]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
4cc96131 MS |
8 | #include "dm-core.h" |
9 | #include "dm-rq.h" | |
51e5b2bd | 10 | #include "dm-uevent.h" |
1da177e4 LT |
11 | |
12 | #include <linux/init.h> | |
13 | #include <linux/module.h> | |
48c9c27b | 14 | #include <linux/mutex.h> |
174cd4b1 | 15 | #include <linux/sched/signal.h> |
1da177e4 LT |
16 | #include <linux/blkpg.h> |
17 | #include <linux/bio.h> | |
1da177e4 | 18 | #include <linux/mempool.h> |
f26c5719 | 19 | #include <linux/dax.h> |
1da177e4 LT |
20 | #include <linux/slab.h> |
21 | #include <linux/idr.h> | |
7e026c8c | 22 | #include <linux/uio.h> |
3ac51e74 | 23 | #include <linux/hdreg.h> |
3f77316d | 24 | #include <linux/delay.h> |
ffcc3936 | 25 | #include <linux/wait.h> |
71cdb697 | 26 | #include <linux/pr.h> |
b0b4d7c6 | 27 | #include <linux/refcount.h> |
55782138 | 28 | |
72d94861 AK |
29 | #define DM_MSG_PREFIX "core" |
30 | ||
60935eb2 MB |
31 | /* |
32 | * Cookies are numeric values sent with CHANGE and REMOVE | |
33 | * uevents while resuming, removing or renaming the device. | |
34 | */ | |
35 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
36 | #define DM_COOKIE_LENGTH 24 | |
37 | ||
1da177e4 LT |
38 | static const char *_name = DM_NAME; |
39 | ||
40 | static unsigned int major = 0; | |
41 | static unsigned int _major = 0; | |
42 | ||
d15b774c AK |
43 | static DEFINE_IDR(_minor_idr); |
44 | ||
f32c10b0 | 45 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
46 | |
47 | static void do_deferred_remove(struct work_struct *w); | |
48 | ||
49 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
50 | ||
acfe0ad7 MP |
51 | static struct workqueue_struct *deferred_remove_workqueue; |
52 | ||
93e6442c MP |
53 | atomic_t dm_global_event_nr = ATOMIC_INIT(0); |
54 | DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq); | |
55 | ||
62e08243 MP |
56 | void dm_issue_global_event(void) |
57 | { | |
58 | atomic_inc(&dm_global_event_nr); | |
59 | wake_up(&dm_global_eventq); | |
60 | } | |
61 | ||
1da177e4 | 62 | /* |
745dc570 | 63 | * One of these is allocated per original bio. |
1da177e4 LT |
64 | */ |
65 | struct dm_io { | |
66 | struct mapped_device *md; | |
4e4cbee9 | 67 | blk_status_t status; |
1da177e4 | 68 | atomic_t io_count; |
745dc570 | 69 | struct bio *orig_bio; |
3eaf840e | 70 | unsigned long start_time; |
f88fb981 | 71 | spinlock_t endio_lock; |
fd2ed4d2 | 72 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
73 | }; |
74 | ||
ba61fdd1 JM |
75 | #define MINOR_ALLOCED ((void *)-1) |
76 | ||
1da177e4 LT |
77 | /* |
78 | * Bits for the md->flags field. | |
79 | */ | |
1eb787ec | 80 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 81 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 82 | #define DMF_FROZEN 2 |
fba9f90e | 83 | #define DMF_FREEING 3 |
5c6bd75d | 84 | #define DMF_DELETING 4 |
2e93ccc1 | 85 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
86 | #define DMF_DEFERRED_REMOVE 6 |
87 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 88 | |
115485e8 | 89 | #define DM_NUMA_NODE NUMA_NO_NODE |
115485e8 | 90 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 91 | |
e6ee8c0b KU |
92 | /* |
93 | * For mempools pre-allocation at the table loading time. | |
94 | */ | |
95 | struct dm_md_mempools { | |
96 | mempool_t *io_pool; | |
e6ee8c0b KU |
97 | struct bio_set *bs; |
98 | }; | |
99 | ||
86f1152b BM |
100 | struct table_device { |
101 | struct list_head list; | |
b0b4d7c6 | 102 | refcount_t count; |
86f1152b BM |
103 | struct dm_dev dm_dev; |
104 | }; | |
105 | ||
e18b890b | 106 | static struct kmem_cache *_io_cache; |
8fbf26ad | 107 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 108 | static struct kmem_cache *_rq_cache; |
94818742 | 109 | |
e8603136 MS |
110 | /* |
111 | * Bio-based DM's mempools' reserved IOs set by the user. | |
112 | */ | |
4cc96131 | 113 | #define RESERVED_BIO_BASED_IOS 16 |
e8603136 MS |
114 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
115 | ||
115485e8 MS |
116 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
117 | { | |
6aa7de05 | 118 | int param = READ_ONCE(*module_param); |
115485e8 MS |
119 | int modified_param = 0; |
120 | bool modified = true; | |
121 | ||
122 | if (param < min) | |
123 | modified_param = min; | |
124 | else if (param > max) | |
125 | modified_param = max; | |
126 | else | |
127 | modified = false; | |
128 | ||
129 | if (modified) { | |
130 | (void)cmpxchg(module_param, param, modified_param); | |
131 | param = modified_param; | |
132 | } | |
133 | ||
134 | return param; | |
135 | } | |
136 | ||
4cc96131 MS |
137 | unsigned __dm_get_module_param(unsigned *module_param, |
138 | unsigned def, unsigned max) | |
f4790826 | 139 | { |
6aa7de05 | 140 | unsigned param = READ_ONCE(*module_param); |
09c2d531 | 141 | unsigned modified_param = 0; |
f4790826 | 142 | |
09c2d531 MS |
143 | if (!param) |
144 | modified_param = def; | |
145 | else if (param > max) | |
146 | modified_param = max; | |
f4790826 | 147 | |
09c2d531 MS |
148 | if (modified_param) { |
149 | (void)cmpxchg(module_param, param, modified_param); | |
150 | param = modified_param; | |
f4790826 MS |
151 | } |
152 | ||
09c2d531 | 153 | return param; |
f4790826 MS |
154 | } |
155 | ||
e8603136 MS |
156 | unsigned dm_get_reserved_bio_based_ios(void) |
157 | { | |
09c2d531 | 158 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 159 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
160 | } |
161 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
162 | ||
115485e8 MS |
163 | static unsigned dm_get_numa_node(void) |
164 | { | |
165 | return __dm_get_module_param_int(&dm_numa_node, | |
166 | DM_NUMA_NODE, num_online_nodes() - 1); | |
167 | } | |
168 | ||
1da177e4 LT |
169 | static int __init local_init(void) |
170 | { | |
51157b4a | 171 | int r = -ENOMEM; |
1da177e4 | 172 | |
1da177e4 | 173 | /* allocate a slab for the dm_ios */ |
028867ac | 174 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 175 | if (!_io_cache) |
51157b4a | 176 | return r; |
1da177e4 | 177 | |
8fbf26ad KU |
178 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
179 | if (!_rq_tio_cache) | |
dba14160 | 180 | goto out_free_io_cache; |
8fbf26ad | 181 | |
eca7ee6d | 182 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
1ae49ea2 MS |
183 | __alignof__(struct request), 0, NULL); |
184 | if (!_rq_cache) | |
185 | goto out_free_rq_tio_cache; | |
186 | ||
51e5b2bd | 187 | r = dm_uevent_init(); |
51157b4a | 188 | if (r) |
1ae49ea2 | 189 | goto out_free_rq_cache; |
51e5b2bd | 190 | |
acfe0ad7 MP |
191 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
192 | if (!deferred_remove_workqueue) { | |
193 | r = -ENOMEM; | |
194 | goto out_uevent_exit; | |
195 | } | |
196 | ||
1da177e4 LT |
197 | _major = major; |
198 | r = register_blkdev(_major, _name); | |
51157b4a | 199 | if (r < 0) |
acfe0ad7 | 200 | goto out_free_workqueue; |
1da177e4 LT |
201 | |
202 | if (!_major) | |
203 | _major = r; | |
204 | ||
205 | return 0; | |
51157b4a | 206 | |
acfe0ad7 MP |
207 | out_free_workqueue: |
208 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
209 | out_uevent_exit: |
210 | dm_uevent_exit(); | |
1ae49ea2 MS |
211 | out_free_rq_cache: |
212 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
213 | out_free_rq_tio_cache: |
214 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
215 | out_free_io_cache: |
216 | kmem_cache_destroy(_io_cache); | |
217 | ||
218 | return r; | |
1da177e4 LT |
219 | } |
220 | ||
221 | static void local_exit(void) | |
222 | { | |
2c140a24 | 223 | flush_scheduled_work(); |
acfe0ad7 | 224 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 225 | |
1ae49ea2 | 226 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 227 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 228 | kmem_cache_destroy(_io_cache); |
00d59405 | 229 | unregister_blkdev(_major, _name); |
51e5b2bd | 230 | dm_uevent_exit(); |
1da177e4 LT |
231 | |
232 | _major = 0; | |
233 | ||
234 | DMINFO("cleaned up"); | |
235 | } | |
236 | ||
b9249e55 | 237 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
238 | local_init, |
239 | dm_target_init, | |
240 | dm_linear_init, | |
241 | dm_stripe_init, | |
952b3557 | 242 | dm_io_init, |
945fa4d2 | 243 | dm_kcopyd_init, |
1da177e4 | 244 | dm_interface_init, |
fd2ed4d2 | 245 | dm_statistics_init, |
1da177e4 LT |
246 | }; |
247 | ||
b9249e55 | 248 | static void (*_exits[])(void) = { |
1da177e4 LT |
249 | local_exit, |
250 | dm_target_exit, | |
251 | dm_linear_exit, | |
252 | dm_stripe_exit, | |
952b3557 | 253 | dm_io_exit, |
945fa4d2 | 254 | dm_kcopyd_exit, |
1da177e4 | 255 | dm_interface_exit, |
fd2ed4d2 | 256 | dm_statistics_exit, |
1da177e4 LT |
257 | }; |
258 | ||
259 | static int __init dm_init(void) | |
260 | { | |
261 | const int count = ARRAY_SIZE(_inits); | |
262 | ||
263 | int r, i; | |
264 | ||
265 | for (i = 0; i < count; i++) { | |
266 | r = _inits[i](); | |
267 | if (r) | |
268 | goto bad; | |
269 | } | |
270 | ||
271 | return 0; | |
272 | ||
273 | bad: | |
274 | while (i--) | |
275 | _exits[i](); | |
276 | ||
277 | return r; | |
278 | } | |
279 | ||
280 | static void __exit dm_exit(void) | |
281 | { | |
282 | int i = ARRAY_SIZE(_exits); | |
283 | ||
284 | while (i--) | |
285 | _exits[i](); | |
d15b774c AK |
286 | |
287 | /* | |
288 | * Should be empty by this point. | |
289 | */ | |
d15b774c | 290 | idr_destroy(&_minor_idr); |
1da177e4 LT |
291 | } |
292 | ||
293 | /* | |
294 | * Block device functions | |
295 | */ | |
432a212c MA |
296 | int dm_deleting_md(struct mapped_device *md) |
297 | { | |
298 | return test_bit(DMF_DELETING, &md->flags); | |
299 | } | |
300 | ||
fe5f9f2c | 301 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
302 | { |
303 | struct mapped_device *md; | |
304 | ||
fba9f90e JM |
305 | spin_lock(&_minor_lock); |
306 | ||
fe5f9f2c | 307 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
308 | if (!md) |
309 | goto out; | |
310 | ||
5c6bd75d | 311 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 312 | dm_deleting_md(md)) { |
fba9f90e JM |
313 | md = NULL; |
314 | goto out; | |
315 | } | |
316 | ||
1da177e4 | 317 | dm_get(md); |
5c6bd75d | 318 | atomic_inc(&md->open_count); |
fba9f90e JM |
319 | out: |
320 | spin_unlock(&_minor_lock); | |
321 | ||
322 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
323 | } |
324 | ||
db2a144b | 325 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 326 | { |
63a4f065 | 327 | struct mapped_device *md; |
6e9624b8 | 328 | |
4a1aeb98 MB |
329 | spin_lock(&_minor_lock); |
330 | ||
63a4f065 MS |
331 | md = disk->private_data; |
332 | if (WARN_ON(!md)) | |
333 | goto out; | |
334 | ||
2c140a24 MP |
335 | if (atomic_dec_and_test(&md->open_count) && |
336 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 337 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 338 | |
1da177e4 | 339 | dm_put(md); |
63a4f065 | 340 | out: |
4a1aeb98 | 341 | spin_unlock(&_minor_lock); |
1da177e4 LT |
342 | } |
343 | ||
5c6bd75d AK |
344 | int dm_open_count(struct mapped_device *md) |
345 | { | |
346 | return atomic_read(&md->open_count); | |
347 | } | |
348 | ||
349 | /* | |
350 | * Guarantees nothing is using the device before it's deleted. | |
351 | */ | |
2c140a24 | 352 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
353 | { |
354 | int r = 0; | |
355 | ||
356 | spin_lock(&_minor_lock); | |
357 | ||
2c140a24 | 358 | if (dm_open_count(md)) { |
5c6bd75d | 359 | r = -EBUSY; |
2c140a24 MP |
360 | if (mark_deferred) |
361 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
362 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
363 | r = -EEXIST; | |
5c6bd75d AK |
364 | else |
365 | set_bit(DMF_DELETING, &md->flags); | |
366 | ||
367 | spin_unlock(&_minor_lock); | |
368 | ||
369 | return r; | |
370 | } | |
371 | ||
2c140a24 MP |
372 | int dm_cancel_deferred_remove(struct mapped_device *md) |
373 | { | |
374 | int r = 0; | |
375 | ||
376 | spin_lock(&_minor_lock); | |
377 | ||
378 | if (test_bit(DMF_DELETING, &md->flags)) | |
379 | r = -EBUSY; | |
380 | else | |
381 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
382 | ||
383 | spin_unlock(&_minor_lock); | |
384 | ||
385 | return r; | |
386 | } | |
387 | ||
388 | static void do_deferred_remove(struct work_struct *w) | |
389 | { | |
390 | dm_deferred_remove(); | |
391 | } | |
392 | ||
fd2ed4d2 MP |
393 | sector_t dm_get_size(struct mapped_device *md) |
394 | { | |
395 | return get_capacity(md->disk); | |
396 | } | |
397 | ||
9974fa2c MS |
398 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
399 | { | |
400 | return md->queue; | |
401 | } | |
402 | ||
fd2ed4d2 MP |
403 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
404 | { | |
405 | return &md->stats; | |
406 | } | |
407 | ||
3ac51e74 DW |
408 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
409 | { | |
410 | struct mapped_device *md = bdev->bd_disk->private_data; | |
411 | ||
412 | return dm_get_geometry(md, geo); | |
413 | } | |
414 | ||
956a4025 MS |
415 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
416 | struct block_device **bdev, | |
