1444 | dm_io_dec_pending(io, BLK_STS_IOERR); | |
1445 | else | |
1446 | dm_io_dec_pending(io, BLK_STS_DM_REQUEUE); | |
846785e6 CH |
1447 | break; |
1448 | default: | |
43e6c111 | 1449 | DMCRIT("unimplemented target map return value: %d", r); |
45cbcd79 | 1450 | BUG(); |
1da177e4 LT |
1451 | } |
1452 | } | |
1da177e4 | 1453 | |
86a3238c | 1454 | static void setup_split_accounting(struct clone_info *ci, unsigned int len) |
7dd76d1f ML |
1455 | { |
1456 | struct dm_io *io = ci->io; | |
1457 | ||
1458 | if (ci->sector_count > len) { | |
1459 | /* | |
1460 | * Split needed, save the mapped part for accounting. | |
1461 | * NOTE: dm_accept_partial_bio() will update accordingly. | |
1462 | */ | |
1463 | dm_io_set_flag(io, DM_IO_WAS_SPLIT); | |
1464 | io->sectors = len; | |
8b211aac | 1465 | io->sector_offset = bio_sectors(ci->bio); |
7dd76d1f ML |
1466 | } |
1467 | } | |
1468 | ||
318716dd | 1469 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
f7b58a69 MS |
1470 | struct dm_target *ti, unsigned int num_bios, |
1471 | unsigned *len) | |
f9ab94ce | 1472 | { |
1d1068ce | 1473 | struct bio *bio; |
318716dd | 1474 | int try; |
dba14160 | 1475 | |
318716dd MS |
1476 | for (try = 0; try < 2; try++) { |
1477 | int bio_nr; | |
318716dd MS |
1478 | |
1479 | if (try) | |
bc02cdbe | 1480 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd | 1481 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
f7b58a69 | 1482 | bio = alloc_tio(ci, ti, bio_nr, len, |
dc8e2021 | 1483 | try ? GFP_NOIO : GFP_NOWAIT); |
1d1068ce | 1484 | if (!bio) |
318716dd MS |
1485 | break; |
1486 | ||
1d1068ce | 1487 | bio_list_add(blist, bio); |
318716dd MS |
1488 | } |
1489 | if (try) | |
bc02cdbe | 1490 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1491 | if (bio_nr == num_bios) |
1492 | return; | |
1493 | ||
6c23f0bd | 1494 | while ((bio = bio_list_pop(blist))) |
1d1068ce | 1495 | free_tio(bio); |
318716dd | 1496 | } |
9015df24 AK |
1497 | } |
1498 | ||
0f14d60a | 1499 | static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
86a3238c | 1500 | unsigned int num_bios, unsigned int *len) |
06a426ce | 1501 | { |
318716dd | 1502 | struct bio_list blist = BIO_EMPTY_LIST; |
8eabf5d0 | 1503 | struct bio *clone; |
564b5c54 | 1504 | unsigned int ret = 0; |
06a426ce | 1505 | |
891fced6 CH |
1506 | switch (num_bios) { |
1507 | case 0: | |
1508 | break; | |
1509 | case 1: | |
7dd76d1f ML |
1510 | if (len) |
1511 | setup_split_accounting(ci, *len); | |
891fced6 | 1512 | clone = alloc_tio(ci, ti, 0, len, GFP_NOIO); |
1561b396 | 1513 | __map_bio(clone); |
0f14d60a | 1514 | ret = 1; |
891fced6 CH |
1515 | break; |
1516 | default: | |
666eed46 MS |
1517 | if (len) |
1518 | setup_split_accounting(ci, *len); | |
7dd06a25 | 1519 | /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */ |
f7b58a69 | 1520 | alloc_multiple_bios(&blist, ci, ti, num_bios, len); |
891fced6 | 1521 | while ((clone = bio_list_pop(&blist))) { |
655f3aad | 1522 | dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO); |
891fced6 | 1523 | __map_bio(clone); |
0f14d60a | 1524 | ret += 1; |
891fced6 CH |
1525 | } |
1526 | break; | |
318716dd | 1527 | } |
0f14d60a ML |
1528 | |
1529 | return ret; | |
06a426ce MS |
1530 | } |
1531 | ||
332f2b1e | 1532 | static void __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1533 | { |
564b5c54 | 1534 | struct dm_table *t = ci->map; |
828678b8 MS |
1535 | struct bio flush_bio; |
1536 | ||
1537 | /* | |
1538 | * Use an on-stack bio for this, it's safe since we don't | |
1539 | * need to reference it after submit. It's just used as | |
1540 | * the basis for the clone(s). | |
1541 | */ | |
49add496 CH |
1542 | bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0, |
1543 | REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC); | |
47d95102 | 1544 | |
828678b8 MS |
1545 | ci->bio = &flush_bio; |
1546 | ci->sector_count = 0; | |
92b914e2 | 1547 | ci->io->tio.clone.bi_iter.bi_size = 0; |
f9ab94ce | 1548 | |
564b5c54 MS |
1549 | for (unsigned int i = 0; i < t->num_targets; i++) { |
1550 | unsigned int bios; | |
1551 | struct dm_target *ti = dm_table_get_target(t, i); | |
0f14d60a ML |
1552 | |
1553 | atomic_add(ti->num_flush_bios, &ci->io->io_count); | |
1554 | bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); | |
1555 | atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count); | |
1556 | } | |
1557 | ||
1558 | /* | |
1559 | * alloc_io() takes one extra reference for submission, so the | |
1560 | * reference won't reach 0 without the following subtraction | |
1561 | */ | |
1562 | atomic_sub(1, &ci->io->io_count); | |
828678b8 MS |
1563 | |
1564 | bio_uninit(ci->bio); | |
f9ab94ce MP |
1565 | } |
1566 | ||
e6fc9f62 | 1567 | static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
86a3238c | 1568 | unsigned int num_bios) |
ba1cbad9 | 1569 | { |
86a3238c | 1570 | unsigned int len, bios; |
ba1cbad9 | 1571 | |
3720281d MS |
1572 | len = min_t(sector_t, ci->sector_count, |
1573 | max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector))); | |
51b86f9a | 1574 | |
0f14d60a ML |
1575 | atomic_add(num_bios, &ci->io->io_count); |
1576 | bios = __send_duplicate_bios(ci, ti, num_bios, &len); | |
1577 | /* | |
1578 | * alloc_io() takes one extra reference for submission, so the | |
1579 | * reference won't reach 0 without the following (+1) subtraction | |
1580 | */ | |
1581 | atomic_sub(num_bios - bios + 1, &ci->io->io_count); | |
7dd06a25 | 1582 | |
3d7f4562 MS |
1583 | ci->sector += len; |
1584 | ci->sector_count -= len; | |
ba1cbad9 MS |
1585 | } |
1586 | ||
568c73a3 MS |
1587 | static bool is_abnormal_io(struct bio *bio) |
1588 | { | |
a3282b43 | 1589 | enum req_op op = bio_op(bio); |
568c73a3 | 1590 | |
4edadf6d MS |
1591 | if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) { |
1592 | switch (op) { | |
1593 | case REQ_OP_DISCARD: | |
1594 | case REQ_OP_SECURE_ERASE: | |
1595 | case REQ_OP_WRITE_ZEROES: | |
1596 | return true; | |
1597 | default: | |
1598 | break; | |
1599 | } | |
568c73a3 MS |
1600 | } |
1601 | ||
4edadf6d | 1602 | return false; |
568c73a3 MS |
1603 | } |
1604 | ||
4edadf6d MS |
1605 | static blk_status_t __process_abnormal_io(struct clone_info *ci, |
1606 | struct dm_target *ti) | |
0519c71e | 1607 | { |
86a3238c | 1608 | unsigned int num_bios = 0; |
0519c71e | 1609 | |
e6fc9f62 | 1610 | switch (bio_op(ci->bio)) { |
9679b5a7 MS |
1611 | case REQ_OP_DISCARD: |
1612 | num_bios = ti->num_discard_bios; | |
1613 | break; | |
1614 | case REQ_OP_SECURE_ERASE: | |
1615 | num_bios = ti->num_secure_erase_bios; | |
1616 | break; | |
9679b5a7 MS |
1617 | case REQ_OP_WRITE_ZEROES: |
1618 | num_bios = ti->num_write_zeroes_bios; | |
1619 | break; | |
2d9b02be BVA |
1620 | default: |
1621 | break; | |
9679b5a7 | 1622 | } |
0519c71e | 1623 | |
e6fc9f62 MS |
1624 | /* |
1625 | * Even though the device advertised support for this type of | |
1626 | * request, that does not mean every target supports it, and | |
1627 | * reconfiguration might also have changed that since the | |
1628 | * check was performed. | |
1629 | */ | |
84b98f4c | 1630 | if (unlikely(!num_bios)) |
4edadf6d MS |
1631 | return BLK_STS_NOTSUPP; |
1632 | ||
1633 | __send_changing_extent_only(ci, ti, num_bios); | |
1634 | return BLK_STS_OK; | |
0519c71e MS |
1635 | } |
1636 | ||
b99fdcdc | 1637 | /* |
ec211631 | 1638 | * Reuse ->bi_private as dm_io list head for storing all dm_io instances |
b99fdcdc ML |
1639 | * associated with this bio, and this bio's bi_private needs to be |
1640 | * stored in dm_io->data before the reuse. | |
1641 | * | |
1642 | * bio->bi_private is owned by fs or upper layer, so block layer won't | |
1643 | * touch it after splitting. Meantime it won't be changed by anyone after | |
1644 | * bio is submitted. So this reuse is safe. | |
1645 | */ | |
ec211631 | 1646 | static inline struct dm_io **dm_poll_list_head(struct bio *bio) |
b99fdcdc | 1647 | { |
ec211631 | 1648 | return (struct dm_io **)&bio->bi_private; |
b99fdcdc ML |
1649 | } |
1650 | ||
1651 | static void dm_queue_poll_io(struct bio *bio, struct dm_io *io) | |
1652 | { | |
ec211631 | 1653 | struct dm_io **head = dm_poll_list_head(bio); |
b99fdcdc ML |
1654 | |
1655 | if (!(bio->bi_opf & REQ_DM_POLL_LIST)) { | |
1656 | bio->bi_opf |= REQ_DM_POLL_LIST; | |
1657 | /* | |
1658 | * Save .bi_private into dm_io, so that we can reuse | |
ec211631 | 1659 | * .bi_private as dm_io list head for storing dm_io list |
b99fdcdc ML |
1660 | */ |
1661 | io->data = bio->bi_private; | |
1662 | ||
b99fdcdc ML |
1663 | /* tell block layer to poll for completion */ |
1664 | bio->bi_cookie = ~BLK_QC_T_NONE; | |
ec211631 ML |
1665 | |
1666 | io->next = NULL; | |
b99fdcdc ML |
1667 | } else { |
1668 | /* | |
1669 | * bio recursed due to split, reuse original poll list, | |
1670 | * and save bio->bi_private too. | |
1671 | */ | |
ec211631 ML |
1672 | io->data = (*head)->data; |
1673 | io->next = *head; | |
b99fdcdc ML |
1674 | } |
1675 | ||
ec211631 | 1676 | *head = io; |
b99fdcdc ML |
1677 | } |
1678 | ||
e4c93811 AK |
