]>
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> |
55782138 | 27 | |
72d94861 AK |
28 | #define DM_MSG_PREFIX "core" |
29 | ||
60935eb2 MB |
30 | /* |
31 | * Cookies are numeric values sent with CHANGE and REMOVE | |
32 | * uevents while resuming, removing or renaming the device. | |
33 | */ | |
34 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
35 | #define DM_COOKIE_LENGTH 24 | |
36 | ||
1da177e4 LT |
37 | static const char *_name = DM_NAME; |
38 | ||
39 | static unsigned int major = 0; | |
40 | static unsigned int _major = 0; | |
41 | ||
d15b774c AK |
42 | static DEFINE_IDR(_minor_idr); |
43 | ||
f32c10b0 | 44 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
45 | |
46 | static void do_deferred_remove(struct work_struct *w); | |
47 | ||
48 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
49 | ||
acfe0ad7 MP |
50 | static struct workqueue_struct *deferred_remove_workqueue; |
51 | ||
93e6442c MP |
52 | atomic_t dm_global_event_nr = ATOMIC_INIT(0); |
53 | DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq); | |
54 | ||
1da177e4 LT |
55 | /* |
56 | * One of these is allocated per bio. | |
57 | */ | |
58 | struct dm_io { | |
59 | struct mapped_device *md; | |
4e4cbee9 | 60 | blk_status_t status; |
1da177e4 | 61 | atomic_t io_count; |
6ae2fa67 | 62 | struct bio *bio; |
3eaf840e | 63 | unsigned long start_time; |
f88fb981 | 64 | spinlock_t endio_lock; |
fd2ed4d2 | 65 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
66 | }; |
67 | ||
ba61fdd1 JM |
68 | #define MINOR_ALLOCED ((void *)-1) |
69 | ||
1da177e4 LT |
70 | /* |
71 | * Bits for the md->flags field. | |
72 | */ | |
1eb787ec | 73 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 74 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 75 | #define DMF_FROZEN 2 |
fba9f90e | 76 | #define DMF_FREEING 3 |
5c6bd75d | 77 | #define DMF_DELETING 4 |
2e93ccc1 | 78 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
79 | #define DMF_DEFERRED_REMOVE 6 |
80 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 81 | |
115485e8 | 82 | #define DM_NUMA_NODE NUMA_NO_NODE |
115485e8 | 83 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 84 | |
e6ee8c0b KU |
85 | /* |
86 | * For mempools pre-allocation at the table loading time. | |
87 | */ | |
88 | struct dm_md_mempools { | |
89 | mempool_t *io_pool; | |
e6ee8c0b KU |
90 | struct bio_set *bs; |
91 | }; | |
92 | ||
86f1152b BM |
93 | struct table_device { |
94 | struct list_head list; | |
95 | atomic_t count; | |
96 | struct dm_dev dm_dev; | |
97 | }; | |
98 | ||
e18b890b | 99 | static struct kmem_cache *_io_cache; |
8fbf26ad | 100 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 101 | static struct kmem_cache *_rq_cache; |
94818742 | 102 | |
e8603136 MS |
103 | /* |
104 | * Bio-based DM's mempools' reserved IOs set by the user. | |
105 | */ | |
4cc96131 | 106 | #define RESERVED_BIO_BASED_IOS 16 |
e8603136 MS |
107 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
108 | ||
115485e8 MS |
109 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
110 | { | |
111 | int param = ACCESS_ONCE(*module_param); | |
112 | int modified_param = 0; | |
113 | bool modified = true; | |
114 | ||
115 | if (param < min) | |
116 | modified_param = min; | |
117 | else if (param > max) | |
118 | modified_param = max; | |
119 | else | |
120 | modified = false; | |
121 | ||
122 | if (modified) { | |
123 | (void)cmpxchg(module_param, param, modified_param); | |
124 | param = modified_param; | |
125 | } | |
126 | ||
127 | return param; | |
128 | } | |
129 | ||
4cc96131 MS |
130 | unsigned __dm_get_module_param(unsigned *module_param, |
131 | unsigned def, unsigned max) | |
f4790826 | 132 | { |
09c2d531 MS |
133 | unsigned param = ACCESS_ONCE(*module_param); |
134 | unsigned modified_param = 0; | |
f4790826 | 135 | |
09c2d531 MS |
136 | if (!param) |
137 | modified_param = def; | |
138 | else if (param > max) | |
139 | modified_param = max; | |
f4790826 | 140 | |
09c2d531 MS |
141 | if (modified_param) { |
142 | (void)cmpxchg(module_param, param, modified_param); | |
143 | param = modified_param; | |
f4790826 MS |
144 | } |
145 | ||
09c2d531 | 146 | return param; |
f4790826 MS |
147 | } |
148 | ||
e8603136 MS |
149 | unsigned dm_get_reserved_bio_based_ios(void) |
150 | { | |
09c2d531 | 151 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 152 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
153 | } |
154 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
155 | ||
115485e8 MS |
156 | static unsigned dm_get_numa_node(void) |
157 | { | |
158 | return __dm_get_module_param_int(&dm_numa_node, | |
159 | DM_NUMA_NODE, num_online_nodes() - 1); | |
160 | } | |
161 | ||
1da177e4 LT |
162 | static int __init local_init(void) |
163 | { | |
51157b4a | 164 | int r = -ENOMEM; |
1da177e4 | 165 | |
1da177e4 | 166 | /* allocate a slab for the dm_ios */ |
028867ac | 167 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 168 | if (!_io_cache) |
51157b4a | 169 | return r; |
1da177e4 | 170 | |
8fbf26ad KU |
171 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
172 | if (!_rq_tio_cache) | |
dba14160 | 173 | goto out_free_io_cache; |
8fbf26ad | 174 | |
eca7ee6d | 175 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
1ae49ea2 MS |
176 | __alignof__(struct request), 0, NULL); |
177 | if (!_rq_cache) | |
178 | goto out_free_rq_tio_cache; | |
179 | ||
51e5b2bd | 180 | r = dm_uevent_init(); |
51157b4a | 181 | if (r) |
1ae49ea2 | 182 | goto out_free_rq_cache; |
51e5b2bd | 183 | |
acfe0ad7 MP |
184 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
185 | if (!deferred_remove_workqueue) { | |
186 | r = -ENOMEM; | |
187 | goto out_uevent_exit; | |
188 | } | |
189 | ||
1da177e4 LT |
190 | _major = major; |
191 | r = register_blkdev(_major, _name); | |
51157b4a | 192 | if (r < 0) |
acfe0ad7 | 193 | goto out_free_workqueue; |
1da177e4 LT |
194 | |
195 | if (!_major) | |
196 | _major = r; | |
197 | ||
198 | return 0; | |
51157b4a | 199 | |
acfe0ad7 MP |
200 | out_free_workqueue: |
201 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
202 | out_uevent_exit: |
203 | dm_uevent_exit(); | |
1ae49ea2 MS |
204 | out_free_rq_cache: |
205 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
206 | out_free_rq_tio_cache: |
207 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
208 | out_free_io_cache: |
209 | kmem_cache_destroy(_io_cache); | |
210 | ||
211 | return r; | |
1da177e4 LT |
212 | } |
213 | ||
214 | static void local_exit(void) | |
215 | { | |
2c140a24 | 216 | flush_scheduled_work(); |
acfe0ad7 | 217 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 218 | |
1ae49ea2 | 219 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 220 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 221 | kmem_cache_destroy(_io_cache); |
00d59405 | 222 | unregister_blkdev(_major, _name); |
51e5b2bd | 223 | dm_uevent_exit(); |
1da177e4 LT |
224 | |
225 | _major = 0; | |
226 | ||
227 | DMINFO("cleaned up"); | |
228 | } | |
229 | ||
b9249e55 | 230 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
231 | local_init, |
232 | dm_target_init, | |
233 | dm_linear_init, | |
234 | dm_stripe_init, | |
952b3557 | 235 | dm_io_init, |
945fa4d2 | 236 | dm_kcopyd_init, |
1da177e4 | 237 | dm_interface_init, |
fd2ed4d2 | 238 | dm_statistics_init, |
1da177e4 LT |
239 | }; |
240 | ||
b9249e55 | 241 | static void (*_exits[])(void) = { |
1da177e4 LT |
242 | local_exit, |
243 | dm_target_exit, | |
244 | dm_linear_exit, | |
245 | dm_stripe_exit, | |
952b3557 | 246 | dm_io_exit, |
945fa4d2 | 247 | dm_kcopyd_exit, |
1da177e4 | 248 | dm_interface_exit, |
fd2ed4d2 | 249 | dm_statistics_exit, |
1da177e4 LT |
250 | }; |
251 | ||
252 | static int __init dm_init(void) | |
253 | { | |
254 | const int count = ARRAY_SIZE(_inits); | |
255 | ||
256 | int r, i; | |
257 | ||
258 | for (i = 0; i < count; i++) { | |
259 | r = _inits[i](); | |
260 | if (r) | |
261 | goto bad; | |
262 | } | |
263 | ||
264 | return 0; | |
265 | ||
266 | bad: | |
267 | while (i--) | |
268 | _exits[i](); | |
269 | ||
270 | return r; | |
271 | } | |
272 | ||
273 | static void __exit dm_exit(void) | |
274 | { | |
275 | int i = ARRAY_SIZE(_exits); | |
276 | ||
277 | while (i--) | |
278 | _exits[i](); | |
d15b774c AK |
279 | |
280 | /* | |
281 | * Should be empty by this point. | |
282 | */ | |
d15b774c | 283 | idr_destroy(&_minor_idr); |
1da177e4 LT |
284 | } |
285 | ||
286 | /* | |
287 | * Block device functions | |
288 | */ | |
432a212c MA |
289 | int dm_deleting_md(struct mapped_device *md) |
290 | { | |
291 | return test_bit(DMF_DELETING, &md->flags); | |
292 | } | |
293 | ||
fe5f9f2c | 294 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
295 | { |
296 | struct mapped_device *md; | |
297 | ||
fba9f90e JM |
298 | spin_lock(&_minor_lock); |
299 | ||
fe5f9f2c | 300 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
301 | if (!md) |
302 | goto out; | |
303 | ||
5c6bd75d | 304 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 305 | dm_deleting_md(md)) { |
fba9f90e JM |
306 | md = NULL; |
307 | goto out; | |
308 | } | |
309 | ||
1da177e4 | 310 | dm_get(md); |
5c6bd75d | 311 | atomic_inc(&md->open_count); |
fba9f90e JM |
312 | out: |
313 | spin_unlock(&_minor_lock); | |
314 | ||
315 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
316 | } |
317 | ||
db2a144b | 318 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 319 | { |
63a4f065 | 320 | struct mapped_device *md; |
6e9624b8 | 321 | |
4a1aeb98 MB |
322 | spin_lock(&_minor_lock); |
323 | ||
63a4f065 MS |
324 | md = disk->private_data; |
325 | if (WARN_ON(!md)) | |
326 | goto out; | |
327 | ||
2c140a24 MP |
328 | if (atomic_dec_and_test(&md->open_count) && |
329 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 330 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 331 | |
1da177e4 | 332 | dm_put(md); |
63a4f065 | 333 | out: |
4a1aeb98 | 334 | spin_unlock(&_minor_lock); |
1da177e4 LT |
335 | } |
336 | ||
5c6bd75d AK |
337 | int dm_open_count(struct mapped_device *md) |
338 | { | |
339 | return atomic_read(&md->open_count); | |
340 | } | |
341 | ||
342 | /* | |
343 | * Guarantees nothing is using the device before it's deleted. | |
344 | */ | |
2c140a24 | 345 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
346 | { |
347 | int r = 0; | |
348 | ||
349 | spin_lock(&_minor_lock); | |
350 | ||
2c140a24 | 351 | if (dm_open_count(md)) { |
5c6bd75d | 352 | r = -EBUSY; |
2c140a24 MP |
353 | if (mark_deferred) |
354 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
355 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
356 | r = -EEXIST; | |
5c6bd75d AK |
357 | else |
358 | set_bit(DMF_DELETING, &md->flags); | |
359 | ||
360 | spin_unlock(&_minor_lock); | |
361 | ||
362 | return r; | |
363 | } | |
364 | ||
2c140a24 MP |
365 | int dm_cancel_deferred_remove(struct mapped_device *md) |
366 | { | |
367 | int r = 0; | |
368 | ||
369 | spin_lock(&_minor_lock); | |
370 | ||
371 | if (test_bit(DMF_DELETING, &md->flags)) | |
372 | r = -EBUSY; | |
373 | else | |
374 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
375 | ||
376 | spin_unlock(&_minor_lock); | |
377 | ||
378 | return r; | |
379 | } | |
380 | ||
381 | static void do_deferred_remove(struct work_struct *w) | |
382 | { | |
383 | dm_deferred_remove(); | |
384 | } | |
385 | ||
fd2ed4d2 MP |
386 | sector_t dm_get_size(struct mapped_device *md) |
387 | { | |
388 | return get_capacity(md->disk); | |
389 | } | |
390 | ||
9974fa2c MS |
391 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
392 | { | |
393 | return md->queue; | |
394 | } | |
395 | ||
fd2ed4d2 MP |
396 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
397 | { | |
398 | return &md->stats; | |
399 | } | |
400 | ||
3ac51e74 DW |
401 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
402 | { | |
403 | struct mapped_device *md = bdev->bd_disk->private_data; | |
404 | ||
405 | return dm_get_geometry(md, geo); | |
406 | } | |
407 | ||
956a4025 MS |
408 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
409 | struct block_device **bdev, | |
410 | fmode_t *mode) | |
aa129a22 | 411 | { |
66482026 | 412 | struct dm_target *tgt; |
6c182cd8 | 413 | struct dm_table *map; |
956a4025 | 414 | int srcu_idx, r; |
aa129a22 | 415 | |
6c182cd8 | 416 | retry: |
e56f81e0 | 417 | r = -ENOTTY; |
956a4025 | 418 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
419 | if (!map || !dm_table_get_size(map)) |
420 | goto out; | |
421 | ||
422 | /* We only support devices that have a single target */ | |
423 | if (dm_table_get_num_targets(map) != 1) | |
424 | goto out; | |
425 | ||
66482026 MS |
426 | tgt = dm_table_get_target(map, 0); |
427 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 428 | goto out; |
aa129a22 | 429 | |
4f186f8b | 430 | if (dm_suspended_md(md)) { |
aa129a22 MB |
431 | r = -EAGAIN; |
432 | goto out; | |
433 | } | |
434 | ||
66482026 | 435 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
436 | if (r < 0) |
437 | goto out; | |
aa129a22 | 438 | |
956a4025 MS |
439 | bdgrab(*bdev); |
440 | dm_put_live_table(md, srcu_idx); | |
e56f81e0 | 441 | return r; |