417 | fmode_t *mode) | |
aa129a22 | 418 | { |
66482026 | 419 | struct dm_target *tgt; |
6c182cd8 | 420 | struct dm_table *map; |
956a4025 | 421 | int srcu_idx, r; |
aa129a22 | 422 | |
6c182cd8 | 423 | retry: |
e56f81e0 | 424 | r = -ENOTTY; |
956a4025 | 425 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
426 | if (!map || !dm_table_get_size(map)) |
427 | goto out; | |
428 | ||
429 | /* We only support devices that have a single target */ | |
430 | if (dm_table_get_num_targets(map) != 1) | |
431 | goto out; | |
432 | ||
66482026 MS |
433 | tgt = dm_table_get_target(map, 0); |
434 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 435 | goto out; |
aa129a22 | 436 | |
4f186f8b | 437 | if (dm_suspended_md(md)) { |
aa129a22 MB |
438 | r = -EAGAIN; |
439 | goto out; | |
440 | } | |
441 | ||
66482026 | 442 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
443 | if (r < 0) |
444 | goto out; | |
aa129a22 | 445 | |
956a4025 MS |
446 | bdgrab(*bdev); |
447 | dm_put_live_table(md, srcu_idx); | |
e56f81e0 | 448 | return r; |
aa129a22 | 449 | |
aa129a22 | 450 | out: |
956a4025 | 451 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 452 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
453 | msleep(10); |
454 | goto retry; | |
455 | } | |
e56f81e0 CH |
456 | return r; |
457 | } | |
458 | ||
459 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
460 | unsigned int cmd, unsigned long arg) | |
461 | { | |
462 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 463 | int r; |
e56f81e0 | 464 | |
956a4025 | 465 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
466 | if (r < 0) |
467 | return r; | |
6c182cd8 | 468 | |
e56f81e0 CH |
469 | if (r > 0) { |
470 | /* | |
e980f623 CH |
471 | * Target determined this ioctl is being issued against a |
472 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 473 | */ |
e980f623 CH |
474 | if (!capable(CAP_SYS_RAWIO)) { |
475 | DMWARN_LIMIT( | |
476 | "%s: sending ioctl %x to DM device without required privilege.", | |
477 | current->comm, cmd); | |
478 | r = -ENOIOCTLCMD; | |
e56f81e0 | 479 | goto out; |
e980f623 | 480 | } |
e56f81e0 | 481 | } |
6c182cd8 | 482 | |
66482026 | 483 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 484 | out: |
956a4025 | 485 | bdput(bdev); |
aa129a22 MB |
486 | return r; |
487 | } | |
488 | ||
028867ac | 489 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
490 | { |
491 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
492 | } | |
493 | ||
028867ac | 494 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
495 | { |
496 | mempool_free(io, md->io_pool); | |
497 | } | |
498 | ||
cfae7529 | 499 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 500 | { |
dba14160 | 501 | bio_put(&tio->clone); |
1da177e4 LT |
502 | } |
503 | ||
4cc96131 | 504 | int md_in_flight(struct mapped_device *md) |
90abb8c4 KU |
505 | { |
506 | return atomic_read(&md->pending[READ]) + | |
507 | atomic_read(&md->pending[WRITE]); | |
508 | } | |
509 | ||
3eaf840e JNN |
510 | static void start_io_acct(struct dm_io *io) |
511 | { | |
512 | struct mapped_device *md = io->md; | |
745dc570 | 513 | struct bio *bio = io->orig_bio; |
c9959059 | 514 | int cpu; |
fd2ed4d2 | 515 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
516 | |
517 | io->start_time = jiffies; | |
518 | ||
074a7aca | 519 | cpu = part_stat_lock(); |
d62e26b3 | 520 | part_round_stats(md->queue, cpu, &dm_disk(md)->part0); |
074a7aca | 521 | part_stat_unlock(); |
1e9bb880 SL |
522 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
523 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
524 | |
525 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
526 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
527 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
528 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
529 | } |
530 | ||
d221d2e7 | 531 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
532 | { |
533 | struct mapped_device *md = io->md; | |
745dc570 | 534 | struct bio *bio = io->orig_bio; |
3eaf840e | 535 | unsigned long duration = jiffies - io->start_time; |
18c0b223 | 536 | int pending; |
3eaf840e JNN |
537 | int rw = bio_data_dir(bio); |
538 | ||
d62e26b3 | 539 | generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 540 | |
fd2ed4d2 | 541 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
542 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
543 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
544 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 545 | |
af7e466a MP |
546 | /* |
547 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 548 | * a flush. |
af7e466a | 549 | */ |
1e9bb880 SL |
550 | pending = atomic_dec_return(&md->pending[rw]); |
551 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 552 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 553 | |
d221d2e7 MP |
554 | /* nudge anyone waiting on suspend queue */ |
555 | if (!pending) | |
556 | wake_up(&md->wait); | |
3eaf840e JNN |
557 | } |
558 | ||
1da177e4 LT |
559 | /* |
560 | * Add the bio to the list of deferred io. | |
561 | */ | |
92c63902 | 562 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 563 | { |
05447420 | 564 | unsigned long flags; |
1da177e4 | 565 | |
05447420 | 566 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 567 | bio_list_add(&md->deferred, bio); |
05447420 | 568 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 569 | queue_work(md->wq, &md->work); |
1da177e4 LT |
570 | } |
571 | ||
572 | /* | |
573 | * Everyone (including functions in this file), should use this | |
574 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 575 | * dm_put_live_table() when finished. |
1da177e4 | 576 | */ |
83d5e5b0 | 577 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 578 | { |
83d5e5b0 MP |
579 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
580 | ||
581 | return srcu_dereference(md->map, &md->io_barrier); | |
582 | } | |
1da177e4 | 583 | |
83d5e5b0 MP |
584 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
585 | { | |
586 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
587 | } | |
588 | ||
589 | void dm_sync_table(struct mapped_device *md) | |
590 | { | |
591 | synchronize_srcu(&md->io_barrier); | |
592 | synchronize_rcu_expedited(); | |
593 | } | |
594 | ||
595 | /* | |
596 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
597 | * The caller must not block between these two functions. | |
598 | */ | |
599 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
600 | { | |
601 | rcu_read_lock(); | |
602 | return rcu_dereference(md->map); | |
603 | } | |
1da177e4 | 604 | |
83d5e5b0 MP |
605 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
606 | { | |
607 | rcu_read_unlock(); | |
1da177e4 LT |
608 | } |
609 | ||
86f1152b BM |
610 | /* |
611 | * Open a table device so we can use it as a map destination. | |
612 | */ | |
613 | static int open_table_device(struct table_device *td, dev_t dev, | |
614 | struct mapped_device *md) | |
615 | { | |
616 | static char *_claim_ptr = "I belong to device-mapper"; | |
617 | struct block_device *bdev; | |
618 | ||
619 | int r; | |
620 | ||
621 | BUG_ON(td->dm_dev.bdev); | |
622 | ||
623 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
624 | if (IS_ERR(bdev)) | |
625 | return PTR_ERR(bdev); | |
626 | ||
627 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
628 | if (r) { | |
629 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
630 | return r; | |
631 | } | |
632 | ||
633 | td->dm_dev.bdev = bdev; | |
817bf402 | 634 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
635 | return 0; |
636 | } | |
637 | ||
638 | /* | |
639 | * Close a table device that we've been using. | |
640 | */ | |
641 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
642 | { | |
643 | if (!td->dm_dev.bdev) | |
644 | return; | |
645 | ||
646 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
647 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 648 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 649 | td->dm_dev.bdev = NULL; |
817bf402 | 650 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
651 | } |
652 | ||
653 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
654 | fmode_t mode) { | |
655 | struct table_device *td; | |
656 | ||
657 | list_for_each_entry(td, l, list) | |
658 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
659 | return td; | |
660 | ||
661 | return NULL; | |
662 | } | |
663 | ||
664 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
665 | struct dm_dev **result) { | |
666 | int r; | |
667 | struct table_device *td; | |
668 | ||
669 | mutex_lock(&md->table_devices_lock); | |
670 | td = find_table_device(&md->table_devices, dev, mode); | |
671 | if (!td) { | |
115485e8 | 672 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
673 | if (!td) { |
674 | mutex_unlock(&md->table_devices_lock); | |
675 | return -ENOMEM; | |
676 | } | |
677 | ||
678 | td->dm_dev.mode = mode; | |
679 | td->dm_dev.bdev = NULL; | |
680 | ||
681 | if ((r = open_table_device(td, dev, md))) { | |
682 | mutex_unlock(&md->table_devices_lock); | |
683 | kfree(td); | |
684 | return r; | |
685 | } | |
686 | ||
687 | format_dev_t(td->dm_dev.name, dev); | |
688 | ||
b0b4d7c6 | 689 | refcount_set(&td->count, 1); |
86f1152b | 690 | list_add(&td->list, &md->table_devices); |
b0b4d7c6 ER |
691 | } else { |
692 | refcount_inc(&td->count); | |
86f1152b | 693 | } |
86f1152b BM |
694 | mutex_unlock(&md->table_devices_lock); |
695 | ||
696 | *result = &td->dm_dev; | |
697 | return 0; | |
698 | } | |
699 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
700 | ||
701 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
702 | { | |
703 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
704 | ||
705 | mutex_lock(&md->table_devices_lock); | |
b0b4d7c6 | 706 | if (refcount_dec_and_test(&td->count)) { |
86f1152b BM |
707 | close_table_device(td, md); |
708 | list_del(&td->list); | |
709 | kfree(td); | |
710 | } | |
711 | mutex_unlock(&md->table_devices_lock); | |
712 | } | |
713 | EXPORT_SYMBOL(dm_put_table_device); | |
714 | ||
715 | static void free_table_devices(struct list_head *devices) | |
716 | { | |
717 | struct list_head *tmp, *next; | |
718 | ||
719 | list_for_each_safe(tmp, next, devices) { | |
720 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
721 | ||
722 | DMWARN("dm_destroy: %s still exists with %d references", | |
b0b4d7c6 | 723 | td->dm_dev.name, refcount_read(&td->count)); |
86f1152b BM |
724 | kfree(td); |
725 | } | |
726 | } | |
727 | ||
3ac51e74 DW |
728 | /* |
729 | * Get the geometry associated with a dm device | |
730 | */ | |
731 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
732 | { | |
733 | *geo = md->geometry; | |
734 | ||
735 | return 0; | |
736 | } | |
737 | ||
738 | /* | |
739 | * Set the geometry of a device. | |
740 | */ | |
741 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
742 | { | |
743 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
744 | ||
745 | if (geo->start > sz) { | |
746 | DMWARN("Start sector is beyond the geometry limits."); | |
747 | return -EINVAL; | |
748 | } | |
749 | ||
750 | md->geometry = *geo; | |
751 | ||
752 | return 0; | |
753 | } | |
754 | ||
2e93ccc1 KU |
755 | static int __noflush_suspending(struct mapped_device *md) |
756 | { | |
757 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
758 | } | |
759 | ||
1da177e4 LT |
760 | /* |
761 | * Decrements the number of outstanding ios that a bio has been | |
762 | * cloned into, completing the original io if necc. | |
763 | */ | |
4e4cbee9 | 764 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 765 | { |
2e93ccc1 | 766 | unsigned long flags; |
4e4cbee9 | 767 | blk_status_t io_error; |
b35f8caa MB |
768 | struct bio *bio; |
769 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
770 | |
771 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
772 | if (unlikely(error)) { |
773 | spin_lock_irqsave(&io->endio_lock, flags); | |
745dc570 | 774 | if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) |
4e4cbee9 | 775 | io->status = error; |
f88fb981 KU |
776 | spin_unlock_irqrestore(&io->endio_lock, flags); |
777 | } | |
1da177e4 LT |
778 | |
779 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 780 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
781 | /* |
782 | * Target requested pushing back the I/O. | |
2e93ccc1 | 783 | */ |
022c2611 | 784 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 | 785 | if (__noflush_suspending(md)) |
745dc570 MS |
786 | /* NOTE early return due to BLK_STS_DM_REQUEUE below */ |
787 | bio_list_add_head(&md->deferred, io->orig_bio); | |
6a8736d1 | 788 | else |
2e93ccc1 | 789 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 790 | io->status = BLK_STS_IOERR; |
022c2611 | 791 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
792 | } |
793 | ||
4e4cbee9 | 794 | io_error = io->status; |
745dc570 | 795 | bio = io->orig_bio; |
6a8736d1 TH |
796 | end_io_acct(io); |
797 | free_io(md, io); | |
798 | ||
4e4cbee9 | 799 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 800 | return; |
2e93ccc1 | 801 | |
