]>
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 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
ffcc3936 | 22 | #include <linux/wait.h> |
2eb6e1e3 | 23 | #include <linux/kthread.h> |
0ce65797 | 24 | #include <linux/ktime.h> |
de3ec86d | 25 | #include <linux/elevator.h> /* for rq_end_sector() */ |
bfebd1cd | 26 | #include <linux/blk-mq.h> |
71cdb697 | 27 | #include <linux/pr.h> |
55782138 LZ |
28 | |
29 | #include <trace/events/block.h> | |
1da177e4 | 30 | |
72d94861 AK |
31 | #define DM_MSG_PREFIX "core" |
32 | ||
71a16736 NK |
33 | #ifdef CONFIG_PRINTK |
34 | /* | |
35 | * ratelimit state to be used in DMXXX_LIMIT(). | |
36 | */ | |
37 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
38 | DEFAULT_RATELIMIT_INTERVAL, | |
39 | DEFAULT_RATELIMIT_BURST); | |
40 | EXPORT_SYMBOL(dm_ratelimit_state); | |
41 | #endif | |
42 | ||
60935eb2 MB |
43 | /* |
44 | * Cookies are numeric values sent with CHANGE and REMOVE | |
45 | * uevents while resuming, removing or renaming the device. | |
46 | */ | |
47 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
48 | #define DM_COOKIE_LENGTH 24 | |
49 | ||
1da177e4 LT |
50 | static const char *_name = DM_NAME; |
51 | ||
52 | static unsigned int major = 0; | |
53 | static unsigned int _major = 0; | |
54 | ||
d15b774c AK |
55 | static DEFINE_IDR(_minor_idr); |
56 | ||
f32c10b0 | 57 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
58 | |
59 | static void do_deferred_remove(struct work_struct *w); | |
60 | ||
61 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
62 | ||
acfe0ad7 MP |
63 | static struct workqueue_struct *deferred_remove_workqueue; |
64 | ||
1da177e4 | 65 | /* |
8fbf26ad | 66 | * For bio-based dm. |
1da177e4 LT |
67 | * One of these is allocated per bio. |
68 | */ | |
69 | struct dm_io { | |
70 | struct mapped_device *md; | |
71 | int error; | |
1da177e4 | 72 | atomic_t io_count; |
6ae2fa67 | 73 | struct bio *bio; |
3eaf840e | 74 | unsigned long start_time; |
f88fb981 | 75 | spinlock_t endio_lock; |
fd2ed4d2 | 76 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
77 | }; |
78 | ||
8fbf26ad KU |
79 | /* |
80 | * For request-based dm. | |
81 | * One of these is allocated per request. | |
82 | */ | |
83 | struct dm_rq_target_io { | |
84 | struct mapped_device *md; | |
85 | struct dm_target *ti; | |
1ae49ea2 | 86 | struct request *orig, *clone; |
2eb6e1e3 | 87 | struct kthread_work work; |
8fbf26ad KU |
88 | int error; |
89 | union map_info info; | |
e262f347 MP |
90 | struct dm_stats_aux stats_aux; |
91 | unsigned long duration_jiffies; | |
92 | unsigned n_sectors; | |
8fbf26ad KU |
93 | }; |
94 | ||
95 | /* | |
94818742 KO |
96 | * For request-based dm - the bio clones we allocate are embedded in these |
97 | * structs. | |
98 | * | |
99 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
100 | * the bioset is created - this means the bio has to come at the end of the | |
101 | * struct. | |
8fbf26ad KU |
102 | */ |
103 | struct dm_rq_clone_bio_info { | |
104 | struct bio *orig; | |
cec47e3d | 105 | struct dm_rq_target_io *tio; |
94818742 | 106 | struct bio clone; |
8fbf26ad KU |
107 | }; |
108 | ||
ba61fdd1 JM |
109 | #define MINOR_ALLOCED ((void *)-1) |
110 | ||
1da177e4 LT |
111 | /* |
112 | * Bits for the md->flags field. | |
113 | */ | |
1eb787ec | 114 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 115 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 116 | #define DMF_FROZEN 2 |
fba9f90e | 117 | #define DMF_FREEING 3 |
5c6bd75d | 118 | #define DMF_DELETING 4 |
2e93ccc1 | 119 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
120 | #define DMF_DEFERRED_REMOVE 6 |
121 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 122 | |
304f3f6a MB |
123 | /* |
124 | * Work processed by per-device workqueue. | |
125 | */ | |
1da177e4 | 126 | struct mapped_device { |
83d5e5b0 | 127 | struct srcu_struct io_barrier; |
e61290a4 | 128 | struct mutex suspend_lock; |
1da177e4 | 129 | |
2a7faeb1 | 130 | /* |
1d3aa6f6 | 131 | * The current mapping (struct dm_table *). |
2a7faeb1 MP |
132 | * Use dm_get_live_table{_fast} or take suspend_lock for |
133 | * dereference. | |
134 | */ | |
1d3aa6f6 | 135 | void __rcu *map; |
2a7faeb1 | 136 | |
86f1152b BM |
137 | struct list_head table_devices; |
138 | struct mutex table_devices_lock; | |
139 | ||
1da177e4 LT |
140 | unsigned long flags; |
141 | ||
165125e1 | 142 | struct request_queue *queue; |
115485e8 MS |
143 | int numa_node_id; |
144 | ||
a5664dad | 145 | unsigned type; |
4a0b4ddf | 146 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
147 | struct mutex type_lock; |
148 | ||
032482fd MS |
149 | atomic_t holders; |
150 | atomic_t open_count; | |
151 | ||
16f12266 | 152 | struct dm_target *immutable_target; |
36a0456f AK |
153 | struct target_type *immutable_target_type; |
154 | ||
1da177e4 | 155 | struct gendisk *disk; |
7e51f257 | 156 | char name[16]; |
1da177e4 LT |
157 | |
158 | void *interface_ptr; | |
159 | ||
160 | /* | |
161 | * A list of ios that arrived while we were suspended. | |
162 | */ | |
316d315b | 163 | atomic_t pending[2]; |
1da177e4 | 164 | wait_queue_head_t wait; |
53d5914f | 165 | struct work_struct work; |
022c2611 | 166 | spinlock_t deferred_lock; |
032482fd MS |
167 | struct bio_list deferred; |
168 | ||
169 | /* | |
170 | * Event handling. | |
171 | */ | |
172 | wait_queue_head_t eventq; | |
173 | atomic_t event_nr; | |
174 | atomic_t uevent_seq; | |
175 | struct list_head uevent_list; | |
176 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
177 | ||
178 | /* the number of internal suspends */ | |
179 | unsigned internal_suspend_count; | |
1da177e4 | 180 | |
af7e466a | 181 | /* |
29e4013d | 182 | * Processing queue (flush) |
304f3f6a MB |
183 | */ |
184 | struct workqueue_struct *wq; | |
185 | ||
1da177e4 LT |
186 | /* |
187 | * io objects are allocated from here. | |
188 | */ | |
189 | mempool_t *io_pool; | |
1ae49ea2 | 190 | mempool_t *rq_pool; |
1da177e4 | 191 | |
9faf400f SB |
192 | struct bio_set *bs; |
193 | ||
1da177e4 LT |
194 | /* |
195 | * freeze/thaw support require holding onto a super block | |
196 | */ | |
197 | struct super_block *frozen_sb; | |
3ac51e74 DW |
198 | |
199 | /* forced geometry settings */ | |
200 | struct hd_geometry geometry; | |
784aae73 | 201 | |
032482fd MS |
202 | struct block_device *bdev; |
203 | ||
2995fa78 MP |
204 | /* kobject and completion */ |
205 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 206 | |
d87f4c14 TH |
207 | /* zero-length flush that will be cloned and submitted to targets */ |
208 | struct bio flush_bio; | |
fd2ed4d2 MP |
209 | |
210 | struct dm_stats stats; | |
2eb6e1e3 KB |
211 | |
212 | struct kthread_worker kworker; | |
213 | struct task_struct *kworker_task; | |
de3ec86d MS |
214 | |
215 | /* for request-based merge heuristic in dm_request_fn() */ | |
0ce65797 | 216 | unsigned seq_rq_merge_deadline_usecs; |
de3ec86d | 217 | int last_rq_rw; |
0ce65797 MS |
218 | sector_t last_rq_pos; |
219 | ktime_t last_rq_start_time; | |
bfebd1cd MS |
220 | |
221 | /* for blk-mq request-based DM support */ | |
1c357a1e | 222 | struct blk_mq_tag_set *tag_set; |
591ddcfc MS |
223 | bool use_blk_mq:1; |
224 | bool init_tio_pdu:1; | |
1da177e4 LT |
225 | }; |
226 | ||
17e149b8 MS |
227 | #ifdef CONFIG_DM_MQ_DEFAULT |
228 | static bool use_blk_mq = true; | |
229 | #else | |
230 | static bool use_blk_mq = false; | |
231 | #endif | |
232 | ||
faad87df MS |
233 | #define DM_MQ_NR_HW_QUEUES 1 |
234 | #define DM_MQ_QUEUE_DEPTH 2048 | |
115485e8 | 235 | #define DM_NUMA_NODE NUMA_NO_NODE |
faad87df MS |
236 | |
237 | static unsigned dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES; | |
238 | static unsigned dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH; | |
115485e8 | 239 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 240 | |
17e149b8 MS |
241 | bool dm_use_blk_mq(struct mapped_device *md) |
242 | { | |
243 | return md->use_blk_mq; | |
244 | } | |
591ddcfc | 245 | EXPORT_SYMBOL_GPL(dm_use_blk_mq); |
17e149b8 | 246 | |
e6ee8c0b KU |
247 | /* |
248 | * For mempools pre-allocation at the table loading time. | |
249 | */ | |
250 | struct dm_md_mempools { | |
251 | mempool_t *io_pool; | |
1ae49ea2 | 252 | mempool_t *rq_pool; |
e6ee8c0b KU |
253 | struct bio_set *bs; |
254 | }; | |
255 | ||
86f1152b BM |
256 | struct table_device { |
257 | struct list_head list; | |
258 | atomic_t count; | |
259 | struct dm_dev dm_dev; | |
260 | }; | |
261 | ||
6cfa5857 MS |
262 | #define RESERVED_BIO_BASED_IOS 16 |
263 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 264 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 265 | static struct kmem_cache *_io_cache; |
8fbf26ad | 266 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 267 | static struct kmem_cache *_rq_cache; |
94818742 | 268 | |
e8603136 MS |
269 | /* |
270 | * Bio-based DM's mempools' reserved IOs set by the user. | |
271 | */ | |
272 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
273 | ||
f4790826 MS |
274 | /* |
275 | * Request-based DM's mempools' reserved IOs set by the user. | |
276 | */ | |
277 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
278 | ||
115485e8 MS |
279 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
280 | { | |
281 | int param = ACCESS_ONCE(*module_param); | |
282 | int modified_param = 0; | |
283 | bool modified = true; | |
284 | ||
285 | if (param < min) | |
286 | modified_param = min; | |
287 | else if (param > max) | |
288 | modified_param = max; | |
289 | else | |
290 | modified = false; | |
291 | ||
292 | if (modified) { | |
293 | (void)cmpxchg(module_param, param, modified_param); | |
294 | param = modified_param; | |
295 | } | |
296 | ||
297 | return param; | |
298 | } | |
299 | ||
09c2d531 | 300 | static unsigned __dm_get_module_param(unsigned *module_param, |
f4790826 MS |
301 | unsigned def, unsigned max) |
302 | { | |
09c2d531 MS |
303 | unsigned param = ACCESS_ONCE(*module_param); |
304 | unsigned modified_param = 0; | |
f4790826 | 305 | |
09c2d531 MS |
306 | if (!param) |
307 | modified_param = def; | |
308 | else if (param > max) | |
309 | modified_param = max; | |
f4790826 | 310 | |
09c2d531 MS |
311 | if (modified_param) { |
312 | (void)cmpxchg(module_param, param, modified_param); | |
313 | param = modified_param; | |
f4790826 MS |
314 | } |
315 | ||
09c2d531 | 316 | return param; |
f4790826 MS |
317 | } |
318 | ||
e8603136 MS |
319 | unsigned dm_get_reserved_bio_based_ios(void) |
320 | { | |
09c2d531 | 321 | return __dm_get_module_param(&reserved_bio_based_ios, |
e8603136 MS |
322 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); |
323 | } | |