aa129a22 | 442 | |
aa129a22 | 443 | out: |
956a4025 | 444 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 445 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
446 | msleep(10); |
447 | goto retry; | |
448 | } | |
e56f81e0 CH |
449 | return r; |
450 | } | |
451 | ||
452 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
453 | unsigned int cmd, unsigned long arg) | |
454 | { | |
455 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 456 | int r; |
e56f81e0 | 457 | |
956a4025 | 458 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
459 | if (r < 0) |
460 | return r; | |
6c182cd8 | 461 | |
e56f81e0 CH |
462 | if (r > 0) { |
463 | /* | |
e980f623 CH |
464 | * Target determined this ioctl is being issued against a |
465 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 466 | */ |
e980f623 CH |
467 | if (!capable(CAP_SYS_RAWIO)) { |
468 | DMWARN_LIMIT( | |
469 | "%s: sending ioctl %x to DM device without required privilege.", | |
470 | current->comm, cmd); | |
471 | r = -ENOIOCTLCMD; | |
e56f81e0 | 472 | goto out; |
e980f623 | 473 | } |
e56f81e0 | 474 | } |
6c182cd8 | 475 | |
66482026 | 476 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 477 | out: |
956a4025 | 478 | bdput(bdev); |
aa129a22 MB |
479 | return r; |
480 | } | |
481 | ||
028867ac | 482 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
483 | { |
484 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
485 | } | |
486 | ||
028867ac | 487 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
488 | { |
489 | mempool_free(io, md->io_pool); | |
490 | } | |
491 | ||
cfae7529 | 492 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 493 | { |
dba14160 | 494 | bio_put(&tio->clone); |
1da177e4 LT |
495 | } |
496 | ||
4cc96131 | 497 | int md_in_flight(struct mapped_device *md) |
90abb8c4 KU |
498 | { |
499 | return atomic_read(&md->pending[READ]) + | |
500 | atomic_read(&md->pending[WRITE]); | |
501 | } | |
502 | ||
3eaf840e JNN |
503 | static void start_io_acct(struct dm_io *io) |
504 | { | |
505 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 506 | struct bio *bio = io->bio; |
c9959059 | 507 | int cpu; |
fd2ed4d2 | 508 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
509 | |
510 | io->start_time = jiffies; | |
511 | ||
074a7aca | 512 | cpu = part_stat_lock(); |
d62e26b3 | 513 | part_round_stats(md->queue, cpu, &dm_disk(md)->part0); |
074a7aca | 514 | part_stat_unlock(); |
1e9bb880 SL |
515 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
516 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
517 | |
518 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
519 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
520 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
521 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
522 | } |
523 | ||
d221d2e7 | 524 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
525 | { |
526 | struct mapped_device *md = io->md; | |
527 | struct bio *bio = io->bio; | |
528 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 529 | int pending; |
3eaf840e JNN |
530 | int rw = bio_data_dir(bio); |
531 | ||
d62e26b3 | 532 | generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 533 | |
fd2ed4d2 | 534 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
535 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
536 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
537 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 538 | |
af7e466a MP |
539 | /* |
540 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 541 | * a flush. |
af7e466a | 542 | */ |
1e9bb880 SL |
543 | pending = atomic_dec_return(&md->pending[rw]); |
544 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 545 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 546 | |
d221d2e7 MP |
547 | /* nudge anyone waiting on suspend queue */ |
548 | if (!pending) | |
549 | wake_up(&md->wait); | |
3eaf840e JNN |
550 | } |
551 | ||
1da177e4 LT |
552 | /* |
553 | * Add the bio to the list of deferred io. | |
554 | */ | |
92c63902 | 555 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 556 | { |
05447420 | 557 | unsigned long flags; |
1da177e4 | 558 | |
05447420 | 559 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 560 | bio_list_add(&md->deferred, bio); |
05447420 | 561 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 562 | queue_work(md->wq, &md->work); |
1da177e4 LT |
563 | } |
564 | ||
565 | /* | |
566 | * Everyone (including functions in this file), should use this | |
567 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 568 | * dm_put_live_table() when finished. |
1da177e4 | 569 | */ |
83d5e5b0 | 570 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 571 | { |
83d5e5b0 MP |
572 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
573 | ||
574 | return srcu_dereference(md->map, &md->io_barrier); | |
575 | } | |
1da177e4 | 576 | |
83d5e5b0 MP |
577 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
578 | { | |
579 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
580 | } | |
581 | ||
582 | void dm_sync_table(struct mapped_device *md) | |
583 | { | |
584 | synchronize_srcu(&md->io_barrier); | |
585 | synchronize_rcu_expedited(); | |
586 | } | |
587 | ||
588 | /* | |
589 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
590 | * The caller must not block between these two functions. | |
591 | */ | |
592 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
593 | { | |
594 | rcu_read_lock(); | |
595 | return rcu_dereference(md->map); | |
596 | } | |
1da177e4 | 597 | |
83d5e5b0 MP |
598 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
599 | { | |
600 | rcu_read_unlock(); | |
1da177e4 LT |
601 | } |
602 | ||
86f1152b BM |
603 | /* |
604 | * Open a table device so we can use it as a map destination. | |
605 | */ | |
606 | static int open_table_device(struct table_device *td, dev_t dev, | |
607 | struct mapped_device *md) | |
608 | { | |
609 | static char *_claim_ptr = "I belong to device-mapper"; | |
610 | struct block_device *bdev; | |
611 | ||
612 | int r; | |
613 | ||
614 | BUG_ON(td->dm_dev.bdev); | |
615 | ||
616 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
617 | if (IS_ERR(bdev)) | |
618 | return PTR_ERR(bdev); | |
619 | ||
620 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
621 | if (r) { | |
622 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
623 | return r; | |
624 | } | |
625 | ||
626 | td->dm_dev.bdev = bdev; | |
817bf402 | 627 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
628 | return 0; |
629 | } | |
630 | ||
631 | /* | |
632 | * Close a table device that we've been using. | |
633 | */ | |
634 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
635 | { | |
636 | if (!td->dm_dev.bdev) | |
637 | return; | |
638 | ||
639 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
640 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 641 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 642 | td->dm_dev.bdev = NULL; |
817bf402 | 643 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
644 | } |
645 | ||
646 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
647 | fmode_t mode) { | |
648 | struct table_device *td; | |
649 | ||
650 | list_for_each_entry(td, l, list) | |
651 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
652 | return td; | |
653 | ||
654 | return NULL; | |
655 | } | |
656 | ||
657 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
658 | struct dm_dev **result) { | |
659 | int r; | |
660 | struct table_device *td; | |
661 | ||
662 | mutex_lock(&md->table_devices_lock); | |
663 | td = find_table_device(&md->table_devices, dev, mode); | |
664 | if (!td) { | |
115485e8 | 665 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
666 | if (!td) { |
667 | mutex_unlock(&md->table_devices_lock); | |
668 | return -ENOMEM; | |
669 | } | |
670 | ||
671 | td->dm_dev.mode = mode; | |
672 | td->dm_dev.bdev = NULL; | |
673 | ||
674 | if ((r = open_table_device(td, dev, md))) { | |
675 | mutex_unlock(&md->table_devices_lock); | |
676 | kfree(td); | |
677 | return r; | |
678 | } | |
679 | ||
680 | format_dev_t(td->dm_dev.name, dev); | |
681 | ||
682 | atomic_set(&td->count, 0); | |
683 | list_add(&td->list, &md->table_devices); | |
684 | } | |
685 | atomic_inc(&td->count); | |
686 | mutex_unlock(&md->table_devices_lock); | |
687 | ||
688 | *result = &td->dm_dev; | |
689 | return 0; | |
690 | } | |
691 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
692 | ||
693 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
694 | { | |
695 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
696 | ||
697 | mutex_lock(&md->table_devices_lock); | |
698 | if (atomic_dec_and_test(&td->count)) { | |
699 | close_table_device(td, md); | |
700 | list_del(&td->list); | |
701 | kfree(td); | |
702 | } | |
703 | mutex_unlock(&md->table_devices_lock); | |
704 | } | |
705 | EXPORT_SYMBOL(dm_put_table_device); | |
706 | ||
707 | static void free_table_devices(struct list_head *devices) | |
708 | { | |
709 | struct list_head *tmp, *next; | |
710 | ||
711 | list_for_each_safe(tmp, next, devices) { | |
712 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
713 | ||
714 | DMWARN("dm_destroy: %s still exists with %d references", | |
715 | td->dm_dev.name, atomic_read(&td->count)); | |
716 | kfree(td); | |
717 | } | |
718 | } | |
719 | ||
3ac51e74 DW |
720 | /* |
721 | * Get the geometry associated with a dm device | |
722 | */ | |
723 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
724 | { | |
725 | *geo = md->geometry; | |
726 | ||
727 | return 0; | |
728 | } | |
729 | ||
730 | /* | |
731 | * Set the geometry of a device. | |
732 | */ | |
733 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
734 | { | |
735 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
736 | ||
737 | if (geo->start > sz) { | |
738 | DMWARN("Start sector is beyond the geometry limits."); | |
739 | return -EINVAL; | |
740 | } | |
741 | ||
742 | md->geometry = *geo; | |
743 | ||
744 | return 0; | |
745 | } | |
746 | ||
1da177e4 LT |
747 | /*----------------------------------------------------------------- |
748 | * CRUD START: | |
749 | * A more elegant soln is in the works that uses the queue | |
750 | * merge fn, unfortunately there are a couple of changes to | |
751 | * the block layer that I want to make for this. So in the | |
752 | * interests of getting something for people to use I give | |
753 | * you this clearly demarcated crap. | |
754 | *---------------------------------------------------------------*/ | |
755 | ||
2e93ccc1 KU |
756 | static int __noflush_suspending(struct mapped_device *md) |
757 | { | |
758 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
759 | } | |
760 | ||
1da177e4 LT |
761 | /* |
762 | * Decrements the number of outstanding ios that a bio has been | |
763 | * cloned into, completing the original io if necc. | |
764 | */ | |
4e4cbee9 | 765 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 766 | { |
2e93ccc1 | 767 | unsigned long flags; |
4e4cbee9 | 768 | blk_status_t io_error; |
b35f8caa MB |
769 | struct bio *bio; |
770 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
771 | |
772 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
773 | if (unlikely(error)) { |
774 | spin_lock_irqsave(&io->endio_lock, flags); | |
4e4cbee9 CH |
775 | if (!(io->status == BLK_STS_DM_REQUEUE && |
776 | __noflush_suspending(md))) | |
777 | io->status = error; | |
f88fb981 KU |
778 | spin_unlock_irqrestore(&io->endio_lock, flags); |
779 | } | |
1da177e4 LT |
780 | |
781 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 782 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
783 | /* |
784 | * Target requested pushing back the I/O. | |
2e93ccc1 | 785 | */ |
022c2611 | 786 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
787 | if (__noflush_suspending(md)) |
788 | bio_list_add_head(&md->deferred, io->bio); | |
789 | else | |
2e93ccc1 | 790 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 791 | io->status = BLK_STS_IOERR; |
022c2611 | 792 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
793 | } |
794 | ||
4e4cbee9 | 795 | io_error = io->status; |
b35f8caa | 796 | bio = io->bio; |
6a8736d1 TH |
797 | end_io_acct(io); |
798 | free_io(md, io); | |
799 | ||
4e4cbee9 | 800 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 801 | return; |
2e93ccc1 | 802 | |
1eff9d32 | 803 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 804 | /* |
6a8736d1 | 805 | * Preflush done for flush with data, reissue |
28a8f0d3 | 806 | * without REQ_PREFLUSH. |
af7e466a | 807 | */ |
1eff9d32 | 808 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 809 | queue_io(md, bio); |
af7e466a | 810 | } else { |
b372d360 | 811 | /* done with normal IO or empty flush */ |
4e4cbee9 | 812 | bio->bi_status = io_error; |
4246a0b6 | 813 | bio_endio(bio); |
b35f8caa | 814 | } |
1da177e4 LT |
815 | } |
816 | } | |
817 | ||
4cc96131 | 818 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
819 | { |
820 | struct queue_limits *limits = dm_get_queue_limits(md); | |
821 | ||
822 | /* device doesn't really support WRITE SAME, disable it */ | |