1eff9d32 | 802 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 803 | /* |
6a8736d1 | 804 | * Preflush done for flush with data, reissue |
28a8f0d3 | 805 | * without REQ_PREFLUSH. |
af7e466a | 806 | */ |
1eff9d32 | 807 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 808 | queue_io(md, bio); |
af7e466a | 809 | } else { |
b372d360 | 810 | /* done with normal IO or empty flush */ |
4e4cbee9 | 811 | bio->bi_status = io_error; |
4246a0b6 | 812 | bio_endio(bio); |
b35f8caa | 813 | } |
1da177e4 LT |
814 | } |
815 | } | |
816 | ||
4cc96131 | 817 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
818 | { |
819 | struct queue_limits *limits = dm_get_queue_limits(md); | |
820 | ||
821 | /* device doesn't really support WRITE SAME, disable it */ | |
822 | limits->max_write_same_sectors = 0; | |
823 | } | |
824 | ||
ac62d620 CH |
825 | void disable_write_zeroes(struct mapped_device *md) |
826 | { | |
827 | struct queue_limits *limits = dm_get_queue_limits(md); | |
828 | ||
829 | /* device doesn't really support WRITE ZEROES, disable it */ | |
830 | limits->max_write_zeroes_sectors = 0; | |
831 | } | |
832 | ||
4246a0b6 | 833 | static void clone_endio(struct bio *bio) |
1da177e4 | 834 | { |
4e4cbee9 | 835 | blk_status_t error = bio->bi_status; |
bfc6d41c | 836 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 837 | struct dm_io *io = tio->io; |
9faf400f | 838 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
839 | dm_endio_fn endio = tio->ti->type->end_io; |
840 | ||
4e4cbee9 | 841 | if (unlikely(error == BLK_STS_TARGET)) { |
ac62d620 | 842 | if (bio_op(bio) == REQ_OP_WRITE_SAME && |
74d46992 | 843 | !bio->bi_disk->queue->limits.max_write_same_sectors) |
ac62d620 CH |
844 | disable_write_same(md); |
845 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && | |
74d46992 | 846 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) |
ac62d620 CH |
847 | disable_write_zeroes(md); |
848 | } | |
7eee4ae2 | 849 | |
1be56909 | 850 | if (endio) { |
4e4cbee9 | 851 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
852 | switch (r) { |
853 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 854 | error = BLK_STS_DM_REQUEUE; |
1be56909 CH |
855 | /*FALLTHRU*/ |
856 | case DM_ENDIO_DONE: | |
857 | break; | |
858 | case DM_ENDIO_INCOMPLETE: | |
859 | /* The target will handle the io */ | |
860 | return; | |
861 | default: | |
862 | DMWARN("unimplemented target endio return value: %d", r); | |
863 | BUG(); | |
864 | } | |
865 | } | |
866 | ||
cfae7529 | 867 | free_tio(tio); |
b35f8caa | 868 | dec_pending(io, error); |
1da177e4 LT |
869 | } |
870 | ||
56a67df7 MS |
871 | /* |
872 | * Return maximum size of I/O possible at the supplied sector up to the current | |
873 | * target boundary. | |
874 | */ | |
875 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
876 | { | |
877 | sector_t target_offset = dm_target_offset(ti, sector); | |
878 | ||
879 | return ti->len - target_offset; | |
880 | } | |
881 | ||
882 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 883 | { |
56a67df7 | 884 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 885 | sector_t offset, max_len; |
1da177e4 LT |
886 | |
887 | /* | |
542f9038 | 888 | * Does the target need to split even further? |
1da177e4 | 889 | */ |
542f9038 MS |
890 | if (ti->max_io_len) { |
891 | offset = dm_target_offset(ti, sector); | |
892 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
893 | max_len = sector_div(offset, ti->max_io_len); | |
894 | else | |
895 | max_len = offset & (ti->max_io_len - 1); | |
896 | max_len = ti->max_io_len - max_len; | |
897 | ||
898 | if (len > max_len) | |
899 | len = max_len; | |
1da177e4 LT |
900 | } |
901 | ||
902 | return len; | |
903 | } | |
904 | ||
542f9038 MS |
905 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
906 | { | |
907 | if (len > UINT_MAX) { | |
908 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
909 | (unsigned long long)len, UINT_MAX); | |
910 | ti->error = "Maximum size of target IO is too large"; | |
911 | return -EINVAL; | |
912 | } | |
913 | ||
914 | ti->max_io_len = (uint32_t) len; | |
915 | ||
916 | return 0; | |
917 | } | |
918 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
919 | ||
f26c5719 DW |
920 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
921 | sector_t sector, int *srcu_idx) | |
545ed20e | 922 | { |
545ed20e TK |
923 | struct dm_table *map; |
924 | struct dm_target *ti; | |
545ed20e | 925 | |
f26c5719 | 926 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 927 | if (!map) |
f26c5719 | 928 | return NULL; |
545ed20e TK |
929 | |
930 | ti = dm_table_find_target(map, sector); | |
931 | if (!dm_target_is_valid(ti)) | |
f26c5719 | 932 | return NULL; |
545ed20e | 933 | |
f26c5719 DW |
934 | return ti; |
935 | } | |
545ed20e | 936 | |
f26c5719 DW |
937 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
938 | long nr_pages, void **kaddr, pfn_t *pfn) | |
939 | { | |
940 | struct mapped_device *md = dax_get_private(dax_dev); | |
941 | sector_t sector = pgoff * PAGE_SECTORS; | |
942 | struct dm_target *ti; | |
943 | long len, ret = -EIO; | |
944 | int srcu_idx; | |
545ed20e | 945 | |
f26c5719 | 946 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 947 | |
f26c5719 DW |
948 | if (!ti) |
949 | goto out; | |
950 | if (!ti->type->direct_access) | |
951 | goto out; | |
952 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
953 | if (len < 1) | |
954 | goto out; | |
955 | nr_pages = min(len, nr_pages); | |
545ed20e | 956 | if (ti->type->direct_access) |
817bf402 DW |
957 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
958 | ||
f26c5719 | 959 | out: |
545ed20e | 960 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
961 | |
962 | return ret; | |
545ed20e TK |
963 | } |
964 | ||
7e026c8c DW |
965 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
966 | void *addr, size_t bytes, struct iov_iter *i) | |
967 | { | |
968 | struct mapped_device *md = dax_get_private(dax_dev); | |
969 | sector_t sector = pgoff * PAGE_SECTORS; | |
970 | struct dm_target *ti; | |
971 | long ret = 0; | |
972 | int srcu_idx; | |
973 | ||
974 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
975 | ||
976 | if (!ti) | |
977 | goto out; | |
978 | if (!ti->type->dax_copy_from_iter) { | |
979 | ret = copy_from_iter(addr, bytes, i); | |
980 | goto out; | |
981 | } | |
982 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
983 | out: | |
984 | dm_put_live_table(md, srcu_idx); | |
985 | ||
986 | return ret; | |
987 | } | |
988 | ||
1dd40c3e MP |
989 | /* |
990 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
c06b3e58 | 991 | * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET. |
1dd40c3e MP |
992 | * |
993 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
994 | * additional n_sectors sectors of the bio and the rest of the data should be | |
995 | * sent in a next bio. | |
996 | * | |
997 | * A diagram that explains the arithmetics: | |
998 | * +--------------------+---------------+-------+ | |
999 | * | 1 | 2 | 3 | | |
1000 | * +--------------------+---------------+-------+ | |
1001 | * | |
1002 | * <-------------- *tio->len_ptr ---------------> | |
1003 | * <------- bi_size -------> | |
1004 | * <-- n_sectors --> | |
1005 | * | |
1006 | * Region 1 was already iterated over with bio_advance or similar function. | |
1007 | * (it may be empty if the target doesn't use bio_advance) | |
1008 | * Region 2 is the remaining bio size that the target wants to process. | |
1009 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1010 | * to make it empty) | |
1011 | * The target requires that region 3 is to be sent in the next bio. | |
1012 | * | |
1013 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1014 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1015 | * copies of the bio. | |
1016 | */ | |
1017 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1018 | { | |
1019 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1020 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1021 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1022 | BUG_ON(bi_size > *tio->len_ptr); |
1023 | BUG_ON(n_sectors > bi_size); | |
1024 | *tio->len_ptr -= bi_size - n_sectors; | |
1025 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1026 | } | |
1027 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1028 | ||
10999307 DLM |
1029 | /* |
1030 | * The zone descriptors obtained with a zone report indicate | |
1031 | * zone positions within the target device. The zone descriptors | |
1032 | * must be remapped to match their position within the dm device. | |
1033 | * A target may call dm_remap_zone_report after completion of a | |
1034 | * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained | |
1035 | * from the target device mapping to the dm device. | |
1036 | */ | |
1037 | void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start) | |
1038 | { | |
1039 | #ifdef CONFIG_BLK_DEV_ZONED | |
1040 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
745dc570 | 1041 | struct bio *report_bio = tio->io->orig_bio; |
10999307 DLM |
1042 | struct blk_zone_report_hdr *hdr = NULL; |
1043 | struct blk_zone *zone; | |
1044 | unsigned int nr_rep = 0; | |
1045 | unsigned int ofst; | |
1046 | struct bio_vec bvec; | |
1047 | struct bvec_iter iter; | |
1048 | void *addr; | |
1049 | ||
1050 | if (bio->bi_status) | |
1051 | return; | |
1052 | ||
1053 | /* | |
1054 | * Remap the start sector of the reported zones. For sequential zones, | |
1055 | * also remap the write pointer position. | |
1056 | */ | |
1057 | bio_for_each_segment(bvec, report_bio, iter) { | |
1058 | addr = kmap_atomic(bvec.bv_page); | |
1059 | ||
1060 | /* Remember the report header in the first page */ | |
1061 | if (!hdr) { | |
1062 | hdr = addr; | |
1063 | ofst = sizeof(struct blk_zone_report_hdr); | |
1064 | } else | |
1065 | ofst = 0; | |
1066 | ||
1067 | /* Set zones start sector */ | |
1068 | while (hdr->nr_zones && ofst < bvec.bv_len) { | |
1069 | zone = addr + ofst; | |
1070 | if (zone->start >= start + ti->len) { | |
1071 | hdr->nr_zones = 0; | |
1072 | break; | |
1073 | } | |
1074 | zone->start = zone->start + ti->begin - start; | |
1075 | if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { | |
1076 | if (zone->cond == BLK_ZONE_COND_FULL) | |
1077 | zone->wp = zone->start + zone->len; | |
1078 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
1079 | zone->wp = zone->start; | |
1080 | else | |
1081 | zone->wp = zone->wp + ti->begin - start; | |
1082 | } | |
1083 | ofst += sizeof(struct blk_zone); | |
1084 | hdr->nr_zones--; | |
1085 | nr_rep++; | |
1086 | } | |
1087 | ||
1088 | if (addr != hdr) | |
1089 | kunmap_atomic(addr); | |
1090 | ||
1091 | if (!hdr->nr_zones) | |
1092 | break; | |
1093 | } | |
1094 | ||
1095 | if (hdr) { | |
1096 | hdr->nr_zones = nr_rep; | |
1097 | kunmap_atomic(hdr); | |
1098 | } | |
1099 | ||
1100 | bio_advance(report_bio, report_bio->bi_iter.bi_size); | |
1101 | ||
1102 | #else /* !CONFIG_BLK_DEV_ZONED */ | |
1103 | bio->bi_status = BLK_STS_NOTSUPP; | |
1104 | #endif | |
1105 | } | |
1106 | EXPORT_SYMBOL_GPL(dm_remap_zone_report); | |
1107 | ||
bd2a49b8 | 1108 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1109 | { |
1110 | int r; | |
2056a782 | 1111 | sector_t sector; |
dba14160 | 1112 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1113 | struct dm_target *ti = tio->ti; |
1da177e4 | 1114 | |
1da177e4 | 1115 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1116 | |
1117 | /* | |
1118 | * Map the clone. If r == 0 we don't need to do | |
1119 | * anything, the target has assumed ownership of | |
1120 | * this io. | |
1121 | */ | |
1122 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1123 | sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1124 | |
7de3ee57 | 1125 | r = ti->type->map(ti, clone); |
846785e6 CH |
1126 | switch (r) { |
1127 | case DM_MAPIO_SUBMITTED: | |
1128 | break; | |
1129 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1130 | /* the bio has been remapped so dispatch it */ |
74d46992 | 1131 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
745dc570 | 1132 | bio_dev(tio->io->orig_bio), sector); |
1da177e4 | 1133 | generic_make_request(clone); |
846785e6 CH |
1134 | break; |
1135 | case DM_MAPIO_KILL: | |
4e4cbee9 CH |
1136 | dec_pending(tio->io, BLK_STS_IOERR); |
1137 | free_tio(tio); | |
1138 | break; | |
846785e6 | 1139 | case DM_MAPIO_REQUEUE: |
4e4cbee9 | 1140 | dec_pending(tio->io, BLK_STS_DM_REQUEUE); |
cfae7529 | 1141 | free_tio(tio); |
846785e6 CH |
1142 | break; |
1143 | default: | |
45cbcd79 KU |
1144 | DMWARN("unimplemented target map return value: %d", r); |
1145 | BUG(); | |
1da177e4 LT |
1146 | } |
1147 | } | |
1148 | ||
1149 | struct clone_info { | |
1150 | struct mapped_device *md; | |
1151 | struct dm_table *map; | |
1152 | struct bio *bio; | |
1153 | struct dm_io *io; | |
1154 | sector_t sector; | |
e0d6609a | 1155 | unsigned sector_count; |
1da177e4 LT |
1156 | }; |
1157 | ||
e0d6609a | 1158 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1159 | { |
4f024f37 KO |