1679 | /* |
1680 | * Select the correct strategy for processing a non-flush bio. | |
1681 | */ | |
84b98f4c | 1682 | static blk_status_t __split_and_process_bio(struct clone_info *ci) |
0ce65797 | 1683 | { |
66bdaa43 | 1684 | struct bio *clone; |
512875bd | 1685 | struct dm_target *ti; |
86a3238c | 1686 | unsigned int len; |
0ce65797 | 1687 | |
512875bd | 1688 | ti = dm_table_find_target(ci->map, ci->sector); |
4edadf6d MS |
1689 | if (unlikely(!ti)) |
1690 | return BLK_STS_IOERR; | |
1ee88de3 MP |
1691 | |
1692 | if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) && | |
1693 | unlikely(!dm_target_supports_nowait(ti->type))) | |
1694 | return BLK_STS_NOTSUPP; | |
1695 | ||
1696 | if (unlikely(ci->is_abnormal_io)) | |
4edadf6d | 1697 | return __process_abnormal_io(ci, ti); |
3d7f4562 | 1698 | |
b99fdcdc ML |
1699 | /* |
1700 | * Only support bio polling for normal IO, and the target io is | |
1701 | * exactly inside the dm_io instance (verified in dm_poll_dm_io) | |
1702 | */ | |
a3282b43 | 1703 | ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED); |
0ce65797 | 1704 | |
3720281d | 1705 | len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count); |
7dd76d1f | 1706 | setup_split_accounting(ci, len); |
66bdaa43 | 1707 | clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO); |
66bdaa43 | 1708 | __map_bio(clone); |
0ce65797 | 1709 | |
1c3b13e6 KO |
1710 | ci->sector += len; |
1711 | ci->sector_count -= len; | |
0ce65797 | 1712 | |
84b98f4c | 1713 | return BLK_STS_OK; |
0ce65797 MS |
1714 | } |
1715 | ||
978e51ba | 1716 | static void init_clone_info(struct clone_info *ci, struct mapped_device *md, |
4edadf6d | 1717 | struct dm_table *map, struct bio *bio, bool is_abnormal) |
978e51ba MS |
1718 | { |
1719 | ci->map = map; | |
1720 | ci->io = alloc_io(md, bio); | |
d41e077a | 1721 | ci->bio = bio; |
4edadf6d | 1722 | ci->is_abnormal_io = is_abnormal; |
b99fdcdc | 1723 | ci->submit_as_polled = false; |
978e51ba | 1724 | ci->sector = bio->bi_iter.bi_sector; |
d41e077a MS |
1725 | ci->sector_count = bio_sectors(bio); |
1726 | ||
1727 | /* Shouldn't happen but sector_count was being set to 0 so... */ | |
442761fd MS |
1728 | if (static_branch_unlikely(&zoned_enabled) && |
1729 | WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count)) | |
d41e077a | 1730 | ci->sector_count = 0; |
978e51ba MS |
1731 | } |
1732 | ||
1da177e4 | 1733 | /* |
14fe594d | 1734 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1735 | */ |
96c9865c MS |
1736 | static void dm_split_and_process_bio(struct mapped_device *md, |
1737 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1738 | { |
1da177e4 | 1739 | struct clone_info ci; |
4857abf6 | 1740 | struct dm_io *io; |
84b98f4c | 1741 | blk_status_t error = BLK_STS_OK; |
4edadf6d MS |
1742 | bool is_abnormal; |
1743 | ||
1744 | is_abnormal = is_abnormal_io(bio); | |
1745 | if (unlikely(is_abnormal)) { | |
1746 | /* | |
5a97806f | 1747 | * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc) |
4edadf6d MS |
1748 | * otherwise associated queue_limits won't be imposed. |
1749 | */ | |
5a97806f | 1750 | bio = bio_split_to_limits(bio); |
613b1488 JA |
1751 | if (!bio) |
1752 | return; | |
4edadf6d | 1753 | } |
1da177e4 | 1754 | |
4edadf6d | 1755 | init_clone_info(&ci, md, map, bio, is_abnormal); |
4857abf6 | 1756 | io = ci.io; |
0ce65797 | 1757 | |
1eff9d32 | 1758 | if (bio->bi_opf & REQ_PREFLUSH) { |
332f2b1e | 1759 | __send_empty_flush(&ci); |
e2736347 | 1760 | /* dm_io_complete submits any data associated with flush */ |
d41e077a | 1761 | goto out; |
d87f4c14 | 1762 | } |
0ce65797 | 1763 | |
d41e077a MS |
1764 | error = __split_and_process_bio(&ci); |
1765 | if (error || !ci.sector_count) | |
1766 | goto out; | |
d41e077a MS |
1767 | /* |
1768 | * Remainder must be passed to submit_bio_noacct() so it gets handled | |
1769 | * *after* bios already submitted have been completely processed. | |
d41e077a | 1770 | */ |
8b211aac ML |
1771 | bio_trim(bio, io->sectors, ci.sector_count); |
1772 | trace_block_split(bio, bio->bi_iter.bi_sector); | |
1773 | bio_inc_remaining(bio); | |
d41e077a MS |
1774 | submit_bio_noacct(bio); |
1775 | out: | |
b99fdcdc ML |
1776 | /* |
1777 | * Drop the extra reference count for non-POLLED bio, and hold one | |
1778 | * reference for POLLED bio, which will be released in dm_poll_bio | |
1779 | * | |
ec211631 ML |
1780 | * Add every dm_io instance into the dm_io list head which is stored |
1781 | * in bio->bi_private, so that dm_poll_bio can poll them all. | |
b99fdcdc | 1782 | */ |
0f14d60a ML |
1783 | if (error || !ci.submit_as_polled) { |
1784 | /* | |
1785 | * In case of submission failure, the extra reference for | |
1786 | * submitting io isn't consumed yet | |
1787 | */ | |
1788 | if (error) | |
1789 | atomic_dec(&io->io_count); | |
1790 | dm_io_dec_pending(io, error); | |
1791 | } else | |
4857abf6 | 1792 | dm_queue_poll_io(bio, io); |
0ce65797 MS |
1793 | } |
1794 | ||
3e08773c | 1795 | static void dm_submit_bio(struct bio *bio) |
cec47e3d | 1796 | { |
309dca30 | 1797 | struct mapped_device *md = bio->bi_bdev->bd_disk->private_data; |
83d5e5b0 MP |
1798 | int srcu_idx; |
1799 | struct dm_table *map; | |
a3282b43 | 1800 | blk_opf_t bio_opf = bio->bi_opf; |
cec47e3d | 1801 | |
5d7362d0 | 1802 | map = dm_get_live_table_bio(md, &srcu_idx, bio_opf); |
29e4013d | 1803 | |
fa247089 MS |
1804 | /* If suspended, or map not yet available, queue this IO for later */ |
1805 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) || | |
1806 | unlikely(!map)) { | |
6abc4946 KK |
1807 | if (bio->bi_opf & REQ_NOWAIT) |
1808 | bio_wouldblock_error(bio); | |
b2abdb1b | 1809 | else if (bio->bi_opf & REQ_RAHEAD) |
54d9a1b4 | 1810 | bio_io_error(bio); |
b2abdb1b MS |
1811 | else |
1812 | queue_io(md, bio); | |
1813 | goto out; | |
cec47e3d | 1814 | } |
1da177e4 | 1815 | |
96c9865c | 1816 | dm_split_and_process_bio(md, map, bio); |
b2abdb1b | 1817 | out: |
5d7362d0 | 1818 | dm_put_live_table_bio(md, srcu_idx, bio_opf); |
978e51ba MS |
1819 | } |
1820 | ||
b99fdcdc ML |
1821 | static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob, |
1822 | unsigned int flags) | |
1823 | { | |
655f3aad | 1824 | WARN_ON_ONCE(!dm_tio_is_normal(&io->tio)); |
b99fdcdc ML |
1825 | |
1826 | /* don't poll if the mapped io is done */ | |
1827 | if (atomic_read(&io->io_count) > 1) | |
1828 | bio_poll(&io->tio.clone, iob, flags); | |
1829 | ||
1830 | /* bio_poll holds the last reference */ | |
1831 | return atomic_read(&io->io_count) == 1; | |
1832 | } | |
1833 | ||
1834 | static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob, | |
1835 | unsigned int flags) | |
1836 | { | |
ec211631 ML |
1837 | struct dm_io **head = dm_poll_list_head(bio); |
1838 | struct dm_io *list = *head; | |
1839 | struct dm_io *tmp = NULL; | |
1840 | struct dm_io *curr, *next; | |
b99fdcdc ML |
1841 | |
1842 | /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */ | |
1843 | if (!(bio->bi_opf & REQ_DM_POLL_LIST)) | |
1844 | return 0; | |
1845 | ||
ec211631 | 1846 | WARN_ON_ONCE(!list); |
b99fdcdc ML |
1847 | |
1848 | /* | |
1849 | * Restore .bi_private before possibly completing dm_io. | |
1850 | * | |
1851 | * bio_poll() is only possible once @bio has been completely | |
1852 | * submitted via submit_bio_noacct()'s depth-first submission. | |
1853 | * So there is no dm_queue_poll_io() race associated with | |
1854 | * clearing REQ_DM_POLL_LIST here. | |
1855 | */ | |
1856 | bio->bi_opf &= ~REQ_DM_POLL_LIST; | |
ec211631 | 1857 | bio->bi_private = list->data; |
b99fdcdc | 1858 | |
ec211631 ML |
1859 | for (curr = list, next = curr->next; curr; curr = next, next = |
1860 | curr ? curr->next : NULL) { | |
1861 | if (dm_poll_dm_io(curr, iob, flags)) { | |
b99fdcdc | 1862 | /* |
84b98f4c MS |
1863 | * clone_endio() has already occurred, so no |
1864 | * error handling is needed here. | |
b99fdcdc | 1865 | */ |
ec211631 ML |
1866 | __dm_io_dec_pending(curr); |
1867 | } else { | |
1868 | curr->next = tmp; | |
1869 | tmp = curr; | |
b99fdcdc ML |
1870 | } |
1871 | } | |
1872 | ||
1873 | /* Not done? */ | |
ec211631 | 1874 | if (tmp) { |
b99fdcdc ML |
1875 | bio->bi_opf |= REQ_DM_POLL_LIST; |
1876 | /* Reset bio->bi_private to dm_io list head */ | |
ec211631 | 1877 | *head = tmp; |
b99fdcdc ML |
1878 | return 0; |
1879 | } | |
1880 | return 1; | |
1881 | } | |
1882 | ||
a4a82ce3 HM |
1883 | /* |
1884 | *--------------------------------------------------------------- | |
1da177e4 | 1885 | * An IDR is used to keep track of allocated minor numbers. |
a4a82ce3 HM |
1886 | *--------------------------------------------------------------- |
1887 | */ | |
2b06cfff | 1888 | static void free_minor(int minor) |
1da177e4 | 1889 | { |
f32c10b0 | 1890 | spin_lock(&_minor_lock); |
1da177e4 | 1891 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1892 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1893 | } |
1894 | ||
1895 | /* | |
1896 | * See if the device with a specific minor # is free. | |
1897 | */ | |