823 | limits->max_write_same_sectors = 0; | |
824 | } | |
825 | ||
ac62d620 CH |
826 | void disable_write_zeroes(struct mapped_device *md) |
827 | { | |
828 | struct queue_limits *limits = dm_get_queue_limits(md); | |
829 | ||
830 | /* device doesn't really support WRITE ZEROES, disable it */ | |
831 | limits->max_write_zeroes_sectors = 0; | |
832 | } | |
833 | ||
4246a0b6 | 834 | static void clone_endio(struct bio *bio) |
1da177e4 | 835 | { |
4e4cbee9 | 836 | blk_status_t error = bio->bi_status; |
bfc6d41c | 837 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 838 | struct dm_io *io = tio->io; |
9faf400f | 839 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
840 | dm_endio_fn endio = tio->ti->type->end_io; |
841 | ||
4e4cbee9 | 842 | if (unlikely(error == BLK_STS_TARGET)) { |
ac62d620 | 843 | if (bio_op(bio) == REQ_OP_WRITE_SAME && |
74d46992 | 844 | !bio->bi_disk->queue->limits.max_write_same_sectors) |
ac62d620 CH |
845 | disable_write_same(md); |
846 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && | |
74d46992 | 847 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) |
ac62d620 CH |
848 | disable_write_zeroes(md); |
849 | } | |
7eee4ae2 | 850 | |
1be56909 | 851 | if (endio) { |
4e4cbee9 | 852 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
853 | switch (r) { |
854 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 855 | error = BLK_STS_DM_REQUEUE; |
1be56909 CH |
856 | /*FALLTHRU*/ |
857 | case DM_ENDIO_DONE: | |
858 | break; | |
859 | case DM_ENDIO_INCOMPLETE: | |
860 | /* The target will handle the io */ | |
861 | return; | |
862 | default: | |
863 | DMWARN("unimplemented target endio return value: %d", r); | |
864 | BUG(); | |
865 | } | |
866 | } | |
867 | ||
cfae7529 | 868 | free_tio(tio); |
b35f8caa | 869 | dec_pending(io, error); |
1da177e4 LT |
870 | } |
871 | ||
56a67df7 MS |
872 | /* |
873 | * Return maximum size of I/O possible at the supplied sector up to the current | |
874 | * target boundary. | |
875 | */ | |
876 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
877 | { | |
878 | sector_t target_offset = dm_target_offset(ti, sector); | |
879 | ||
880 | return ti->len - target_offset; | |
881 | } | |
882 | ||
883 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 884 | { |
56a67df7 | 885 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 886 | sector_t offset, max_len; |
1da177e4 LT |
887 | |
888 | /* | |
542f9038 | 889 | * Does the target need to split even further? |
1da177e4 | 890 | */ |
542f9038 MS |
891 | if (ti->max_io_len) { |
892 | offset = dm_target_offset(ti, sector); | |
893 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
894 | max_len = sector_div(offset, ti->max_io_len); | |
895 | else | |
896 | max_len = offset & (ti->max_io_len - 1); | |
897 | max_len = ti->max_io_len - max_len; | |
898 | ||
899 | if (len > max_len) | |
900 | len = max_len; | |
1da177e4 LT |
901 | } |
902 | ||
903 | return len; | |
904 | } | |
905 | ||
542f9038 MS |
906 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
907 | { | |
908 | if (len > UINT_MAX) { | |
909 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
910 | (unsigned long long)len, UINT_MAX); | |
911 | ti->error = "Maximum size of target IO is too large"; | |
912 | return -EINVAL; | |
913 | } | |
914 | ||
915 | ti->max_io_len = (uint32_t) len; | |
916 | ||
917 | return 0; | |
918 | } | |
919 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
920 | ||
f26c5719 DW |
921 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
922 | sector_t sector, int *srcu_idx) | |
545ed20e | 923 | { |
545ed20e TK |
924 | struct dm_table *map; |
925 | struct dm_target *ti; | |
545ed20e | 926 | |
f26c5719 | 927 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 928 | if (!map) |
f26c5719 | 929 | return NULL; |
545ed20e TK |
930 | |
931 | ti = dm_table_find_target(map, sector); | |
932 | if (!dm_target_is_valid(ti)) | |
f26c5719 | 933 | return NULL; |
545ed20e | 934 | |
f26c5719 DW |
935 | return ti; |
936 | } | |
545ed20e | 937 | |
f26c5719 DW |
938 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
939 | long nr_pages, void **kaddr, pfn_t *pfn) | |
940 | { | |
941 | struct mapped_device *md = dax_get_private(dax_dev); | |
942 | sector_t sector = pgoff * PAGE_SECTORS; | |
943 | struct dm_target *ti; | |
944 | long len, ret = -EIO; | |
945 | int srcu_idx; | |
545ed20e | 946 | |
f26c5719 | 947 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 948 | |
f26c5719 DW |
949 | if (!ti) |
950 | goto out; | |
951 | if (!ti->type->direct_access) | |
952 | goto out; | |
953 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
954 | if (len < 1) | |
955 | goto out; | |
956 | nr_pages = min(len, nr_pages); | |
545ed20e | 957 | if (ti->type->direct_access) |
817bf402 DW |
958 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
959 | ||
f26c5719 | 960 | out: |
545ed20e | 961 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
962 | |
963 | return ret; | |
545ed20e TK |
964 | } |
965 | ||
7e026c8c DW |
966 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
967 | void *addr, size_t bytes, struct iov_iter *i) | |
968 | { | |
969 | struct mapped_device *md = dax_get_private(dax_dev); | |
970 | sector_t sector = pgoff * PAGE_SECTORS; | |
971 | struct dm_target *ti; | |
972 | long ret = 0; | |
973 | int srcu_idx; | |
974 | ||
975 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
976 | ||
977 | if (!ti) | |
978 | goto out; | |
979 | if (!ti->type->dax_copy_from_iter) { | |
980 | ret = copy_from_iter(addr, bytes, i); | |
981 | goto out; | |
982 | } | |
983 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
984 | out: | |
985 | dm_put_live_table(md, srcu_idx); | |
986 | ||
987 | return ret; | |
988 | } | |
989 | ||
1dd40c3e MP |
990 | /* |
991 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
28a8f0d3 | 992 | * allowed for all bio types except REQ_PREFLUSH. |
1dd40c3e MP |
993 | * |
994 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
995 | * additional n_sectors sectors of the bio and the rest of the data should be | |
996 | * sent in a next bio. | |
997 | * | |
998 | * A diagram that explains the arithmetics: | |
999 | * +--------------------+---------------+-------+ | |
1000 | * | 1 | 2 | 3 | | |
1001 | * +--------------------+---------------+-------+ | |
1002 | * | |
1003 | * <-------------- *tio->len_ptr ---------------> | |
1004 | * <------- bi_size -------> | |
1005 | * <-- n_sectors --> | |
1006 | * | |
1007 | * Region 1 was already iterated over with bio_advance or similar function. | |
1008 | * (it may be empty if the target doesn't use bio_advance) | |
1009 | * Region 2 is the remaining bio size that the target wants to process. | |
1010 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1011 | * to make it empty) | |
1012 | * The target requires that region 3 is to be sent in the next bio. | |
1013 | * | |
1014 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1015 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1016 | * copies of the bio. | |
1017 | */ | |
1018 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1019 | { | |
1020 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1021 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1022 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1023 | BUG_ON(bi_size > *tio->len_ptr); |
1024 | BUG_ON(n_sectors > bi_size); | |
1025 | *tio->len_ptr -= bi_size - n_sectors; | |
1026 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1027 | } | |
1028 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1029 | ||
10999307 DLM |
1030 | /* |
1031 | * The zone descriptors obtained with a zone report indicate | |
1032 | * zone positions within the target device. The zone descriptors | |
1033 | * must be remapped to match their position within the dm device. | |
1034 | * A target may call dm_remap_zone_report after completion of a | |
1035 | * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained | |
1036 | * from the target device mapping to the dm device. | |
1037 | */ | |
1038 | void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start) | |
1039 | { | |
1040 | #ifdef CONFIG_BLK_DEV_ZONED | |
1041 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1042 | struct bio *report_bio = tio->io->bio; | |
1043 | struct blk_zone_report_hdr *hdr = NULL; | |
1044 | struct blk_zone *zone; | |
1045 | unsigned int nr_rep = 0; | |
1046 | unsigned int ofst; | |
1047 | struct bio_vec bvec; | |
1048 | struct bvec_iter iter; | |
1049 | void *addr; | |
1050 | ||
1051 | if (bio->bi_status) | |
1052 | return; | |
1053 | ||
1054 | /* | |
1055 | * Remap the start sector of the reported zones. For sequential zones, | |
1056 | * also remap the write pointer position. | |
1057 | */ | |
1058 | bio_for_each_segment(bvec, report_bio, iter) { | |
1059 | addr = kmap_atomic(bvec.bv_page); | |
1060 | ||
1061 | /* Remember the report header in the first page */ | |
1062 | if (!hdr) { | |
1063 | hdr = addr; | |
1064 | ofst = sizeof(struct blk_zone_report_hdr); | |
1065 | } else | |
1066 | ofst = 0; | |
1067 | ||
1068 | /* Set zones start sector */ | |
1069 | while (hdr->nr_zones && ofst < bvec.bv_len) { | |
1070 | zone = addr + ofst; | |
1071 | if (zone->start >= start + ti->len) { | |
1072 | hdr->nr_zones = 0; | |
1073 | break; | |
1074 | } | |
1075 | zone->start = zone->start + ti->begin - start; | |
1076 | if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { | |
1077 | if (zone->cond == BLK_ZONE_COND_FULL) | |
1078 | zone->wp = zone->start + zone->len; | |
1079 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
1080 | zone->wp = zone->start; | |
1081 | else | |
1082 | zone->wp = zone->wp + ti->begin - start; | |
1083 | } | |
1084 | ofst += sizeof(struct blk_zone); | |
1085 | hdr->nr_zones--; | |
1086 | nr_rep++; | |
1087 | } | |
1088 | ||
1089 | if (addr != hdr) | |
1090 | kunmap_atomic(addr); | |
1091 | ||
1092 | if (!hdr->nr_zones) | |
1093 | break; | |
1094 | } | |
1095 | ||
1096 | if (hdr) { | |
1097 | hdr->nr_zones = nr_rep; | |
1098 | kunmap_atomic(hdr); | |
1099 | } | |
1100 | ||
1101 | bio_advance(report_bio, report_bio->bi_iter.bi_size); | |
1102 | ||
1103 | #else /* !CONFIG_BLK_DEV_ZONED */ | |
1104 | bio->bi_status = BLK_STS_NOTSUPP; | |
1105 | #endif | |
1106 | } | |
1107 | EXPORT_SYMBOL_GPL(dm_remap_zone_report); | |
1108 | ||
d67a5f4b MP |
1109 | /* |
1110 | * Flush current->bio_list when the target map method blocks. | |
1111 | * This fixes deadlocks in snapshot and possibly in other targets. | |
1112 | */ | |
1113 | struct dm_offload { | |
1114 | struct blk_plug plug; | |
1115 | struct blk_plug_cb cb; | |
1116 | }; | |
1117 | ||
1118 | static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule) | |
1119 | { | |
1120 | struct dm_offload *o = container_of(cb, struct dm_offload, cb); | |
1121 | struct bio_list list; | |
1122 | struct bio *bio; | |
f5fe1b51 | 1123 | int i; |
d67a5f4b MP |
1124 | |
1125 | INIT_LIST_HEAD(&o->cb.list); | |
1126 | ||
1127 | if (unlikely(!current->bio_list)) | |
1128 | return; | |
1129 | ||
f5fe1b51 N |
1130 | for (i = 0; i < 2; i++) { |
1131 | list = current->bio_list[i]; | |
1132 | bio_list_init(¤t->bio_list[i]); | |
1133 | ||
1134 | while ((bio = bio_list_pop(&list))) { | |
1135 | struct bio_set *bs = bio->bi_pool; | |
47e0fb46 N |
1136 | if (unlikely(!bs) || bs == fs_bio_set || |
1137 | !bs->rescue_workqueue) { | |
f5fe1b51 N |
1138 | bio_list_add(¤t->bio_list[i], bio); |
1139 | continue; | |
1140 | } | |
1141 | ||
1142 | spin_lock(&bs->rescue_lock); | |
1143 | bio_list_add(&bs->rescue_list, bio); | |
1144 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
1145 | spin_unlock(&bs->rescue_lock); | |
d67a5f4b | 1146 | } |
d67a5f4b MP |
1147 | } |
1148 | } | |
1149 | ||
1150 | static void dm_offload_start(struct dm_offload *o) | |
1151 | { | |
1152 | blk_start_plug(&o->plug); | |
1153 | o->cb.callback = flush_current_bio_list; | |
1154 | list_add(&o->cb.list, ¤t->plug->cb_list); | |
1155 | } | |
1156 | ||
1157 | static void dm_offload_end(struct dm_offload *o) | |
1158 | { | |
1159 | list_del(&o->cb.list); | |
1160 | blk_finish_plug(&o->plug); | |
1161 | } | |
1162 | ||
bd2a49b8 | 1163 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1164 | { |
1165 | int r; | |
2056a782 | 1166 | sector_t sector; |
d67a5f4b | 1167 | struct dm_offload o; |
dba14160 | 1168 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1169 | struct dm_target *ti = tio->ti; |
1da177e4 | 1170 | |
1da177e4 | 1171 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1172 | |
1173 | /* | |
1174 | * Map the clone. If r == 0 we don't need to do | |
1175 | * anything, the target has assumed ownership of | |
1176 | * this io. | |
1177 | */ | |
1178 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1179 | sector = clone->bi_iter.bi_sector; |
d67a5f4b MP |
1180 | |
1181 | dm_offload_start(&o); | |
7de3ee57 | 1182 | r = ti->type->map(ti, clone); |
d67a5f4b MP |
1183 | dm_offload_end(&o); |
1184 | ||
846785e6 CH |
1185 | switch (r) { |
1186 | case DM_MAPIO_SUBMITTED: | |
1187 | break; | |