1160 | bio->bi_iter.bi_sector = sector; |
1161 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1162 | } |
1163 | ||
1164 | /* | |
1165 | * Creates a bio that consists of range of complete bvecs. | |
1166 | */ | |
c80914e8 MS |
1167 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1168 | sector_t sector, unsigned len) | |
1da177e4 | 1169 | { |
dba14160 | 1170 | struct bio *clone = &tio->clone; |
1da177e4 | 1171 | |
1c3b13e6 KO |
1172 | __bio_clone_fast(clone, bio); |
1173 | ||
e2460f2a MP |
1174 | if (unlikely(bio_integrity(bio) != NULL)) { |
1175 | int r; | |
1176 | ||
1177 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1178 | !dm_target_passes_integrity(tio->ti->type))) { | |
1179 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1180 | dm_device_name(tio->io->md), | |
1181 | tio->ti->type->name); | |
1182 | return -EIO; | |
1183 | } | |
1184 | ||
1185 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1186 | if (r < 0) |
1187 | return r; | |
1188 | } | |
bd2a49b8 | 1189 | |
264c869d DLM |
1190 | if (bio_op(bio) != REQ_OP_ZONE_REPORT) |
1191 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); | |
1c3b13e6 KO |
1192 | clone->bi_iter.bi_size = to_bytes(len); |
1193 | ||
e2460f2a | 1194 | if (unlikely(bio_integrity(bio) != NULL)) |
fbd08e76 | 1195 | bio_integrity_trim(clone); |
c80914e8 MS |
1196 | |
1197 | return 0; | |
1da177e4 LT |
1198 | } |
1199 | ||
318716dd MS |
1200 | static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, |
1201 | unsigned target_bio_nr, gfp_t gfp_mask) | |
f9ab94ce | 1202 | { |
dba14160 MP |
1203 | struct dm_target_io *tio; |
1204 | struct bio *clone; | |
1205 | ||
318716dd MS |
1206 | clone = bio_alloc_bioset(gfp_mask, 0, ci->md->bs); |
1207 | if (!clone) | |
1208 | return NULL; | |
f9ab94ce | 1209 | |
318716dd | 1210 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1211 | tio->io = ci->io; |
1212 | tio->ti = ti; | |
55a62eef | 1213 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1214 | |
1215 | return tio; | |
1216 | } | |
1217 | ||
318716dd MS |
1218 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
1219 | struct dm_target *ti, unsigned num_bios) | |
1220 | { | |
1221 | struct dm_target_io *tio; | |
1222 | int try; | |
1223 | ||
1224 | if (!num_bios) | |
1225 | return; | |
1226 | ||
1227 | if (num_bios == 1) { | |
1228 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); | |
1229 | bio_list_add(blist, &tio->clone); | |
1230 | return; | |
1231 | } | |
1232 | ||
1233 | for (try = 0; try < 2; try++) { | |
1234 | int bio_nr; | |
1235 | struct bio *bio; | |
1236 | ||
1237 | if (try) | |
1238 | mutex_lock(&ci->md->table_devices_lock); | |
1239 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { | |
1240 | tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); | |
1241 | if (!tio) | |
1242 | break; | |
1243 | ||
1244 | bio_list_add(blist, &tio->clone); | |
1245 | } | |
1246 | if (try) | |
1247 | mutex_unlock(&ci->md->table_devices_lock); | |
1248 | if (bio_nr == num_bios) | |
1249 | return; | |
1250 | ||
1251 | while ((bio = bio_list_pop(blist))) { | |
1252 | tio = container_of(bio, struct dm_target_io, clone); | |
1253 | free_tio(tio); | |
1254 | } | |
1255 | } | |
1256 | } | |
1257 | ||
14fe594d | 1258 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
318716dd | 1259 | struct dm_target_io *tio, unsigned *len) |
9015df24 | 1260 | { |
dba14160 | 1261 | struct bio *clone = &tio->clone; |
9015df24 | 1262 | |
1dd40c3e MP |
1263 | tio->len_ptr = len; |
1264 | ||
99778273 | 1265 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1266 | if (len) |
1dd40c3e | 1267 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1268 | |
bd2a49b8 | 1269 | __map_bio(tio); |
f9ab94ce MP |
1270 | } |
1271 | ||
14fe594d | 1272 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1273 | unsigned num_bios, unsigned *len) |
06a426ce | 1274 | { |
318716dd MS |
1275 | struct bio_list blist = BIO_EMPTY_LIST; |
1276 | struct bio *bio; | |
1277 | struct dm_target_io *tio; | |
1278 | ||
1279 | alloc_multiple_bios(&blist, ci, ti, num_bios); | |
06a426ce | 1280 | |
318716dd MS |
1281 | while ((bio = bio_list_pop(&blist))) { |
1282 | tio = container_of(bio, struct dm_target_io, clone); | |
1283 | __clone_and_map_simple_bio(ci, tio, len); | |
1284 | } | |
06a426ce MS |
1285 | } |
1286 | ||
14fe594d | 1287 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1288 | { |
06a426ce | 1289 | unsigned target_nr = 0; |
f9ab94ce MP |
1290 | struct dm_target *ti; |
1291 | ||
b372d360 | 1292 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1293 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1294 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1295 | |
f9ab94ce MP |
1296 | return 0; |
1297 | } | |
1298 | ||
c80914e8 | 1299 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
f31c21e4 | 1300 | sector_t sector, unsigned *len) |
5ae89a87 | 1301 | { |
dba14160 | 1302 | struct bio *bio = ci->bio; |
5ae89a87 | 1303 | struct dm_target_io *tio; |
f31c21e4 | 1304 | int r; |
5ae89a87 | 1305 | |
318716dd | 1306 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); |
f31c21e4 N |
1307 | tio->len_ptr = len; |
1308 | r = clone_bio(tio, bio, sector, *len); | |
1309 | if (r < 0) { | |
1310 | free_tio(tio); | |
1311 | return r; | |
b0d8ed4d | 1312 | } |
f31c21e4 | 1313 | __map_bio(tio); |
c80914e8 | 1314 | |
f31c21e4 | 1315 | return 0; |
5ae89a87 MS |
1316 | } |
1317 | ||
55a62eef | 1318 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1319 | |
55a62eef | 1320 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1321 | { |
55a62eef | 1322 | return ti->num_discard_bios; |
23508a96 MS |
1323 | } |
1324 | ||
55a62eef | 1325 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1326 | { |
55a62eef | 1327 | return ti->num_write_same_bios; |
23508a96 MS |
1328 | } |
1329 | ||
ac62d620 CH |
1330 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1331 | { | |
1332 | return ti->num_write_zeroes_bios; | |
1333 | } | |
1334 | ||
23508a96 | 1335 | typedef bool (*is_split_required_fn)(struct dm_target *ti); |
9eef87da | 1336 | |
23508a96 MS |
1337 | static bool is_split_required_for_discard(struct dm_target *ti) |
1338 | { | |
55a62eef | 1339 | return ti->split_discard_bios; |
cec47e3d KU |
1340 | } |
1341 | ||
3d7f4562 | 1342 | static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
14fe594d AK |
1343 | get_num_bios_fn get_num_bios, |
1344 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1345 | { |
e0d6609a | 1346 | unsigned len; |
55a62eef | 1347 | unsigned num_bios; |
ba1cbad9 | 1348 | |
3d7f4562 MS |
1349 | /* |
1350 | * Even though the device advertised support for this type of | |
1351 | * request, that does not mean every target supports it, and | |
1352 | * reconfiguration might also have changed that since the | |
1353 | * check was performed. | |
1354 | */ | |
1355 | num_bios = get_num_bios ? get_num_bios(ti) : 0; | |
1356 | if (!num_bios) | |
1357 | return -EOPNOTSUPP; | |
ba1cbad9 | 1358 | |
3d7f4562 MS |
1359 | if (is_split_required && !is_split_required(ti)) |
1360 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); | |
1361 | else | |
1362 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); | |
de3ec86d | 1363 | |
3d7f4562 | 1364 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1365 | |
3d7f4562 MS |
1366 | ci->sector += len; |
1367 | ci->sector_count -= len; | |
5ae89a87 MS |
1368 | |
1369 | return 0; | |
ba1cbad9 MS |
1370 | } |
1371 | ||
3d7f4562 | 1372 | static int __send_discard(struct clone_info *ci, struct dm_target *ti) |
23508a96 | 1373 | { |
3d7f4562 | 1374 | return __send_changing_extent_only(ci, ti, get_num_discard_bios, |
14fe594d | 1375 | is_split_required_for_discard); |
23508a96 | 1376 | } |
0ce65797 | 1377 | |
3d7f4562 | 1378 | static int __send_write_same(struct clone_info *ci, struct dm_target *ti) |
0ce65797 | 1379 | { |
3d7f4562 | 1380 | return __send_changing_extent_only(ci, ti, get_num_write_same_bios, NULL); |
0ce65797 MS |
1381 | } |
1382 | ||
3d7f4562 | 1383 | static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti) |
ac62d620 | 1384 | { |
3d7f4562 | 1385 | return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios, NULL); |
ac62d620 CH |
1386 | } |
1387 | ||
e4c93811 AK |
1388 | /* |
1389 | * Select the correct strategy for processing a non-flush bio. | |
1390 | */ | |
14fe594d | 1391 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1392 | { |
dba14160 | 1393 | struct bio *bio = ci->bio; |
512875bd | 1394 | struct dm_target *ti; |
1c3b13e6 | 1395 | unsigned len; |
c80914e8 | 1396 | int r; |
0ce65797 | 1397 | |
512875bd JN |
1398 | ti = dm_table_find_target(ci->map, ci->sector); |
1399 | if (!dm_target_is_valid(ti)) | |
1400 | return -EIO; | |
1401 | ||
3d7f4562 MS |
1402 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
1403 | return __send_discard(ci, ti); | |
1404 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) | |
1405 | return __send_write_same(ci, ti); | |
1406 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES)) | |
1407 | return __send_write_zeroes(ci, ti); | |
1408 | ||
264c869d DLM |
1409 | if (bio_op(bio) == REQ_OP_ZONE_REPORT) |
1410 | len = ci->sector_count; | |
1411 | else | |
1412 | len = min_t(sector_t, max_io_len(ci->sector, ti), | |
1413 | ci->sector_count); | |
0ce65797 | 1414 | |
c80914e8 MS |
1415 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1416 | if (r < 0) | |
1417 | return r; | |
0ce65797 | 1418 | |
1c3b13e6 KO |
1419 | ci->sector += len; |
1420 | ci->sector_count -= len; | |
0ce65797 | 1421 | |
1c3b13e6 | 1422 | return 0; |
0ce65797 MS |
1423 | } |
1424 | ||
1da177e4 | 1425 | /* |
14fe594d | 1426 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1427 | */ |
83d5e5b0 MP |
1428 | static void __split_and_process_bio(struct mapped_device *md, |
1429 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1430 | { |
1da177e4 | 1431 | struct clone_info ci; |
512875bd | 1432 | int error = 0; |
1da177e4 | 1433 | |
83d5e5b0 | 1434 | if (unlikely(!map)) { |
6a8736d1 | 1435 | bio_io_error(bio); |
f0b9a450 MP |
1436 | return; |
1437 | } | |
692d0eb9 | 1438 | |
83d5e5b0 | 1439 | ci.map = map; |
1da177e4 | 1440 | ci.md = md; |
1da177e4 | 1441 | ci.io = alloc_io(md); |
4e4cbee9 | 1442 | ci.io->status = 0; |
1da177e4 | 1443 | atomic_set(&ci.io->io_count, 1); |
745dc570 | 1444 | ci.io->orig_bio = bio; |
1da177e4 | 1445 | ci.io->md = md; |
f88fb981 | 1446 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1447 | ci.sector = bio->bi_iter.bi_sector; |
0ce65797 | 1448 | |
3eaf840e | 1449 | start_io_acct(ci.io); |
0ce65797 | 1450 | |
1eff9d32 | 1451 | if (bio->bi_opf & REQ_PREFLUSH) { |
b372d360 MS |
1452 | ci.bio = &ci.md->flush_bio; |
1453 | ci.sector_count = 0; | |
14fe594d | 1454 | error = __send_empty_flush(&ci); |
b372d360 | 1455 | /* dec_pending submits any data associated with flush */ |
a4aa5e56 DLM |
1456 | } else if (bio_op(bio) == REQ_OP_ZONE_RESET) { |
1457 | ci.bio = bio; | |
1458 | ci.sector_count = 0; | |
1459 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1460 | } else { |
6a8736d1 | 1461 | ci.bio = bio; |
d87f4c14 | 1462 | ci.sector_count = bio_sectors(bio); |
18a25da8 | 1463 | while (ci.sector_count && !error) { |
14fe594d | 1464 | error = __split_and_process_non_flush(&ci); |
18a25da8 N |
1465 | if (current->bio_list && ci.sector_count && !error) { |
1466 | /* | |
1467 | * Remainder must be passed to generic_make_request() | |
1468 | * so that it gets handled *after* bios already submitted | |
1469 | * have been completely processed. | |
1470 | * We take a clone of the original to store in | |
745dc570 | 1471 | * ci.io->orig_bio to be used by end_io_acct() and |
18a25da8 N |
1472 | * for dec_pending to use for completion handling. |
1473 | * As this path is not used for REQ_OP_ZONE_REPORT, | |
745dc570 | 1474 | * the usage of io->orig_bio in dm_remap_zone_report() |
18a25da8 N |
1475 | * won't be affected by this reassignment. |
1476 | */ | |
1477 | struct bio *b = bio_clone_bioset(bio, GFP_NOIO, | |
1478 | md->queue->bio_split); | |
745dc570 | 1479 | ci.io->orig_bio = b; |
18a25da8 N |
1480 | bio_advance(bio, (bio_sectors(bio) - ci.sector_count) << 9); |
1481 | bio_chain(b, bio); | |
1482 | generic_make_request(bio); | |
1483 | break; | |
1484 | } | |
1485 | } | |
d87f4c14 | 1486 | } |
0ce65797 | 1487 | |
1da177e4 | 1488 | /* drop the extra reference count */ |
54385bf7 | 1489 | dec_pending(ci.io, errno_to_blk_status(error)); |
0ce65797 MS |
1490 | } |
1491 | ||
cec47e3d | 1492 | /* |
18a25da8 N |
1493 | * The request function that remaps the bio to one target and |
1494 | * splits off any remainder. | |
cec47e3d | 1495 | */ |
dece1635 | 1496 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
cec47e3d | 1497 | { |
12f03a49 | 1498 | int rw = bio_data_dir(bio); |
cec47e3d | 1499 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1500 | int srcu_idx; |
1501 | struct dm_table *map; | |
cec47e3d | 1502 | |
83d5e5b0 | 1503 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1504 | |
d62e26b3 | 1505 | generic_start_io_acct(q, rw, bio_sectors(bio), &dm_disk(md)->part0); |