324 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
325 | ||
f4790826 MS |
326 | unsigned dm_get_reserved_rq_based_ios(void) |
327 | { | |
09c2d531 | 328 | return __dm_get_module_param(&reserved_rq_based_ios, |
f4790826 MS |
329 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); |
330 | } | |
331 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
332 | ||
faad87df MS |
333 | static unsigned dm_get_blk_mq_nr_hw_queues(void) |
334 | { | |
335 | return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32); | |
336 | } | |
337 | ||
338 | static unsigned dm_get_blk_mq_queue_depth(void) | |
339 | { | |
340 | return __dm_get_module_param(&dm_mq_queue_depth, | |
341 | DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH); | |
342 | } | |
343 | ||
115485e8 MS |
344 | static unsigned dm_get_numa_node(void) |
345 | { | |
346 | return __dm_get_module_param_int(&dm_numa_node, | |
347 | DM_NUMA_NODE, num_online_nodes() - 1); | |
348 | } | |
349 | ||
1da177e4 LT |
350 | static int __init local_init(void) |
351 | { | |
51157b4a | 352 | int r = -ENOMEM; |
1da177e4 | 353 | |
1da177e4 | 354 | /* allocate a slab for the dm_ios */ |
028867ac | 355 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 356 | if (!_io_cache) |
51157b4a | 357 | return r; |
1da177e4 | 358 | |
8fbf26ad KU |
359 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
360 | if (!_rq_tio_cache) | |
dba14160 | 361 | goto out_free_io_cache; |
8fbf26ad | 362 | |
eca7ee6d | 363 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
1ae49ea2 MS |
364 | __alignof__(struct request), 0, NULL); |
365 | if (!_rq_cache) | |
366 | goto out_free_rq_tio_cache; | |
367 | ||
51e5b2bd | 368 | r = dm_uevent_init(); |
51157b4a | 369 | if (r) |
1ae49ea2 | 370 | goto out_free_rq_cache; |
51e5b2bd | 371 | |
acfe0ad7 MP |
372 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
373 | if (!deferred_remove_workqueue) { | |
374 | r = -ENOMEM; | |
375 | goto out_uevent_exit; | |
376 | } | |
377 | ||
1da177e4 LT |
378 | _major = major; |
379 | r = register_blkdev(_major, _name); | |
51157b4a | 380 | if (r < 0) |
acfe0ad7 | 381 | goto out_free_workqueue; |
1da177e4 LT |
382 | |
383 | if (!_major) | |
384 | _major = r; | |
385 | ||
386 | return 0; | |
51157b4a | 387 | |
acfe0ad7 MP |
388 | out_free_workqueue: |
389 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
390 | out_uevent_exit: |
391 | dm_uevent_exit(); | |
1ae49ea2 MS |
392 | out_free_rq_cache: |
393 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
394 | out_free_rq_tio_cache: |
395 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
396 | out_free_io_cache: |
397 | kmem_cache_destroy(_io_cache); | |
398 | ||
399 | return r; | |
1da177e4 LT |
400 | } |
401 | ||
402 | static void local_exit(void) | |
403 | { | |
2c140a24 | 404 | flush_scheduled_work(); |
acfe0ad7 | 405 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 406 | |
1ae49ea2 | 407 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 408 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 409 | kmem_cache_destroy(_io_cache); |
00d59405 | 410 | unregister_blkdev(_major, _name); |
51e5b2bd | 411 | dm_uevent_exit(); |
1da177e4 LT |
412 | |
413 | _major = 0; | |
414 | ||
415 | DMINFO("cleaned up"); | |
416 | } | |
417 | ||
b9249e55 | 418 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
419 | local_init, |
420 | dm_target_init, | |
421 | dm_linear_init, | |
422 | dm_stripe_init, | |
952b3557 | 423 | dm_io_init, |
945fa4d2 | 424 | dm_kcopyd_init, |
1da177e4 | 425 | dm_interface_init, |
fd2ed4d2 | 426 | dm_statistics_init, |
1da177e4 LT |
427 | }; |
428 | ||
b9249e55 | 429 | static void (*_exits[])(void) = { |
1da177e4 LT |
430 | local_exit, |
431 | dm_target_exit, | |
432 | dm_linear_exit, | |
433 | dm_stripe_exit, | |
952b3557 | 434 | dm_io_exit, |
945fa4d2 | 435 | dm_kcopyd_exit, |
1da177e4 | 436 | dm_interface_exit, |
fd2ed4d2 | 437 | dm_statistics_exit, |
1da177e4 LT |
438 | }; |
439 | ||
440 | static int __init dm_init(void) | |
441 | { | |
442 | const int count = ARRAY_SIZE(_inits); | |
443 | ||
444 | int r, i; | |
445 | ||
446 | for (i = 0; i < count; i++) { | |
447 | r = _inits[i](); | |
448 | if (r) | |
449 | goto bad; | |
450 | } | |
451 | ||
452 | return 0; | |
453 | ||
454 | bad: | |
455 | while (i--) | |
456 | _exits[i](); | |
457 | ||
458 | return r; | |
459 | } | |
460 | ||
461 | static void __exit dm_exit(void) | |
462 | { | |
463 | int i = ARRAY_SIZE(_exits); | |
464 | ||
465 | while (i--) | |
466 | _exits[i](); | |
d15b774c AK |
467 | |
468 | /* | |
469 | * Should be empty by this point. | |
470 | */ | |
d15b774c | 471 | idr_destroy(&_minor_idr); |
1da177e4 LT |
472 | } |
473 | ||
474 | /* | |
475 | * Block device functions | |
476 | */ | |
432a212c MA |
477 | int dm_deleting_md(struct mapped_device *md) |
478 | { | |
479 | return test_bit(DMF_DELETING, &md->flags); | |
480 | } | |
481 | ||
fe5f9f2c | 482 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
483 | { |
484 | struct mapped_device *md; | |
485 | ||
fba9f90e JM |
486 | spin_lock(&_minor_lock); |
487 | ||
fe5f9f2c | 488 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
489 | if (!md) |
490 | goto out; | |
491 | ||
5c6bd75d | 492 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 493 | dm_deleting_md(md)) { |
fba9f90e JM |
494 | md = NULL; |
495 | goto out; | |
496 | } | |
497 | ||
1da177e4 | 498 | dm_get(md); |
5c6bd75d | 499 | atomic_inc(&md->open_count); |
fba9f90e JM |
500 | out: |
501 | spin_unlock(&_minor_lock); | |
502 | ||
503 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
504 | } |
505 | ||
db2a144b | 506 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 507 | { |
63a4f065 | 508 | struct mapped_device *md; |
6e9624b8 | 509 | |
4a1aeb98 MB |
510 | spin_lock(&_minor_lock); |
511 | ||
63a4f065 MS |
512 | md = disk->private_data; |
513 | if (WARN_ON(!md)) | |
514 | goto out; | |
515 | ||
2c140a24 MP |
516 | if (atomic_dec_and_test(&md->open_count) && |
517 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 518 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 519 | |
1da177e4 | 520 | dm_put(md); |
63a4f065 | 521 | out: |
4a1aeb98 | 522 | spin_unlock(&_minor_lock); |
1da177e4 LT |
523 | } |
524 | ||
5c6bd75d AK |
525 | int dm_open_count(struct mapped_device *md) |
526 | { | |
527 | return atomic_read(&md->open_count); | |
528 | } | |
529 | ||
530 | /* | |
531 | * Guarantees nothing is using the device before it's deleted. | |
532 | */ | |
2c140a24 | 533 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
534 | { |
535 | int r = 0; | |
536 | ||
537 | spin_lock(&_minor_lock); | |
538 | ||
2c140a24 | 539 | if (dm_open_count(md)) { |
5c6bd75d | 540 | r = -EBUSY; |
2c140a24 MP |
541 | if (mark_deferred) |
542 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
543 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
544 | r = -EEXIST; | |
5c6bd75d AK |
545 | else |
546 | set_bit(DMF_DELETING, &md->flags); | |
547 | ||
548 | spin_unlock(&_minor_lock); | |
549 | ||
550 | return r; | |
551 | } | |
552 | ||
2c140a24 MP |
553 | int dm_cancel_deferred_remove(struct mapped_device *md) |
554 | { | |
555 | int r = 0; | |
556 | ||
557 | spin_lock(&_minor_lock); | |
558 | ||
559 | if (test_bit(DMF_DELETING, &md->flags)) | |
560 | r = -EBUSY; | |
561 | else | |
562 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
563 | ||
564 | spin_unlock(&_minor_lock); | |
565 | ||
566 | return r; | |
567 | } | |
568 | ||
569 | static void do_deferred_remove(struct work_struct *w) | |
570 | { | |
571 | dm_deferred_remove(); | |
572 | } | |
573 | ||
fd2ed4d2 MP |
574 | sector_t dm_get_size(struct mapped_device *md) |
575 | { | |
576 | return get_capacity(md->disk); | |
577 | } | |
578 | ||
9974fa2c MS |
579 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
580 | { | |
581 | return md->queue; | |
582 | } | |
583 | ||
fd2ed4d2 MP |
584 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
585 | { | |
586 | return &md->stats; | |
587 | } | |
588 | ||
3ac51e74 DW |
589 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
590 | { | |
591 | struct mapped_device *md = bdev->bd_disk->private_data; | |
592 | ||
593 | return dm_get_geometry(md, geo); | |
594 | } | |
595 | ||
956a4025 MS |
596 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
597 | struct block_device **bdev, | |
598 | fmode_t *mode) | |
aa129a22 | 599 | { |
66482026 | 600 | struct dm_target *tgt; |
6c182cd8 | 601 | struct dm_table *map; |
956a4025 | 602 | int srcu_idx, r; |
aa129a22 | 603 | |
6c182cd8 | 604 | retry: |
e56f81e0 | 605 | r = -ENOTTY; |
956a4025 | 606 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
607 | if (!map || !dm_table_get_size(map)) |
608 | goto out; | |
609 | ||
610 | /* We only support devices that have a single target */ | |
611 | if (dm_table_get_num_targets(map) != 1) | |
612 | goto out; | |
613 | ||
66482026 MS |
614 | tgt = dm_table_get_target(map, 0); |
615 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 616 | goto out; |
aa129a22 | 617 | |
4f186f8b | 618 | if (dm_suspended_md(md)) { |
aa129a22 MB |
619 | r = -EAGAIN; |
620 | goto out; | |
621 | } | |
622 | ||
66482026 | 623 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
624 | if (r < 0) |
625 | goto out; | |
aa129a22 | 626 | |
956a4025 MS |
627 | bdgrab(*bdev); |
628 | dm_put_live_table(md, srcu_idx); | |
e56f81e0 | 629 | return r; |
aa129a22 | 630 | |
aa129a22 | 631 | out: |
956a4025 | 632 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 633 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
634 | msleep(10); |
635 | goto retry; | |
636 | } | |
e56f81e0 CH |
637 | return r; |
638 | } | |
639 | ||
640 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
641 | unsigned int cmd, unsigned long arg) | |
642 | { | |
643 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 644 | int r; |
e56f81e0 | 645 | |
956a4025 | 646 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
647 | if (r < 0) |
648 | return r; | |
6c182cd8 | 649 | |
e56f81e0 CH |
650 | if (r > 0) { |
651 | /* | |
652 | * Target determined this ioctl is being issued against | |
653 | * a logical partition of the parent bdev; so extra | |
654 | * validation is needed. | |
655 | */ | |
656 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
657 | if (r) | |
658 | goto out; | |
659 | } | |
6c182cd8 | 660 | |
66482026 | 661 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 662 | out: |
956a4025 | 663 | bdput(bdev); |
aa129a22 MB |
664 | return r; |
665 | } | |
666 | ||
028867ac | 667 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
668 | { |
669 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
670 | } | |
671 | ||