1188 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1189 | /* the bio has been remapped so dispatch it */ |
74d46992 CH |
1190 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
1191 | bio_dev(tio->io->bio), sector); | |
1da177e4 | 1192 | generic_make_request(clone); |
846785e6 CH |
1193 | break; |
1194 | case DM_MAPIO_KILL: | |
4e4cbee9 CH |
1195 | dec_pending(tio->io, BLK_STS_IOERR); |
1196 | free_tio(tio); | |
1197 | break; | |
846785e6 | 1198 | case DM_MAPIO_REQUEUE: |
4e4cbee9 | 1199 | dec_pending(tio->io, BLK_STS_DM_REQUEUE); |
cfae7529 | 1200 | free_tio(tio); |
846785e6 CH |
1201 | break; |
1202 | default: | |
45cbcd79 KU |
1203 | DMWARN("unimplemented target map return value: %d", r); |
1204 | BUG(); | |
1da177e4 LT |
1205 | } |
1206 | } | |
1207 | ||
1208 | struct clone_info { | |
1209 | struct mapped_device *md; | |
1210 | struct dm_table *map; | |
1211 | struct bio *bio; | |
1212 | struct dm_io *io; | |
1213 | sector_t sector; | |
e0d6609a | 1214 | unsigned sector_count; |
1da177e4 LT |
1215 | }; |
1216 | ||
e0d6609a | 1217 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1218 | { |
4f024f37 KO |
1219 | bio->bi_iter.bi_sector = sector; |
1220 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | /* | |
1224 | * Creates a bio that consists of range of complete bvecs. | |
1225 | */ | |
c80914e8 MS |
1226 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1227 | sector_t sector, unsigned len) | |
1da177e4 | 1228 | { |
dba14160 | 1229 | struct bio *clone = &tio->clone; |
1da177e4 | 1230 | |
1c3b13e6 KO |
1231 | __bio_clone_fast(clone, bio); |
1232 | ||
e2460f2a MP |
1233 | if (unlikely(bio_integrity(bio) != NULL)) { |
1234 | int r; | |
1235 | ||
1236 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1237 | !dm_target_passes_integrity(tio->ti->type))) { | |
1238 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1239 | dm_device_name(tio->io->md), | |
1240 | tio->ti->type->name); | |
1241 | return -EIO; | |
1242 | } | |
1243 | ||
1244 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1245 | if (r < 0) |
1246 | return r; | |
1247 | } | |
bd2a49b8 | 1248 | |
264c869d DLM |
1249 | if (bio_op(bio) != REQ_OP_ZONE_REPORT) |
1250 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); | |
1c3b13e6 KO |
1251 | clone->bi_iter.bi_size = to_bytes(len); |
1252 | ||
e2460f2a | 1253 | if (unlikely(bio_integrity(bio) != NULL)) |
fbd08e76 | 1254 | bio_integrity_trim(clone); |
c80914e8 MS |
1255 | |
1256 | return 0; | |
1da177e4 LT |
1257 | } |
1258 | ||
9015df24 | 1259 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1260 | struct dm_target *ti, |
55a62eef | 1261 | unsigned target_bio_nr) |
f9ab94ce | 1262 | { |
dba14160 MP |
1263 | struct dm_target_io *tio; |
1264 | struct bio *clone; | |
1265 | ||
99778273 | 1266 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1267 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1268 | |
1269 | tio->io = ci->io; | |
1270 | tio->ti = ti; | |
55a62eef | 1271 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1272 | |
1273 | return tio; | |
1274 | } | |
1275 | ||
14fe594d AK |
1276 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1277 | struct dm_target *ti, | |
1dd40c3e | 1278 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1279 | { |
99778273 | 1280 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1281 | struct bio *clone = &tio->clone; |
9015df24 | 1282 | |
1dd40c3e MP |
1283 | tio->len_ptr = len; |
1284 | ||
99778273 | 1285 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1286 | if (len) |
1dd40c3e | 1287 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1288 | |
bd2a49b8 | 1289 | __map_bio(tio); |
f9ab94ce MP |
1290 | } |
1291 | ||
14fe594d | 1292 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1293 | unsigned num_bios, unsigned *len) |
06a426ce | 1294 | { |
55a62eef | 1295 | unsigned target_bio_nr; |
06a426ce | 1296 | |
55a62eef | 1297 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1298 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1299 | } |
1300 | ||
14fe594d | 1301 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1302 | { |
06a426ce | 1303 | unsigned target_nr = 0; |
f9ab94ce MP |
1304 | struct dm_target *ti; |
1305 | ||
b372d360 | 1306 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1307 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1308 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1309 | |
f9ab94ce MP |
1310 | return 0; |
1311 | } | |
1312 | ||
c80914e8 | 1313 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1314 | sector_t sector, unsigned *len) |
5ae89a87 | 1315 | { |
dba14160 | 1316 | struct bio *bio = ci->bio; |
5ae89a87 | 1317 | struct dm_target_io *tio; |
b0d8ed4d AK |
1318 | unsigned target_bio_nr; |
1319 | unsigned num_target_bios = 1; | |
c80914e8 | 1320 | int r = 0; |
5ae89a87 | 1321 | |
b0d8ed4d AK |
1322 | /* |
1323 | * Does the target want to receive duplicate copies of the bio? | |
1324 | */ | |
1325 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1326 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1327 | |
b0d8ed4d | 1328 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1329 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e | 1330 | tio->len_ptr = len; |
c80914e8 | 1331 | r = clone_bio(tio, bio, sector, *len); |
072623de | 1332 | if (r < 0) { |
cfae7529 | 1333 | free_tio(tio); |
c80914e8 | 1334 | break; |
072623de | 1335 | } |
b0d8ed4d AK |
1336 | __map_bio(tio); |
1337 | } | |
c80914e8 MS |
1338 | |
1339 | return r; | |
5ae89a87 MS |
1340 | } |
1341 | ||
55a62eef | 1342 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1343 | |
55a62eef | 1344 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1345 | { |
55a62eef | 1346 | return ti->num_discard_bios; |
23508a96 MS |
1347 | } |
1348 | ||
55a62eef | 1349 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1350 | { |
55a62eef | 1351 | return ti->num_write_same_bios; |
23508a96 MS |
1352 | } |
1353 | ||
ac62d620 CH |
1354 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1355 | { | |
1356 | return ti->num_write_zeroes_bios; | |
1357 | } | |
1358 | ||
23508a96 | 1359 | typedef bool (*is_split_required_fn)(struct dm_target *ti); |
9eef87da | 1360 | |
23508a96 MS |
1361 | static bool is_split_required_for_discard(struct dm_target *ti) |
1362 | { | |
55a62eef | 1363 | return ti->split_discard_bios; |
cec47e3d KU |
1364 | } |
1365 | ||
14fe594d AK |
1366 | static int __send_changing_extent_only(struct clone_info *ci, |
1367 | get_num_bios_fn get_num_bios, | |
1368 | is_split_required_fn is_split_required) | |
ba1cbad9 | 1369 | { |
5ae89a87 | 1370 | struct dm_target *ti; |
e0d6609a | 1371 | unsigned len; |
55a62eef | 1372 | unsigned num_bios; |
ba1cbad9 | 1373 | |
a79245b3 MS |
1374 | do { |
1375 | ti = dm_table_find_target(ci->map, ci->sector); | |
1376 | if (!dm_target_is_valid(ti)) | |
1377 | return -EIO; | |
2eb6e1e3 | 1378 | |
5ae89a87 | 1379 | /* |
23508a96 MS |
1380 | * Even though the device advertised support for this type of |
1381 | * request, that does not mean every target supports it, and | |
936688d7 | 1382 | * reconfiguration might also have changed that since the |
a79245b3 | 1383 | * check was performed. |
5ae89a87 | 1384 | */ |
55a62eef AK |
1385 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1386 | if (!num_bios) | |
a79245b3 | 1387 | return -EOPNOTSUPP; |
ba1cbad9 | 1388 | |
23508a96 | 1389 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1390 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1391 | else |
e0d6609a | 1392 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
de3ec86d | 1393 | |
1dd40c3e | 1394 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1395 | |
a79245b3 MS |
1396 | ci->sector += len; |
1397 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1398 | |
1399 | return 0; | |
ba1cbad9 MS |
1400 | } |
1401 | ||
14fe594d | 1402 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1403 | { |
14fe594d AK |
1404 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1405 | is_split_required_for_discard); | |
23508a96 | 1406 | } |
0ce65797 | 1407 | |
14fe594d | 1408 | static int __send_write_same(struct clone_info *ci) |
0ce65797 | 1409 | { |
14fe594d | 1410 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
0ce65797 MS |
1411 | } |
1412 | ||
ac62d620 CH |
1413 | static int __send_write_zeroes(struct clone_info *ci) |
1414 | { | |
1415 | return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL); | |
1416 | } | |
1417 | ||
e4c93811 AK |
1418 | /* |
1419 | * Select the correct strategy for processing a non-flush bio. | |
1420 | */ | |
14fe594d | 1421 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1422 | { |
dba14160 | 1423 | struct bio *bio = ci->bio; |
512875bd | 1424 | struct dm_target *ti; |
1c3b13e6 | 1425 | unsigned len; |
c80914e8 | 1426 | int r; |
0ce65797 | 1427 | |
e6047149 | 1428 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
14fe594d | 1429 | return __send_discard(ci); |
e6047149 | 1430 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
14fe594d | 1431 | return __send_write_same(ci); |
ac62d620 CH |
1432 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES)) |
1433 | return __send_write_zeroes(ci); | |
0ce65797 | 1434 | |
512875bd JN |
1435 | ti = dm_table_find_target(ci->map, ci->sector); |
1436 | if (!dm_target_is_valid(ti)) | |
1437 | return -EIO; | |
1438 | ||
264c869d DLM |
1439 | if (bio_op(bio) == REQ_OP_ZONE_REPORT) |
1440 | len = ci->sector_count; | |
1441 | else | |
1442 | len = min_t(sector_t, max_io_len(ci->sector, ti), | |
1443 | ci->sector_count); | |
0ce65797 | 1444 | |
c80914e8 MS |
1445 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1446 | if (r < 0) | |
1447 | return r; | |
0ce65797 | 1448 | |
1c3b13e6 KO |
1449 | ci->sector += len; |
1450 | ci->sector_count -= len; | |
0ce65797 | 1451 | |
1c3b13e6 | 1452 | return 0; |
0ce65797 MS |
1453 | } |
1454 | ||
1da177e4 | 1455 | /* |
14fe594d | 1456 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1457 | */ |
83d5e5b0 MP |
1458 | static void __split_and_process_bio(struct mapped_device *md, |
1459 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1460 | { |
1da177e4 | 1461 | struct clone_info ci; |
512875bd | 1462 | int error = 0; |
1da177e4 | 1463 | |
83d5e5b0 | 1464 | if (unlikely(!map)) { |
6a8736d1 | 1465 | bio_io_error(bio); |
f0b9a450 MP |
1466 | return; |
1467 | } | |
692d0eb9 | 1468 | |
83d5e5b0 | 1469 | ci.map = map; |
1da177e4 | 1470 | ci.md = md; |
1da177e4 | 1471 | ci.io = alloc_io(md); |
4e4cbee9 | 1472 | ci.io->status = 0; |
1da177e4 LT |
1473 | atomic_set(&ci.io->io_count, 1); |
1474 | ci.io->bio = bio; | |
1475 | ci.io->md = md; | |
f88fb981 | 1476 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1477 | ci.sector = bio->bi_iter.bi_sector; |
0ce65797 | 1478 | |
3eaf840e | 1479 | start_io_acct(ci.io); |
0ce65797 | 1480 | |
1eff9d32 | 1481 | if (bio->bi_opf & REQ_PREFLUSH) { |
b372d360 MS |
1482 | ci.bio = &ci.md->flush_bio; |
1483 | ci.sector_count = 0; | |
14fe594d | 1484 | error = __send_empty_flush(&ci); |
b372d360 | 1485 | /* dec_pending submits any data associated with flush */ |
a4aa5e56 DLM |
1486 | } else if (bio_op(bio) == REQ_OP_ZONE_RESET) { |
1487 | ci.bio = bio; | |
1488 | ci.sector_count = 0; | |
1489 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1490 | } else { |
6a8736d1 | 1491 | ci.bio = bio; |
d87f4c14 | 1492 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1493 | while (ci.sector_count && !error) |
14fe594d | 1494 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1495 | } |
0ce65797 | 1496 | |
1da177e4 | 1497 | /* drop the extra reference count */ |
54385bf7 | 1498 | dec_pending(ci.io, errno_to_blk_status(error)); |
0ce65797 | 1499 | } |
1da177e4 LT |
1500 | /*----------------------------------------------------------------- |
1501 | * CRUD END | |
1502 | *---------------------------------------------------------------*/ | |
0ce65797 | 1503 | |
cec47e3d | 1504 | /* |
1da177e4 LT |
1505 | * The request function that just remaps the bio built up by |
1506 | * dm_merge_bvec. | |
cec47e3d | 1507 | */ |
dece1635 | 1508 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
cec47e3d | 1509 | { |
12f03a49 | 1510 | int rw = bio_data_dir(bio); |
cec47e3d | 1511 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1512 | int srcu_idx; |
1513 | struct dm_table *map; | |
cec47e3d | 1514 | |
83d5e5b0 | 1515 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1516 | |
d62e26b3 | 1517 | generic_start_io_acct(q, rw, bio_sectors(bio), &dm_disk(md)->part0); |
d0bcb878 | 1518 | |
6a8736d1 TH |
1519 | /* if we're suspended, we have to queue this io for later */ |
1520 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1521 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1522 | |
1eff9d32 | 1523 | if (!(bio->bi_opf & REQ_RAHEAD)) |
6a8736d1 TH |
1524 | queue_io(md, bio); |
1525 | else | |
54d9a1b4 | 1526 | bio_io_error(bio); |
dece1635 | 1527 | return BLK_QC_T_NONE; |
cec47e3d | 1528 | } |
1da177e4 | 1529 | |
83d5e5b0 MP |
1530 | __split_and_process_bio(md, map, bio); |