d0bcb878 | 1506 | |
6a8736d1 TH |
1507 | /* if we're suspended, we have to queue this io for later */ |
1508 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1509 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1510 | |
1eff9d32 | 1511 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1512 | queue_io(md, bio); |
1513 | else | |
54d9a1b4 | 1514 | bio_io_error(bio); |
dece1635 | 1515 | return BLK_QC_T_NONE; |
cec47e3d | 1516 | } |
1da177e4 | 1517 | |
83d5e5b0 MP |
1518 | __split_and_process_bio(md, map, bio); |
1519 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1520 | return BLK_QC_T_NONE; |
cec47e3d KU |
1521 | } |
1522 | ||
1da177e4 LT |
1523 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1524 | { | |
8a57dfc6 CS |
1525 | int r = bdi_bits; |
1526 | struct mapped_device *md = congested_data; | |
1527 | struct dm_table *map; | |
1da177e4 | 1528 | |
1eb787ec | 1529 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1530 | if (dm_request_based(md)) { |
cec47e3d | 1531 | /* |
e522c039 MS |
1532 | * With request-based DM we only need to check the |
1533 | * top-level queue for congestion. | |
cec47e3d | 1534 | */ |
dc3b17cc | 1535 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1536 | } else { |
1537 | map = dm_get_live_table_fast(md); | |
1538 | if (map) | |
cec47e3d | 1539 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1540 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1541 | } |
1542 | } | |
1543 | ||
1da177e4 LT |
1544 | return r; |
1545 | } | |
1546 | ||
1547 | /*----------------------------------------------------------------- | |
1548 | * An IDR is used to keep track of allocated minor numbers. | |
1549 | *---------------------------------------------------------------*/ | |
2b06cfff | 1550 | static void free_minor(int minor) |
1da177e4 | 1551 | { |
f32c10b0 | 1552 | spin_lock(&_minor_lock); |
1da177e4 | 1553 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1554 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1555 | } |
1556 | ||
1557 | /* | |
1558 | * See if the device with a specific minor # is free. | |
1559 | */ | |
cf13ab8e | 1560 | static int specific_minor(int minor) |
1da177e4 | 1561 | { |
c9d76be6 | 1562 | int r; |
1da177e4 LT |
1563 | |
1564 | if (minor >= (1 << MINORBITS)) | |
1565 | return -EINVAL; | |
1566 | ||
c9d76be6 | 1567 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1568 | spin_lock(&_minor_lock); |
1da177e4 | 1569 | |
c9d76be6 | 1570 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1571 | |
f32c10b0 | 1572 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1573 | idr_preload_end(); |
1574 | if (r < 0) | |
1575 | return r == -ENOSPC ? -EBUSY : r; | |
1576 | return 0; | |
1da177e4 LT |
1577 | } |
1578 | ||
cf13ab8e | 1579 | static int next_free_minor(int *minor) |
1da177e4 | 1580 | { |
c9d76be6 | 1581 | int r; |
62f75c2f | 1582 | |
c9d76be6 | 1583 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1584 | spin_lock(&_minor_lock); |
1da177e4 | 1585 | |
c9d76be6 | 1586 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1587 | |
f32c10b0 | 1588 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1589 | idr_preload_end(); |
1590 | if (r < 0) | |
1591 | return r; | |
1592 | *minor = r; | |
1593 | return 0; | |
1da177e4 LT |
1594 | } |
1595 | ||
83d5cde4 | 1596 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1597 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1598 | |
53d5914f MP |
1599 | static void dm_wq_work(struct work_struct *work); |
1600 | ||
4cc96131 | 1601 | void dm_init_md_queue(struct mapped_device *md) |
4a0b4ddf | 1602 | { |
ad5f498f MP |
1603 | /* |
1604 | * Initialize data that will only be used by a non-blk-mq DM queue | |
1605 | * - must do so here (in alloc_dev callchain) before queue is used | |
1606 | */ | |
1607 | md->queue->queuedata = md; | |
dc3b17cc | 1608 | md->queue->backing_dev_info->congested_data = md; |
bfebd1cd | 1609 | } |
4a0b4ddf | 1610 | |
4cc96131 | 1611 | void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1612 | { |
17e149b8 | 1613 | md->use_blk_mq = false; |
bfebd1cd MS |
1614 | dm_init_md_queue(md); |
1615 | ||
1616 | /* | |
1617 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1618 | */ | |
dc3b17cc | 1619 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf MS |
1620 | } |
1621 | ||
0f20972f MS |
1622 | static void cleanup_mapped_device(struct mapped_device *md) |
1623 | { | |
0f20972f MS |
1624 | if (md->wq) |
1625 | destroy_workqueue(md->wq); | |
1626 | if (md->kworker_task) | |
1627 | kthread_stop(md->kworker_task); | |
6f65985e | 1628 | mempool_destroy(md->io_pool); |
0f20972f MS |
1629 | if (md->bs) |
1630 | bioset_free(md->bs); | |
1631 | ||
f26c5719 DW |
1632 | if (md->dax_dev) { |
1633 | kill_dax(md->dax_dev); | |
1634 | put_dax(md->dax_dev); | |
1635 | md->dax_dev = NULL; | |
1636 | } | |
1637 | ||
0f20972f MS |
1638 | if (md->disk) { |
1639 | spin_lock(&_minor_lock); | |
1640 | md->disk->private_data = NULL; | |
1641 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1642 | del_gendisk(md->disk); |
1643 | put_disk(md->disk); | |
1644 | } | |
1645 | ||
1646 | if (md->queue) | |
1647 | blk_cleanup_queue(md->queue); | |
1648 | ||
d09960b0 TE |
1649 | cleanup_srcu_struct(&md->io_barrier); |
1650 | ||
0f20972f MS |
1651 | if (md->bdev) { |
1652 | bdput(md->bdev); | |
1653 | md->bdev = NULL; | |
1654 | } | |
4cc96131 MS |
1655 | |
1656 | dm_mq_cleanup_mapped_device(md); | |
0f20972f MS |
1657 | } |
1658 | ||
1da177e4 LT |
1659 | /* |
1660 | * Allocate and initialise a blank device with a given minor. | |
1661 | */ | |
2b06cfff | 1662 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1663 | { |
115485e8 | 1664 | int r, numa_node_id = dm_get_numa_node(); |
f26c5719 | 1665 | struct dax_device *dax_dev; |
115485e8 | 1666 | struct mapped_device *md; |
ba61fdd1 | 1667 | void *old_md; |
1da177e4 | 1668 | |
856eb091 | 1669 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1670 | if (!md) { |
1671 | DMWARN("unable to allocate device, out of memory."); | |
1672 | return NULL; | |
1673 | } | |
1674 | ||
10da4f79 | 1675 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1676 | goto bad_module_get; |
10da4f79 | 1677 | |
1da177e4 | 1678 | /* get a minor number for the dev */ |
2b06cfff | 1679 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1680 | r = next_free_minor(&minor); |
2b06cfff | 1681 | else |
cf13ab8e | 1682 | r = specific_minor(minor); |
1da177e4 | 1683 | if (r < 0) |
6ed7ade8 | 1684 | goto bad_minor; |
1da177e4 | 1685 | |
83d5e5b0 MP |
1686 | r = init_srcu_struct(&md->io_barrier); |
1687 | if (r < 0) | |
1688 | goto bad_io_barrier; | |
1689 | ||
115485e8 | 1690 | md->numa_node_id = numa_node_id; |
4cc96131 | 1691 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1692 | md->init_tio_pdu = false; |
a5664dad | 1693 | md->type = DM_TYPE_NONE; |
e61290a4 | 1694 | mutex_init(&md->suspend_lock); |
a5664dad | 1695 | mutex_init(&md->type_lock); |
86f1152b | 1696 | mutex_init(&md->table_devices_lock); |
022c2611 | 1697 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1698 | atomic_set(&md->holders, 1); |
5c6bd75d | 1699 | atomic_set(&md->open_count, 0); |
1da177e4 | 1700 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1701 | atomic_set(&md->uevent_seq, 0); |
1702 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1703 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1704 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1705 | |
115485e8 | 1706 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1707 | if (!md->queue) |
0f20972f | 1708 | goto bad; |
1da177e4 | 1709 | |
4a0b4ddf | 1710 | dm_init_md_queue(md); |
9faf400f | 1711 | |
115485e8 | 1712 | md->disk = alloc_disk_node(1, numa_node_id); |
1da177e4 | 1713 | if (!md->disk) |
0f20972f | 1714 | goto bad; |
1da177e4 | 1715 | |
316d315b NK |
1716 | atomic_set(&md->pending[0], 0); |
1717 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1718 | init_waitqueue_head(&md->wait); |
53d5914f | 1719 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1720 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1721 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1722 | md->kworker_task = NULL; |
f0b04115 | 1723 | |
1da177e4 LT |
1724 | md->disk->major = _major; |
1725 | md->disk->first_minor = minor; | |
1726 | md->disk->fops = &dm_blk_dops; | |
1727 | md->disk->queue = md->queue; | |
1728 | md->disk->private_data = md; | |
1729 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 DW |
1730 | |
1731 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); | |
1732 | if (!dax_dev) | |
1733 | goto bad; | |
1734 | md->dax_dev = dax_dev; | |
1735 | ||
1da177e4 | 1736 | add_disk(md->disk); |
7e51f257 | 1737 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1738 | |
670368a8 | 1739 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1740 | if (!md->wq) |
0f20972f | 1741 | goto bad; |
304f3f6a | 1742 | |
32a926da MP |
1743 | md->bdev = bdget_disk(md->disk, 0); |
1744 | if (!md->bdev) | |
0f20972f | 1745 | goto bad; |
32a926da | 1746 | |
3a83f467 | 1747 | bio_init(&md->flush_bio, NULL, 0); |
74d46992 | 1748 | bio_set_dev(&md->flush_bio, md->bdev); |
ff0361b3 | 1749 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; |
6a8736d1 | 1750 | |
fd2ed4d2 MP |
1751 | dm_stats_init(&md->stats); |
1752 | ||
ba61fdd1 | 1753 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1754 | spin_lock(&_minor_lock); |
ba61fdd1 | 1755 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1756 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1757 | |
1758 | BUG_ON(old_md != MINOR_ALLOCED); | |
1759 | ||
1da177e4 LT |
1760 | return md; |
1761 | ||
0f20972f MS |
1762 | bad: |
1763 | cleanup_mapped_device(md); | |
83d5e5b0 | 1764 | bad_io_barrier: |
1da177e4 | 1765 | free_minor(minor); |
6ed7ade8 | 1766 | bad_minor: |
10da4f79 | 1767 | module_put(THIS_MODULE); |
6ed7ade8 | 1768 | bad_module_get: |
856eb091 | 1769 | kvfree(md); |
1da177e4 LT |
1770 | return NULL; |
1771 | } | |
1772 | ||
ae9da83f JN |
1773 | static void unlock_fs(struct mapped_device *md); |
1774 | ||
1da177e4 LT |
1775 | static void free_dev(struct mapped_device *md) |
1776 | { | |
f331c029 | 1777 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1778 | |
32a926da | 1779 | unlock_fs(md); |
2eb6e1e3 | 1780 | |
0f20972f | 1781 | cleanup_mapped_device(md); |
63a4f065 | 1782 | |
86f1152b | 1783 | free_table_devices(&md->table_devices); |
63a4f065 | 1784 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1785 | free_minor(minor); |
1786 | ||
10da4f79 | 1787 | module_put(THIS_MODULE); |
856eb091 | 1788 | kvfree(md); |
1da177e4 LT |
1789 | } |
1790 | ||
e6ee8c0b KU |
1791 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1792 | { | |
c0820cf5 | 1793 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1794 | |
0776aa0e MS |
1795 | if (dm_table_bio_based(t)) { |
1796 | /* The md may already have mempools that need changing. */ | |
1797 | if (md->bs) { | |
16245bdc JN |
1798 | /* |
1799 | * Reload bioset because front_pad may have changed | |
1800 | * because a different table was loaded. | |
1801 | */ | |
1802 | bioset_free(md->bs); | |
0776aa0e | 1803 | md->bs = NULL; |
16245bdc | 1804 | } |
0776aa0e MS |
1805 | if (md->io_pool) { |
1806 | /* | |
1807 | * Reload io_pool because pool_size may have changed | |
1808 | * because a different table was loaded. | |
1809 | */ | |
1810 | mempool_destroy(md->io_pool); | |
1811 | md->io_pool = NULL; | |
1812 | } | |
1813 | ||
1814 | } else if (md->bs) { | |
4e6e36c3 MS |
1815 | /* |
1816 | * There's no need to reload with request-based dm | |
1817 | * because the size of front_pad doesn't change. | |
1818 | * Note for future: If you are to reload bioset, | |
1819 | * prep-ed requests in the queue may refer | |
1820 | * to bio from the old bioset, so you must walk | |
1821 | * through the queue to unprep. | |
1822 | */ | |
1823 | goto out; | |
c0820cf5 | 1824 | } |
e6ee8c0b | 1825 | |
eb8db831 | 1826 | BUG_ON(!p || md->io_pool || md->bs); |
cbc4e3c1 | 1827 | |
e6ee8c0b KU |
1828 | md->io_pool = p->io_pool; |
1829 | p->io_pool = NULL; | |
e6ee8c0b KU |
1830 | md->bs = p->bs; |
1831 | p->bs = NULL; | |
e6ee8c0b | 1832 | out: |
02233342 | 1833 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1834 | dm_table_free_md_mempools(t); |
1835 | } | |
1836 | ||
1da177e4 LT |
1837 | /* |
1838 | * Bind a table to the device. | |
1839 | */ | |
1840 | static void event_callback(void *context) | |
1841 | { | |
7a8c3d3b MA |
1842 | unsigned long flags; |
1843 | LIST_HEAD(uevents); | |
1da177e4 LT |
1844 | struct mapped_device *md = (struct mapped_device *) context; |
1845 | ||
7a8c3d3b MA |
1846 | spin_lock_irqsave(&md->uevent_lock, flags); |
1847 | list_splice_init(&md->uevent_list, &uevents); | |
1848 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1849 | ||
ed9e1982 | 1850 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1851 | |
1da177e4 LT |
1852 | atomic_inc(&md->event_nr); |
1853 | wake_up(&md->eventq); | |
62e08243 | 1854 | dm_issue_global_event(); |