cf13ab8e | 1898 | static int specific_minor(int minor) |
1da177e4 | 1899 | { |
c9d76be6 | 1900 | int r; |
1da177e4 LT |
1901 | |
1902 | if (minor >= (1 << MINORBITS)) | |
1903 | return -EINVAL; | |
1904 | ||
c9d76be6 | 1905 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1906 | spin_lock(&_minor_lock); |
1da177e4 | 1907 | |
c9d76be6 | 1908 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1909 | |
f32c10b0 | 1910 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1911 | idr_preload_end(); |
1912 | if (r < 0) | |
1913 | return r == -ENOSPC ? -EBUSY : r; | |
1914 | return 0; | |
1da177e4 LT |
1915 | } |
1916 | ||
cf13ab8e | 1917 | static int next_free_minor(int *minor) |
1da177e4 | 1918 | { |
c9d76be6 | 1919 | int r; |
62f75c2f | 1920 | |
c9d76be6 | 1921 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1922 | spin_lock(&_minor_lock); |
1da177e4 | 1923 | |
c9d76be6 | 1924 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1925 | |
f32c10b0 | 1926 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1927 | idr_preload_end(); |
1928 | if (r < 0) | |
1929 | return r; | |
1930 | *minor = r; | |
1931 | return 0; | |
1da177e4 LT |
1932 | } |
1933 | ||
83d5cde4 | 1934 | static const struct block_device_operations dm_blk_dops; |
681cc5e8 | 1935 | static const struct block_device_operations dm_rq_blk_dops; |
f26c5719 | 1936 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1937 | |
53d5914f MP |
1938 | static void dm_wq_work(struct work_struct *work); |
1939 | ||
aa6ce87a | 1940 | #ifdef CONFIG_BLK_INLINE_ENCRYPTION |
cb77cb5a | 1941 | static void dm_queue_destroy_crypto_profile(struct request_queue *q) |
aa6ce87a | 1942 | { |
cb77cb5a | 1943 | dm_destroy_crypto_profile(q->crypto_profile); |
aa6ce87a ST |
1944 | } |
1945 | ||
1946 | #else /* CONFIG_BLK_INLINE_ENCRYPTION */ | |
1947 | ||
cb77cb5a | 1948 | static inline void dm_queue_destroy_crypto_profile(struct request_queue *q) |
aa6ce87a ST |
1949 | { |
1950 | } | |
1951 | #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */ | |
1952 | ||
0f20972f MS |
1953 | static void cleanup_mapped_device(struct mapped_device *md) |
1954 | { | |
0f20972f MS |
1955 | if (md->wq) |
1956 | destroy_workqueue(md->wq); | |
29dec90a | 1957 | dm_free_md_mempools(md->mempools); |
0f20972f | 1958 | |
f26c5719 | 1959 | if (md->dax_dev) { |
fb08a190 | 1960 | dax_remove_host(md->disk); |
f26c5719 DW |
1961 | kill_dax(md->dax_dev); |
1962 | put_dax(md->dax_dev); | |
1963 | md->dax_dev = NULL; | |
1964 | } | |
1965 | ||
588b7f5d | 1966 | dm_cleanup_zoned_dev(md); |
0f20972f MS |
1967 | if (md->disk) { |
1968 | spin_lock(&_minor_lock); | |
1969 | md->disk->private_data = NULL; | |
1970 | spin_unlock(&_minor_lock); | |
89f871af | 1971 | if (dm_get_md_type(md) != DM_TYPE_NONE) { |
1a581b72 CH |
1972 | struct table_device *td; |
1973 | ||
89f871af | 1974 | dm_sysfs_exit(md); |
1a581b72 CH |
1975 | list_for_each_entry(td, &md->table_devices, list) { |
1976 | bd_unlink_disk_holder(td->dm_dev.bdev, | |
1977 | md->disk); | |
1978 | } | |
d563792c YK |
1979 | |
1980 | /* | |
1981 | * Hold lock to make sure del_gendisk() won't concurrent | |
1982 | * with open/close_table_device(). | |
1983 | */ | |
1984 | mutex_lock(&md->table_devices_lock); | |
89f871af | 1985 | del_gendisk(md->disk); |
d563792c | 1986 | mutex_unlock(&md->table_devices_lock); |
89f871af | 1987 | } |
cb77cb5a | 1988 | dm_queue_destroy_crypto_profile(md->queue); |
8b9ab626 | 1989 | put_disk(md->disk); |
74a2b6ec | 1990 | } |
0f20972f | 1991 | |
9f6dc633 MS |
1992 | if (md->pending_io) { |
1993 | free_percpu(md->pending_io); | |
1994 | md->pending_io = NULL; | |
1995 | } | |
1996 | ||
d09960b0 TE |
1997 | cleanup_srcu_struct(&md->io_barrier); |
1998 | ||
d5ffebdd MS |
1999 | mutex_destroy(&md->suspend_lock); |
2000 | mutex_destroy(&md->type_lock); | |
2001 | mutex_destroy(&md->table_devices_lock); | |
a666e5c0 | 2002 | mutex_destroy(&md->swap_bios_lock); |
d5ffebdd | 2003 | |
4cc96131 | 2004 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
2005 | } |
2006 | ||
1da177e4 LT |
2007 | /* |
2008 | * Allocate and initialise a blank device with a given minor. | |
2009 | */ | |
2b06cfff | 2010 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 2011 | { |
115485e8 MS |
2012 | int r, numa_node_id = dm_get_numa_node(); |
2013 | struct mapped_device *md; | |
ba61fdd1 | 2014 | void *old_md; |
1da177e4 | 2015 | |
856eb091 | 2016 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 | 2017 | if (!md) { |
43e6c111 | 2018 | DMERR("unable to allocate device, out of memory."); |
1da177e4 LT |
2019 | return NULL; |
2020 | } | |
2021 | ||
10da4f79 | 2022 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2023 | goto bad_module_get; |
10da4f79 | 2024 | |
1da177e4 | 2025 | /* get a minor number for the dev */ |
2b06cfff | 2026 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2027 | r = next_free_minor(&minor); |
2b06cfff | 2028 | else |
cf13ab8e | 2029 | r = specific_minor(minor); |
1da177e4 | 2030 | if (r < 0) |
6ed7ade8 | 2031 | goto bad_minor; |
1da177e4 | 2032 | |
83d5e5b0 MP |
2033 | r = init_srcu_struct(&md->io_barrier); |
2034 | if (r < 0) | |
2035 | goto bad_io_barrier; | |
2036 | ||
115485e8 | 2037 | md->numa_node_id = numa_node_id; |
591ddcfc | 2038 | md->init_tio_pdu = false; |
a5664dad | 2039 | md->type = DM_TYPE_NONE; |
e61290a4 | 2040 | mutex_init(&md->suspend_lock); |
a5664dad | 2041 | mutex_init(&md->type_lock); |
86f1152b | 2042 | mutex_init(&md->table_devices_lock); |
022c2611 | 2043 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2044 | atomic_set(&md->holders, 1); |
5c6bd75d | 2045 | atomic_set(&md->open_count, 0); |
1da177e4 | 2046 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2047 | atomic_set(&md->uevent_seq, 0); |
2048 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2049 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2050 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2051 | |
47ace7e0 | 2052 | /* |
c62b37d9 CH |
2053 | * default to bio-based until DM table is loaded and md->type |
2054 | * established. If request-based table is loaded: blk-mq will | |
2055 | * override accordingly. | |
47ace7e0 | 2056 | */ |
74fe6ba9 | 2057 | md->disk = blk_alloc_disk(md->numa_node_id); |
1da177e4 | 2058 | if (!md->disk) |
0f20972f | 2059 | goto bad; |
74fe6ba9 | 2060 | md->queue = md->disk->queue; |
1da177e4 | 2061 | |
f0b04115 | 2062 | init_waitqueue_head(&md->wait); |
53d5914f | 2063 | INIT_WORK(&md->work, dm_wq_work); |
8b211aac | 2064 | INIT_WORK(&md->requeue_work, dm_wq_requeue_work); |
f0b04115 | 2065 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2066 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 2067 | |
8b211aac | 2068 | md->requeue_list = NULL; |
a666e5c0 MP |
2069 | md->swap_bios = get_swap_bios(); |
2070 | sema_init(&md->swap_bios_semaphore, md->swap_bios); | |
2071 | mutex_init(&md->swap_bios_lock); | |
2072 | ||
1da177e4 LT |
2073 | md->disk->major = _major; |
2074 | md->disk->first_minor = minor; | |
74fe6ba9 | 2075 | md->disk->minors = 1; |
1ebe2e5f | 2076 | md->disk->flags |= GENHD_FL_NO_PART; |
1da177e4 | 2077 | md->disk->fops = &dm_blk_dops; |
1da177e4 LT |
2078 | md->disk->private_data = md; |
2079 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 | 2080 | |
5d2a228b | 2081 | if (IS_ENABLED(CONFIG_FS_DAX)) { |
30c6828a | 2082 | md->dax_dev = alloc_dax(md, &dm_dax_ops); |
d7519392 CH |
2083 | if (IS_ERR(md->dax_dev)) { |
2084 | md->dax_dev = NULL; | |
976431b0 | 2085 | goto bad; |
d7519392 | 2086 | } |
7ac5360c CH |
2087 | set_dax_nocache(md->dax_dev); |
2088 | set_dax_nomc(md->dax_dev); | |
fb08a190 | 2089 | if (dax_add_host(md->dax_dev, md->disk)) |
976431b0 DW |
2090 | goto bad; |
2091 | } | |
f26c5719 | 2092 | |
7e51f257 | 2093 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2094 | |
c7c879ee | 2095 | md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name); |
304f3f6a | 2096 | if (!md->wq) |
0f20972f | 2097 | goto bad; |
304f3f6a | 2098 | |
9f6dc633 MS |
2099 | md->pending_io = alloc_percpu(unsigned long); |
2100 | if (!md->pending_io) | |
2101 | goto bad; | |
2102 | ||
d3aa3e06 JJ |
2103 | r = dm_stats_init(&md->stats); |
2104 | if (r < 0) | |
2105 | goto bad; | |
fd2ed4d2 | 2106 | |
ba61fdd1 | 2107 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2108 | spin_lock(&_minor_lock); |
ba61fdd1 | 2109 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2110 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2111 | |
2112 | BUG_ON(old_md != MINOR_ALLOCED); | |
2113 | ||
1da177e4 LT |
2114 | return md; |
2115 | ||
0f20972f MS |
2116 | bad: |
2117 | cleanup_mapped_device(md); | |
83d5e5b0 | 2118 | bad_io_barrier: |
1da177e4 | 2119 | free_minor(minor); |
6ed7ade8 | 2120 | bad_minor: |
10da4f79 | 2121 | module_put(THIS_MODULE); |
6ed7ade8 | 2122 | bad_module_get: |
856eb091 | 2123 | kvfree(md); |
1da177e4 LT |
2124 | return NULL; |
2125 | } | |
2126 | ||
ae9da83f JN |
2127 | static void unlock_fs(struct mapped_device *md); |
2128 | ||
1da177e4 LT |