028867ac | 672 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
673 | { |
674 | mempool_free(io, md->io_pool); | |
675 | } | |
676 | ||
cfae7529 | 677 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 678 | { |
dba14160 | 679 | bio_put(&tio->clone); |
1da177e4 LT |
680 | } |
681 | ||
eca7ee6d MS |
682 | static struct dm_rq_target_io *alloc_old_rq_tio(struct mapped_device *md, |
683 | gfp_t gfp_mask) | |
cec47e3d | 684 | { |
5f015204 | 685 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
686 | } |
687 | ||
eca7ee6d | 688 | static void free_old_rq_tio(struct dm_rq_target_io *tio) |
cec47e3d | 689 | { |
5f015204 | 690 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
691 | } |
692 | ||
eca7ee6d MS |
693 | static struct request *alloc_old_clone_request(struct mapped_device *md, |
694 | gfp_t gfp_mask) | |
1ae49ea2 MS |
695 | { |
696 | return mempool_alloc(md->rq_pool, gfp_mask); | |
697 | } | |
698 | ||
eca7ee6d | 699 | static void free_old_clone_request(struct mapped_device *md, struct request *rq) |
1ae49ea2 MS |
700 | { |
701 | mempool_free(rq, md->rq_pool); | |
702 | } | |
703 | ||
90abb8c4 KU |
704 | static int md_in_flight(struct mapped_device *md) |
705 | { | |
706 | return atomic_read(&md->pending[READ]) + | |
707 | atomic_read(&md->pending[WRITE]); | |
708 | } | |
709 | ||
3eaf840e JNN |
710 | static void start_io_acct(struct dm_io *io) |
711 | { | |
712 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 713 | struct bio *bio = io->bio; |
c9959059 | 714 | int cpu; |
fd2ed4d2 | 715 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
716 | |
717 | io->start_time = jiffies; | |
718 | ||
074a7aca TH |
719 | cpu = part_stat_lock(); |
720 | part_round_stats(cpu, &dm_disk(md)->part0); | |
721 | part_stat_unlock(); | |
1e9bb880 SL |
722 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
723 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
724 | |
725 | if (unlikely(dm_stats_used(&md->stats))) | |
528ec5ab MC |
726 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
727 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
728 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
729 | } |
730 | ||
d221d2e7 | 731 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
732 | { |
733 | struct mapped_device *md = io->md; | |
734 | struct bio *bio = io->bio; | |
735 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 736 | int pending; |
3eaf840e JNN |
737 | int rw = bio_data_dir(bio); |
738 | ||
18c0b223 | 739 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 740 | |
fd2ed4d2 | 741 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
742 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
743 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
744 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 745 | |
af7e466a MP |
746 | /* |
747 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 748 | * a flush. |
af7e466a | 749 | */ |
1e9bb880 SL |
750 | pending = atomic_dec_return(&md->pending[rw]); |
751 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 752 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 753 | |
d221d2e7 MP |
754 | /* nudge anyone waiting on suspend queue */ |
755 | if (!pending) | |
756 | wake_up(&md->wait); | |
3eaf840e JNN |
757 | } |
758 | ||
1da177e4 LT |
759 | /* |
760 | * Add the bio to the list of deferred io. | |
761 | */ | |
92c63902 | 762 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 763 | { |
05447420 | 764 | unsigned long flags; |
1da177e4 | 765 | |
05447420 | 766 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 767 | bio_list_add(&md->deferred, bio); |
05447420 | 768 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 769 | queue_work(md->wq, &md->work); |
1da177e4 LT |
770 | } |
771 | ||
772 | /* | |
773 | * Everyone (including functions in this file), should use this | |
774 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 775 | * dm_put_live_table() when finished. |
1da177e4 | 776 | */ |
83d5e5b0 | 777 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 778 | { |
83d5e5b0 MP |
779 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
780 | ||
781 | return srcu_dereference(md->map, &md->io_barrier); | |
782 | } | |
1da177e4 | 783 | |
83d5e5b0 MP |
784 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
785 | { | |
786 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
787 | } | |
788 | ||
789 | void dm_sync_table(struct mapped_device *md) | |
790 | { | |
791 | synchronize_srcu(&md->io_barrier); | |
792 | synchronize_rcu_expedited(); | |
793 | } | |
794 | ||
795 | /* | |
796 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
797 | * The caller must not block between these two functions. | |
798 | */ | |
799 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
800 | { | |
801 | rcu_read_lock(); | |
802 | return rcu_dereference(md->map); | |
803 | } | |
1da177e4 | 804 | |
83d5e5b0 MP |
805 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
806 | { | |
807 | rcu_read_unlock(); | |
1da177e4 LT |
808 | } |
809 | ||
86f1152b BM |
810 | /* |
811 | * Open a table device so we can use it as a map destination. | |
812 | */ | |
813 | static int open_table_device(struct table_device *td, dev_t dev, | |
814 | struct mapped_device *md) | |
815 | { | |
816 | static char *_claim_ptr = "I belong to device-mapper"; | |
817 | struct block_device *bdev; | |
818 | ||
819 | int r; | |
820 | ||
821 | BUG_ON(td->dm_dev.bdev); | |
822 | ||
823 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
824 | if (IS_ERR(bdev)) | |
825 | return PTR_ERR(bdev); | |
826 | ||
827 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
828 | if (r) { | |
829 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
830 | return r; | |
831 | } | |
832 | ||
833 | td->dm_dev.bdev = bdev; | |
834 | return 0; | |
835 | } | |
836 | ||
837 | /* | |
838 | * Close a table device that we've been using. | |
839 | */ | |
840 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
841 | { | |
842 | if (!td->dm_dev.bdev) | |
843 | return; | |
844 | ||
845 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
846 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
847 | td->dm_dev.bdev = NULL; | |
848 | } | |
849 | ||
850 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
851 | fmode_t mode) { | |
852 | struct table_device *td; | |
853 | ||
854 | list_for_each_entry(td, l, list) | |
855 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
856 | return td; | |
857 | ||
858 | return NULL; | |
859 | } | |
860 | ||
861 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
862 | struct dm_dev **result) { | |
863 | int r; | |
864 | struct table_device *td; | |
865 | ||
866 | mutex_lock(&md->table_devices_lock); | |
867 | td = find_table_device(&md->table_devices, dev, mode); | |
868 | if (!td) { | |
115485e8 | 869 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
870 | if (!td) { |
871 | mutex_unlock(&md->table_devices_lock); | |
872 | return -ENOMEM; | |
873 | } | |
874 | ||
875 | td->dm_dev.mode = mode; | |
876 | td->dm_dev.bdev = NULL; | |
877 | ||
878 | if ((r = open_table_device(td, dev, md))) { | |
879 | mutex_unlock(&md->table_devices_lock); | |
880 | kfree(td); | |
881 | return r; | |
882 | } | |
883 | ||
884 | format_dev_t(td->dm_dev.name, dev); | |
885 | ||
886 | atomic_set(&td->count, 0); | |
887 | list_add(&td->list, &md->table_devices); | |
888 | } | |
889 | atomic_inc(&td->count); | |
890 | mutex_unlock(&md->table_devices_lock); | |
891 | ||
892 | *result = &td->dm_dev; | |
893 | return 0; | |
894 | } | |
895 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
896 | ||
897 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
898 | { | |
899 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
900 | ||
901 | mutex_lock(&md->table_devices_lock); | |
902 | if (atomic_dec_and_test(&td->count)) { | |
903 | close_table_device(td, md); | |
904 | list_del(&td->list); | |
905 | kfree(td); | |
906 | } | |
907 | mutex_unlock(&md->table_devices_lock); | |
908 | } | |
909 | EXPORT_SYMBOL(dm_put_table_device); | |
910 | ||
911 | static void free_table_devices(struct list_head *devices) | |
912 | { | |
913 | struct list_head *tmp, *next; | |
914 | ||
915 | list_for_each_safe(tmp, next, devices) { | |
916 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
917 | ||
918 | DMWARN("dm_destroy: %s still exists with %d references", | |
919 | td->dm_dev.name, atomic_read(&td->count)); | |
920 | kfree(td); | |
921 | } | |
922 | } | |
923 | ||
3ac51e74 DW |
924 | /* |
925 | * Get the geometry associated with a dm device | |
926 | */ | |
927 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
928 | { | |
929 | *geo = md->geometry; | |
930 | ||
931 | return 0; | |
932 | } | |
933 | ||
934 | /* | |
935 | * Set the geometry of a device. | |
936 | */ | |
937 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
938 | { | |
939 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
940 | ||
941 | if (geo->start > sz) { | |
942 | DMWARN("Start sector is beyond the geometry limits."); | |
943 | return -EINVAL; | |
944 | } | |
945 | ||
946 | md->geometry = *geo; | |
947 | ||
948 | return 0; | |
949 | } | |
950 | ||
1da177e4 LT |
951 | /*----------------------------------------------------------------- |
952 | * CRUD START: | |
953 | * A more elegant soln is in the works that uses the queue | |
954 | * merge fn, unfortunately there are a couple of changes to | |
955 | * the block layer that I want to make for this. So in the | |
956 | * interests of getting something for people to use I give | |
957 | * you this clearly demarcated crap. | |
958 | *---------------------------------------------------------------*/ | |
959 | ||
2e93ccc1 KU |
960 | static int __noflush_suspending(struct mapped_device *md) |
961 | { | |
962 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
963 | } | |
964 | ||
1da177e4 LT |
965 | /* |
966 | * Decrements the number of outstanding ios that a bio has been | |
967 | * cloned into, completing the original io if necc. | |
968 | */ | |
858119e1 | 969 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 970 | { |
2e93ccc1 | 971 | unsigned long flags; |
b35f8caa MB |
972 | int io_error; |
973 | struct bio *bio; | |
974 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
975 | |
976 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
977 | if (unlikely(error)) { |
978 | spin_lock_irqsave(&io->endio_lock, flags); | |
979 | if (!(io->error > 0 && __noflush_suspending(md))) | |
980 | io->error = error; | |
981 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
982 | } | |
1da177e4 LT |
983 | |
984 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