1da177e4 LT |
1855 | } |
1856 | ||
c217649b MS |
1857 | /* |
1858 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1859 | */ | |
4e90188b | 1860 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1861 | { |
1ea0654e BVA |
1862 | lockdep_assert_held(&md->suspend_lock); |
1863 | ||
4e90188b | 1864 | set_capacity(md->disk, size); |
1da177e4 | 1865 | |
db8fef4f | 1866 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
1867 | } |
1868 | ||
042d2a9b AK |
1869 | /* |
1870 | * Returns old map, which caller must destroy. | |
1871 | */ | |
1872 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
1873 | struct queue_limits *limits) | |
1da177e4 | 1874 | { |
042d2a9b | 1875 | struct dm_table *old_map; |
165125e1 | 1876 | struct request_queue *q = md->queue; |
1da177e4 LT |
1877 | sector_t size; |
1878 | ||
5a8f1f80 BVA |
1879 | lockdep_assert_held(&md->suspend_lock); |
1880 | ||
1da177e4 | 1881 | size = dm_table_get_size(t); |
3ac51e74 DW |
1882 | |
1883 | /* | |
1884 | * Wipe any geometry if the size of the table changed. | |
1885 | */ | |
fd2ed4d2 | 1886 | if (size != dm_get_size(md)) |
3ac51e74 DW |
1887 | memset(&md->geometry, 0, sizeof(md->geometry)); |
1888 | ||
32a926da | 1889 | __set_size(md, size); |
d5816876 | 1890 | |
2ca3310e AK |
1891 | dm_table_event_callback(t, event_callback, md); |
1892 | ||
e6ee8c0b KU |
1893 | /* |
1894 | * The queue hasn't been stopped yet, if the old table type wasn't | |
1895 | * for request-based during suspension. So stop it to prevent | |
1896 | * I/O mapping before resume. | |
1897 | * This must be done before setting the queue restrictions, | |
1898 | * because request-based dm may be run just after the setting. | |
1899 | */ | |
16f12266 | 1900 | if (dm_table_request_based(t)) { |
eca7ee6d | 1901 | dm_stop_queue(q); |
16f12266 MS |
1902 | /* |
1903 | * Leverage the fact that request-based DM targets are | |
1904 | * immutable singletons and establish md->immutable_target | |
1905 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
1906 | */ | |
1907 | md->immutable_target = dm_table_get_immutable_target(t); | |
1908 | } | |
e6ee8c0b KU |
1909 | |
1910 | __bind_mempools(md, t); | |
1911 | ||
a12f5d48 | 1912 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 1913 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
1914 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
1915 | ||
754c5fc7 | 1916 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
1917 | if (old_map) |
1918 | dm_sync_table(md); | |
1da177e4 | 1919 | |
042d2a9b | 1920 | return old_map; |
1da177e4 LT |
1921 | } |
1922 | ||
a7940155 AK |
1923 | /* |
1924 | * Returns unbound table for the caller to free. | |
1925 | */ | |
1926 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 1927 | { |
a12f5d48 | 1928 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
1929 | |
1930 | if (!map) | |
a7940155 | 1931 | return NULL; |
1da177e4 LT |
1932 | |
1933 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 1934 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 1935 | dm_sync_table(md); |
a7940155 AK |
1936 | |
1937 | return map; | |
1da177e4 LT |
1938 | } |
1939 | ||
1940 | /* | |
1941 | * Constructor for a new device. | |
1942 | */ | |
2b06cfff | 1943 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1944 | { |
1945 | struct mapped_device *md; | |
1946 | ||
2b06cfff | 1947 | md = alloc_dev(minor); |
1da177e4 LT |
1948 | if (!md) |
1949 | return -ENXIO; | |
1950 | ||
784aae73 MB |
1951 | dm_sysfs_init(md); |
1952 | ||
1da177e4 LT |
1953 | *result = md; |
1954 | return 0; | |
1955 | } | |
1956 | ||
a5664dad MS |
1957 | /* |
1958 | * Functions to manage md->type. | |
1959 | * All are required to hold md->type_lock. | |
1960 | */ | |
1961 | void dm_lock_md_type(struct mapped_device *md) | |
1962 | { | |
1963 | mutex_lock(&md->type_lock); | |
1964 | } | |
1965 | ||
1966 | void dm_unlock_md_type(struct mapped_device *md) | |
1967 | { | |
1968 | mutex_unlock(&md->type_lock); | |
1969 | } | |
1970 | ||
7e0d574f | 1971 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 1972 | { |
00c4fc3b | 1973 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
1974 | md->type = type; |
1975 | } | |
1976 | ||
7e0d574f | 1977 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
1978 | { |
1979 | return md->type; | |
1980 | } | |
1981 | ||
36a0456f AK |
1982 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
1983 | { | |
1984 | return md->immutable_target_type; | |
1985 | } | |
1986 | ||
f84cb8a4 MS |
1987 | /* |
1988 | * The queue_limits are only valid as long as you have a reference | |
1989 | * count on 'md'. | |
1990 | */ | |
1991 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
1992 | { | |
1993 | BUG_ON(!atomic_read(&md->holders)); | |
1994 | return &md->queue->limits; | |
1995 | } | |
1996 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
1997 | ||
4a0b4ddf MS |
1998 | /* |
1999 | * Setup the DM device's queue based on md's type | |
2000 | */ | |
591ddcfc | 2001 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2002 | { |
bfebd1cd | 2003 | int r; |
7e0d574f | 2004 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2005 | |
545ed20e | 2006 | switch (type) { |
bfebd1cd | 2007 | case DM_TYPE_REQUEST_BASED: |
eb8db831 | 2008 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 2009 | if (r) { |
eca7ee6d | 2010 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2011 | return r; |
ff36ab34 | 2012 | } |
bfebd1cd MS |
2013 | break; |
2014 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 2015 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2016 | if (r) { |
eca7ee6d | 2017 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2018 | return r; |
2019 | } | |
2020 | break; | |
2021 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2022 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 2023 | dm_init_normal_md_queue(md); |
ff36ab34 | 2024 | blk_queue_make_request(md->queue, dm_make_request); |
bfebd1cd | 2025 | break; |
7e0d574f BVA |
2026 | case DM_TYPE_NONE: |
2027 | WARN_ON_ONCE(true); | |
2028 | break; | |
4a0b4ddf MS |
2029 | } |
2030 | ||
2031 | return 0; | |
2032 | } | |
2033 | ||
2bec1f4a | 2034 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2035 | { |
2036 | struct mapped_device *md; | |
1da177e4 LT |
2037 | unsigned minor = MINOR(dev); |
2038 | ||
2039 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2040 | return NULL; | |
2041 | ||
f32c10b0 | 2042 | spin_lock(&_minor_lock); |
1da177e4 LT |
2043 | |
2044 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2045 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2046 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2047 | md = NULL; | |
2048 | goto out; | |
fba9f90e | 2049 | } |
49de5769 | 2050 | dm_get(md); |
fba9f90e | 2051 | out: |
f32c10b0 | 2052 | spin_unlock(&_minor_lock); |
1da177e4 | 2053 | |
637842cf DT |
2054 | return md; |
2055 | } | |
3cf2e4ba | 2056 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2057 | |
9ade92a9 | 2058 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2059 | { |
9ade92a9 | 2060 | return md->interface_ptr; |
1da177e4 LT |
2061 | } |
2062 | ||
2063 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2064 | { | |
2065 | md->interface_ptr = ptr; | |
2066 | } | |
2067 | ||
2068 | void dm_get(struct mapped_device *md) | |
2069 | { | |
2070 | atomic_inc(&md->holders); | |
3f77316d | 2071 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2072 | } |
2073 | ||
09ee96b2 MP |
2074 | int dm_hold(struct mapped_device *md) |
2075 | { | |
2076 | spin_lock(&_minor_lock); | |
2077 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2078 | spin_unlock(&_minor_lock); | |
2079 | return -EBUSY; | |
2080 | } | |
2081 | dm_get(md); | |
2082 | spin_unlock(&_minor_lock); | |
2083 | return 0; | |
2084 | } | |
2085 | EXPORT_SYMBOL_GPL(dm_hold); | |
2086 | ||
72d94861 AK |
2087 | const char *dm_device_name(struct mapped_device *md) |
2088 | { | |
2089 | return md->name; | |
2090 | } | |
2091 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2092 | ||
3f77316d | 2093 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2094 | { |
3b785fbc | 2095 | struct request_queue *q = dm_get_md_queue(md); |
1134e5ae | 2096 | struct dm_table *map; |
83d5e5b0 | 2097 | int srcu_idx; |
1da177e4 | 2098 | |
3f77316d | 2099 | might_sleep(); |
fba9f90e | 2100 | |
63a4f065 | 2101 | spin_lock(&_minor_lock); |
3f77316d KU |
2102 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2103 | set_bit(DMF_FREEING, &md->flags); | |
2104 | spin_unlock(&_minor_lock); | |
3b785fbc | 2105 | |
2e91c369 | 2106 | blk_set_queue_dying(q); |
3f77316d | 2107 | |
02233342 | 2108 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 2109 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2110 | |
ab7c7bb6 MP |
2111 | /* |
2112 | * Take suspend_lock so that presuspend and postsuspend methods | |
2113 | * do not race with internal suspend. | |
2114 | */ | |
2115 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2116 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2117 | if (!dm_suspended_md(md)) { |
2118 | dm_table_presuspend_targets(map); | |
2119 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2120 | } |
83d5e5b0 MP |
2121 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2122 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2123 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2124 | |
3f77316d KU |
2125 | /* |
2126 | * Rare, but there may be I/O requests still going to complete, | |
2127 | * for example. Wait for all references to disappear. | |
2128 | * No one should increment the reference count of the mapped_device, | |
2129 | * after the mapped_device state becomes DMF_FREEING. | |
2130 | */ | |
2131 | if (wait) | |
2132 | while (atomic_read(&md->holders)) | |
2133 | msleep(1); | |
2134 | else if (atomic_read(&md->holders)) | |
2135 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2136 | dm_device_name(md), atomic_read(&md->holders)); | |
2137 | ||
2138 | dm_sysfs_exit(md); | |
3f77316d KU |
2139 | dm_table_destroy(__unbind(md)); |
2140 | free_dev(md); | |
2141 | } | |
2142 | ||
2143 | void dm_destroy(struct mapped_device *md) | |
2144 | { | |
2145 | __dm_destroy(md, true); | |
2146 | } | |
2147 | ||
2148 | void dm_destroy_immediate(struct mapped_device *md) | |
2149 | { | |
2150 | __dm_destroy(md, false); | |
2151 | } | |
2152 | ||
2153 | void dm_put(struct mapped_device *md) | |
2154 | { | |
2155 | atomic_dec(&md->holders); | |
1da177e4 | 2156 | } |
79eb885c | 2157 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2158 | |
b48633f8 | 2159 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2160 | { |
2161 | int r = 0; | |
9f4c3f87 | 2162 | DEFINE_WAIT(wait); |
46125c1c MB |
2163 | |
2164 | while (1) { | |
9f4c3f87 | 2165 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2166 | |
b4324fee | 2167 | if (!md_in_flight(md)) |
46125c1c MB |
2168 | break; |
2169 | ||
e3fabdfd | 2170 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2171 | r = -EINTR; |
2172 | break; | |
2173 | } | |
2174 | ||
2175 | io_schedule(); | |
2176 | } | |
9f4c3f87 | 2177 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2178 | |
46125c1c MB |
2179 | return r; |
2180 | } | |
2181 | ||
1da177e4 LT |
2182 | /* |
2183 | * Process the deferred bios | |
2184 | */ | |
ef208587 | 2185 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2186 | { |
ef208587 MP |
2187 | struct mapped_device *md = container_of(work, struct mapped_device, |
2188 | work); | |
6d6f10df | 2189 | struct bio *c; |
83d5e5b0 MP |
2190 | int srcu_idx; |
2191 | struct dm_table *map; | |
1da177e4 | 2192 | |
83d5e5b0 | 2193 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2194 | |
3b00b203 | 2195 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2196 | spin_lock_irq(&md->deferred_lock); |
2197 | c = bio_list_pop(&md->deferred); | |
2198 | spin_unlock_irq(&md->deferred_lock); | |
2199 | ||
6a8736d1 | 2200 | if (!c) |
df12ee99 | 2201 | break; |
022c2611 | 2202 | |
e6ee8c0b KU |
2203 | if (dm_request_based(md)) |
2204 | generic_make_request(c); | |
6a8736d1 | 2205 | else |
83d5e5b0 | 2206 | __split_and_process_bio(md, map, c); |
022c2611 | 2207 | } |
73d410c0 | 2208 | |
83d5e5b0 | 2209 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2210 | } |
2211 | ||
9a1fb464 | 2212 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2213 | { |
3b00b203 | 2214 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2215 | smp_mb__after_atomic(); |
53d5914f | 2216 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2217 | } |
2218 | ||
1da177e4 | 2219 | /* |
042d2a9b | 2220 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2221 | */ |
042d2a9b | 2222 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2223 | { |
87eb5b21 | 2224 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2225 | struct queue_limits limits; |
042d2a9b | 2226 | int r; |
1da177e4 | 2227 | |
e61290a4 | 2228 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2229 | |
2230 | /* device must be suspended */ | |
4f186f8b | 2231 | if (!dm_suspended_md(md)) |