2129 | static void free_dev(struct mapped_device *md) |
2130 | { | |
f331c029 | 2131 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2132 | |
32a926da | 2133 | unlock_fs(md); |
2eb6e1e3 | 2134 | |
0f20972f | 2135 | cleanup_mapped_device(md); |
63a4f065 | 2136 | |
992ec6a9 | 2137 | WARN_ON_ONCE(!list_empty(&md->table_devices)); |
63a4f065 | 2138 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2139 | free_minor(minor); |
2140 | ||
10da4f79 | 2141 | module_put(THIS_MODULE); |
856eb091 | 2142 | kvfree(md); |
1da177e4 LT |
2143 | } |
2144 | ||
2145 | /* | |
2146 | * Bind a table to the device. | |
2147 | */ | |
2148 | static void event_callback(void *context) | |
2149 | { | |
7a8c3d3b MA |
2150 | unsigned long flags; |
2151 | LIST_HEAD(uevents); | |
d695e441 | 2152 | struct mapped_device *md = context; |
1da177e4 | 2153 | |
7a8c3d3b MA |
2154 | spin_lock_irqsave(&md->uevent_lock, flags); |
2155 | list_splice_init(&md->uevent_list, &uevents); | |
2156 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2157 | ||
ed9e1982 | 2158 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2159 | |
1da177e4 LT |
2160 | atomic_inc(&md->event_nr); |
2161 | wake_up(&md->eventq); | |
62e08243 | 2162 | dm_issue_global_event(); |
1da177e4 LT |
2163 | } |
2164 | ||
042d2a9b AK |
2165 | /* |
2166 | * Returns old map, which caller must destroy. | |
2167 | */ | |
2168 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2169 | struct queue_limits *limits) | |
1da177e4 | 2170 | { |
042d2a9b | 2171 | struct dm_table *old_map; |
1da177e4 | 2172 | sector_t size; |
2a2a4c51 | 2173 | int ret; |
1da177e4 | 2174 | |
5a8f1f80 BVA |
2175 | lockdep_assert_held(&md->suspend_lock); |
2176 | ||
1da177e4 | 2177 | size = dm_table_get_size(t); |
3ac51e74 DW |
2178 | |
2179 | /* | |
2180 | * Wipe any geometry if the size of the table changed. | |
2181 | */ | |
fd2ed4d2 | 2182 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2183 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2184 | ||
7533afa1 | 2185 | set_capacity(md->disk, size); |
d5816876 | 2186 | |
2ca3310e AK |
2187 | dm_table_event_callback(t, event_callback, md); |
2188 | ||
f5b4aee1 | 2189 | if (dm_table_request_based(t)) { |
16f12266 | 2190 | /* |
9c37de29 MS |
2191 | * Leverage the fact that request-based DM targets are |
2192 | * immutable singletons - used to optimize dm_mq_queue_rq. | |
16f12266 MS |
2193 | */ |
2194 | md->immutable_target = dm_table_get_immutable_target(t); | |
e6ee8c0b | 2195 | |
29dec90a CH |
2196 | /* |
2197 | * There is no need to reload with request-based dm because the | |
2198 | * size of front_pad doesn't change. | |
2199 | * | |
2200 | * Note for future: If you are to reload bioset, prep-ed | |
2201 | * requests in the queue may refer to bio from the old bioset, | |
2202 | * so you must walk through the queue to unprep. | |
2203 | */ | |
2204 | if (!md->mempools) { | |
2205 | md->mempools = t->mempools; | |
2206 | t->mempools = NULL; | |
2207 | } | |
2208 | } else { | |
2209 | /* | |
2210 | * The md may already have mempools that need changing. | |
2211 | * If so, reload bioset because front_pad may have changed | |
2212 | * because a different table was loaded. | |
2213 | */ | |
2214 | dm_free_md_mempools(md->mempools); | |
2215 | md->mempools = t->mempools; | |
2216 | t->mempools = NULL; | |
2a2a4c51 | 2217 | } |
e6ee8c0b | 2218 | |
f5b4aee1 | 2219 | ret = dm_table_set_restrictions(t, md->queue, limits); |
bb37d772 DLM |
2220 | if (ret) { |
2221 | old_map = ERR_PTR(ret); | |
2222 | goto out; | |
2223 | } | |
2224 | ||
a12f5d48 | 2225 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2226 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2227 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2228 | ||
41abc4e1 HR |
2229 | if (old_map) |
2230 | dm_sync_table(md); | |
2a2a4c51 | 2231 | out: |
042d2a9b | 2232 | return old_map; |
1da177e4 LT |
2233 | } |
2234 | ||
a7940155 AK |
2235 | /* |
2236 | * Returns unbound table for the caller to free. | |
2237 | */ | |
2238 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2239 | { |
a12f5d48 | 2240 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2241 | |
2242 | if (!map) | |
a7940155 | 2243 | return NULL; |
1da177e4 LT |
2244 | |
2245 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2246 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2247 | dm_sync_table(md); |
a7940155 AK |
2248 | |
2249 | return map; | |
1da177e4 LT |
2250 | } |
2251 | ||
2252 | /* | |
2253 | * Constructor for a new device. | |
2254 | */ | |
2b06cfff | 2255 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2256 | { |
2257 | struct mapped_device *md; | |
2258 | ||
2b06cfff | 2259 | md = alloc_dev(minor); |
1da177e4 LT |
2260 | if (!md) |
2261 | return -ENXIO; | |
2262 | ||
91ccbbac TS |
2263 | dm_ima_reset_data(md); |
2264 | ||
1da177e4 LT |
2265 | *result = md; |
2266 | return 0; | |
2267 | } | |
2268 | ||
a5664dad MS |
2269 | /* |
2270 | * Functions to manage md->type. | |
2271 | * All are required to hold md->type_lock. | |
2272 | */ | |
2273 | void dm_lock_md_type(struct mapped_device *md) | |
2274 | { | |
2275 | mutex_lock(&md->type_lock); | |
2276 | } | |
2277 | ||
2278 | void dm_unlock_md_type(struct mapped_device *md) | |
2279 | { | |
2280 | mutex_unlock(&md->type_lock); | |
2281 | } | |
2282 | ||
7e0d574f | 2283 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2284 | { |
00c4fc3b | 2285 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2286 | md->type = type; |
2287 | } | |
2288 | ||
7e0d574f | 2289 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2290 | { |
2291 | return md->type; | |
2292 | } | |
2293 | ||
36a0456f AK |
2294 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2295 | { | |
2296 | return md->immutable_target_type; | |
2297 | } | |
2298 | ||
f84cb8a4 MS |
2299 | /* |
2300 | * The queue_limits are only valid as long as you have a reference | |
2301 | * count on 'md'. | |
2302 | */ | |
2303 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2304 | { | |
2305 | BUG_ON(!atomic_read(&md->holders)); | |
2306 | return &md->queue->limits; | |
2307 | } | |
2308 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2309 | ||
4a0b4ddf MS |
2310 | /* |
2311 | * Setup the DM device's queue based on md's type | |
2312 | */ | |
591ddcfc | 2313 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2314 | { |
ba305859 | 2315 | enum dm_queue_mode type = dm_table_get_type(t); |
c100ec49 | 2316 | struct queue_limits limits; |
1a581b72 | 2317 | struct table_device *td; |
ba305859 | 2318 | int r; |
bfebd1cd | 2319 | |
545ed20e | 2320 | switch (type) { |
bfebd1cd | 2321 | case DM_TYPE_REQUEST_BASED: |
681cc5e8 | 2322 | md->disk->fops = &dm_rq_blk_dops; |
e83068a5 | 2323 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2324 | if (r) { |
681cc5e8 | 2325 | DMERR("Cannot initialize queue for request-based dm mapped device"); |
bfebd1cd MS |
2326 | return r; |
2327 | } | |
2328 | break; | |
2329 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2330 | case DM_TYPE_DAX_BIO_BASED: |
bfebd1cd | 2331 | break; |
7e0d574f BVA |
2332 | case DM_TYPE_NONE: |
2333 | WARN_ON_ONCE(true); | |
2334 | break; | |
4a0b4ddf MS |
2335 | } |
2336 | ||
c100ec49 MS |
2337 | r = dm_calculate_queue_limits(t, &limits); |
2338 | if (r) { | |
2339 | DMERR("Cannot calculate initial queue limits"); | |
2340 | return r; | |
2341 | } | |
bb37d772 DLM |
2342 | r = dm_table_set_restrictions(t, md->queue, &limits); |
2343 | if (r) | |
2344 | return r; | |
2345 | ||
d563792c YK |
2346 | /* |
2347 | * Hold lock to make sure add_disk() and del_gendisk() won't concurrent | |
2348 | * with open_table_device() and close_table_device(). | |
2349 | */ | |
2350 | mutex_lock(&md->table_devices_lock); | |
e7089f65 | 2351 | r = add_disk(md->disk); |
d563792c | 2352 | mutex_unlock(&md->table_devices_lock); |
e7089f65 LC |
2353 | if (r) |
2354 | return r; | |
c100ec49 | 2355 | |
1a581b72 CH |
2356 | /* |
2357 | * Register the holder relationship for devices added before the disk | |
2358 | * was live. | |
2359 | */ | |
2360 | list_for_each_entry(td, &md->table_devices, list) { | |
2361 | r = bd_link_disk_holder(td->dm_dev.bdev, md->disk); | |
2362 | if (r) | |
2363 | goto out_undo_holders; | |
89f871af | 2364 | } |
d563792c | 2365 | |
1a581b72 CH |
2366 | r = dm_sysfs_init(md); |
2367 | if (r) | |
2368 | goto out_undo_holders; | |
2369 | ||
89f871af | 2370 | md->type = type; |
4a0b4ddf | 2371 | return 0; |
1a581b72 CH |
2372 | |
2373 | out_undo_holders: | |
2374 | list_for_each_entry_continue_reverse(td, &md->table_devices, list) | |
2375 | bd_unlink_disk_holder(td->dm_dev.bdev, md->disk); | |
2376 | mutex_lock(&md->table_devices_lock); | |
2377 | del_gendisk(md->disk); | |
2378 | mutex_unlock(&md->table_devices_lock); | |
2379 | return r; | |
4a0b4ddf MS |
2380 | } |
2381 | ||
2bec1f4a | 2382 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2383 | { |
2384 | struct mapped_device *md; | |
86a3238c | 2385 | unsigned int minor = MINOR(dev); |
1da177e4 LT |
2386 | |