985 | if (io->error == DM_ENDIO_REQUEUE) { |
986 | /* | |
987 | * Target requested pushing back the I/O. | |
2e93ccc1 | 988 | */ |
022c2611 | 989 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
990 | if (__noflush_suspending(md)) |
991 | bio_list_add_head(&md->deferred, io->bio); | |
992 | else | |
2e93ccc1 KU |
993 | /* noflush suspend was interrupted. */ |
994 | io->error = -EIO; | |
022c2611 | 995 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
996 | } |
997 | ||
b35f8caa MB |
998 | io_error = io->error; |
999 | bio = io->bio; | |
6a8736d1 TH |
1000 | end_io_acct(io); |
1001 | free_io(md, io); | |
1002 | ||
1003 | if (io_error == DM_ENDIO_REQUEUE) | |
1004 | return; | |
2e93ccc1 | 1005 | |
28a8f0d3 | 1006 | if ((bio->bi_rw & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 1007 | /* |
6a8736d1 | 1008 | * Preflush done for flush with data, reissue |
28a8f0d3 | 1009 | * without REQ_PREFLUSH. |
af7e466a | 1010 | */ |
28a8f0d3 | 1011 | bio->bi_rw &= ~REQ_PREFLUSH; |
6a8736d1 | 1012 | queue_io(md, bio); |
af7e466a | 1013 | } else { |
b372d360 | 1014 | /* done with normal IO or empty flush */ |
0a82a8d1 | 1015 | trace_block_bio_complete(md->queue, bio, io_error); |
4246a0b6 CH |
1016 | bio->bi_error = io_error; |
1017 | bio_endio(bio); | |
b35f8caa | 1018 | } |
1da177e4 LT |
1019 | } |
1020 | } | |
1021 | ||
7eee4ae2 MS |
1022 | static void disable_write_same(struct mapped_device *md) |
1023 | { | |
1024 | struct queue_limits *limits = dm_get_queue_limits(md); | |
1025 | ||
1026 | /* device doesn't really support WRITE SAME, disable it */ | |
1027 | limits->max_write_same_sectors = 0; | |
1028 | } | |
1029 | ||
4246a0b6 | 1030 | static void clone_endio(struct bio *bio) |
1da177e4 | 1031 | { |
4246a0b6 | 1032 | int error = bio->bi_error; |
5164bece | 1033 | int r = error; |
bfc6d41c | 1034 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 1035 | struct dm_io *io = tio->io; |
9faf400f | 1036 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
1037 | dm_endio_fn endio = tio->ti->type->end_io; |
1038 | ||
1da177e4 | 1039 | if (endio) { |
7de3ee57 | 1040 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
1041 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
1042 | /* | |
1043 | * error and requeue request are handled | |
1044 | * in dec_pending(). | |
1045 | */ | |
1da177e4 | 1046 | error = r; |
45cbcd79 KU |
1047 | else if (r == DM_ENDIO_INCOMPLETE) |
1048 | /* The target will handle the io */ | |
6712ecf8 | 1049 | return; |
45cbcd79 KU |
1050 | else if (r) { |
1051 | DMWARN("unimplemented target endio return value: %d", r); | |
1052 | BUG(); | |
1053 | } | |
1da177e4 LT |
1054 | } |
1055 | ||
e6047149 | 1056 | if (unlikely(r == -EREMOTEIO && (bio_op(bio) == REQ_OP_WRITE_SAME) && |
7eee4ae2 MS |
1057 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) |
1058 | disable_write_same(md); | |
1059 | ||
cfae7529 | 1060 | free_tio(tio); |
b35f8caa | 1061 | dec_pending(io, error); |
1da177e4 LT |
1062 | } |
1063 | ||
78d8e58a MS |
1064 | /* |
1065 | * Partial completion handling for request-based dm | |
1066 | */ | |
4246a0b6 | 1067 | static void end_clone_bio(struct bio *clone) |
78d8e58a MS |
1068 | { |
1069 | struct dm_rq_clone_bio_info *info = | |
1070 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
1071 | struct dm_rq_target_io *tio = info->tio; | |
1072 | struct bio *bio = info->orig; | |
1073 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; | |
50887bd1 | 1074 | int error = clone->bi_error; |
78d8e58a MS |
1075 | |
1076 | bio_put(clone); | |
1077 | ||
1078 | if (tio->error) | |
1079 | /* | |
1080 | * An error has already been detected on the request. | |
1081 | * Once error occurred, just let clone->end_io() handle | |
1082 | * the remainder. | |
1083 | */ | |
1084 | return; | |
50887bd1 | 1085 | else if (error) { |
78d8e58a MS |
1086 | /* |
1087 | * Don't notice the error to the upper layer yet. | |
1088 | * The error handling decision is made by the target driver, | |
1089 | * when the request is completed. | |
1090 | */ | |
50887bd1 | 1091 | tio->error = error; |
78d8e58a MS |
1092 | return; |
1093 | } | |
1094 | ||
1095 | /* | |
1096 | * I/O for the bio successfully completed. | |
1097 | * Notice the data completion to the upper layer. | |
1098 | */ | |
1099 | ||
1100 | /* | |
1101 | * bios are processed from the head of the list. | |
1102 | * So the completing bio should always be rq->bio. | |
1103 | * If it's not, something wrong is happening. | |
1104 | */ | |
1105 | if (tio->orig->bio != bio) | |
1106 | DMERR("bio completion is going in the middle of the request"); | |
1107 | ||
1108 | /* | |
1109 | * Update the original request. | |
1110 | * Do not use blk_end_request() here, because it may complete | |
1111 | * the original request before the clone, and break the ordering. | |
1112 | */ | |
1113 | blk_update_request(tio->orig, 0, nr_bytes); | |
1114 | } | |
1115 | ||
bfebd1cd MS |
1116 | static struct dm_rq_target_io *tio_from_request(struct request *rq) |
1117 | { | |
1118 | return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special); | |
1119 | } | |
1120 | ||
e262f347 MP |
1121 | static void rq_end_stats(struct mapped_device *md, struct request *orig) |
1122 | { | |
1123 | if (unlikely(dm_stats_used(&md->stats))) { | |
1124 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
1125 | tio->duration_jiffies = jiffies - tio->duration_jiffies; | |
528ec5ab MC |
1126 | dm_stats_account_io(&md->stats, rq_data_dir(orig), |
1127 | blk_rq_pos(orig), tio->n_sectors, true, | |
1128 | tio->duration_jiffies, &tio->stats_aux); | |
e262f347 MP |
1129 | } |
1130 | } | |
1131 | ||
cec47e3d KU |
1132 | /* |
1133 | * Don't touch any member of the md after calling this function because | |
1134 | * the md may be freed in dm_put() at the end of this function. | |
1135 | * Or do dm_get() before calling this function and dm_put() later. | |
1136 | */ | |
466d89a6 | 1137 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
cec47e3d | 1138 | { |
b4324fee | 1139 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
1140 | |
1141 | /* nudge anyone waiting on suspend queue */ | |
621739b0 | 1142 | if (!md_in_flight(md)) |
cec47e3d KU |
1143 | wake_up(&md->wait); |
1144 | ||
a8c32a5c JA |
1145 | /* |
1146 | * Run this off this callpath, as drivers could invoke end_io while | |
1147 | * inside their request_fn (and holding the queue lock). Calling | |
1148 | * back into ->request_fn() could deadlock attempting to grab the | |
1149 | * queue lock again. | |
1150 | */ | |
6acfe68b MS |
1151 | if (!md->queue->mq_ops && run_queue) |
1152 | blk_run_queue_async(md->queue); | |
cec47e3d KU |
1153 | |
1154 | /* | |
1155 | * dm_put() must be at the end of this function. See the comment above | |
1156 | */ | |
1157 | dm_put(md); | |
1158 | } | |
1159 | ||
e5d8de32 | 1160 | static void free_rq_clone(struct request *clone) |
a77e28c7 KU |
1161 | { |
1162 | struct dm_rq_target_io *tio = clone->end_io_data; | |
bfebd1cd | 1163 | struct mapped_device *md = tio->md; |
a77e28c7 | 1164 | |
78d8e58a MS |
1165 | blk_rq_unprep_clone(clone); |
1166 | ||
aa6df8dd MS |
1167 | if (md->type == DM_TYPE_MQ_REQUEST_BASED) |
1168 | /* stacked on blk-mq queue(s) */ | |
e5863d9a | 1169 | tio->ti->type->release_clone_rq(clone); |
02233342 MS |
1170 | else if (!md->queue->mq_ops) |
1171 | /* request_fn queue stacked on request_fn queue(s) */ | |
eca7ee6d | 1172 | free_old_clone_request(md, clone); |
bfebd1cd MS |
1173 | |
1174 | if (!md->queue->mq_ops) | |
eca7ee6d | 1175 | free_old_rq_tio(tio); |
a77e28c7 KU |
1176 | } |
1177 | ||
980691e5 KU |
1178 | /* |
1179 | * Complete the clone and the original request. | |
466d89a6 KB |
1180 | * Must be called without clone's queue lock held, |
1181 | * see end_clone_request() for more details. | |
980691e5 KU |
1182 | */ |
1183 | static void dm_end_request(struct request *clone, int error) | |
1184 | { | |
1185 | int rw = rq_data_dir(clone); | |
1186 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1187 | struct mapped_device *md = tio->md; | |
1188 | struct request *rq = tio->orig; | |
1189 | ||
29e4013d | 1190 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1191 | rq->errors = clone->errors; |
1192 | rq->resid_len = clone->resid_len; | |
1193 | ||
1194 | if (rq->sense) | |
1195 | /* | |
1196 | * We are using the sense buffer of the original | |
1197 | * request. | |
1198 | * So setting the length of the sense data is enough. | |
1199 | */ | |
1200 | rq->sense_len = clone->sense_len; | |
1201 | } | |
1202 | ||
e5d8de32 | 1203 | free_rq_clone(clone); |
e262f347 | 1204 | rq_end_stats(md, rq); |
bfebd1cd MS |
1205 | if (!rq->q->mq_ops) |
1206 | blk_end_request_all(rq, error); | |
1207 | else | |
1208 | blk_mq_end_request(rq, error); | |
29e4013d | 1209 | rq_completed(md, rw, true); |
980691e5 KU |
1210 | } |
1211 | ||
cec47e3d KU |
1212 | static void dm_unprep_request(struct request *rq) |
1213 | { | |
bfebd1cd | 1214 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1215 | struct request *clone = tio->clone; |
cec47e3d | 1216 | |
bfebd1cd MS |
1217 | if (!rq->q->mq_ops) { |
1218 | rq->special = NULL; | |
1219 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1220 | } | |
cec47e3d | 1221 | |
e5863d9a | 1222 | if (clone) |
e5d8de32 | 1223 | free_rq_clone(clone); |
4328daa2 | 1224 | else if (!tio->md->queue->mq_ops) |
eca7ee6d | 1225 | free_old_rq_tio(tio); |
cec47e3d KU |
1226 | } |
1227 | ||
1228 | /* | |
1229 | * Requeue the original request of a clone. | |
1230 | */ | |
eca7ee6d | 1231 | static void dm_old_requeue_request(struct request *rq) |
cec47e3d | 1232 | { |
cec47e3d KU |
1233 | struct request_queue *q = rq->q; |
1234 | unsigned long flags; | |
1235 | ||
cec47e3d | 1236 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d | 1237 | blk_requeue_request(q, rq); |
4ae9944d | 1238 | blk_run_queue_async(q); |
cec47e3d | 1239 | spin_unlock_irqrestore(q->queue_lock, flags); |
bfebd1cd MS |
1240 | } |
1241 | ||
818c5f3b MS |
1242 | static void dm_mq_requeue_request(struct request *rq) |
1243 | { | |
1244 | struct request_queue *q = rq->q; | |
1245 | unsigned long flags; | |
1246 | ||
1247 | blk_mq_requeue_request(rq); | |
1248 | spin_lock_irqsave(q->queue_lock, flags); | |
1249 | if (!blk_queue_stopped(q)) | |
1250 | blk_mq_kick_requeue_list(q); | |
1251 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1252 | } | |
1253 | ||
2d76fff1 MS |
1254 | static void dm_requeue_original_request(struct mapped_device *md, |
1255 | struct request *rq) | |
bfebd1cd MS |
1256 | { |
1257 | int rw = rq_data_dir(rq); | |
1258 | ||
98dbc9c6 | 1259 | rq_end_stats(md, rq); |