1531 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1532 | return BLK_QC_T_NONE; |
cec47e3d KU |
1533 | } |
1534 | ||
1da177e4 LT |
1535 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1536 | { | |
8a57dfc6 CS |
1537 | int r = bdi_bits; |
1538 | struct mapped_device *md = congested_data; | |
1539 | struct dm_table *map; | |
1da177e4 | 1540 | |
1eb787ec | 1541 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 1542 | if (dm_request_based(md)) { |
cec47e3d | 1543 | /* |
e522c039 MS |
1544 | * With request-based DM we only need to check the |
1545 | * top-level queue for congestion. | |
cec47e3d | 1546 | */ |
dc3b17cc | 1547 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
e522c039 MS |
1548 | } else { |
1549 | map = dm_get_live_table_fast(md); | |
1550 | if (map) | |
cec47e3d | 1551 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 1552 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1553 | } |
1554 | } | |
1555 | ||
1da177e4 LT |
1556 | return r; |
1557 | } | |
1558 | ||
1559 | /*----------------------------------------------------------------- | |
1560 | * An IDR is used to keep track of allocated minor numbers. | |
1561 | *---------------------------------------------------------------*/ | |
2b06cfff | 1562 | static void free_minor(int minor) |
1da177e4 | 1563 | { |
f32c10b0 | 1564 | spin_lock(&_minor_lock); |
1da177e4 | 1565 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1566 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1567 | } |
1568 | ||
1569 | /* | |
1570 | * See if the device with a specific minor # is free. | |
1571 | */ | |
cf13ab8e | 1572 | static int specific_minor(int minor) |
1da177e4 | 1573 | { |
c9d76be6 | 1574 | int r; |
1da177e4 LT |
1575 | |
1576 | if (minor >= (1 << MINORBITS)) | |
1577 | return -EINVAL; | |
1578 | ||
c9d76be6 | 1579 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1580 | spin_lock(&_minor_lock); |
1da177e4 | 1581 | |
c9d76be6 | 1582 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1583 | |
f32c10b0 | 1584 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1585 | idr_preload_end(); |
1586 | if (r < 0) | |
1587 | return r == -ENOSPC ? -EBUSY : r; | |
1588 | return 0; | |
1da177e4 LT |
1589 | } |
1590 | ||
cf13ab8e | 1591 | static int next_free_minor(int *minor) |
1da177e4 | 1592 | { |
c9d76be6 | 1593 | int r; |
62f75c2f | 1594 | |
c9d76be6 | 1595 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1596 | spin_lock(&_minor_lock); |
1da177e4 | 1597 | |
c9d76be6 | 1598 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1599 | |
f32c10b0 | 1600 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1601 | idr_preload_end(); |
1602 | if (r < 0) | |
1603 | return r; | |
1604 | *minor = r; | |
1605 | return 0; | |
1da177e4 LT |
1606 | } |
1607 | ||
83d5cde4 | 1608 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1609 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1610 | |
53d5914f MP |
1611 | static void dm_wq_work(struct work_struct *work); |
1612 | ||
4cc96131 | 1613 | void dm_init_md_queue(struct mapped_device *md) |
4a0b4ddf MS |
1614 | { |
1615 | /* | |
1616 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 1617 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
1618 | * The type is decided at the first table loading time. |
1619 | * To prevent problematic device stacking, clear the queue flag | |
1620 | * for request stacking support until then. | |
1621 | * | |
1622 | * This queue is new, so no concurrency on the queue_flags. | |
1623 | */ | |
1624 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
1625 | |
1626 | /* | |
1627 | * Initialize data that will only be used by a non-blk-mq DM queue | |
1628 | * - must do so here (in alloc_dev callchain) before queue is used | |
1629 | */ | |
1630 | md->queue->queuedata = md; | |
dc3b17cc | 1631 | md->queue->backing_dev_info->congested_data = md; |
bfebd1cd | 1632 | } |
4a0b4ddf | 1633 | |
4cc96131 | 1634 | void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 1635 | { |
17e149b8 | 1636 | md->use_blk_mq = false; |
bfebd1cd MS |
1637 | dm_init_md_queue(md); |
1638 | ||
1639 | /* | |
1640 | * Initialize aspects of queue that aren't relevant for blk-mq | |
1641 | */ | |
dc3b17cc | 1642 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
4a0b4ddf MS |
1643 | } |
1644 | ||
0f20972f MS |
1645 | static void cleanup_mapped_device(struct mapped_device *md) |
1646 | { | |
0f20972f MS |
1647 | if (md->wq) |
1648 | destroy_workqueue(md->wq); | |
1649 | if (md->kworker_task) | |
1650 | kthread_stop(md->kworker_task); | |
6f65985e | 1651 | mempool_destroy(md->io_pool); |
0f20972f MS |
1652 | if (md->bs) |
1653 | bioset_free(md->bs); | |
1654 | ||
f26c5719 DW |
1655 | if (md->dax_dev) { |
1656 | kill_dax(md->dax_dev); | |
1657 | put_dax(md->dax_dev); | |
1658 | md->dax_dev = NULL; | |
1659 | } | |
1660 | ||
0f20972f MS |
1661 | if (md->disk) { |
1662 | spin_lock(&_minor_lock); | |
1663 | md->disk->private_data = NULL; | |
1664 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1665 | del_gendisk(md->disk); |
1666 | put_disk(md->disk); | |
1667 | } | |
1668 | ||
1669 | if (md->queue) | |
1670 | blk_cleanup_queue(md->queue); | |
1671 | ||
d09960b0 TE |
1672 | cleanup_srcu_struct(&md->io_barrier); |
1673 | ||
0f20972f MS |
1674 | if (md->bdev) { |
1675 | bdput(md->bdev); | |
1676 | md->bdev = NULL; | |
1677 | } | |
4cc96131 MS |
1678 | |
1679 | dm_mq_cleanup_mapped_device(md); | |
0f20972f MS |
1680 | } |
1681 | ||
1da177e4 LT |
1682 | /* |
1683 | * Allocate and initialise a blank device with a given minor. | |
1684 | */ | |
2b06cfff | 1685 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1686 | { |
115485e8 | 1687 | int r, numa_node_id = dm_get_numa_node(); |
f26c5719 | 1688 | struct dax_device *dax_dev; |
115485e8 | 1689 | struct mapped_device *md; |
ba61fdd1 | 1690 | void *old_md; |
1da177e4 | 1691 | |
115485e8 | 1692 | md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1693 | if (!md) { |
1694 | DMWARN("unable to allocate device, out of memory."); | |
1695 | return NULL; | |
1696 | } | |
1697 | ||
10da4f79 | 1698 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1699 | goto bad_module_get; |
10da4f79 | 1700 | |
1da177e4 | 1701 | /* get a minor number for the dev */ |
2b06cfff | 1702 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1703 | r = next_free_minor(&minor); |
2b06cfff | 1704 | else |
cf13ab8e | 1705 | r = specific_minor(minor); |
1da177e4 | 1706 | if (r < 0) |
6ed7ade8 | 1707 | goto bad_minor; |
1da177e4 | 1708 | |
83d5e5b0 MP |
1709 | r = init_srcu_struct(&md->io_barrier); |
1710 | if (r < 0) | |
1711 | goto bad_io_barrier; | |
1712 | ||
115485e8 | 1713 | md->numa_node_id = numa_node_id; |
4cc96131 | 1714 | md->use_blk_mq = dm_use_blk_mq_default(); |
591ddcfc | 1715 | md->init_tio_pdu = false; |
a5664dad | 1716 | md->type = DM_TYPE_NONE; |
e61290a4 | 1717 | mutex_init(&md->suspend_lock); |
a5664dad | 1718 | mutex_init(&md->type_lock); |
86f1152b | 1719 | mutex_init(&md->table_devices_lock); |
022c2611 | 1720 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1721 | atomic_set(&md->holders, 1); |
5c6bd75d | 1722 | atomic_set(&md->open_count, 0); |
1da177e4 | 1723 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1724 | atomic_set(&md->uevent_seq, 0); |
1725 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1726 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1727 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1728 | |
115485e8 | 1729 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 1730 | if (!md->queue) |
0f20972f | 1731 | goto bad; |
1da177e4 | 1732 | |
4a0b4ddf | 1733 | dm_init_md_queue(md); |
9faf400f | 1734 | |
115485e8 | 1735 | md->disk = alloc_disk_node(1, numa_node_id); |
1da177e4 | 1736 | if (!md->disk) |
0f20972f | 1737 | goto bad; |
1da177e4 | 1738 | |
316d315b NK |
1739 | atomic_set(&md->pending[0], 0); |
1740 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1741 | init_waitqueue_head(&md->wait); |
53d5914f | 1742 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1743 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1744 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 1745 | md->kworker_task = NULL; |
f0b04115 | 1746 | |
1da177e4 LT |
1747 | md->disk->major = _major; |
1748 | md->disk->first_minor = minor; | |
1749 | md->disk->fops = &dm_blk_dops; | |
1750 | md->disk->queue = md->queue; | |
1751 | md->disk->private_data = md; | |
1752 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 DW |
1753 | |
1754 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); | |
1755 | if (!dax_dev) | |
1756 | goto bad; | |
1757 | md->dax_dev = dax_dev; | |
1758 | ||
1da177e4 | 1759 | add_disk(md->disk); |
7e51f257 | 1760 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1761 | |
670368a8 | 1762 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1763 | if (!md->wq) |
0f20972f | 1764 | goto bad; |
304f3f6a | 1765 | |
32a926da MP |
1766 | md->bdev = bdget_disk(md->disk, 0); |
1767 | if (!md->bdev) | |
0f20972f | 1768 | goto bad; |
32a926da | 1769 | |
3a83f467 | 1770 | bio_init(&md->flush_bio, NULL, 0); |
74d46992 | 1771 | bio_set_dev(&md->flush_bio, md->bdev); |
ff0361b3 | 1772 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; |
6a8736d1 | 1773 | |
fd2ed4d2 MP |
1774 | dm_stats_init(&md->stats); |
1775 | ||
ba61fdd1 | 1776 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1777 | spin_lock(&_minor_lock); |
ba61fdd1 | 1778 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1779 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1780 | |
1781 | BUG_ON(old_md != MINOR_ALLOCED); | |
1782 | ||
1da177e4 LT |
1783 | return md; |
1784 | ||
0f20972f MS |
1785 | bad: |
1786 | cleanup_mapped_device(md); | |
83d5e5b0 | 1787 | bad_io_barrier: |
1da177e4 | 1788 | free_minor(minor); |
6ed7ade8 | 1789 | bad_minor: |
10da4f79 | 1790 | module_put(THIS_MODULE); |
6ed7ade8 | 1791 | bad_module_get: |
1da177e4 LT |
1792 | kfree(md); |
1793 | return NULL; | |
1794 | } | |
1795 | ||
ae9da83f JN |
1796 | static void unlock_fs(struct mapped_device *md); |
1797 | ||
1da177e4 LT |
1798 | static void free_dev(struct mapped_device *md) |
1799 | { | |
f331c029 | 1800 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1801 | |
32a926da | 1802 | unlock_fs(md); |
2eb6e1e3 | 1803 | |
0f20972f | 1804 | cleanup_mapped_device(md); |
63a4f065 | 1805 | |
86f1152b | 1806 | free_table_devices(&md->table_devices); |
63a4f065 | 1807 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1808 | free_minor(minor); |
1809 | ||
10da4f79 | 1810 | module_put(THIS_MODULE); |
1da177e4 LT |
1811 | kfree(md); |
1812 | } | |
1813 | ||
e6ee8c0b KU |
1814 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1815 | { | |
c0820cf5 | 1816 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1817 | |
4e6e36c3 MS |
1818 | if (md->bs) { |
1819 | /* The md already has necessary mempools. */ | |
545ed20e | 1820 | if (dm_table_bio_based(t)) { |
16245bdc JN |
1821 | /* |
1822 | * Reload bioset because front_pad may have changed | |
1823 | * because a different table was loaded. | |
1824 | */ | |
1825 | bioset_free(md->bs); | |
1826 | md->bs = p->bs; | |
1827 | p->bs = NULL; | |
16245bdc | 1828 | } |
4e6e36c3 MS |
1829 | /* |
1830 | * There's no need to reload with request-based dm | |
1831 | * because the size of front_pad doesn't change. | |
1832 | * Note for future: If you are to reload bioset, | |
1833 | * prep-ed requests in the queue may refer | |
1834 | * to bio from the old bioset, so you must walk | |
1835 | * through the queue to unprep. | |
1836 | */ | |
1837 | goto out; | |
c0820cf5 | 1838 | } |
e6ee8c0b | 1839 | |
eb8db831 | 1840 | BUG_ON(!p || md->io_pool || md->bs); |
cbc4e3c1 | 1841 | |
e6ee8c0b KU |
1842 | md->io_pool = p->io_pool; |
1843 | p->io_pool = NULL; | |
e6ee8c0b KU |
1844 | md->bs = p->bs; |
1845 | p->bs = NULL; | |
4e6e36c3 | 1846 | |
e6ee8c0b | 1847 | out: |
02233342 | 1848 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
1849 | dm_table_free_md_mempools(t); |
1850 | } | |
1851 | ||
1da177e4 LT |
1852 | /* |
1853 | * Bind a table to the device. | |
1854 | */ | |
1855 | static void event_callback(void *context) | |
1856 | { | |
7a8c3d3b MA |
1857 | unsigned long flags; |
1858 | LIST_HEAD(uevents); | |
1da177e4 LT |
1859 | struct mapped_device *md = (struct mapped_device *) context; |
1860 | ||
7a8c3d3b MA |
1861 | spin_lock_irqsave(&md->uevent_lock, flags); |
1862 | list_splice_init(&md->uevent_list, &uevents); | |
1863 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1864 | ||
ed9e1982 | 1865 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1866 | |
1da177e4 | 1867 | atomic_inc(&md->event_nr); |
93e6442c | 1868 | atomic_inc(&dm_global_event_nr); |
1da177e4 | 1869 | wake_up(&md->eventq); |
93e6442c | 1870 | wake_up(&dm_global_eventq); |
1da177e4 LT |
1871 | } |
1872 | ||
c217649b MS |
1873 | /* |
1874 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1875 | */ | |
4e90188b | 1876 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1877 | { |
1ea0654e BVA |
1878 | lockdep_assert_held(&md->suspend_lock); |
1879 | ||