2387 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2388 | return NULL; | |
2389 | ||
f32c10b0 | 2390 | spin_lock(&_minor_lock); |
1da177e4 LT |
2391 | |
2392 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2393 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2394 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2395 | md = NULL; | |
2396 | goto out; | |
fba9f90e | 2397 | } |
49de5769 | 2398 | dm_get(md); |
fba9f90e | 2399 | out: |
f32c10b0 | 2400 | spin_unlock(&_minor_lock); |
1da177e4 | 2401 | |
637842cf DT |
2402 | return md; |
2403 | } | |
3cf2e4ba | 2404 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2405 | |
9ade92a9 | 2406 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2407 | { |
9ade92a9 | 2408 | return md->interface_ptr; |
1da177e4 LT |
2409 | } |
2410 | ||
2411 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2412 | { | |
2413 | md->interface_ptr = ptr; | |
2414 | } | |
2415 | ||
2416 | void dm_get(struct mapped_device *md) | |
2417 | { | |
2418 | atomic_inc(&md->holders); | |
3f77316d | 2419 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2420 | } |
2421 | ||
09ee96b2 MP |
2422 | int dm_hold(struct mapped_device *md) |
2423 | { | |
2424 | spin_lock(&_minor_lock); | |
2425 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2426 | spin_unlock(&_minor_lock); | |
2427 | return -EBUSY; | |
2428 | } | |
2429 | dm_get(md); | |
2430 | spin_unlock(&_minor_lock); | |
2431 | return 0; | |
2432 | } | |
2433 | EXPORT_SYMBOL_GPL(dm_hold); | |
2434 | ||
72d94861 AK |
2435 | const char *dm_device_name(struct mapped_device *md) |
2436 | { | |
2437 | return md->name; | |
2438 | } | |
2439 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2440 | ||
3f77316d | 2441 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2442 | { |
1134e5ae | 2443 | struct dm_table *map; |
83d5e5b0 | 2444 | int srcu_idx; |
1da177e4 | 2445 | |
3f77316d | 2446 | might_sleep(); |
fba9f90e | 2447 | |
63a4f065 | 2448 | spin_lock(&_minor_lock); |
3f77316d KU |
2449 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2450 | set_bit(DMF_FREEING, &md->flags); | |
2451 | spin_unlock(&_minor_lock); | |
3b785fbc | 2452 | |
7a5428dc | 2453 | blk_mark_disk_dead(md->disk); |
3f77316d | 2454 | |
ab7c7bb6 MP |
2455 | /* |
2456 | * Take suspend_lock so that presuspend and postsuspend methods | |
2457 | * do not race with internal suspend. | |
2458 | */ | |
2459 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2460 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2461 | if (!dm_suspended_md(md)) { |
2462 | dm_table_presuspend_targets(map); | |
adc0daad | 2463 | set_bit(DMF_SUSPENDED, &md->flags); |
5df96f2b | 2464 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
3f77316d | 2465 | dm_table_postsuspend_targets(map); |
1da177e4 | 2466 | } |
238d991f | 2467 | /* dm_put_live_table must be before fsleep, otherwise deadlock is possible */ |
83d5e5b0 | 2468 | dm_put_live_table(md, srcu_idx); |
2a708cff | 2469 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2470 | |
3f77316d KU |
2471 | /* |
2472 | * Rare, but there may be I/O requests still going to complete, | |
2473 | * for example. Wait for all references to disappear. | |
2474 | * No one should increment the reference count of the mapped_device, | |
2475 | * after the mapped_device state becomes DMF_FREEING. | |
2476 | */ | |
2477 | if (wait) | |
2478 | while (atomic_read(&md->holders)) | |
238d991f | 2479 | fsleep(1000); |
3f77316d KU |
2480 | else if (atomic_read(&md->holders)) |
2481 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2482 | dm_device_name(md), atomic_read(&md->holders)); | |
2483 | ||
3f77316d KU |
2484 | dm_table_destroy(__unbind(md)); |
2485 | free_dev(md); | |
2486 | } | |
2487 | ||
2488 | void dm_destroy(struct mapped_device *md) | |
2489 | { | |
2490 | __dm_destroy(md, true); | |
2491 | } | |
2492 | ||
2493 | void dm_destroy_immediate(struct mapped_device *md) | |
2494 | { | |
2495 | __dm_destroy(md, false); | |
2496 | } | |
2497 | ||
2498 | void dm_put(struct mapped_device *md) | |
2499 | { | |
2500 | atomic_dec(&md->holders); | |
1da177e4 | 2501 | } |
79eb885c | 2502 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2503 | |
9f6dc633 | 2504 | static bool dm_in_flight_bios(struct mapped_device *md) |
85067747 ML |
2505 | { |
2506 | int cpu; | |
9f6dc633 | 2507 | unsigned long sum = 0; |
85067747 | 2508 | |
9f6dc633 MS |
2509 | for_each_possible_cpu(cpu) |
2510 | sum += *per_cpu_ptr(md->pending_io, cpu); | |
85067747 ML |
2511 | |
2512 | return sum != 0; | |
2513 | } | |
2514 | ||
2f064a59 | 2515 | static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state) |
46125c1c MB |
2516 | { |
2517 | int r = 0; | |
9f4c3f87 | 2518 | DEFINE_WAIT(wait); |
46125c1c | 2519 | |
85067747 | 2520 | while (true) { |
9f4c3f87 | 2521 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2522 | |
9f6dc633 | 2523 | if (!dm_in_flight_bios(md)) |
46125c1c MB |
2524 | break; |
2525 | ||
e3fabdfd | 2526 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2527 | r = -EINTR; |
2528 | break; | |
2529 | } | |
2530 | ||
2531 | io_schedule(); | |
2532 | } | |
9f4c3f87 | 2533 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2534 | |
9f6dc633 MS |
2535 | smp_rmb(); |
2536 | ||
46125c1c MB |
2537 | return r; |
2538 | } | |
2539 | ||
2f064a59 | 2540 | static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state) |
85067747 ML |
2541 | { |
2542 | int r = 0; | |
2543 | ||
2544 | if (!queue_is_mq(md->queue)) | |
2545 | return dm_wait_for_bios_completion(md, task_state); | |
2546 | ||
2547 | while (true) { | |
2548 | if (!blk_mq_queue_inflight(md->queue)) | |
2549 | break; | |
2550 | ||
2551 | if (signal_pending_state(task_state, current)) { | |
2552 | r = -EINTR; | |
2553 | break; | |
2554 | } | |
2555 | ||
238d991f | 2556 | fsleep(5000); |
85067747 ML |
2557 | } |
2558 | ||
2559 | return r; | |
2560 | } | |
2561 | ||
1da177e4 LT |
2562 | /* |
2563 | * Process the deferred bios | |
2564 | */ | |
ef208587 | 2565 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2566 | { |
0c2915b8 MS |
2567 | struct mapped_device *md = container_of(work, struct mapped_device, work); |
2568 | struct bio *bio; | |
ef208587 | 2569 | |
3b00b203 | 2570 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 | 2571 | spin_lock_irq(&md->deferred_lock); |
0c2915b8 | 2572 | bio = bio_list_pop(&md->deferred); |
df12ee99 AK |
2573 | spin_unlock_irq(&md->deferred_lock); |
2574 | ||
0c2915b8 | 2575 | if (!bio) |
df12ee99 | 2576 | break; |
022c2611 | 2577 | |
0c2915b8 | 2578 | submit_bio_noacct(bio); |
0ca44fce | 2579 | cond_resched(); |
022c2611 | 2580 | } |
1da177e4 LT |
2581 | } |
2582 | ||
9a1fb464 | 2583 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2584 | { |
3b00b203 | 2585 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2586 | smp_mb__after_atomic(); |
53d5914f | 2587 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2588 | } |
2589 | ||
1da177e4 | 2590 | /* |
042d2a9b | 2591 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2592 | */ |
042d2a9b | 2593 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2594 | { |
87eb5b21 | 2595 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2596 | struct queue_limits limits; |
042d2a9b | 2597 | int r; |
1da177e4 | 2598 | |
e61290a4 | 2599 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2600 | |
2601 | /* device must be suspended */ | |
4f186f8b | 2602 | if (!dm_suspended_md(md)) |
93c534ae | 2603 | goto out; |
1da177e4 | 2604 | |
3ae70656 MS |
2605 | /* |
2606 | * If the new table has no data devices, retain the existing limits. | |
2607 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2608 | * then new I/O is queued based on these limits, and then some paths | |
2609 | * reappear. | |
2610 | */ | |
2611 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2612 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2613 | if (live_map) |
2614 | limits = md->queue->limits; | |
83d5e5b0 | 2615 | dm_put_live_table_fast(md); |
3ae70656 MS |
2616 | } |
2617 | ||
87eb5b21 MC |
2618 | if (!live_map) { |
2619 | r = dm_calculate_queue_limits(table, &limits); | |
2620 | if (r) { | |
2621 | map = ERR_PTR(r); | |
2622 | goto out; | |
2623 | } | |
042d2a9b | 2624 | } |
754c5fc7 | 2625 | |
042d2a9b | 2626 | map = __bind(md, table, &limits); |
62e08243 | 2627 | dm_issue_global_event(); |
1da177e4 | 2628 | |
93c534ae | 2629 | out: |
e61290a4 | 2630 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2631 | return map; |
1da177e4 LT |
2632 | } |
2633 | ||
2634 | /* | |
2635 | * Functions to lock and unlock any filesystem running on the | |
2636 | * device. | |
2637 | */ | |
2ca3310e | 2638 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2639 | { |
e39e2e95 | 2640 | int r; |
1da177e4 | 2641 | |
040f04bd | 2642 | WARN_ON(test_bit(DMF_FROZEN, &md->flags)); |
aa8d7c2f | 2643 | |
977115c0 | 2644 | r = freeze_bdev(md->disk->part0); |
040f04bd CH |
2645 | if (!r) |
2646 | set_bit(DMF_FROZEN, &md->flags); | |
2647 | return r; | |
1da177e4 LT |