93c534ae | 2232 | goto out; |
1da177e4 | 2233 | |
3ae70656 MS |
2234 | /* |
2235 | * If the new table has no data devices, retain the existing limits. | |
2236 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2237 | * then new I/O is queued based on these limits, and then some paths | |
2238 | * reappear. | |
2239 | */ | |
2240 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2241 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2242 | if (live_map) |
2243 | limits = md->queue->limits; | |
83d5e5b0 | 2244 | dm_put_live_table_fast(md); |
3ae70656 MS |
2245 | } |
2246 | ||
87eb5b21 MC |
2247 | if (!live_map) { |
2248 | r = dm_calculate_queue_limits(table, &limits); | |
2249 | if (r) { | |
2250 | map = ERR_PTR(r); | |
2251 | goto out; | |
2252 | } | |
042d2a9b | 2253 | } |
754c5fc7 | 2254 | |
042d2a9b | 2255 | map = __bind(md, table, &limits); |
62e08243 | 2256 | dm_issue_global_event(); |
1da177e4 | 2257 | |
93c534ae | 2258 | out: |
e61290a4 | 2259 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2260 | return map; |
1da177e4 LT |
2261 | } |
2262 | ||
2263 | /* | |
2264 | * Functions to lock and unlock any filesystem running on the | |
2265 | * device. | |
2266 | */ | |
2ca3310e | 2267 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2268 | { |
e39e2e95 | 2269 | int r; |
1da177e4 LT |
2270 | |
2271 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2272 | |
db8fef4f | 2273 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2274 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2275 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2276 | md->frozen_sb = NULL; |
2277 | return r; | |
dfbe03f6 AK |
2278 | } |
2279 | ||
aa8d7c2f AK |
2280 | set_bit(DMF_FROZEN, &md->flags); |
2281 | ||
1da177e4 LT |
2282 | return 0; |
2283 | } | |
2284 | ||
2ca3310e | 2285 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2286 | { |
aa8d7c2f AK |
2287 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2288 | return; | |
2289 | ||
db8fef4f | 2290 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2291 | md->frozen_sb = NULL; |
aa8d7c2f | 2292 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2293 | } |
2294 | ||
2295 | /* | |
b48633f8 BVA |
2296 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2297 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2298 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2299 | * | |
ffcc3936 MS |
2300 | * If __dm_suspend returns 0, the device is completely quiescent |
2301 | * now. There is no request-processing activity. All new requests | |
2302 | * are being added to md->deferred list. | |
cec47e3d | 2303 | */ |
ffcc3936 | 2304 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2305 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2306 | int dmf_suspended_flag) |
1da177e4 | 2307 | { |
ffcc3936 MS |
2308 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2309 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2310 | int r; | |
1da177e4 | 2311 | |
5a8f1f80 BVA |
2312 | lockdep_assert_held(&md->suspend_lock); |
2313 | ||
2e93ccc1 KU |
2314 | /* |
2315 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2316 | * This flag is cleared before dm_suspend returns. | |
2317 | */ | |
2318 | if (noflush) | |
2319 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 BVA |
2320 | else |
2321 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); | |
2e93ccc1 | 2322 | |
d67ee213 MS |
2323 | /* |
2324 | * This gets reverted if there's an error later and the targets | |
2325 | * provide the .presuspend_undo hook. | |
2326 | */ | |
cf222b37 AK |
2327 | dm_table_presuspend_targets(map); |
2328 | ||
32a926da | 2329 | /* |
9f518b27 KU |
2330 | * Flush I/O to the device. |
2331 | * Any I/O submitted after lock_fs() may not be flushed. | |
2332 | * noflush takes precedence over do_lockfs. | |
2333 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2334 | */ |
2335 | if (!noflush && do_lockfs) { | |
2336 | r = lock_fs(md); | |
d67ee213 MS |
2337 | if (r) { |
2338 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2339 | return r; |
d67ee213 | 2340 | } |
aa8d7c2f | 2341 | } |
1da177e4 LT |
2342 | |
2343 | /* | |
3b00b203 MP |
2344 | * Here we must make sure that no processes are submitting requests |
2345 | * to target drivers i.e. no one may be executing | |
2346 | * __split_and_process_bio. This is called from dm_request and | |
2347 | * dm_wq_work. | |
2348 | * | |
2349 | * To get all processes out of __split_and_process_bio in dm_request, | |
2350 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2351 | * __split_and_process_bio from dm_request and quiesce the thread |
2352 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2353 | * flush_workqueue(md->wq). | |
1da177e4 | 2354 | */ |
1eb787ec | 2355 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2356 | if (map) |
2357 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2358 | |
d0bcb878 | 2359 | /* |
29e4013d TH |
2360 | * Stop md->queue before flushing md->wq in case request-based |
2361 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2362 | */ |
2eb6e1e3 | 2363 | if (dm_request_based(md)) { |
eca7ee6d | 2364 | dm_stop_queue(md->queue); |
02233342 | 2365 | if (md->kworker_task) |
3989144f | 2366 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2367 | } |
cec47e3d | 2368 | |
d0bcb878 KU |
2369 | flush_workqueue(md->wq); |
2370 | ||
1da177e4 | 2371 | /* |
3b00b203 MP |
2372 | * At this point no more requests are entering target request routines. |
2373 | * We call dm_wait_for_completion to wait for all existing requests | |
2374 | * to finish. | |
1da177e4 | 2375 | */ |
b48633f8 | 2376 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2377 | if (!r) |
2378 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2379 | |
6d6f10df | 2380 | if (noflush) |
022c2611 | 2381 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2382 | if (map) |
2383 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2384 | |
1da177e4 | 2385 | /* were we interrupted ? */ |
46125c1c | 2386 | if (r < 0) { |
9a1fb464 | 2387 | dm_queue_flush(md); |
73d410c0 | 2388 | |
cec47e3d | 2389 | if (dm_request_based(md)) |
eca7ee6d | 2390 | dm_start_queue(md->queue); |
cec47e3d | 2391 | |
2ca3310e | 2392 | unlock_fs(md); |
d67ee213 | 2393 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2394 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2395 | } |
1da177e4 | 2396 | |
ffcc3936 MS |
2397 | return r; |
2398 | } | |
2399 | ||
2400 | /* | |
2401 | * We need to be able to change a mapping table under a mounted | |
2402 | * filesystem. For example we might want to move some data in | |
2403 | * the background. Before the table can be swapped with | |
2404 | * dm_bind_table, dm_suspend must be called to flush any in | |
2405 | * flight bios and ensure that any further io gets deferred. | |
2406 | */ | |
2407 | /* | |
2408 | * Suspend mechanism in request-based dm. | |
2409 | * | |
2410 | * 1. Flush all I/Os by lock_fs() if needed. | |
2411 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2412 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2413 | * | |
2414 | * To abort suspend, start the request_queue. | |
2415 | */ | |
2416 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2417 | { | |
2418 | struct dm_table *map = NULL; | |
2419 | int r = 0; | |
2420 | ||
2421 | retry: | |
2422 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2423 | ||
2424 | if (dm_suspended_md(md)) { | |
2425 | r = -EINVAL; | |
2426 | goto out_unlock; | |
2427 | } | |
2428 | ||
2429 | if (dm_suspended_internally_md(md)) { | |
2430 | /* already internally suspended, wait for internal resume */ | |
2431 | mutex_unlock(&md->suspend_lock); | |
2432 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2433 | if (r) | |
2434 | return r; | |
2435 | goto retry; | |
2436 | } | |
2437 | ||
a12f5d48 | 2438 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2439 | |
eaf9a736 | 2440 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2441 | if (r) |
2442 | goto out_unlock; | |
3b00b203 | 2443 | |
4d4471cb KU |
2444 | dm_table_postsuspend_targets(map); |
2445 | ||
d287483d | 2446 | out_unlock: |
e61290a4 | 2447 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2448 | return r; |
1da177e4 LT |
2449 | } |
2450 | ||
ffcc3936 MS |
2451 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2452 | { | |
2453 | if (map) { | |
2454 | int r = dm_table_resume_targets(map); | |
2455 | if (r) | |
2456 | return r; | |
2457 | } | |
2458 | ||
2459 | dm_queue_flush(md); | |
2460 | ||
2461 | /* | |
2462 | * Flushing deferred I/Os must be done after targets are resumed | |
2463 | * so that mapping of targets can work correctly. | |
2464 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2465 | */ | |
2466 | if (dm_request_based(md)) | |
eca7ee6d | 2467 | dm_start_queue(md->queue); |
ffcc3936 MS |
2468 | |
2469 | unlock_fs(md); | |
2470 | ||
2471 | return 0; | |
2472 | } | |
2473 | ||
1da177e4 LT |
2474 | int dm_resume(struct mapped_device *md) |
2475 | { | |
8dc23658 | 2476 | int r; |
cf222b37 | 2477 | struct dm_table *map = NULL; |
1da177e4 | 2478 | |
ffcc3936 | 2479 | retry: |
8dc23658 | 2480 | r = -EINVAL; |
ffcc3936 MS |
2481 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2482 | ||
4f186f8b | 2483 | if (!dm_suspended_md(md)) |
cf222b37 | 2484 | goto out; |
cf222b37 | 2485 | |
ffcc3936 MS |
2486 | if (dm_suspended_internally_md(md)) { |
2487 | /* already internally suspended, wait for internal resume */ | |
2488 | mutex_unlock(&md->suspend_lock); | |
2489 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2490 | if (r) | |
2491 | return r; | |
2492 | goto retry; | |
2493 | } | |
2494 | ||
a12f5d48 | 2495 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2496 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2497 | goto out; |
1da177e4 | 2498 | |
ffcc3936 | 2499 | r = __dm_resume(md, map); |
8757b776 MB |
2500 | if (r) |
2501 | goto out; | |
2ca3310e | 2502 | |
2ca3310e | 2503 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2504 | out: |
e61290a4 | 2505 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2506 | |
cf222b37 | 2507 | return r; |
1da177e4 LT |
2508 | } |
2509 | ||
fd2ed4d2 MP |
2510 | /* |
2511 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2512 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2513 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2514 | */ |
2515 | ||
ffcc3936 | 2516 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2517 | { |
ffcc3936 MS |
2518 | struct dm_table *map = NULL; |
2519 | ||
1ea0654e BVA |
2520 | lockdep_assert_held(&md->suspend_lock); |
2521 | ||
96b26c8c | 2522 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2523 | return; /* nested internal suspend */ |
2524 | ||
2525 | if (dm_suspended_md(md)) { | |
2526 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2527 | return; /* nest suspend */ | |
2528 | } | |
2529 | ||
a12f5d48 | 2530 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2531 | |
2532 | /* | |
2533 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2534 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2535 | * would require changing .presuspend to return an error -- avoid this | |
2536 | * until there is a need for more elaborate variants of internal suspend. | |
2537 | */ | |
eaf9a736 MS |
2538 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2539 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2540 | |
2541 | dm_table_postsuspend_targets(map); | |
2542 | } | |
2543 | ||
2544 | static void __dm_internal_resume(struct mapped_device *md) | |
2545 | { | |
96b26c8c MP |
2546 | BUG_ON(!md->internal_suspend_count); |
2547 | ||
2548 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2549 | return; /* resume from nested internal suspend */ |
2550 | ||
fd2ed4d2 | 2551 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2552 | goto done; /* resume from nested suspend */ |
2553 | ||
2554 | /* | |
2555 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2556 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2557 | */ | |
2558 | (void) __dm_resume(md, NULL); | |
2559 | ||
2560 | done: | |
2561 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2562 | smp_mb__after_atomic(); | |
2563 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2564 | } | |
2565 | ||
2566 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2567 | { | |
2568 | mutex_lock(&md->suspend_lock); | |
2569 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2570 | mutex_unlock(&md->suspend_lock); | |
2571 | } | |
2572 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2573 | ||
2574 | void dm_internal_resume(struct mapped_device *md) | |
2575 | { | |
2576 | mutex_lock(&md->suspend_lock); | |
2577 | __dm_internal_resume(md); | |
2578 | mutex_unlock(&md->suspend_lock); | |
2579 | } | |
2580 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2581 | ||
2582 | /* | |
2583 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2584 | * which prevents interaction with userspace-driven suspend. | |
2585 | */ | |
2586 | ||