bfebd1cd MS |
1260 | dm_unprep_request(rq); |
1261 | ||
1262 | if (!rq->q->mq_ops) | |
eca7ee6d | 1263 | dm_old_requeue_request(rq); |
818c5f3b MS |
1264 | else |
1265 | dm_mq_requeue_request(rq); | |
cec47e3d | 1266 | |
466d89a6 KB |
1267 | rq_completed(md, rw, false); |
1268 | } | |
1269 | ||
eca7ee6d | 1270 | static void dm_old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1271 | { |
1272 | unsigned long flags; | |
1273 | ||
818c5f3b MS |
1274 | spin_lock_irqsave(q->queue_lock, flags); |
1275 | if (blk_queue_stopped(q)) { | |
1276 | spin_unlock_irqrestore(q->queue_lock, flags); | |
bfebd1cd | 1277 | return; |
818c5f3b | 1278 | } |
bfebd1cd | 1279 | |
bfebd1cd | 1280 | blk_stop_queue(q); |
cec47e3d KU |
1281 | spin_unlock_irqrestore(q->queue_lock, flags); |
1282 | } | |
1283 | ||
eca7ee6d | 1284 | static void dm_stop_queue(struct request_queue *q) |
cec47e3d | 1285 | { |
bfebd1cd | 1286 | if (!q->mq_ops) |
eca7ee6d | 1287 | dm_old_stop_queue(q); |
bfebd1cd MS |
1288 | else |
1289 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1290 | } |
1291 | ||
eca7ee6d | 1292 | static void dm_old_start_queue(struct request_queue *q) |
cec47e3d KU |
1293 | { |
1294 | unsigned long flags; | |
1295 | ||
1296 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1297 | if (blk_queue_stopped(q)) |
1298 | blk_start_queue(q); | |
cec47e3d KU |
1299 | spin_unlock_irqrestore(q->queue_lock, flags); |
1300 | } | |
1301 | ||
eca7ee6d | 1302 | static void dm_start_queue(struct request_queue *q) |
bfebd1cd MS |
1303 | { |
1304 | if (!q->mq_ops) | |
eca7ee6d | 1305 | dm_old_start_queue(q); |
818c5f3b | 1306 | else { |
bfebd1cd | 1307 | blk_mq_start_stopped_hw_queues(q, true); |
818c5f3b MS |
1308 | blk_mq_kick_requeue_list(q); |
1309 | } | |
bfebd1cd MS |
1310 | } |
1311 | ||
11a68244 | 1312 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1313 | { |
11a68244 | 1314 | int r = error; |
cec47e3d | 1315 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1316 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1317 | |
ba1cbad9 MS |
1318 | if (tio->ti) { |
1319 | rq_end_io = tio->ti->type->rq_end_io; | |
1320 | ||
1321 | if (mapped && rq_end_io) | |
1322 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1323 | } | |
cec47e3d | 1324 | |
c2df40df | 1325 | if (unlikely(r == -EREMOTEIO && (req_op(clone) == REQ_OP_WRITE_SAME) && |
7eee4ae2 MS |
1326 | !clone->q->limits.max_write_same_sectors)) |
1327 | disable_write_same(tio->md); | |
1328 | ||
11a68244 | 1329 | if (r <= 0) |
cec47e3d | 1330 | /* The target wants to complete the I/O */ |
11a68244 KU |
1331 | dm_end_request(clone, r); |
1332 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1333 | /* The target will handle the I/O */ |
1334 | return; | |
11a68244 | 1335 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d | 1336 | /* The target wants to requeue the I/O */ |
2d76fff1 | 1337 | dm_requeue_original_request(tio->md, tio->orig); |
cec47e3d | 1338 | else { |
11a68244 | 1339 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1340 | BUG(); |
1341 | } | |
1342 | } | |
1343 | ||
11a68244 KU |
1344 | /* |
1345 | * Request completion handler for request-based dm | |
1346 | */ | |
1347 | static void dm_softirq_done(struct request *rq) | |
1348 | { | |
1349 | bool mapped = true; | |
bfebd1cd | 1350 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1351 | struct request *clone = tio->clone; |
bfebd1cd | 1352 | int rw; |
11a68244 | 1353 | |
e5863d9a | 1354 | if (!clone) { |
e262f347 | 1355 | rq_end_stats(tio->md, rq); |
bfebd1cd MS |
1356 | rw = rq_data_dir(rq); |
1357 | if (!rq->q->mq_ops) { | |
1358 | blk_end_request_all(rq, tio->error); | |
1359 | rq_completed(tio->md, rw, false); | |
eca7ee6d | 1360 | free_old_rq_tio(tio); |
bfebd1cd MS |
1361 | } else { |
1362 | blk_mq_end_request(rq, tio->error); | |
1363 | rq_completed(tio->md, rw, false); | |
1364 | } | |
e5863d9a MS |
1365 | return; |
1366 | } | |
11a68244 KU |
1367 | |
1368 | if (rq->cmd_flags & REQ_FAILED) | |
1369 | mapped = false; | |
1370 | ||
1371 | dm_done(clone, tio->error, mapped); | |
1372 | } | |
1373 | ||
cec47e3d KU |
1374 | /* |
1375 | * Complete the clone and the original request with the error status | |
1376 | * through softirq context. | |
1377 | */ | |
466d89a6 | 1378 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1379 | { |
bfebd1cd | 1380 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1381 | |
1382 | tio->error = error; | |
6acfe68b MS |
1383 | if (!rq->q->mq_ops) |
1384 | blk_complete_request(rq); | |
1385 | else | |
1386 | blk_mq_complete_request(rq, error); | |
cec47e3d KU |
1387 | } |
1388 | ||
1389 | /* | |
1390 | * Complete the not-mapped clone and the original request with the error status | |
1391 | * through softirq context. | |
1392 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1393 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1394 | */ |
466d89a6 | 1395 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1396 | { |
cec47e3d | 1397 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1398 | dm_complete_request(rq, error); |
cec47e3d | 1399 | } |
cec47e3d KU |
1400 | |
1401 | /* | |
eca7ee6d | 1402 | * Called with the clone's queue lock held (in the case of .request_fn) |
cec47e3d KU |
1403 | */ |
1404 | static void end_clone_request(struct request *clone, int error) | |
1405 | { | |
466d89a6 KB |
1406 | struct dm_rq_target_io *tio = clone->end_io_data; |
1407 | ||
e5863d9a MS |
1408 | if (!clone->q->mq_ops) { |
1409 | /* | |
1410 | * For just cleaning up the information of the queue in which | |
1411 | * the clone was dispatched. | |
1412 | * The clone is *NOT* freed actually here because it is alloced | |
1413 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1414 | */ | |
1415 | __blk_put_request(clone->q, clone); | |
1416 | } | |
cec47e3d KU |
1417 | |
1418 | /* | |
1419 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1420 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1421 | * - another request may be submitted by the upper level driver |
1422 | * of the stacking during the completion | |
1423 | * - the submission which requires queue lock may be done | |
466d89a6 | 1424 | * against this clone's queue |
cec47e3d | 1425 | */ |
466d89a6 | 1426 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1427 | } |
1428 | ||
56a67df7 MS |
1429 | /* |
1430 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1431 | * target boundary. | |
1432 | */ | |
1433 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1434 | { | |
1435 | sector_t target_offset = dm_target_offset(ti, sector); | |
1436 | ||
1437 | return ti->len - target_offset; | |
1438 | } | |
1439 | ||
1440 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1441 | { |
56a67df7 | 1442 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1443 | sector_t offset, max_len; |
1da177e4 LT |
1444 | |
1445 | /* | |
542f9038 | 1446 | * Does the target need to split even further? |
1da177e4 | 1447 | */ |
542f9038 MS |
1448 | if (ti->max_io_len) { |
1449 | offset = dm_target_offset(ti, sector); | |
1450 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1451 | max_len = sector_div(offset, ti->max_io_len); | |
1452 | else | |
1453 | max_len = offset & (ti->max_io_len - 1); | |
1454 | max_len = ti->max_io_len - max_len; | |
1455 | ||
1456 | if (len > max_len) | |
1457 | len = max_len; | |
1da177e4 LT |
1458 | } |
1459 | ||
1460 | return len; | |
1461 | } | |
1462 | ||
542f9038 MS |
1463 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1464 | { | |
1465 | if (len > UINT_MAX) { | |
1466 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1467 | (unsigned long long)len, UINT_MAX); | |
1468 | ti->error = "Maximum size of target IO is too large"; | |
1469 | return -EINVAL; | |
1470 | } | |
1471 | ||
1472 | ti->max_io_len = (uint32_t) len; | |
1473 | ||
1474 | return 0; | |
1475 | } | |
1476 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1477 | ||
1dd40c3e MP |
1478 | /* |
1479 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
28a8f0d3 | 1480 | * allowed for all bio types except REQ_PREFLUSH. |
1dd40c3e MP |
1481 | * |
1482 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1483 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1484 | * sent in a next bio. | |
1485 | * | |
1486 | * A diagram that explains the arithmetics: | |
1487 | * +--------------------+---------------+-------+ | |
1488 | * | 1 | 2 | 3 | | |
1489 | * +--------------------+---------------+-------+ | |
1490 | * | |
1491 | * <-------------- *tio->len_ptr ---------------> | |
1492 | * <------- bi_size -------> | |
1493 | * <-- n_sectors --> | |
1494 | * | |
1495 | * Region 1 was already iterated over with bio_advance or similar function. | |
1496 | * (it may be empty if the target doesn't use bio_advance) | |
1497 | * Region 2 is the remaining bio size that the target wants to process. | |
1498 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1499 | * to make it empty) | |
1500 | * The target requires that region 3 is to be sent in the next bio. | |
1501 | * | |
1502 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1503 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1504 | * copies of the bio. | |
1505 | */ | |
1506 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1507 | { | |
1508 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1509 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
28a8f0d3 | 1510 | BUG_ON(bio->bi_rw & REQ_PREFLUSH); |
1dd40c3e MP |
1511 | BUG_ON(bi_size > *tio->len_ptr); |
1512 | BUG_ON(n_sectors > bi_size); | |
1513 | *tio->len_ptr -= bi_size - n_sectors; | |
1514 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1515 | } | |
1516 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1517 | ||
bd2a49b8 | 1518 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1519 | { |
1520 | int r; | |
2056a782 | 1521 | sector_t sector; |
dba14160 | 1522 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1523 | struct dm_target *ti = tio->ti; |
1da177e4 | 1524 | |
1da177e4 | 1525 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1526 | |
1527 | /* | |
1528 | * Map the clone. If r == 0 we don't need to do | |
1529 | * anything, the target has assumed ownership of | |
1530 | * this io. | |
1531 | */ | |
1532 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1533 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1534 | r = ti->type->map(ti, clone); |
45cbcd79 | 1535 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1536 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1537 | |
d07335e5 MS |
1538 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1539 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1540 | |