2648 | } |
2649 | ||
2ca3310e | 2650 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2651 | { |
aa8d7c2f AK |
2652 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2653 | return; | |
977115c0 | 2654 | thaw_bdev(md->disk->part0); |
aa8d7c2f | 2655 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2656 | } |
2657 | ||
2658 | /* | |
b48633f8 BVA |
2659 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2660 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2661 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2662 | * | |
ffcc3936 MS |
2663 | * If __dm_suspend returns 0, the device is completely quiescent |
2664 | * now. There is no request-processing activity. All new requests | |
2665 | * are being added to md->deferred list. | |
cec47e3d | 2666 | */ |
ffcc3936 | 2667 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
86a3238c | 2668 | unsigned int suspend_flags, unsigned int task_state, |
eaf9a736 | 2669 | int dmf_suspended_flag) |
1da177e4 | 2670 | { |
ffcc3936 MS |
2671 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2672 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2673 | int r; | |
1da177e4 | 2674 | |
5a8f1f80 BVA |
2675 | lockdep_assert_held(&md->suspend_lock); |
2676 | ||
2e93ccc1 KU |
2677 | /* |
2678 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2679 | * This flag is cleared before dm_suspend returns. | |
2680 | */ | |
2681 | if (noflush) | |
2682 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 | 2683 | else |
ac75b09f | 2684 | DMDEBUG("%s: suspending with flush", dm_device_name(md)); |
2e93ccc1 | 2685 | |
d67ee213 MS |
2686 | /* |
2687 | * This gets reverted if there's an error later and the targets | |
2688 | * provide the .presuspend_undo hook. | |
2689 | */ | |
cf222b37 AK |
2690 | dm_table_presuspend_targets(map); |
2691 | ||
32a926da | 2692 | /* |
9f518b27 KU |
2693 | * Flush I/O to the device. |
2694 | * Any I/O submitted after lock_fs() may not be flushed. | |
2695 | * noflush takes precedence over do_lockfs. | |
2696 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2697 | */ |
2698 | if (!noflush && do_lockfs) { | |
2699 | r = lock_fs(md); | |
d67ee213 MS |
2700 | if (r) { |
2701 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2702 | return r; |
d67ee213 | 2703 | } |
aa8d7c2f | 2704 | } |
1da177e4 LT |
2705 | |
2706 | /* | |
3b00b203 MP |
2707 | * Here we must make sure that no processes are submitting requests |
2708 | * to target drivers i.e. no one may be executing | |
96c9865c | 2709 | * dm_split_and_process_bio from dm_submit_bio. |
3b00b203 | 2710 | * |
96c9865c | 2711 | * To get all processes out of dm_split_and_process_bio in dm_submit_bio, |
3b00b203 | 2712 | * we take the write lock. To prevent any process from reentering |
96c9865c | 2713 | * dm_split_and_process_bio from dm_submit_bio and quiesce the thread |
0cede372 | 2714 | * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call |
6a8736d1 | 2715 | * flush_workqueue(md->wq). |
1da177e4 | 2716 | */ |
1eb787ec | 2717 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2718 | if (map) |
2719 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2720 | |
d0bcb878 | 2721 | /* |
29e4013d TH |
2722 | * Stop md->queue before flushing md->wq in case request-based |
2723 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2724 | */ |
6a23e05c | 2725 | if (dm_request_based(md)) |
eca7ee6d | 2726 | dm_stop_queue(md->queue); |
cec47e3d | 2727 | |
d0bcb878 KU |
2728 | flush_workqueue(md->wq); |
2729 | ||
1da177e4 | 2730 | /* |
3b00b203 MP |
2731 | * At this point no more requests are entering target request routines. |
2732 | * We call dm_wait_for_completion to wait for all existing requests | |
2733 | * to finish. | |
1da177e4 | 2734 | */ |
b48633f8 | 2735 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2736 | if (!r) |
2737 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2738 | |
6d6f10df | 2739 | if (noflush) |
022c2611 | 2740 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2741 | if (map) |
2742 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2743 | |
1da177e4 | 2744 | /* were we interrupted ? */ |
46125c1c | 2745 | if (r < 0) { |
9a1fb464 | 2746 | dm_queue_flush(md); |
73d410c0 | 2747 | |
cec47e3d | 2748 | if (dm_request_based(md)) |
eca7ee6d | 2749 | dm_start_queue(md->queue); |
cec47e3d | 2750 | |
2ca3310e | 2751 | unlock_fs(md); |
d67ee213 | 2752 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2753 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2754 | } |
1da177e4 | 2755 | |
ffcc3936 MS |
2756 | return r; |
2757 | } | |
2758 | ||
2759 | /* | |
2760 | * We need to be able to change a mapping table under a mounted | |
2761 | * filesystem. For example we might want to move some data in | |
2762 | * the background. Before the table can be swapped with | |
2763 | * dm_bind_table, dm_suspend must be called to flush any in | |
2764 | * flight bios and ensure that any further io gets deferred. | |
2765 | */ | |
2766 | /* | |
2767 | * Suspend mechanism in request-based dm. | |
2768 | * | |
2769 | * 1. Flush all I/Os by lock_fs() if needed. | |
2770 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2771 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2772 | * | |
2773 | * To abort suspend, start the request_queue. | |
2774 | */ | |
86a3238c | 2775 | int dm_suspend(struct mapped_device *md, unsigned int suspend_flags) |
ffcc3936 MS |
2776 | { |
2777 | struct dm_table *map = NULL; | |
2778 | int r = 0; | |
2779 | ||
2780 | retry: | |
2781 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2782 | ||
2783 | if (dm_suspended_md(md)) { | |
2784 | r = -EINVAL; | |
2785 | goto out_unlock; | |
2786 | } | |
2787 | ||
2788 | if (dm_suspended_internally_md(md)) { | |
2789 | /* already internally suspended, wait for internal resume */ | |
2790 | mutex_unlock(&md->suspend_lock); | |
2791 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2792 | if (r) | |
2793 | return r; | |
2794 | goto retry; | |
2795 | } | |
2796 | ||
a12f5d48 | 2797 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2798 | |
eaf9a736 | 2799 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2800 | if (r) |
2801 | goto out_unlock; | |
3b00b203 | 2802 | |
5df96f2b | 2803 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
4d4471cb | 2804 | dm_table_postsuspend_targets(map); |
5df96f2b | 2805 | clear_bit(DMF_POST_SUSPENDING, &md->flags); |
4d4471cb | 2806 | |
d287483d | 2807 | out_unlock: |
e61290a4 | 2808 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2809 | return r; |
1da177e4 LT |
2810 | } |
2811 | ||
ffcc3936 MS |
2812 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2813 | { | |
2814 | if (map) { | |
2815 | int r = dm_table_resume_targets(map); | |
b30f1607 | 2816 | |
ffcc3936 MS |
2817 | if (r) |
2818 | return r; | |
2819 | } | |
2820 | ||
2821 | dm_queue_flush(md); | |
2822 | ||
2823 | /* | |
2824 | * Flushing deferred I/Os must be done after targets are resumed | |
2825 | * so that mapping of targets can work correctly. | |
2826 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2827 | */ | |
2828 | if (dm_request_based(md)) | |
eca7ee6d | 2829 | dm_start_queue(md->queue); |
ffcc3936 MS |
2830 | |
2831 | unlock_fs(md); | |
2832 | ||
2833 | return 0; | |
2834 | } | |
2835 | ||
1da177e4 LT |
2836 | int dm_resume(struct mapped_device *md) |
2837 | { | |
8dc23658 | 2838 | int r; |
cf222b37 | 2839 | struct dm_table *map = NULL; |
1da177e4 | 2840 | |
ffcc3936 | 2841 | retry: |
8dc23658 | 2842 | r = -EINVAL; |
ffcc3936 MS |
2843 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2844 | ||
4f186f8b | 2845 | if (!dm_suspended_md(md)) |
cf222b37 | 2846 | goto out; |
cf222b37 | 2847 | |
ffcc3936 MS |
2848 | if (dm_suspended_internally_md(md)) { |
2849 | /* already internally suspended, wait for internal resume */ | |
2850 | mutex_unlock(&md->suspend_lock); | |
2851 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2852 | if (r) | |
2853 | return r; | |
2854 | goto retry; | |
2855 | } | |
2856 | ||
a12f5d48 | 2857 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2858 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2859 | goto out; |
1da177e4 | 2860 | |
ffcc3936 | 2861 | r = __dm_resume(md, map); |
8757b776 MB |
2862 | if (r) |
2863 | goto out; | |
2ca3310e | 2864 | |
2ca3310e | 2865 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2866 | out: |
e61290a4 | 2867 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2868 | |
cf222b37 | 2869 | return r; |
1da177e4 LT |
2870 | } |
2871 | ||
fd2ed4d2 MP |
2872 | /* |
2873 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2874 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2875 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2876 | */ |
2877 | ||
86a3238c | 2878 | static void __dm_internal_suspend(struct mapped_device *md, unsigned int suspend_flags) |
fd2ed4d2 | 2879 | { |
ffcc3936 MS |
2880 | struct dm_table *map = NULL; |
2881 | ||
1ea0654e BVA |
2882 | lockdep_assert_held(&md->suspend_lock); |
2883 | ||