2587 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2588 | { | |
2589 | mutex_lock(&md->suspend_lock); | |
2590 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2591 | return; |
2592 | ||
2593 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2594 | synchronize_srcu(&md->io_barrier); | |
2595 | flush_workqueue(md->wq); | |
2596 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2597 | } | |
b735fede | 2598 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2599 | |
ffcc3936 | 2600 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2601 | { |
ffcc3936 | 2602 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2603 | goto done; |
2604 | ||
2605 | dm_queue_flush(md); | |
2606 | ||
2607 | done: | |
2608 | mutex_unlock(&md->suspend_lock); | |
2609 | } | |
b735fede | 2610 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2611 | |
1da177e4 LT |
2612 | /*----------------------------------------------------------------- |
2613 | * Event notification. | |
2614 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2615 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2616 | unsigned cookie) |
69267a30 | 2617 | { |
60935eb2 MB |
2618 | char udev_cookie[DM_COOKIE_LENGTH]; |
2619 | char *envp[] = { udev_cookie, NULL }; | |
2620 | ||
2621 | if (!cookie) | |
3abf85b5 | 2622 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2623 | else { |
2624 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2625 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2626 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2627 | action, envp); | |
60935eb2 | 2628 | } |
69267a30 AK |
2629 | } |
2630 | ||
7a8c3d3b MA |
2631 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2632 | { | |
2633 | return atomic_add_return(1, &md->uevent_seq); | |
2634 | } | |
2635 | ||
1da177e4 LT |
2636 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2637 | { | |
2638 | return atomic_read(&md->event_nr); | |
2639 | } | |
2640 | ||
2641 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2642 | { | |
2643 | return wait_event_interruptible(md->eventq, | |
2644 | (event_nr != atomic_read(&md->event_nr))); | |
2645 | } | |
2646 | ||
7a8c3d3b MA |
2647 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2648 | { | |
2649 | unsigned long flags; | |
2650 | ||
2651 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2652 | list_add(elist, &md->uevent_list); | |
2653 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2654 | } | |
2655 | ||
1da177e4 LT |
2656 | /* |
2657 | * The gendisk is only valid as long as you have a reference | |
2658 | * count on 'md'. | |
2659 | */ | |
2660 | struct gendisk *dm_disk(struct mapped_device *md) | |
2661 | { | |
2662 | return md->disk; | |
2663 | } | |
65ff5b7d | 2664 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2665 | |
784aae73 MB |
2666 | struct kobject *dm_kobject(struct mapped_device *md) |
2667 | { | |
2995fa78 | 2668 | return &md->kobj_holder.kobj; |
784aae73 MB |
2669 | } |
2670 | ||
784aae73 MB |
2671 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2672 | { | |
2673 | struct mapped_device *md; | |
2674 | ||
2995fa78 | 2675 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2676 | |
b9a41d21 HT |
2677 | spin_lock(&_minor_lock); |
2678 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2679 | md = NULL; | |
2680 | goto out; | |
2681 | } | |
784aae73 | 2682 | dm_get(md); |
b9a41d21 HT |
2683 | out: |
2684 | spin_unlock(&_minor_lock); | |
2685 | ||
784aae73 MB |
2686 | return md; |
2687 | } | |
2688 | ||
4f186f8b | 2689 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2690 | { |
2691 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2692 | } | |
2693 | ||
ffcc3936 MS |
2694 | int dm_suspended_internally_md(struct mapped_device *md) |
2695 | { | |
2696 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2697 | } | |
2698 | ||
2c140a24 MP |
2699 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2700 | { | |
2701 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2702 | } | |
2703 | ||
64dbce58 KU |
2704 | int dm_suspended(struct dm_target *ti) |
2705 | { | |
ecdb2e25 | 2706 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2707 | } |
2708 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2709 | ||
2e93ccc1 KU |
2710 | int dm_noflush_suspending(struct dm_target *ti) |
2711 | { | |
ecdb2e25 | 2712 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2713 | } |
2714 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2715 | ||
7e0d574f | 2716 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
0776aa0e MS |
2717 | unsigned integrity, unsigned per_io_data_size, |
2718 | unsigned min_pool_size) | |
e6ee8c0b | 2719 | { |
115485e8 | 2720 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2721 | unsigned int pool_size = 0; |
5f015204 | 2722 | unsigned int front_pad; |
e6ee8c0b KU |
2723 | |
2724 | if (!pools) | |
4e6e36c3 | 2725 | return NULL; |
e6ee8c0b | 2726 | |
78d8e58a MS |
2727 | switch (type) { |
2728 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2729 | case DM_TYPE_DAX_BIO_BASED: |
0776aa0e | 2730 | pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); |
30187e1d | 2731 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
eb8db831 CH |
2732 | pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); |
2733 | if (!pools->io_pool) | |
2734 | goto out; | |
78d8e58a MS |
2735 | break; |
2736 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2737 | case DM_TYPE_MQ_REQUEST_BASED: |
0776aa0e | 2738 | pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); |
78d8e58a | 2739 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2740 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2741 | break; |
2742 | default: | |
2743 | BUG(); | |
2744 | } | |
2745 | ||
4a3f54d9 | 2746 | pools->bs = bioset_create(pool_size, front_pad, 0); |
e6ee8c0b | 2747 | if (!pools->bs) |
5f015204 | 2748 | goto out; |
e6ee8c0b | 2749 | |
a91a2785 | 2750 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2751 | goto out; |
a91a2785 | 2752 | |
e6ee8c0b | 2753 | return pools; |
5f1b670d | 2754 | |
5f1b670d CH |
2755 | out: |
2756 | dm_free_md_mempools(pools); | |
78d8e58a | 2757 | |
4e6e36c3 | 2758 | return NULL; |
e6ee8c0b KU |
2759 | } |
2760 | ||
2761 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2762 | { | |
2763 | if (!pools) | |
2764 | return; | |
2765 | ||
6f65985e | 2766 | mempool_destroy(pools->io_pool); |
1ae49ea2 | 2767 | |
e6ee8c0b KU |
2768 | if (pools->bs) |
2769 | bioset_free(pools->bs); | |
2770 | ||
2771 | kfree(pools); | |
2772 | } | |
2773 | ||
9c72bad1 CH |
2774 | struct dm_pr { |
2775 | u64 old_key; | |
2776 | u64 new_key; | |
2777 | u32 flags; | |
2778 | bool fail_early; | |
2779 | }; | |
2780 | ||
2781 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2782 | void *data) | |
71cdb697 CH |
2783 | { |
2784 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2785 | struct dm_table *table; |
2786 | struct dm_target *ti; | |
2787 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2788 | |
9c72bad1 CH |
2789 | table = dm_get_live_table(md, &srcu_idx); |
2790 | if (!table || !dm_table_get_size(table)) | |
2791 | goto out; | |
71cdb697 | 2792 | |
9c72bad1 CH |
2793 | /* We only support devices that have a single target */ |
2794 | if (dm_table_get_num_targets(table) != 1) | |
2795 | goto out; | |
2796 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2797 | |
9c72bad1 CH |
2798 | ret = -EINVAL; |
2799 | if (!ti->type->iterate_devices) | |
2800 | goto out; | |
2801 | ||
2802 | ret = ti->type->iterate_devices(ti, fn, data); | |
2803 | out: | |
2804 | dm_put_live_table(md, srcu_idx); | |
2805 | return ret; | |
2806 | } | |
2807 | ||
2808 | /* | |
2809 | * For register / unregister we need to manually call out to every path. | |
2810 | */ | |
2811 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
2812 | sector_t start, sector_t len, void *data) | |
2813 | { | |
2814 | struct dm_pr *pr = data; | |
2815 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
2816 | ||
2817 | if (!ops || !ops->pr_register) | |
2818 | return -EOPNOTSUPP; | |
2819 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
2820 | } | |
2821 | ||
2822 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
2823 | u32 flags) | |
2824 | { | |
2825 | struct dm_pr pr = { | |
2826 | .old_key = old_key, | |
2827 | .new_key = new_key, | |
2828 | .flags = flags, | |
2829 | .fail_early = true, | |
2830 | }; | |
2831 | int ret; | |
2832 | ||
2833 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
2834 | if (ret && new_key) { | |
2835 | /* unregister all paths if we failed to register any path */ | |
2836 | pr.old_key = new_key; | |
2837 | pr.new_key = 0; | |
2838 | pr.flags = 0; | |
2839 | pr.fail_early = false; | |
2840 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
2841 | } | |
2842 | ||
2843 | return ret; | |
71cdb697 CH |
2844 | } |
2845 | ||
2846 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 2847 | u32 flags) |
71cdb697 CH |
2848 | { |
2849 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2850 | const struct pr_ops *ops; | |
71cdb697 | 2851 | fmode_t mode; |
956a4025 | 2852 | int r; |
71cdb697 | 2853 | |
956a4025 | 2854 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2855 | if (r < 0) |
2856 | return r; | |
2857 | ||
2858 | ops = bdev->bd_disk->fops->pr_ops; | |
2859 | if (ops && ops->pr_reserve) | |
2860 | r = ops->pr_reserve(bdev, key, type, flags); | |
2861 | else | |
2862 | r = -EOPNOTSUPP; | |
2863 | ||
956a4025 | 2864 | bdput(bdev); |
71cdb697 CH |
2865 | return r; |
2866 | } | |
2867 | ||
2868 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
2869 | { | |
2870 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2871 | const struct pr_ops *ops; | |
71cdb697 | 2872 | fmode_t mode; |
956a4025 | 2873 | int r; |
71cdb697 | 2874 | |
956a4025 | 2875 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2876 | if (r < 0) |
2877 | return r; | |
2878 | ||
2879 | ops = bdev->bd_disk->fops->pr_ops; | |
2880 | if (ops && ops->pr_release) | |
2881 | r = ops->pr_release(bdev, key, type); | |
2882 | else | |
2883 | r = -EOPNOTSUPP; | |
2884 | ||
956a4025 | 2885 | bdput(bdev); |
71cdb697 CH |
2886 | return r; |
2887 | } | |
2888 | ||
2889 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 2890 | enum pr_type type, bool abort) |
71cdb697 CH |
2891 | { |
2892 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2893 | const struct pr_ops *ops; | |
71cdb697 | 2894 | fmode_t mode; |
956a4025 | 2895 | int r; |
71cdb697 | 2896 | |
956a4025 | 2897 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2898 | if (r < 0) |
2899 | return r; | |
2900 | ||
2901 | ops = bdev->bd_disk->fops->pr_ops; | |
2902 | if (ops && ops->pr_preempt) | |
2903 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
2904 | else | |
2905 | r = -EOPNOTSUPP; | |
2906 | ||
956a4025 | 2907 | bdput(bdev); |
71cdb697 CH |
2908 | return r; |
2909 | } | |
2910 | ||
2911 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
2912 | { | |
2913 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2914 | const struct pr_ops *ops; | |
71cdb697 | 2915 | fmode_t mode; |
956a4025 | 2916 | int r; |
71cdb697 | 2917 | |
956a4025 | 2918 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2919 | if (r < 0) |
2920 | return r; | |
2921 | ||
2922 | ops = bdev->bd_disk->fops->pr_ops; | |
2923 | if (ops && ops->pr_clear) | |
2924 | r = ops->pr_clear(bdev, key); | |
2925 | else | |
2926 | r = -EOPNOTSUPP; | |
2927 | ||
956a4025 | 2928 | bdput(bdev); |
71cdb697 CH |
2929 | return r; |
2930 | } | |
2931 | ||
2932 | static const struct pr_ops dm_pr_ops = { | |
2933 | .pr_register = dm_pr_register, | |
2934 | .pr_reserve = dm_pr_reserve, | |
2935 | .pr_release = dm_pr_release, | |
2936 | .pr_preempt = dm_pr_preempt, | |
2937 | .pr_clear = dm_pr_clear, | |
2938 | }; | |
2939 | ||
83d5cde4 | 2940 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2941 | .open = dm_blk_open, |
2942 | .release = dm_blk_close, | |
aa129a22 | 2943 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2944 | .getgeo = dm_blk_getgeo, |
71cdb697 | 2945 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
2946 | .owner = THIS_MODULE |
2947 | }; | |
2948 | ||
f26c5719 DW |
2949 | static const struct dax_operations dm_dax_ops = { |
2950 | .direct_access = dm_dax_direct_access, | |
7e026c8c | 2951 | .copy_from_iter = dm_dax_copy_from_iter, |
f26c5719 DW |
2952 | }; |
2953 | ||
1da177e4 LT |
2954 | /* |
2955 | * module hooks | |
2956 | */ | |
2957 | module_init(dm_init); | |
2958 | module_exit(dm_exit); | |
2959 | ||
2960 | module_param(major, uint, 0); | |
2961 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 2962 | |
e8603136 MS |
2963 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
2964 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
2965 | ||
115485e8 MS |
2966 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
2967 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
2968 | ||
1da177e4 LT |
2969 | MODULE_DESCRIPTION(DM_NAME " driver"); |
2970 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2971 | MODULE_LICENSE("GPL"); |