1da177e4 | 1541 | generic_make_request(clone); |
2e93ccc1 KU |
1542 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1543 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f | 1544 | dec_pending(tio->io, r); |
cfae7529 | 1545 | free_tio(tio); |
ab37844d | 1546 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1547 | DMWARN("unimplemented target map return value: %d", r); |
1548 | BUG(); | |
1da177e4 LT |
1549 | } |
1550 | } | |
1551 | ||
1552 | struct clone_info { | |
1553 | struct mapped_device *md; | |
1554 | struct dm_table *map; | |
1555 | struct bio *bio; | |
1556 | struct dm_io *io; | |
1557 | sector_t sector; | |
e0d6609a | 1558 | unsigned sector_count; |
1da177e4 LT |
1559 | }; |
1560 | ||
e0d6609a | 1561 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1562 | { |
4f024f37 KO |
1563 | bio->bi_iter.bi_sector = sector; |
1564 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1565 | } |
1566 | ||
1567 | /* | |
1568 | * Creates a bio that consists of range of complete bvecs. | |
1569 | */ | |
c80914e8 MS |
1570 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1571 | sector_t sector, unsigned len) | |
1da177e4 | 1572 | { |
dba14160 | 1573 | struct bio *clone = &tio->clone; |
1da177e4 | 1574 | |
1c3b13e6 KO |
1575 | __bio_clone_fast(clone, bio); |
1576 | ||
c80914e8 MS |
1577 | if (bio_integrity(bio)) { |
1578 | int r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
1579 | if (r < 0) | |
1580 | return r; | |
1581 | } | |
bd2a49b8 | 1582 | |
1c3b13e6 KO |
1583 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1584 | clone->bi_iter.bi_size = to_bytes(len); | |
1585 | ||
1586 | if (bio_integrity(bio)) | |
1587 | bio_integrity_trim(clone, 0, len); | |
c80914e8 MS |
1588 | |
1589 | return 0; | |
1da177e4 LT |
1590 | } |
1591 | ||
9015df24 | 1592 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1593 | struct dm_target *ti, |
55a62eef | 1594 | unsigned target_bio_nr) |
f9ab94ce | 1595 | { |
dba14160 MP |
1596 | struct dm_target_io *tio; |
1597 | struct bio *clone; | |
1598 | ||
99778273 | 1599 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1600 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1601 | |
1602 | tio->io = ci->io; | |
1603 | tio->ti = ti; | |
55a62eef | 1604 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1605 | |
1606 | return tio; | |
1607 | } | |
1608 | ||
14fe594d AK |
1609 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1610 | struct dm_target *ti, | |
1dd40c3e | 1611 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1612 | { |
99778273 | 1613 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1614 | struct bio *clone = &tio->clone; |
9015df24 | 1615 | |
1dd40c3e MP |
1616 | tio->len_ptr = len; |
1617 | ||
99778273 | 1618 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1619 | if (len) |
1dd40c3e | 1620 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1621 | |
bd2a49b8 | 1622 | __map_bio(tio); |
f9ab94ce MP |
1623 | } |
1624 | ||
14fe594d | 1625 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1626 | unsigned num_bios, unsigned *len) |
06a426ce | 1627 | { |
55a62eef | 1628 | unsigned target_bio_nr; |
06a426ce | 1629 | |
55a62eef | 1630 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1631 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1632 | } |
1633 | ||
14fe594d | 1634 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1635 | { |
06a426ce | 1636 | unsigned target_nr = 0; |
f9ab94ce MP |
1637 | struct dm_target *ti; |
1638 | ||
b372d360 | 1639 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1640 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1641 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1642 | |
f9ab94ce MP |
1643 | return 0; |
1644 | } | |
1645 | ||
c80914e8 | 1646 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1647 | sector_t sector, unsigned *len) |
5ae89a87 | 1648 | { |
dba14160 | 1649 | struct bio *bio = ci->bio; |
5ae89a87 | 1650 | struct dm_target_io *tio; |
b0d8ed4d AK |
1651 | unsigned target_bio_nr; |
1652 | unsigned num_target_bios = 1; | |
c80914e8 | 1653 | int r = 0; |
5ae89a87 | 1654 | |
b0d8ed4d AK |
1655 | /* |
1656 | * Does the target want to receive duplicate copies of the bio? | |
1657 | */ | |
1658 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1659 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1660 | |
b0d8ed4d | 1661 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1662 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e | 1663 | tio->len_ptr = len; |
c80914e8 | 1664 | r = clone_bio(tio, bio, sector, *len); |
072623de | 1665 | if (r < 0) { |
cfae7529 | 1666 | free_tio(tio); |
c80914e8 | 1667 | break; |
072623de | 1668 | } |
b0d8ed4d AK |
1669 | __map_bio(tio); |
1670 | } | |
c80914e8 MS |
1671 | |
1672 | return r; | |
5ae89a87 MS |
1673 | } |
1674 | ||
55a62eef | 1675 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1676 | |
55a62eef | 1677 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1678 | { |
55a62eef | 1679 | return ti->num_discard_bios; |
23508a96 MS |
1680 | } |
1681 | ||
55a62eef | 1682 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1683 | { |
55a62eef | 1684 | return ti->num_write_same_bios; |
23508a96 MS |
1685 | } |
1686 | ||
1687 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1688 | ||
1689 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1690 | { | |
55a62eef | 1691 | return ti->split_discard_bios; |
23508a96 MS |
1692 | } |
1693 | ||
14fe594d AK |
1694 | static int __send_changing_extent_only(struct clone_info *ci, |
1695 | get_num_bios_fn get_num_bios, | |
1696 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1697 | { |
1698 | struct dm_target *ti; | |
e0d6609a | 1699 | unsigned len; |
55a62eef | 1700 | unsigned num_bios; |
5ae89a87 | 1701 | |
a79245b3 MS |
1702 | do { |
1703 | ti = dm_table_find_target(ci->map, ci->sector); | |
1704 | if (!dm_target_is_valid(ti)) | |
1705 | return -EIO; | |
5ae89a87 | 1706 | |
5ae89a87 | 1707 | /* |
23508a96 MS |
1708 | * Even though the device advertised support for this type of |
1709 | * request, that does not mean every target supports it, and | |
936688d7 | 1710 | * reconfiguration might also have changed that since the |
a79245b3 | 1711 | * check was performed. |
5ae89a87 | 1712 | */ |
55a62eef AK |
1713 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1714 | if (!num_bios) | |
a79245b3 | 1715 | return -EOPNOTSUPP; |
5ae89a87 | 1716 | |
23508a96 | 1717 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1718 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1719 | else |
e0d6609a | 1720 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1721 | |
1dd40c3e | 1722 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1723 | |
1724 | ci->sector += len; | |
1725 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1726 | |
1727 | return 0; | |
1728 | } | |
1729 | ||
14fe594d | 1730 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1731 | { |
14fe594d AK |
1732 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1733 | is_split_required_for_discard); | |
23508a96 MS |
1734 | } |
1735 | ||
14fe594d | 1736 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1737 | { |
14fe594d | 1738 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1739 | } |
1740 | ||
e4c93811 AK |
1741 | /* |
1742 | * Select the correct strategy for processing a non-flush bio. | |
1743 | */ | |
14fe594d | 1744 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1745 | { |
dba14160 | 1746 | struct bio *bio = ci->bio; |
512875bd | 1747 | struct dm_target *ti; |
1c3b13e6 | 1748 | unsigned len; |
c80914e8 | 1749 | int r; |
1da177e4 | 1750 | |
e6047149 | 1751 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
14fe594d | 1752 | return __send_discard(ci); |
e6047149 | 1753 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
14fe594d | 1754 | return __send_write_same(ci); |
5ae89a87 | 1755 | |
512875bd JN |
1756 | ti = dm_table_find_target(ci->map, ci->sector); |
1757 | if (!dm_target_is_valid(ti)) | |
1758 | return -EIO; | |
1759 | ||
1c3b13e6 | 1760 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1761 | |
c80914e8 MS |
1762 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1763 | if (r < 0) | |
1764 | return r; | |
1da177e4 | 1765 | |
1c3b13e6 KO |
1766 | ci->sector += len; |
1767 | ci->sector_count -= len; | |
1da177e4 | 1768 | |
1c3b13e6 | 1769 | return 0; |
1da177e4 LT |
1770 | } |
1771 | ||
1772 | /* | |
14fe594d | 1773 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1774 | */ |
83d5e5b0 MP |
1775 | static void __split_and_process_bio(struct mapped_device *md, |
1776 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1777 | { |
1778 | struct clone_info ci; | |
512875bd | 1779 | int error = 0; |
1da177e4 | 1780 | |
83d5e5b0 | 1781 | if (unlikely(!map)) { |
6a8736d1 | 1782 | bio_io_error(bio); |
f0b9a450 MP |
1783 | return; |
1784 | } | |
692d0eb9 | 1785 | |
83d5e5b0 | 1786 | ci.map = map; |
1da177e4 | 1787 | ci.md = md; |
1da177e4 LT |
1788 | ci.io = alloc_io(md); |
1789 | ci.io->error = 0; | |
1790 | atomic_set(&ci.io->io_count, 1); | |
1791 | ci.io->bio = bio; | |
1792 | ci.io->md = md; | |
f88fb981 | 1793 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1794 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1795 | |
3eaf840e | 1796 | start_io_acct(ci.io); |
bd2a49b8 | 1797 | |
28a8f0d3 | 1798 | if (bio->bi_rw & REQ_PREFLUSH) { |
b372d360 MS |
1799 | ci.bio = &ci.md->flush_bio; |
1800 | ci.sector_count = 0; | |
14fe594d | 1801 | error = __send_empty_flush(&ci); |
b372d360 MS |
1802 | /* dec_pending submits any data associated with flush */ |
1803 | } else { | |
6a8736d1 | 1804 | ci.bio = bio; |
d87f4c14 | 1805 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1806 | while (ci.sector_count && !error) |
14fe594d | 1807 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1808 | } |
1da177e4 LT |
1809 | |
1810 | /* drop the extra reference count */ | |
512875bd | 1811 | dec_pending(ci.io, error); |
1da177e4 LT |
1812 | } |
1813 | /*----------------------------------------------------------------- | |
1814 | * CRUD END | |
1815 | *---------------------------------------------------------------*/ | |
1816 | ||
1817 | /* | |
1818 | * The request function that just remaps the bio built up by | |
1819 | * dm_merge_bvec. | |
1820 | */ | |
dece1635 | 1821 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1822 | { |
12f03a49 | 1823 | int rw = bio_data_dir(bio); |
1da177e4 | 1824 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1825 | int srcu_idx; |
1826 | struct dm_table *map; | |
1da177e4 | 1827 | |