96b26c8c | 2884 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2885 | return; /* nested internal suspend */ |
2886 | ||
2887 | if (dm_suspended_md(md)) { | |
2888 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2889 | return; /* nest suspend */ | |
2890 | } | |
2891 | ||
a12f5d48 | 2892 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2893 | |
2894 | /* | |
2895 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2896 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2897 | * would require changing .presuspend to return an error -- avoid this | |
2898 | * until there is a need for more elaborate variants of internal suspend. | |
2899 | */ | |
eaf9a736 MS |
2900 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2901 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 | 2902 | |
5df96f2b | 2903 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
ffcc3936 | 2904 | dm_table_postsuspend_targets(map); |
5df96f2b | 2905 | clear_bit(DMF_POST_SUSPENDING, &md->flags); |
ffcc3936 MS |
2906 | } |
2907 | ||
2908 | static void __dm_internal_resume(struct mapped_device *md) | |
2909 | { | |
96b26c8c MP |
2910 | BUG_ON(!md->internal_suspend_count); |
2911 | ||
2912 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2913 | return; /* resume from nested internal suspend */ |
2914 | ||
fd2ed4d2 | 2915 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2916 | goto done; /* resume from nested suspend */ |
2917 | ||
2918 | /* | |
2919 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2920 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2921 | */ | |
2922 | (void) __dm_resume(md, NULL); | |
2923 | ||
2924 | done: | |
2925 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2926 | smp_mb__after_atomic(); | |
2927 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2928 | } | |
2929 | ||
2930 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2931 | { | |
2932 | mutex_lock(&md->suspend_lock); | |
2933 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2934 | mutex_unlock(&md->suspend_lock); | |
2935 | } | |
2936 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2937 | ||
2938 | void dm_internal_resume(struct mapped_device *md) | |
2939 | { | |
2940 | mutex_lock(&md->suspend_lock); | |
2941 | __dm_internal_resume(md); | |
2942 | mutex_unlock(&md->suspend_lock); | |
2943 | } | |
2944 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2945 | ||
2946 | /* | |
2947 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2948 | * which prevents interaction with userspace-driven suspend. | |
2949 | */ | |
2950 | ||
2951 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2952 | { | |
2953 | mutex_lock(&md->suspend_lock); | |
2954 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2955 | return; |
2956 | ||
2957 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2958 | synchronize_srcu(&md->io_barrier); | |
2959 | flush_workqueue(md->wq); | |
2960 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2961 | } | |
b735fede | 2962 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2963 | |
ffcc3936 | 2964 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2965 | { |
ffcc3936 | 2966 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2967 | goto done; |
2968 | ||
2969 | dm_queue_flush(md); | |
2970 | ||
2971 | done: | |
2972 | mutex_unlock(&md->suspend_lock); | |
2973 | } | |
b735fede | 2974 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2975 | |
a4a82ce3 HM |
2976 | /* |
2977 | *--------------------------------------------------------------- | |
1da177e4 | 2978 | * Event notification. |
a4a82ce3 HM |
2979 | *--------------------------------------------------------------- |
2980 | */ | |
3abf85b5 | 2981 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
86a3238c | 2982 | unsigned int cookie, bool need_resize_uevent) |
69267a30 | 2983 | { |
6958c1c6 | 2984 | int r; |
86a3238c | 2985 | unsigned int noio_flag; |
60935eb2 | 2986 | char udev_cookie[DM_COOKIE_LENGTH]; |
7533afa1 MP |
2987 | char *envp[3] = { NULL, NULL, NULL }; |
2988 | char **envpp = envp; | |
2989 | if (cookie) { | |
60935eb2 MB |
2990 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", |
2991 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
7533afa1 | 2992 | *envpp++ = udev_cookie; |
60935eb2 | 2993 | } |
7533afa1 MP |
2994 | if (need_resize_uevent) { |
2995 | *envpp++ = "RESIZE=1"; | |
2996 | } | |
2997 | ||
2998 | noio_flag = memalloc_noio_save(); | |
2999 | ||
3000 | r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp); | |
6958c1c6 MP |
3001 | |
3002 | memalloc_noio_restore(noio_flag); | |
3003 | ||
3004 | return r; | |
69267a30 AK |
3005 | } |
3006 | ||
7a8c3d3b MA |
3007 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3008 | { | |
3009 | return atomic_add_return(1, &md->uevent_seq); | |
3010 | } | |
3011 | ||
1da177e4 LT |
3012 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3013 | { | |
3014 | return atomic_read(&md->event_nr); | |
3015 | } | |
3016 | ||
3017 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3018 | { | |
3019 | return wait_event_interruptible(md->eventq, | |
3020 | (event_nr != atomic_read(&md->event_nr))); | |
3021 | } | |
3022 | ||
7a8c3d3b MA |
3023 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3024 | { | |
3025 | unsigned long flags; | |
3026 | ||
3027 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3028 | list_add(elist, &md->uevent_list); | |
3029 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3030 | } | |
3031 | ||
1da177e4 LT |
3032 | /* |
3033 | * The gendisk is only valid as long as you have a reference | |
3034 | * count on 'md'. | |
3035 | */ | |
3036 | struct gendisk *dm_disk(struct mapped_device *md) | |
3037 | { | |
3038 | return md->disk; | |
3039 | } | |
65ff5b7d | 3040 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3041 | |
784aae73 MB |
3042 | struct kobject *dm_kobject(struct mapped_device *md) |
3043 | { | |
2995fa78 | 3044 | return &md->kobj_holder.kobj; |
784aae73 MB |
3045 | } |
3046 | ||
784aae73 MB |
3047 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3048 | { | |
3049 | struct mapped_device *md; | |
3050 | ||
2995fa78 | 3051 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3052 | |
b9a41d21 HT |
3053 | spin_lock(&_minor_lock); |
3054 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
3055 | md = NULL; | |
3056 | goto out; | |
3057 | } | |
784aae73 | 3058 | dm_get(md); |
b9a41d21 HT |
3059 | out: |
3060 | spin_unlock(&_minor_lock); | |
3061 | ||
784aae73 MB |
3062 | return md; |
3063 | } | |
3064 | ||
4f186f8b | 3065 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3066 | { |
3067 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3068 | } | |
3069 | ||
5df96f2b MP |
3070 | static int dm_post_suspending_md(struct mapped_device *md) |
3071 | { | |
3072 | return test_bit(DMF_POST_SUSPENDING, &md->flags); | |
3073 | } | |
3074 | ||
ffcc3936 MS |
3075 | int dm_suspended_internally_md(struct mapped_device *md) |
3076 | { | |
3077 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3078 | } | |
3079 | ||
2c140a24 MP |
3080 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3081 | { | |
3082 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3083 | } | |
3084 | ||
64dbce58 KU |
3085 | int dm_suspended(struct dm_target *ti) |
3086 | { | |
33bd6f06 | 3087 | return dm_suspended_md(ti->table->md); |
64dbce58 KU |
3088 | } |
3089 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3090 | ||
5df96f2b MP |
3091 | int dm_post_suspending(struct dm_target *ti) |
3092 | { | |
33bd6f06 | 3093 | return dm_post_suspending_md(ti->table->md); |
5df96f2b MP |
3094 | } |
3095 | EXPORT_SYMBOL_GPL(dm_post_suspending); | |
3096 | ||
2e93ccc1 KU |
3097 | int dm_noflush_suspending(struct dm_target *ti) |
3098 | { | |
33bd6f06 | 3099 | return __noflush_suspending(ti->table->md); |
2e93ccc1 KU |
3100 | } |
3101 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3102 | ||
e6ee8c0b KU |
3103 | void dm_free_md_mempools(struct dm_md_mempools *pools) |
3104 | { | |
3105 | if (!pools) | |
3106 | return; | |
3107 | ||
6f1c819c KO |
3108 | bioset_exit(&pools->bs); |
3109 | bioset_exit(&pools->io_bs); | |
e6ee8c0b KU |
3110 | |
3111 | kfree(pools); | |
3112 | } | |
3113 | ||
9c72bad1 CH |
3114 | struct dm_pr { |
3115 | u64 old_key; | |
3116 | u64 new_key; | |
3117 | u32 flags; | |
c6adada5 | 3118 | bool abort; |
9c72bad1 | 3119 | bool fail_early; |
8dd87f3c | 3120 | int ret; |
70151087 | 3121 | enum pr_type type; |
9c72bad1 CH |
3122 | }; |
3123 | ||
3124 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
8dd87f3c | 3125 | struct dm_pr *pr) |
71cdb697 CH |
3126 | { |
3127 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
3128 | struct dm_table *table; |
3129 | struct dm_target *ti; | |
3130 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 3131 | |
9c72bad1 CH |
3132 | table = dm_get_live_table(md, &srcu_idx); |
3133 | if (!table || !dm_table_get_size(table)) | |
3134 | goto out; | |
71cdb697 | 3135 | |
9c72bad1 | 3136 | /* We only support devices that have a single target */ |