4e90188b | 1880 | set_capacity(md->disk, size); |
1da177e4 | 1881 | |
db8fef4f | 1882 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
1883 | } |
1884 | ||
042d2a9b AK |
1885 | /* |
1886 | * Returns old map, which caller must destroy. | |
1887 | */ | |
1888 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
1889 | struct queue_limits *limits) | |
1da177e4 | 1890 | { |
042d2a9b | 1891 | struct dm_table *old_map; |
165125e1 | 1892 | struct request_queue *q = md->queue; |
1da177e4 LT |
1893 | sector_t size; |
1894 | ||
5a8f1f80 BVA |
1895 | lockdep_assert_held(&md->suspend_lock); |
1896 | ||
1da177e4 | 1897 | size = dm_table_get_size(t); |
3ac51e74 DW |
1898 | |
1899 | /* | |
1900 | * Wipe any geometry if the size of the table changed. | |
1901 | */ | |
fd2ed4d2 | 1902 | if (size != dm_get_size(md)) |
3ac51e74 DW |
1903 | memset(&md->geometry, 0, sizeof(md->geometry)); |
1904 | ||
32a926da | 1905 | __set_size(md, size); |
d5816876 | 1906 | |
2ca3310e AK |
1907 | dm_table_event_callback(t, event_callback, md); |
1908 | ||
e6ee8c0b KU |
1909 | /* |
1910 | * The queue hasn't been stopped yet, if the old table type wasn't | |
1911 | * for request-based during suspension. So stop it to prevent | |
1912 | * I/O mapping before resume. | |
1913 | * This must be done before setting the queue restrictions, | |
1914 | * because request-based dm may be run just after the setting. | |
1915 | */ | |
16f12266 | 1916 | if (dm_table_request_based(t)) { |
eca7ee6d | 1917 | dm_stop_queue(q); |
16f12266 MS |
1918 | /* |
1919 | * Leverage the fact that request-based DM targets are | |
1920 | * immutable singletons and establish md->immutable_target | |
1921 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
1922 | */ | |
1923 | md->immutable_target = dm_table_get_immutable_target(t); | |
1924 | } | |
e6ee8c0b KU |
1925 | |
1926 | __bind_mempools(md, t); | |
1927 | ||
a12f5d48 | 1928 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 1929 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
1930 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
1931 | ||
754c5fc7 | 1932 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
1933 | if (old_map) |
1934 | dm_sync_table(md); | |
1da177e4 | 1935 | |
042d2a9b | 1936 | return old_map; |
1da177e4 LT |
1937 | } |
1938 | ||
a7940155 AK |
1939 | /* |
1940 | * Returns unbound table for the caller to free. | |
1941 | */ | |
1942 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 1943 | { |
a12f5d48 | 1944 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
1945 | |
1946 | if (!map) | |
a7940155 | 1947 | return NULL; |
1da177e4 LT |
1948 | |
1949 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 1950 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 1951 | dm_sync_table(md); |
a7940155 AK |
1952 | |
1953 | return map; | |
1da177e4 LT |
1954 | } |
1955 | ||
1956 | /* | |
1957 | * Constructor for a new device. | |
1958 | */ | |
2b06cfff | 1959 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1960 | { |
1961 | struct mapped_device *md; | |
1962 | ||
2b06cfff | 1963 | md = alloc_dev(minor); |
1da177e4 LT |
1964 | if (!md) |
1965 | return -ENXIO; | |
1966 | ||
784aae73 MB |
1967 | dm_sysfs_init(md); |
1968 | ||
1da177e4 LT |
1969 | *result = md; |
1970 | return 0; | |
1971 | } | |
1972 | ||
a5664dad MS |
1973 | /* |
1974 | * Functions to manage md->type. | |
1975 | * All are required to hold md->type_lock. | |
1976 | */ | |
1977 | void dm_lock_md_type(struct mapped_device *md) | |
1978 | { | |
1979 | mutex_lock(&md->type_lock); | |
1980 | } | |
1981 | ||
1982 | void dm_unlock_md_type(struct mapped_device *md) | |
1983 | { | |
1984 | mutex_unlock(&md->type_lock); | |
1985 | } | |
1986 | ||
7e0d574f | 1987 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 1988 | { |
00c4fc3b | 1989 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
1990 | md->type = type; |
1991 | } | |
1992 | ||
7e0d574f | 1993 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
1994 | { |
1995 | return md->type; | |
1996 | } | |
1997 | ||
36a0456f AK |
1998 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
1999 | { | |
2000 | return md->immutable_target_type; | |
2001 | } | |
2002 | ||
f84cb8a4 MS |
2003 | /* |
2004 | * The queue_limits are only valid as long as you have a reference | |
2005 | * count on 'md'. | |
2006 | */ | |
2007 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2008 | { | |
2009 | BUG_ON(!atomic_read(&md->holders)); | |
2010 | return &md->queue->limits; | |
2011 | } | |
2012 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2013 | ||
4a0b4ddf MS |
2014 | /* |
2015 | * Setup the DM device's queue based on md's type | |
2016 | */ | |
591ddcfc | 2017 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2018 | { |
bfebd1cd | 2019 | int r; |
7e0d574f | 2020 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2021 | |
545ed20e | 2022 | switch (type) { |
bfebd1cd | 2023 | case DM_TYPE_REQUEST_BASED: |
eb8db831 | 2024 | r = dm_old_init_request_queue(md, t); |
bfebd1cd | 2025 | if (r) { |
eca7ee6d | 2026 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2027 | return r; |
ff36ab34 | 2028 | } |
bfebd1cd MS |
2029 | break; |
2030 | case DM_TYPE_MQ_REQUEST_BASED: | |
e83068a5 | 2031 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2032 | if (r) { |
eca7ee6d | 2033 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2034 | return r; |
2035 | } | |
2036 | break; | |
2037 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2038 | case DM_TYPE_DAX_BIO_BASED: |
eca7ee6d | 2039 | dm_init_normal_md_queue(md); |
ff36ab34 | 2040 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
2041 | /* |
2042 | * DM handles splitting bios as needed. Free the bio_split bioset | |
2043 | * since it won't be used (saves 1 process per bio-based DM device). | |
2044 | */ | |
2045 | bioset_free(md->queue->bio_split); | |
2046 | md->queue->bio_split = NULL; | |
545ed20e TK |
2047 | |
2048 | if (type == DM_TYPE_DAX_BIO_BASED) | |
2049 | queue_flag_set_unlocked(QUEUE_FLAG_DAX, md->queue); | |
bfebd1cd | 2050 | break; |
7e0d574f BVA |
2051 | case DM_TYPE_NONE: |
2052 | WARN_ON_ONCE(true); | |
2053 | break; | |
4a0b4ddf MS |
2054 | } |
2055 | ||
2056 | return 0; | |
2057 | } | |
2058 | ||
2bec1f4a | 2059 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2060 | { |
2061 | struct mapped_device *md; | |
1da177e4 LT |
2062 | unsigned minor = MINOR(dev); |
2063 | ||
2064 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2065 | return NULL; | |
2066 | ||
f32c10b0 | 2067 | spin_lock(&_minor_lock); |
1da177e4 LT |
2068 | |
2069 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2070 | if (md) { |
2071 | if ((md == MINOR_ALLOCED || | |
2072 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2073 | dm_deleting_md(md) || | |
2074 | test_bit(DMF_FREEING, &md->flags))) { | |
2075 | md = NULL; | |
2076 | goto out; | |
2077 | } | |
2078 | dm_get(md); | |
fba9f90e | 2079 | } |
1da177e4 | 2080 | |
fba9f90e | 2081 | out: |
f32c10b0 | 2082 | spin_unlock(&_minor_lock); |
1da177e4 | 2083 | |
637842cf DT |
2084 | return md; |
2085 | } | |
3cf2e4ba | 2086 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2087 | |
9ade92a9 | 2088 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2089 | { |
9ade92a9 | 2090 | return md->interface_ptr; |
1da177e4 LT |
2091 | } |
2092 | ||
2093 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2094 | { | |
2095 | md->interface_ptr = ptr; | |
2096 | } | |
2097 | ||
2098 | void dm_get(struct mapped_device *md) | |
2099 | { | |
2100 | atomic_inc(&md->holders); | |
3f77316d | 2101 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2102 | } |
2103 | ||
09ee96b2 MP |
2104 | int dm_hold(struct mapped_device *md) |
2105 | { | |
2106 | spin_lock(&_minor_lock); | |
2107 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2108 | spin_unlock(&_minor_lock); | |
2109 | return -EBUSY; | |
2110 | } | |
2111 | dm_get(md); | |
2112 | spin_unlock(&_minor_lock); | |
2113 | return 0; | |
2114 | } | |
2115 | EXPORT_SYMBOL_GPL(dm_hold); | |
2116 | ||
72d94861 AK |
2117 | const char *dm_device_name(struct mapped_device *md) |
2118 | { | |
2119 | return md->name; | |
2120 | } | |
2121 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2122 | ||
3f77316d | 2123 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2124 | { |
3b785fbc | 2125 | struct request_queue *q = dm_get_md_queue(md); |
1134e5ae | 2126 | struct dm_table *map; |
83d5e5b0 | 2127 | int srcu_idx; |
1da177e4 | 2128 | |
3f77316d | 2129 | might_sleep(); |
fba9f90e | 2130 | |
63a4f065 | 2131 | spin_lock(&_minor_lock); |
3f77316d KU |
2132 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2133 | set_bit(DMF_FREEING, &md->flags); | |
2134 | spin_unlock(&_minor_lock); | |
3b785fbc | 2135 | |
2e91c369 | 2136 | blk_set_queue_dying(q); |
3f77316d | 2137 | |
02233342 | 2138 | if (dm_request_based(md) && md->kworker_task) |
3989144f | 2139 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2140 | |
ab7c7bb6 MP |
2141 | /* |
2142 | * Take suspend_lock so that presuspend and postsuspend methods | |
2143 | * do not race with internal suspend. | |
2144 | */ | |
2145 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2146 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2147 | if (!dm_suspended_md(md)) { |
2148 | dm_table_presuspend_targets(map); | |
2149 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2150 | } |
83d5e5b0 MP |
2151 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2152 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2153 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2154 | |
3f77316d KU |
2155 | /* |
2156 | * Rare, but there may be I/O requests still going to complete, | |
2157 | * for example. Wait for all references to disappear. | |
2158 | * No one should increment the reference count of the mapped_device, | |
2159 | * after the mapped_device state becomes DMF_FREEING. | |
2160 | */ | |
2161 | if (wait) | |
2162 | while (atomic_read(&md->holders)) | |
2163 | msleep(1); | |
2164 | else if (atomic_read(&md->holders)) | |
2165 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2166 | dm_device_name(md), atomic_read(&md->holders)); | |
2167 | ||
2168 | dm_sysfs_exit(md); | |
3f77316d KU |
2169 | dm_table_destroy(__unbind(md)); |
2170 | free_dev(md); | |
2171 | } | |
2172 | ||
2173 | void dm_destroy(struct mapped_device *md) | |
2174 | { | |
2175 | __dm_destroy(md, true); | |
2176 | } | |
2177 | ||
2178 | void dm_destroy_immediate(struct mapped_device *md) | |
2179 | { | |
2180 | __dm_destroy(md, false); | |
2181 | } | |
2182 | ||
2183 | void dm_put(struct mapped_device *md) | |
2184 | { | |
2185 | atomic_dec(&md->holders); | |
1da177e4 | 2186 | } |
79eb885c | 2187 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2188 | |
b48633f8 | 2189 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
46125c1c MB |
2190 | { |
2191 | int r = 0; | |
9f4c3f87 | 2192 | DEFINE_WAIT(wait); |
46125c1c MB |
2193 | |
2194 | while (1) { | |
9f4c3f87 | 2195 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2196 | |
b4324fee | 2197 | if (!md_in_flight(md)) |
46125c1c MB |
2198 | break; |
2199 | ||
e3fabdfd | 2200 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2201 | r = -EINTR; |
2202 | break; | |
2203 | } | |
2204 | ||
2205 | io_schedule(); | |
2206 | } | |
9f4c3f87 | 2207 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2208 | |
46125c1c MB |
2209 | return r; |
2210 | } | |
2211 | ||
1da177e4 LT |
2212 | /* |
2213 | * Process the deferred bios | |
2214 | */ | |
ef208587 | 2215 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2216 | { |
ef208587 MP |
2217 | struct mapped_device *md = container_of(work, struct mapped_device, |
2218 | work); | |
6d6f10df | 2219 | struct bio *c; |
83d5e5b0 MP |
2220 | int srcu_idx; |
2221 | struct dm_table *map; | |
1da177e4 | 2222 | |
83d5e5b0 | 2223 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2224 | |
3b00b203 | 2225 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2226 | spin_lock_irq(&md->deferred_lock); |
2227 | c = bio_list_pop(&md->deferred); | |
2228 | spin_unlock_irq(&md->deferred_lock); | |
2229 | ||
6a8736d1 | 2230 | if (!c) |
df12ee99 | 2231 | break; |
022c2611 | 2232 | |
e6ee8c0b KU |
2233 | if (dm_request_based(md)) |
2234 | generic_make_request(c); | |
6a8736d1 | 2235 | else |
83d5e5b0 | 2236 | __split_and_process_bio(md, map, c); |
022c2611 | 2237 | } |
73d410c0 | 2238 | |
83d5e5b0 | 2239 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2240 | } |
2241 | ||
9a1fb464 | 2242 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2243 | { |
3b00b203 | 2244 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2245 | smp_mb__after_atomic(); |
53d5914f | 2246 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2247 | } |
2248 | ||
1da177e4 | 2249 | /* |
042d2a9b | 2250 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2251 | */ |
042d2a9b | 2252 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2253 | { |
87eb5b21 | 2254 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2255 | struct queue_limits limits; |
042d2a9b | 2256 | int r; |