83d5e5b0 | 1828 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1829 | |
18c0b223 | 1830 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1831 | |
6a8736d1 TH |
1832 | /* if we're suspended, we have to queue this io for later */ |
1833 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1834 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1835 | |
6a8736d1 TH |
1836 | if (bio_rw(bio) != READA) |
1837 | queue_io(md, bio); | |
1838 | else | |
54d9a1b4 | 1839 | bio_io_error(bio); |
dece1635 | 1840 | return BLK_QC_T_NONE; |
1da177e4 LT |
1841 | } |
1842 | ||
83d5e5b0 MP |
1843 | __split_and_process_bio(md, map, bio); |
1844 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1845 | return BLK_QC_T_NONE; |
cec47e3d KU |
1846 | } |
1847 | ||
fd2ed4d2 | 1848 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1849 | { |
1850 | return blk_queue_stackable(md->queue); | |
1851 | } | |
1852 | ||
466d89a6 | 1853 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1854 | { |
1855 | int r; | |
1856 | ||
466d89a6 KB |
1857 | if (blk_queue_io_stat(clone->q)) |
1858 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1859 | |
466d89a6 KB |
1860 | clone->start_time = jiffies; |
1861 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1862 | if (r) |
466d89a6 | 1863 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1864 | dm_complete_request(rq, r); |
1865 | } | |
cec47e3d | 1866 | |
78d8e58a MS |
1867 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1868 | void *data) | |
cec47e3d | 1869 | { |
78d8e58a MS |
1870 | struct dm_rq_target_io *tio = data; |
1871 | struct dm_rq_clone_bio_info *info = | |
1872 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
1873 | ||
1874 | info->orig = bio_orig; | |
1875 | info->tio = tio; | |
1876 | bio->bi_end_io = end_clone_bio; | |
1877 | ||
1878 | return 0; | |
1879 | } | |
1880 | ||
1881 | static int setup_clone(struct request *clone, struct request *rq, | |
1882 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
cec47e3d | 1883 | { |
78d8e58a MS |
1884 | int r; |
1885 | ||
1886 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, | |
1887 | dm_rq_bio_constructor, tio); | |
1888 | if (r) | |
1889 | return r; | |
1890 | ||
1891 | clone->cmd = rq->cmd; | |
1892 | clone->cmd_len = rq->cmd_len; | |
1893 | clone->sense = rq->sense; | |
cec47e3d KU |
1894 | clone->end_io = end_clone_request; |
1895 | clone->end_io_data = tio; | |
78d8e58a | 1896 | |
1ae49ea2 | 1897 | tio->clone = clone; |
78d8e58a MS |
1898 | |
1899 | return 0; | |
cec47e3d KU |
1900 | } |
1901 | ||
eca7ee6d MS |
1902 | static struct request *clone_old_rq(struct request *rq, struct mapped_device *md, |
1903 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
1ae49ea2 | 1904 | { |
02233342 | 1905 | /* |
c5248f79 | 1906 | * Create clone for use with .request_fn request_queue |
02233342 | 1907 | */ |
02233342 | 1908 | struct request *clone; |
1ae49ea2 | 1909 | |
eca7ee6d | 1910 | clone = alloc_old_clone_request(md, gfp_mask); |
c5248f79 MS |
1911 | if (!clone) |
1912 | return NULL; | |
1ae49ea2 MS |
1913 | |
1914 | blk_rq_init(NULL, clone); | |
78d8e58a MS |
1915 | if (setup_clone(clone, rq, tio, gfp_mask)) { |
1916 | /* -ENOMEM */ | |
eca7ee6d | 1917 | free_old_clone_request(md, clone); |
78d8e58a MS |
1918 | return NULL; |
1919 | } | |
1ae49ea2 MS |
1920 | |
1921 | return clone; | |
1922 | } | |
1923 | ||
2eb6e1e3 KB |
1924 | static void map_tio_request(struct kthread_work *work); |
1925 | ||
bfebd1cd MS |
1926 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1927 | struct mapped_device *md) | |
1928 | { | |
1929 | tio->md = md; | |
1930 | tio->ti = NULL; | |
1931 | tio->clone = NULL; | |
1932 | tio->orig = rq; | |
1933 | tio->error = 0; | |
591ddcfc MS |
1934 | /* |
1935 | * Avoid initializing info for blk-mq; it passes | |
1936 | * target-specific data through info.ptr | |
1937 | * (see: dm_mq_init_request) | |
1938 | */ | |
1939 | if (!md->init_tio_pdu) | |
1940 | memset(&tio->info, 0, sizeof(tio->info)); | |
02233342 MS |
1941 | if (md->kworker_task) |
1942 | init_kthread_work(&tio->work, map_tio_request); | |
bfebd1cd MS |
1943 | } |
1944 | ||
eca7ee6d MS |
1945 | static struct dm_rq_target_io *dm_old_prep_tio(struct request *rq, |
1946 | struct mapped_device *md, | |
1947 | gfp_t gfp_mask) | |
6facdaff | 1948 | { |
6facdaff | 1949 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1950 | int srcu_idx; |
1951 | struct dm_table *table; | |
6facdaff | 1952 | |
eca7ee6d | 1953 | tio = alloc_old_rq_tio(md, gfp_mask); |
6facdaff KU |
1954 | if (!tio) |
1955 | return NULL; | |
1956 | ||
bfebd1cd | 1957 | init_tio(tio, rq, md); |
6facdaff | 1958 | |
e5863d9a | 1959 | table = dm_get_live_table(md, &srcu_idx); |
eca7ee6d MS |
1960 | /* |
1961 | * Must clone a request if this .request_fn DM device | |
1962 | * is stacked on .request_fn device(s). | |
1963 | */ | |
e5863d9a | 1964 | if (!dm_table_mq_request_based(table)) { |
eca7ee6d | 1965 | if (!clone_old_rq(rq, md, tio, gfp_mask)) { |
e5863d9a | 1966 | dm_put_live_table(md, srcu_idx); |
eca7ee6d | 1967 | free_old_rq_tio(tio); |
e5863d9a MS |
1968 | return NULL; |
1969 | } | |
6facdaff | 1970 | } |
e5863d9a | 1971 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1972 | |
466d89a6 | 1973 | return tio; |
6facdaff KU |
1974 | } |
1975 | ||
cec47e3d KU |
1976 | /* |
1977 | * Called with the queue lock held. | |
1978 | */ | |
eca7ee6d | 1979 | static int dm_old_prep_fn(struct request_queue *q, struct request *rq) |
cec47e3d KU |
1980 | { |
1981 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1982 | struct dm_rq_target_io *tio; |
cec47e3d | 1983 | |
cec47e3d KU |
1984 | if (unlikely(rq->special)) { |
1985 | DMWARN("Already has something in rq->special."); | |
1986 | return BLKPREP_KILL; | |
1987 | } | |
1988 | ||
eca7ee6d | 1989 | tio = dm_old_prep_tio(rq, md, GFP_ATOMIC); |
466d89a6 | 1990 | if (!tio) |
cec47e3d | 1991 | return BLKPREP_DEFER; |
cec47e3d | 1992 | |
466d89a6 | 1993 | rq->special = tio; |
cec47e3d KU |
1994 | rq->cmd_flags |= REQ_DONTPREP; |
1995 | ||
1996 | return BLKPREP_OK; | |
1997 | } | |
1998 | ||
9eef87da KU |
1999 | /* |
2000 | * Returns: | |
e5863d9a MS |
2001 | * 0 : the request has been processed |
2002 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
2003 | * < 0 : the request was completed due to failure | |
9eef87da | 2004 | */ |
bfebd1cd | 2005 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 2006 | struct mapped_device *md) |
cec47e3d | 2007 | { |
e5863d9a | 2008 | int r; |
bfebd1cd | 2009 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
2010 | struct request *clone = NULL; |
2011 | ||
2012 | if (tio->clone) { | |
2013 | clone = tio->clone; | |
2014 | r = ti->type->map_rq(ti, clone, &tio->info); | |
2015 | } else { | |
2016 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
2017 | if (r < 0) { | |
2018 | /* The target wants to complete the I/O */ | |
2019 | dm_kill_unmapped_request(rq, r); | |
2020 | return r; | |
2021 | } | |
3a140755 JN |
2022 | if (r != DM_MAPIO_REMAPPED) |
2023 | return r; | |
78d8e58a MS |
2024 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
2025 | /* -ENOMEM */ | |
2026 | ti->type->release_clone_rq(clone); | |
2027 | return DM_MAPIO_REQUEUE; | |
2028 | } | |
e5863d9a | 2029 | } |
cec47e3d | 2030 | |
cec47e3d KU |
2031 | switch (r) { |
2032 | case DM_MAPIO_SUBMITTED: | |
2033 | /* The target has taken the I/O to submit by itself later */ | |
2034 | break; | |
2035 | case DM_MAPIO_REMAPPED: | |
2036 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 2037 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
2038 | blk_rq_pos(rq)); |
2039 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
2040 | break; |
2041 | case DM_MAPIO_REQUEUE: | |
2042 | /* The target wants to requeue the I/O */ | |
2d76fff1 | 2043 | dm_requeue_original_request(md, tio->orig); |
cec47e3d KU |
2044 | break; |
2045 | default: | |
2046 | if (r > 0) { | |
2047 | DMWARN("unimplemented target map return value: %d", r); | |
2048 | BUG(); | |
2049 | } | |
2050 | ||
2051 | /* The target wants to complete the I/O */ | |
466d89a6 | 2052 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 2053 | return r; |
cec47e3d | 2054 | } |
9eef87da | 2055 | |
e5863d9a | 2056 | return 0; |
cec47e3d KU |
2057 | } |
2058 | ||
2eb6e1e3 | 2059 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 2060 | { |
2eb6e1e3 | 2061 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
2062 | struct request *rq = tio->orig; |
2063 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 2064 | |
bfebd1cd | 2065 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
2d76fff1 | 2066 | dm_requeue_original_request(md, rq); |
2eb6e1e3 KB |
2067 | } |
2068 | ||
466d89a6 | 2069 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 2070 | { |
bfebd1cd MS |
2071 | if (!orig->q->mq_ops) |
2072 | blk_start_request(orig); | |
2073 | else | |
2074 | blk_mq_start_request(orig); | |
466d89a6 | 2075 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 2076 | |
0ce65797 MS |
2077 | if (md->seq_rq_merge_deadline_usecs) { |
2078 | md->last_rq_pos = rq_end_sector(orig); | |
2079 | md->last_rq_rw = rq_data_dir(orig); | |
2080 | md->last_rq_start_time = ktime_get(); | |
2081 | } | |
de3ec86d | 2082 | |
e262f347 MP |
2083 | if (unlikely(dm_stats_used(&md->stats))) { |
2084 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
2085 | tio->duration_jiffies = jiffies; | |
2086 | tio->n_sectors = blk_rq_sectors(orig); | |
528ec5ab MC |
2087 | dm_stats_account_io(&md->stats, rq_data_dir(orig), |
2088 | blk_rq_pos(orig), tio->n_sectors, false, 0, | |
2089 | &tio->stats_aux); | |
e262f347 MP |
2090 | } |
2091 | ||
ba1cbad9 MS |
2092 | /* |
2093 | * Hold the md reference here for the in-flight I/O. | |
2094 | * We can't rely on the reference count by device opener, | |
2095 | * because the device may be closed during the request completion | |
2096 | * when all bios are completed. | |
2097 | * See the comment in rq_completed() too. | |
2098 | */ | |
2099 | dm_get(md); | |
ba1cbad9 MS |
2100 | } |
2101 | ||
0ce65797 MS |
2102 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
2103 | ||
2104 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
2105 | { | |
2106 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
2107 | } | |
2108 | ||
2109 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
2110 | const char *buf, size_t count) | |
2111 | { | |
2112 | unsigned deadline; | |
2113 | ||
17e149b8 | 2114 | if (!dm_request_based(md) || md->use_blk_mq) |
0ce65797 MS |
2115 | return count; |
2116 | ||
2117 | if (kstrtouint(buf, 10, &deadline)) | |