2aec377a | 3137 | if (table->num_targets != 1) |
9c72bad1 CH |
3138 | goto out; |
3139 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 3140 | |
e120a5f1 MS |
3141 | if (dm_suspended_md(md)) { |
3142 | ret = -EAGAIN; | |
3143 | goto out; | |
3144 | } | |
3145 | ||
9c72bad1 CH |
3146 | ret = -EINVAL; |
3147 | if (!ti->type->iterate_devices) | |
3148 | goto out; | |
3149 | ||
8dd87f3c MC |
3150 | ti->type->iterate_devices(ti, fn, pr); |
3151 | ret = 0; | |
9c72bad1 CH |
3152 | out: |
3153 | dm_put_live_table(md, srcu_idx); | |
3154 | return ret; | |
3155 | } | |
3156 | ||
3157 | /* | |
3158 | * For register / unregister we need to manually call out to every path. | |
3159 | */ | |
3160 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
3161 | sector_t start, sector_t len, void *data) | |
3162 | { | |
3163 | struct dm_pr *pr = data; | |
3164 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
8dd87f3c MC |
3165 | int ret; |
3166 | ||
3167 | if (!ops || !ops->pr_register) { | |
3168 | pr->ret = -EOPNOTSUPP; | |
3169 | return -1; | |
3170 | } | |
9c72bad1 | 3171 | |
8dd87f3c MC |
3172 | ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); |
3173 | if (!ret) | |
3174 | return 0; | |
3175 | ||
3176 | if (!pr->ret) | |
3177 | pr->ret = ret; | |
9c72bad1 | 3178 | |
8dd87f3c MC |
3179 | if (pr->fail_early) |
3180 | return -1; | |
3181 | ||
3182 | return 0; | |
9c72bad1 CH |
3183 | } |
3184 | ||
3185 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
3186 | u32 flags) | |
3187 | { | |
3188 | struct dm_pr pr = { | |
3189 | .old_key = old_key, | |
3190 | .new_key = new_key, | |
3191 | .flags = flags, | |
3192 | .fail_early = true, | |
8dd87f3c | 3193 | .ret = 0, |
9c72bad1 CH |
3194 | }; |
3195 | int ret; | |
3196 | ||
3197 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
8dd87f3c MC |
3198 | if (ret) { |
3199 | /* Didn't even get to register a path */ | |
3200 | return ret; | |
9c72bad1 CH |
3201 | } |
3202 | ||
8dd87f3c MC |
3203 | if (!pr.ret) |
3204 | return 0; | |
3205 | ret = pr.ret; | |
3206 | ||
3207 | if (!new_key) | |
3208 | return ret; | |
3209 | ||
3210 | /* unregister all paths if we failed to register any path */ | |
3211 | pr.old_key = new_key; | |
3212 | pr.new_key = 0; | |
3213 | pr.flags = 0; | |
3214 | pr.fail_early = false; | |
3215 | (void) dm_call_pr(bdev, __dm_pr_register, &pr); | |
9c72bad1 | 3216 | return ret; |
71cdb697 CH |
3217 | } |
3218 | ||
70151087 MC |
3219 | |
3220 | static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev, | |
3221 | sector_t start, sector_t len, void *data) | |
3222 | { | |
3223 | struct dm_pr *pr = data; | |
3224 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3225 | ||
3226 | if (!ops || !ops->pr_reserve) { | |
3227 | pr->ret = -EOPNOTSUPP; | |
3228 | return -1; | |
3229 | } | |
3230 | ||
3231 | pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags); | |
3232 | if (!pr->ret) | |
3233 | return -1; | |
3234 | ||
3235 | return 0; | |
3236 | } | |
3237 | ||
71cdb697 | 3238 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, |
956a4025 | 3239 | u32 flags) |
71cdb697 | 3240 | { |
70151087 MC |
3241 | struct dm_pr pr = { |
3242 | .old_key = key, | |
3243 | .flags = flags, | |
3244 | .type = type, | |
3245 | .fail_early = false, | |
3246 | .ret = 0, | |
3247 | }; | |
3248 | int ret; | |
71cdb697 | 3249 | |
70151087 MC |
3250 | ret = dm_call_pr(bdev, __dm_pr_reserve, &pr); |
3251 | if (ret) | |
3252 | return ret; | |
71cdb697 | 3253 | |
70151087 | 3254 | return pr.ret; |
71cdb697 CH |
3255 | } |
3256 | ||
08a3c338 MC |
3257 | /* |
3258 | * If there is a non-All Registrants type of reservation, the release must be | |
3259 | * sent down the holding path. For the cases where there is no reservation or | |
3260 | * the path is not the holder the device will also return success, so we must | |
3261 | * try each path to make sure we got the correct path. | |
3262 | */ | |
3263 | static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev, | |
3264 | sector_t start, sector_t len, void *data) | |
3265 | { | |
3266 | struct dm_pr *pr = data; | |
3267 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3268 | ||
3269 | if (!ops || !ops->pr_release) { | |
3270 | pr->ret = -EOPNOTSUPP; | |
3271 | return -1; | |
3272 | } | |
3273 | ||
3274 | pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type); | |
3275 | if (pr->ret) | |
3276 | return -1; | |
3277 | ||
3278 | return 0; | |
71cdb697 CH |
3279 | } |
3280 | ||
3281 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3282 | { | |
08a3c338 MC |
3283 | struct dm_pr pr = { |
3284 | .old_key = key, | |
3285 | .type = type, | |
3286 | .fail_early = false, | |
3287 | }; | |
3288 | int ret; | |
71cdb697 | 3289 | |
08a3c338 MC |
3290 | ret = dm_call_pr(bdev, __dm_pr_release, &pr); |
3291 | if (ret) | |
3292 | return ret; | |
71cdb697 | 3293 | |
08a3c338 | 3294 | return pr.ret; |
71cdb697 CH |
3295 | } |
3296 | ||
c6adada5 MC |
3297 | static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev, |
3298 | sector_t start, sector_t len, void *data) | |
3299 | { | |
3300 | struct dm_pr *pr = data; | |
3301 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3302 | ||
3303 | if (!ops || !ops->pr_preempt) { | |
3304 | pr->ret = -EOPNOTSUPP; | |
3305 | return -1; | |
3306 | } | |
3307 | ||
3308 | pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type, | |
3309 | pr->abort); | |
3310 | if (!pr->ret) | |
3311 | return -1; | |
3312 | ||
3313 | return 0; | |
71cdb697 CH |
3314 | } |
3315 | ||
3316 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3317 | enum pr_type type, bool abort) |
71cdb697 | 3318 | { |
c6adada5 MC |
3319 | struct dm_pr pr = { |
3320 | .new_key = new_key, | |
3321 | .old_key = old_key, | |
3322 | .type = type, | |
3323 | .fail_early = false, | |
3324 | }; | |
3325 | int ret; | |
71cdb697 | 3326 | |
c6adada5 MC |
3327 | ret = dm_call_pr(bdev, __dm_pr_preempt, &pr); |
3328 | if (ret) | |
3329 | return ret; | |
71cdb697 | 3330 | |
c6adada5 | 3331 | return pr.ret; |
71cdb697 CH |
3332 | } |
3333 | ||
3334 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3335 | { | |
3336 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3337 | const struct pr_ops *ops; | |
971888c4 | 3338 | int r, srcu_idx; |
71cdb697 | 3339 | |
5bd5e8d8 | 3340 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3341 | if (r < 0) |
971888c4 | 3342 | goto out; |
71cdb697 CH |
3343 | |
3344 | ops = bdev->bd_disk->fops->pr_ops; | |
3345 | if (ops && ops->pr_clear) | |
3346 | r = ops->pr_clear(bdev, key); | |
3347 | else | |
3348 | r = -EOPNOTSUPP; | |
971888c4 MS |
3349 | out: |
3350 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3351 | return r; |
3352 | } | |
3353 | ||
3354 | static const struct pr_ops dm_pr_ops = { | |
3355 | .pr_register = dm_pr_register, | |
3356 | .pr_reserve = dm_pr_reserve, | |
3357 | .pr_release = dm_pr_release, | |
3358 | .pr_preempt = dm_pr_preempt, | |
3359 | .pr_clear = dm_pr_clear, | |
3360 | }; | |
3361 | ||
83d5cde4 | 3362 | static const struct block_device_operations dm_blk_dops = { |
c62b37d9 | 3363 | .submit_bio = dm_submit_bio, |
b99fdcdc | 3364 | .poll_bio = dm_poll_bio, |
1da177e4 LT |
3365 | .open = dm_blk_open, |
3366 | .release = dm_blk_close, | |
aa129a22 | 3367 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3368 | .getgeo = dm_blk_getgeo, |
e76239a3 | 3369 | .report_zones = dm_blk_report_zones, |
71cdb697 | 3370 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3371 | .owner = THIS_MODULE |
3372 | }; | |
3373 | ||
681cc5e8 MS |
3374 | static const struct block_device_operations dm_rq_blk_dops = { |
3375 | .open = dm_blk_open, | |
3376 | .release = dm_blk_close, | |
3377 | .ioctl = dm_blk_ioctl, | |
3378 | .getgeo = dm_blk_getgeo, | |
3379 | .pr_ops = &dm_pr_ops, | |
3380 | .owner = THIS_MODULE | |
3381 | }; | |
3382 | ||
f26c5719 DW |
3383 | static const struct dax_operations dm_dax_ops = { |
3384 | .direct_access = dm_dax_direct_access, | |
cdf6cdcd | 3385 | .zero_page_range = dm_dax_zero_page_range, |
047218ec | 3386 | .recovery_write = dm_dax_recovery_write, |
f26c5719 DW |
3387 | }; |
3388 | ||
1da177e4 LT |
3389 | /* |
3390 | * module hooks | |
3391 | */ | |
3392 | module_init(dm_init); | |
3393 | module_exit(dm_exit); | |
3394 | ||
3395 | module_param(major, uint, 0); | |
3396 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3397 | |
6a808034 | 3398 | module_param(reserved_bio_based_ios, uint, 0644); |
e8603136 MS |
3399 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); |
3400 | ||
6a808034 | 3401 | module_param(dm_numa_node, int, 0644); |
115485e8 MS |
3402 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); |
3403 | ||
6a808034 | 3404 | module_param(swap_bios, int, 0644); |
a666e5c0 MP |
3405 | MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs"); |
3406 | ||
1da177e4 LT |
3407 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3408 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3409 | MODULE_LICENSE("GPL"); |