1da177e4 | 2257 | |
e61290a4 | 2258 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2259 | |
2260 | /* device must be suspended */ | |
4f186f8b | 2261 | if (!dm_suspended_md(md)) |
93c534ae | 2262 | goto out; |
1da177e4 | 2263 | |
3ae70656 MS |
2264 | /* |
2265 | * If the new table has no data devices, retain the existing limits. | |
2266 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2267 | * then new I/O is queued based on these limits, and then some paths | |
2268 | * reappear. | |
2269 | */ | |
2270 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2271 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2272 | if (live_map) |
2273 | limits = md->queue->limits; | |
83d5e5b0 | 2274 | dm_put_live_table_fast(md); |
3ae70656 MS |
2275 | } |
2276 | ||
87eb5b21 MC |
2277 | if (!live_map) { |
2278 | r = dm_calculate_queue_limits(table, &limits); | |
2279 | if (r) { | |
2280 | map = ERR_PTR(r); | |
2281 | goto out; | |
2282 | } | |
042d2a9b | 2283 | } |
754c5fc7 | 2284 | |
042d2a9b | 2285 | map = __bind(md, table, &limits); |
1da177e4 | 2286 | |
93c534ae | 2287 | out: |
e61290a4 | 2288 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2289 | return map; |
1da177e4 LT |
2290 | } |
2291 | ||
2292 | /* | |
2293 | * Functions to lock and unlock any filesystem running on the | |
2294 | * device. | |
2295 | */ | |
2ca3310e | 2296 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2297 | { |
e39e2e95 | 2298 | int r; |
1da177e4 LT |
2299 | |
2300 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2301 | |
db8fef4f | 2302 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2303 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2304 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2305 | md->frozen_sb = NULL; |
2306 | return r; | |
dfbe03f6 AK |
2307 | } |
2308 | ||
aa8d7c2f AK |
2309 | set_bit(DMF_FROZEN, &md->flags); |
2310 | ||
1da177e4 LT |
2311 | return 0; |
2312 | } | |
2313 | ||
2ca3310e | 2314 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2315 | { |
aa8d7c2f AK |
2316 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2317 | return; | |
2318 | ||
db8fef4f | 2319 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2320 | md->frozen_sb = NULL; |
aa8d7c2f | 2321 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2322 | } |
2323 | ||
2324 | /* | |
b48633f8 BVA |
2325 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2326 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2327 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2328 | * | |
ffcc3936 MS |
2329 | * If __dm_suspend returns 0, the device is completely quiescent |
2330 | * now. There is no request-processing activity. All new requests | |
2331 | * are being added to md->deferred list. | |
cec47e3d | 2332 | */ |
ffcc3936 | 2333 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2334 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2335 | int dmf_suspended_flag) |
1da177e4 | 2336 | { |
ffcc3936 MS |
2337 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2338 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2339 | int r; | |
1da177e4 | 2340 | |
5a8f1f80 BVA |
2341 | lockdep_assert_held(&md->suspend_lock); |
2342 | ||
2e93ccc1 KU |
2343 | /* |
2344 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2345 | * This flag is cleared before dm_suspend returns. | |
2346 | */ | |
2347 | if (noflush) | |
2348 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 BVA |
2349 | else |
2350 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); | |
2e93ccc1 | 2351 | |
d67ee213 MS |
2352 | /* |
2353 | * This gets reverted if there's an error later and the targets | |
2354 | * provide the .presuspend_undo hook. | |
2355 | */ | |
cf222b37 AK |
2356 | dm_table_presuspend_targets(map); |
2357 | ||
32a926da | 2358 | /* |
9f518b27 KU |
2359 | * Flush I/O to the device. |
2360 | * Any I/O submitted after lock_fs() may not be flushed. | |
2361 | * noflush takes precedence over do_lockfs. | |
2362 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2363 | */ |
2364 | if (!noflush && do_lockfs) { | |
2365 | r = lock_fs(md); | |
d67ee213 MS |
2366 | if (r) { |
2367 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2368 | return r; |
d67ee213 | 2369 | } |
aa8d7c2f | 2370 | } |
1da177e4 LT |
2371 | |
2372 | /* | |
3b00b203 MP |
2373 | * Here we must make sure that no processes are submitting requests |
2374 | * to target drivers i.e. no one may be executing | |
2375 | * __split_and_process_bio. This is called from dm_request and | |
2376 | * dm_wq_work. | |
2377 | * | |
2378 | * To get all processes out of __split_and_process_bio in dm_request, | |
2379 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2380 | * __split_and_process_bio from dm_request and quiesce the thread |
2381 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2382 | * flush_workqueue(md->wq). | |
1da177e4 | 2383 | */ |
1eb787ec | 2384 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2385 | if (map) |
2386 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2387 | |
d0bcb878 | 2388 | /* |
29e4013d TH |
2389 | * Stop md->queue before flushing md->wq in case request-based |
2390 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2391 | */ |
2eb6e1e3 | 2392 | if (dm_request_based(md)) { |
eca7ee6d | 2393 | dm_stop_queue(md->queue); |
02233342 | 2394 | if (md->kworker_task) |
3989144f | 2395 | kthread_flush_worker(&md->kworker); |
2eb6e1e3 | 2396 | } |
cec47e3d | 2397 | |
d0bcb878 KU |
2398 | flush_workqueue(md->wq); |
2399 | ||
1da177e4 | 2400 | /* |
3b00b203 MP |
2401 | * At this point no more requests are entering target request routines. |
2402 | * We call dm_wait_for_completion to wait for all existing requests | |
2403 | * to finish. | |
1da177e4 | 2404 | */ |
b48633f8 | 2405 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2406 | if (!r) |
2407 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2408 | |
6d6f10df | 2409 | if (noflush) |
022c2611 | 2410 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2411 | if (map) |
2412 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2413 | |
1da177e4 | 2414 | /* were we interrupted ? */ |
46125c1c | 2415 | if (r < 0) { |
9a1fb464 | 2416 | dm_queue_flush(md); |
73d410c0 | 2417 | |
cec47e3d | 2418 | if (dm_request_based(md)) |
eca7ee6d | 2419 | dm_start_queue(md->queue); |
cec47e3d | 2420 | |
2ca3310e | 2421 | unlock_fs(md); |
d67ee213 | 2422 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2423 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2424 | } |
1da177e4 | 2425 | |
ffcc3936 MS |
2426 | return r; |
2427 | } | |
2428 | ||
2429 | /* | |
2430 | * We need to be able to change a mapping table under a mounted | |
2431 | * filesystem. For example we might want to move some data in | |
2432 | * the background. Before the table can be swapped with | |
2433 | * dm_bind_table, dm_suspend must be called to flush any in | |
2434 | * flight bios and ensure that any further io gets deferred. | |
2435 | */ | |
2436 | /* | |
2437 | * Suspend mechanism in request-based dm. | |
2438 | * | |
2439 | * 1. Flush all I/Os by lock_fs() if needed. | |
2440 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2441 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2442 | * | |
2443 | * To abort suspend, start the request_queue. | |
2444 | */ | |
2445 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2446 | { | |
2447 | struct dm_table *map = NULL; | |
2448 | int r = 0; | |
2449 | ||
2450 | retry: | |
2451 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2452 | ||
2453 | if (dm_suspended_md(md)) { | |
2454 | r = -EINVAL; | |
2455 | goto out_unlock; | |
2456 | } | |
2457 | ||
2458 | if (dm_suspended_internally_md(md)) { | |
2459 | /* already internally suspended, wait for internal resume */ | |
2460 | mutex_unlock(&md->suspend_lock); | |
2461 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2462 | if (r) | |
2463 | return r; | |
2464 | goto retry; | |
2465 | } | |
2466 | ||
a12f5d48 | 2467 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2468 | |
eaf9a736 | 2469 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2470 | if (r) |
2471 | goto out_unlock; | |
3b00b203 | 2472 | |
4d4471cb KU |
2473 | dm_table_postsuspend_targets(map); |
2474 | ||
d287483d | 2475 | out_unlock: |
e61290a4 | 2476 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2477 | return r; |
1da177e4 LT |
2478 | } |
2479 | ||
ffcc3936 MS |
2480 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2481 | { | |
2482 | if (map) { | |
2483 | int r = dm_table_resume_targets(map); | |
2484 | if (r) | |
2485 | return r; | |
2486 | } | |
2487 | ||
2488 | dm_queue_flush(md); | |
2489 | ||
2490 | /* | |
2491 | * Flushing deferred I/Os must be done after targets are resumed | |
2492 | * so that mapping of targets can work correctly. | |
2493 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2494 | */ | |
2495 | if (dm_request_based(md)) | |
eca7ee6d | 2496 | dm_start_queue(md->queue); |
ffcc3936 MS |
2497 | |
2498 | unlock_fs(md); | |
2499 | ||
2500 | return 0; | |
2501 | } | |
2502 | ||
1da177e4 LT |
2503 | int dm_resume(struct mapped_device *md) |
2504 | { | |
8dc23658 | 2505 | int r; |
cf222b37 | 2506 | struct dm_table *map = NULL; |
1da177e4 | 2507 | |
ffcc3936 | 2508 | retry: |
8dc23658 | 2509 | r = -EINVAL; |
ffcc3936 MS |
2510 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2511 | ||
4f186f8b | 2512 | if (!dm_suspended_md(md)) |
cf222b37 | 2513 | goto out; |
cf222b37 | 2514 | |
ffcc3936 MS |
2515 | if (dm_suspended_internally_md(md)) { |
2516 | /* already internally suspended, wait for internal resume */ | |
2517 | mutex_unlock(&md->suspend_lock); | |
2518 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2519 | if (r) | |
2520 | return r; | |
2521 | goto retry; | |
2522 | } | |
2523 | ||
a12f5d48 | 2524 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2525 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2526 | goto out; |
1da177e4 | 2527 | |
ffcc3936 | 2528 | r = __dm_resume(md, map); |
8757b776 MB |
2529 | if (r) |
2530 | goto out; | |
2ca3310e | 2531 | |
2ca3310e | 2532 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2533 | out: |
e61290a4 | 2534 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2535 | |
cf222b37 | 2536 | return r; |
1da177e4 LT |
2537 | } |
2538 | ||
fd2ed4d2 MP |
2539 | /* |
2540 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2541 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2542 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2543 | */ |
2544 | ||
ffcc3936 | 2545 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2546 | { |
ffcc3936 MS |
2547 | struct dm_table *map = NULL; |
2548 | ||
1ea0654e BVA |
2549 | lockdep_assert_held(&md->suspend_lock); |
2550 | ||
96b26c8c | 2551 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2552 | return; /* nested internal suspend */ |
2553 | ||
2554 | if (dm_suspended_md(md)) { | |
2555 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2556 | return; /* nest suspend */ | |
2557 | } | |
2558 | ||
a12f5d48 | 2559 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2560 | |
2561 | /* | |
2562 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2563 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2564 | * would require changing .presuspend to return an error -- avoid this | |
2565 | * until there is a need for more elaborate variants of internal suspend. | |
2566 | */ | |
eaf9a736 MS |
2567 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2568 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 MS |
2569 | |
2570 | dm_table_postsuspend_targets(map); | |
2571 | } | |
2572 | ||
2573 | static void __dm_internal_resume(struct mapped_device *md) | |
2574 | { | |
96b26c8c MP |
2575 | BUG_ON(!md->internal_suspend_count); |
2576 | ||
2577 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2578 | return; /* resume from nested internal suspend */ |
2579 | ||
fd2ed4d2 | 2580 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2581 | goto done; /* resume from nested suspend */ |
2582 | ||
2583 | /* | |
2584 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2585 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2586 | */ | |
2587 | (void) __dm_resume(md, NULL); | |
2588 | ||
2589 | done: | |
2590 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2591 | smp_mb__after_atomic(); | |
2592 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2593 | } | |
2594 | ||
2595 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2596 | { | |
2597 | mutex_lock(&md->suspend_lock); | |
2598 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2599 | mutex_unlock(&md->suspend_lock); | |
2600 | } | |
2601 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2602 | ||
2603 | void dm_internal_resume(struct mapped_device *md) | |
2604 | { | |
2605 | mutex_lock(&md->suspend_lock); | |
2606 | __dm_internal_resume(md); | |
2607 | mutex_unlock(&md->suspend_lock); | |
2608 | } | |
2609 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2610 | ||
2611 | /* | |
2612 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2613 | * which prevents interaction with userspace-driven suspend. | |
2614 | */ | |
2615 | ||
2616 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2617 | { | |
2618 | mutex_lock(&md->suspend_lock); | |
2619 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2620 | return; |
2621 | ||
2622 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2623 | synchronize_srcu(&md->io_barrier); | |
2624 | flush_workqueue(md->wq); | |
2625 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2626 | } | |
b735fede | 2627 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2628 | |
ffcc3936 | 2629 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2630 | { |
ffcc3936 | 2631 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2632 | goto done; |
2633 | ||
2634 | dm_queue_flush(md); | |
2635 | ||
2636 | done: | |
2637 | mutex_unlock(&md->suspend_lock); | |
2638 | } | |
b735fede | 2639 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2640 | |
1da177e4 LT |
2641 | /*----------------------------------------------------------------- |
2642 | * Event notification. | |
2643 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2644 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2645 | unsigned cookie) |
69267a30 | 2646 | { |
60935eb2 MB |
2647 | char udev_cookie[DM_COOKIE_LENGTH]; |
2648 | char *envp[] = { udev_cookie, NULL }; | |
2649 | ||
2650 | if (!cookie) | |
3abf85b5 | 2651 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2652 | else { |
2653 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2654 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2655 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2656 | action, envp); | |
60935eb2 | 2657 | } |
69267a30 AK |
2658 | } |
2659 | ||
7a8c3d3b MA |
2660 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2661 | { | |
2662 | return atomic_add_return(1, &md->uevent_seq); | |
2663 | } | |
2664 | ||
1da177e4 LT |
2665 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2666 | { | |
2667 | return atomic_read(&md->event_nr); | |
2668 | } | |
2669 | ||
2670 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2671 | { | |
2672 | return wait_event_interruptible(md->eventq, | |
2673 | (event_nr != atomic_read(&md->event_nr))); | |
2674 | } | |
2675 | ||
7a8c3d3b MA |
2676 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2677 | { | |
2678 | unsigned long flags; | |
2679 | ||
2680 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2681 | list_add(elist, &md->uevent_list); | |
2682 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2683 | } | |
2684 | ||
1da177e4 LT |
2685 | /* |
2686 | * The gendisk is only valid as long as you have a reference | |
2687 | * count on 'md'. | |
2688 | */ | |
2689 | struct gendisk *dm_disk(struct mapped_device *md) | |
2690 | { | |
2691 | return md->disk; | |
2692 | } | |
65ff5b7d | 2693 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2694 | |
784aae73 MB |
2695 | struct kobject *dm_kobject(struct mapped_device *md) |
2696 | { | |
2995fa78 | 2697 | return &md->kobj_holder.kobj; |
784aae73 MB |
2698 | } |
2699 | ||
784aae73 MB |
2700 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2701 | { | |
2702 | struct mapped_device *md; | |
2703 | ||
2995fa78 | 2704 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2705 | |
4d89b7b4 | 2706 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2707 | dm_deleting_md(md)) |
4d89b7b4 MB |
2708 | return NULL; |
2709 | ||
784aae73 MB |
2710 | dm_get(md); |
2711 | return md; | |
2712 | } | |
2713 | ||
4f186f8b | 2714 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2715 | { |
2716 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2717 | } | |
2718 | ||
ffcc3936 MS |
2719 | int dm_suspended_internally_md(struct mapped_device *md) |
2720 | { | |
2721 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2722 | } | |
2723 | ||
2c140a24 MP |
2724 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2725 | { | |
2726 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2727 | } | |
2728 | ||
64dbce58 KU |
2729 | int dm_suspended(struct dm_target *ti) |
2730 | { | |
ecdb2e25 | 2731 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2732 | } |
2733 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2734 | ||
2e93ccc1 KU |
2735 | int dm_noflush_suspending(struct dm_target *ti) |
2736 | { | |
ecdb2e25 | 2737 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2738 | } |
2739 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2740 | ||
7e0d574f | 2741 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
30187e1d | 2742 | unsigned integrity, unsigned per_io_data_size) |
e6ee8c0b | 2743 | { |
115485e8 | 2744 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2745 | unsigned int pool_size = 0; |
5f015204 | 2746 | unsigned int front_pad; |
e6ee8c0b KU |
2747 | |
2748 | if (!pools) | |
4e6e36c3 | 2749 | return NULL; |
e6ee8c0b | 2750 | |
78d8e58a MS |
2751 | switch (type) { |
2752 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2753 | case DM_TYPE_DAX_BIO_BASED: |
78d8e58a | 2754 | pool_size = dm_get_reserved_bio_based_ios(); |
30187e1d | 2755 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
eb8db831 CH |
2756 | |
2757 | pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); | |
2758 | if (!pools->io_pool) | |
2759 | goto out; | |
78d8e58a MS |
2760 | break; |
2761 | case DM_TYPE_REQUEST_BASED: | |
78d8e58a | 2762 | case DM_TYPE_MQ_REQUEST_BASED: |
eb8db831 | 2763 | pool_size = dm_get_reserved_rq_based_ios(); |
78d8e58a | 2764 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2765 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2766 | break; |
2767 | default: | |
2768 | BUG(); | |
2769 | } | |
2770 | ||
47e0fb46 | 2771 | pools->bs = bioset_create(pool_size, front_pad, BIOSET_NEED_RESCUER); |
e6ee8c0b | 2772 | if (!pools->bs) |
5f015204 | 2773 | goto out; |
e6ee8c0b | 2774 | |
a91a2785 | 2775 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2776 | goto out; |
a91a2785 | 2777 | |
e6ee8c0b | 2778 | return pools; |
5f1b670d | 2779 | |
5f1b670d CH |
2780 | out: |
2781 | dm_free_md_mempools(pools); | |
78d8e58a | 2782 | |
4e6e36c3 | 2783 | return NULL; |
e6ee8c0b KU |
2784 | } |
2785 | ||
2786 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2787 | { | |
2788 | if (!pools) | |
2789 | return; | |
2790 | ||
6f65985e | 2791 | mempool_destroy(pools->io_pool); |
1ae49ea2 | 2792 | |
e6ee8c0b KU |
2793 | if (pools->bs) |
2794 | bioset_free(pools->bs); | |
2795 | ||
2796 | kfree(pools); | |
2797 | } | |
2798 | ||
9c72bad1 CH |
2799 | struct dm_pr { |
2800 | u64 old_key; | |
2801 | u64 new_key; | |
2802 | u32 flags; | |
2803 | bool fail_early; | |
2804 | }; | |
2805 | ||
2806 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2807 | void *data) | |
71cdb697 CH |
2808 | { |
2809 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2810 | struct dm_table *table; |
2811 | struct dm_target *ti; | |
2812 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2813 | |
9c72bad1 CH |
2814 | table = dm_get_live_table(md, &srcu_idx); |
2815 | if (!table || !dm_table_get_size(table)) | |
2816 | goto out; | |
71cdb697 | 2817 | |
9c72bad1 CH |
2818 | /* We only support devices that have a single target */ |
2819 | if (dm_table_get_num_targets(table) != 1) | |
2820 | goto out; | |
2821 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2822 | |
9c72bad1 CH |
2823 | ret = -EINVAL; |
2824 | if (!ti->type->iterate_devices) | |
2825 | goto out; | |
2826 | ||
2827 | ret = ti->type->iterate_devices(ti, fn, data); | |
2828 | out: | |
2829 | dm_put_live_table(md, srcu_idx); | |
2830 | return ret; | |
2831 | } | |
2832 | ||
2833 | /* | |
2834 | * For register / unregister we need to manually call out to every path. | |
2835 | */ | |
2836 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
2837 | sector_t start, sector_t len, void *data) | |
2838 | { | |
2839 | struct dm_pr *pr = data; | |
2840 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
2841 | ||
2842 | if (!ops || !ops->pr_register) | |
2843 | return -EOPNOTSUPP; | |
2844 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
2845 | } | |
2846 | ||
2847 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
2848 | u32 flags) | |
2849 | { | |
2850 | struct dm_pr pr = { | |
2851 | .old_key = old_key, | |
2852 | .new_key = new_key, | |
2853 | .flags = flags, | |
2854 | .fail_early = true, | |
2855 | }; | |
2856 | int ret; | |
2857 | ||
2858 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
2859 | if (ret && new_key) { | |
2860 | /* unregister all paths if we failed to register any path */ | |
2861 | pr.old_key = new_key; | |
2862 | pr.new_key = 0; | |
2863 | pr.flags = 0; | |
2864 | pr.fail_early = false; | |
2865 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
2866 | } | |
2867 | ||
2868 | return ret; | |
71cdb697 CH |
2869 | } |
2870 | ||
2871 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 2872 | u32 flags) |
71cdb697 CH |
2873 | { |
2874 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2875 | const struct pr_ops *ops; | |
71cdb697 | 2876 | fmode_t mode; |
956a4025 | 2877 | int r; |
71cdb697 | 2878 | |
956a4025 | 2879 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2880 | if (r < 0) |
2881 | return r; | |
2882 | ||
2883 | ops = bdev->bd_disk->fops->pr_ops; | |
2884 | if (ops && ops->pr_reserve) | |
2885 | r = ops->pr_reserve(bdev, key, type, flags); | |
2886 | else | |
2887 | r = -EOPNOTSUPP; | |
2888 | ||
956a4025 | 2889 | bdput(bdev); |
71cdb697 CH |
2890 | return r; |
2891 | } | |
2892 | ||
2893 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
2894 | { | |
2895 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2896 | const struct pr_ops *ops; | |
71cdb697 | 2897 | fmode_t mode; |
956a4025 | 2898 | int r; |
71cdb697 | 2899 | |
956a4025 | 2900 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2901 | if (r < 0) |
2902 | return r; | |
2903 | ||
2904 | ops = bdev->bd_disk->fops->pr_ops; | |
2905 | if (ops && ops->pr_release) | |
2906 | r = ops->pr_release(bdev, key, type); | |
2907 | else | |
2908 | r = -EOPNOTSUPP; | |
2909 | ||
956a4025 | 2910 | bdput(bdev); |
71cdb697 CH |
2911 | return r; |
2912 | } | |
2913 | ||
2914 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 2915 | enum pr_type type, bool abort) |
71cdb697 CH |
2916 | { |
2917 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2918 | const struct pr_ops *ops; | |
71cdb697 | 2919 | fmode_t mode; |
956a4025 | 2920 | int r; |
71cdb697 | 2921 | |
956a4025 | 2922 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2923 | if (r < 0) |
2924 | return r; | |
2925 | ||
2926 | ops = bdev->bd_disk->fops->pr_ops; | |
2927 | if (ops && ops->pr_preempt) | |
2928 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
2929 | else | |
2930 | r = -EOPNOTSUPP; | |
2931 | ||
956a4025 | 2932 | bdput(bdev); |
71cdb697 CH |
2933 | return r; |
2934 | } | |
2935 | ||
2936 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
2937 | { | |
2938 | struct mapped_device *md = bdev->bd_disk->private_data; | |
2939 | const struct pr_ops *ops; | |
71cdb697 | 2940 | fmode_t mode; |
956a4025 | 2941 | int r; |
71cdb697 | 2942 | |
956a4025 | 2943 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
2944 | if (r < 0) |
2945 | return r; | |
2946 | ||
2947 | ops = bdev->bd_disk->fops->pr_ops; | |
2948 | if (ops && ops->pr_clear) | |
2949 | r = ops->pr_clear(bdev, key); | |
2950 | else | |
2951 | r = -EOPNOTSUPP; | |
2952 | ||
956a4025 | 2953 | bdput(bdev); |
71cdb697 CH |
2954 | return r; |
2955 | } | |
2956 | ||
2957 | static const struct pr_ops dm_pr_ops = { | |
2958 | .pr_register = dm_pr_register, | |
2959 | .pr_reserve = dm_pr_reserve, | |
2960 | .pr_release = dm_pr_release, | |
2961 | .pr_preempt = dm_pr_preempt, | |
2962 | .pr_clear = dm_pr_clear, | |
2963 | }; | |
2964 | ||
83d5cde4 | 2965 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2966 | .open = dm_blk_open, |
2967 | .release = dm_blk_close, | |
aa129a22 | 2968 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2969 | .getgeo = dm_blk_getgeo, |
71cdb697 | 2970 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
2971 | .owner = THIS_MODULE |
2972 | }; | |
2973 | ||
f26c5719 DW |
2974 | static const struct dax_operations dm_dax_ops = { |
2975 | .direct_access = dm_dax_direct_access, | |
7e026c8c | 2976 | .copy_from_iter = dm_dax_copy_from_iter, |
f26c5719 DW |
2977 | }; |
2978 | ||
1da177e4 LT |
2979 | /* |
2980 | * module hooks | |
2981 | */ | |
2982 | module_init(dm_init); | |
2983 | module_exit(dm_exit); | |
2984 | ||
2985 | module_param(major, uint, 0); | |
2986 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 2987 | |
e8603136 MS |
2988 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
2989 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
2990 | ||
115485e8 MS |
2991 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
2992 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
2993 | ||
1da177e4 LT |
2994 | MODULE_DESCRIPTION(DM_NAME " driver"); |
2995 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2996 | MODULE_LICENSE("GPL"); |