2118 | return -EINVAL; | |
2119 | ||
2120 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2121 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2122 | ||
2123 | md->seq_rq_merge_deadline_usecs = deadline; | |
2124 | ||
2125 | return count; | |
2126 | } | |
2127 | ||
2128 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2129 | { | |
2130 | ktime_t kt_deadline; | |
2131 | ||
2132 | if (!md->seq_rq_merge_deadline_usecs) | |
2133 | return false; | |
2134 | ||
2135 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2136 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2137 | ||
2138 | return !ktime_after(ktime_get(), kt_deadline); | |
2139 | } | |
2140 | ||
cec47e3d KU |
2141 | /* |
2142 | * q->request_fn for request-based dm. | |
2143 | * Called with the queue lock held. | |
2144 | */ | |
2145 | static void dm_request_fn(struct request_queue *q) | |
2146 | { | |
2147 | struct mapped_device *md = q->queuedata; | |
c91852ff | 2148 | struct dm_target *ti = md->immutable_target; |
466d89a6 | 2149 | struct request *rq; |
2eb6e1e3 | 2150 | struct dm_rq_target_io *tio; |
c91852ff MS |
2151 | sector_t pos = 0; |
2152 | ||
2153 | if (unlikely(!ti)) { | |
2154 | int srcu_idx; | |
2155 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2156 | ||
2157 | ti = dm_table_find_target(map, pos); | |
2158 | dm_put_live_table(md, srcu_idx); | |
2159 | } | |
cec47e3d KU |
2160 | |
2161 | /* | |
b4324fee KU |
2162 | * For suspend, check blk_queue_stopped() and increment |
2163 | * ->pending within a single queue_lock not to increment the | |
2164 | * number of in-flight I/Os after the queue is stopped in | |
2165 | * dm_suspend(). | |
cec47e3d | 2166 | */ |
7eaceacc | 2167 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2168 | rq = blk_peek_request(q); |
2169 | if (!rq) | |
c91852ff | 2170 | return; |
cec47e3d | 2171 | |
29e4013d TH |
2172 | /* always use block 0 to find the target for flushes for now */ |
2173 | pos = 0; | |
3a5e02ce | 2174 | if (req_op(rq) != REQ_OP_FLUSH) |
29e4013d TH |
2175 | pos = blk_rq_pos(rq); |
2176 | ||
c91852ff MS |
2177 | if ((dm_request_peeked_before_merge_deadline(md) && |
2178 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
2179 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) || | |
2180 | (ti->type->busy && ti->type->busy(ti))) { | |
2181 | blk_delay_queue(q, HZ / 100); | |
2182 | return; | |
ba1cbad9 | 2183 | } |
d0bcb878 | 2184 | |
466d89a6 | 2185 | dm_start_request(md, rq); |
9eef87da | 2186 | |
bfebd1cd | 2187 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2188 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2189 | tio->ti = ti; | |
2190 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2191 | BUG_ON(!irqs_disabled()); |
cec47e3d | 2192 | } |
cec47e3d KU |
2193 | } |
2194 | ||
1da177e4 LT |
2195 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2196 | { | |
8a57dfc6 CS |
2197 | int r = bdi_bits; |
2198 | struct mapped_device *md = congested_data; | |
2199 | struct dm_table *map; | |
1da177e4 | 2200 | |
1eb787ec | 2201 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 2202 | if (dm_request_based(md)) { |
cec47e3d | 2203 | /* |
e522c039 MS |
2204 | * With request-based DM we only need to check the |
2205 | * top-level queue for congestion. | |
cec47e3d | 2206 | */ |
e522c039 MS |
2207 | r = md->queue->backing_dev_info.wb.state & bdi_bits; |
2208 | } else { | |
2209 | map = dm_get_live_table_fast(md); | |
2210 | if (map) | |
cec47e3d | 2211 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 2212 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2213 | } |
2214 | } | |
2215 | ||
1da177e4 LT |
2216 | return r; |
2217 | } | |
2218 | ||
2219 | /*----------------------------------------------------------------- | |
2220 | * An IDR is used to keep track of allocated minor numbers. | |
2221 | *---------------------------------------------------------------*/ | |
2b06cfff | 2222 | static void free_minor(int minor) |
1da177e4 | 2223 | { |
f32c10b0 | 2224 | spin_lock(&_minor_lock); |
1da177e4 | 2225 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2226 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2227 | } |
2228 | ||
2229 | /* | |
2230 | * See if the device with a specific minor # is free. | |
2231 | */ | |
cf13ab8e | 2232 | static int specific_minor(int minor) |
1da177e4 | 2233 | { |
c9d76be6 | 2234 | int r; |
1da177e4 LT |
2235 | |
2236 | if (minor >= (1 << MINORBITS)) | |
2237 | return -EINVAL; | |
2238 | ||
c9d76be6 | 2239 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2240 | spin_lock(&_minor_lock); |
1da177e4 | 2241 | |
c9d76be6 | 2242 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2243 | |
f32c10b0 | 2244 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2245 | idr_preload_end(); |
2246 | if (r < 0) | |
2247 | return r == -ENOSPC ? -EBUSY : r; | |
2248 | return 0; | |
1da177e4 LT |
2249 | } |
2250 | ||
cf13ab8e | 2251 | static int next_free_minor(int *minor) |
1da177e4 | 2252 | { |
c9d76be6 | 2253 | int r; |
62f75c2f | 2254 | |
c9d76be6 | 2255 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2256 | spin_lock(&_minor_lock); |
1da177e4 | 2257 | |
c9d76be6 | 2258 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2259 | |
f32c10b0 | 2260 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2261 | idr_preload_end(); |
2262 | if (r < 0) | |
2263 | return r; | |
2264 | *minor = r; | |
2265 | return 0; | |
1da177e4 LT |
2266 | } |
2267 | ||
83d5cde4 | 2268 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2269 | |
53d5914f MP |
2270 | static void dm_wq_work(struct work_struct *work); |
2271 | ||
4a0b4ddf MS |
2272 | static void dm_init_md_queue(struct mapped_device *md) |
2273 | { | |
2274 | /* | |
2275 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2276 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2277 | * The type is decided at the first table loading time. |
2278 | * To prevent problematic device stacking, clear the queue flag | |
2279 | * for request stacking support until then. | |
2280 | * | |
2281 | * This queue is new, so no concurrency on the queue_flags. | |
2282 | */ | |
2283 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
2284 | |
2285 | /* | |
2286 | * Initialize data that will only be used by a non-blk-mq DM queue | |
2287 | * - must do so here (in alloc_dev callchain) before queue is used | |
2288 | */ | |
2289 | md->queue->queuedata = md; | |
2290 | md->queue->backing_dev_info.congested_data = md; | |
bfebd1cd | 2291 | } |
4a0b4ddf | 2292 | |
eca7ee6d | 2293 | static void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 2294 | { |
17e149b8 | 2295 | md->use_blk_mq = false; |
bfebd1cd MS |
2296 | dm_init_md_queue(md); |
2297 | ||
2298 | /* | |
2299 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2300 | */ | |
4a0b4ddf | 2301 | md->queue->backing_dev_info.congested_fn = dm_any_congested; |
4a0b4ddf | 2302 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2303 | } |
2304 | ||
0f20972f MS |
2305 | static void cleanup_mapped_device(struct mapped_device *md) |
2306 | { | |
0f20972f MS |
2307 | if (md->wq) |
2308 | destroy_workqueue(md->wq); | |
2309 | if (md->kworker_task) | |
2310 | kthread_stop(md->kworker_task); | |
6f65985e JL |
2311 | mempool_destroy(md->io_pool); |
2312 | mempool_destroy(md->rq_pool); | |
0f20972f MS |
2313 | if (md->bs) |
2314 | bioset_free(md->bs); | |
2315 | ||
b06075a9 MP |
2316 | cleanup_srcu_struct(&md->io_barrier); |
2317 | ||
0f20972f MS |
2318 | if (md->disk) { |
2319 | spin_lock(&_minor_lock); | |
2320 | md->disk->private_data = NULL; | |
2321 | spin_unlock(&_minor_lock); | |
0f20972f MS |
2322 | del_gendisk(md->disk); |
2323 | put_disk(md->disk); | |
2324 | } | |
2325 | ||
2326 | if (md->queue) | |
2327 | blk_cleanup_queue(md->queue); | |
2328 | ||
2329 | if (md->bdev) { | |
2330 | bdput(md->bdev); | |
2331 | md->bdev = NULL; | |
2332 | } | |
2333 | } | |
2334 | ||
1da177e4 LT |
2335 | /* |
2336 | * Allocate and initialise a blank device with a given minor. | |
2337 | */ | |
2b06cfff | 2338 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 2339 | { |
115485e8 MS |
2340 | int r, numa_node_id = dm_get_numa_node(); |
2341 | struct mapped_device *md; | |
ba61fdd1 | 2342 | void *old_md; |
1da177e4 | 2343 | |
115485e8 | 2344 | md = kzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
2345 | if (!md) { |
2346 | DMWARN("unable to allocate device, out of memory."); | |
2347 | return NULL; | |
2348 | } | |
2349 | ||
10da4f79 | 2350 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2351 | goto bad_module_get; |
10da4f79 | 2352 | |
1da177e4 | 2353 | /* get a minor number for the dev */ |
2b06cfff | 2354 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2355 | r = next_free_minor(&minor); |
2b06cfff | 2356 | else |
cf13ab8e | 2357 | r = specific_minor(minor); |
1da177e4 | 2358 | if (r < 0) |
6ed7ade8 | 2359 | goto bad_minor; |
1da177e4 | 2360 | |
83d5e5b0 MP |
2361 | r = init_srcu_struct(&md->io_barrier); |
2362 | if (r < 0) | |
2363 | goto bad_io_barrier; | |
2364 | ||
115485e8 | 2365 | md->numa_node_id = numa_node_id; |
17e149b8 | 2366 | md->use_blk_mq = use_blk_mq; |
591ddcfc | 2367 | md->init_tio_pdu = false; |
a5664dad | 2368 | md->type = DM_TYPE_NONE; |
e61290a4 | 2369 | mutex_init(&md->suspend_lock); |
a5664dad | 2370 | mutex_init(&md->type_lock); |
86f1152b | 2371 | mutex_init(&md->table_devices_lock); |
022c2611 | 2372 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2373 | atomic_set(&md->holders, 1); |
5c6bd75d | 2374 | atomic_set(&md->open_count, 0); |
1da177e4 | 2375 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2376 | atomic_set(&md->uevent_seq, 0); |
2377 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2378 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2379 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2380 | |
115485e8 | 2381 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
1da177e4 | 2382 | if (!md->queue) |
0f20972f | 2383 | goto bad; |
1da177e4 | 2384 | |
4a0b4ddf | 2385 | dm_init_md_queue(md); |
9faf400f | 2386 | |
115485e8 | 2387 | md->disk = alloc_disk_node(1, numa_node_id); |
1da177e4 | 2388 | if (!md->disk) |
0f20972f | 2389 | goto bad; |
1da177e4 | 2390 | |
316d315b NK |
2391 | atomic_set(&md->pending[0], 0); |
2392 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2393 | init_waitqueue_head(&md->wait); |
53d5914f | 2394 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2395 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2396 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2397 | md->kworker_task = NULL; |
f0b04115 | 2398 | |
1da177e4 LT |
2399 | md->disk->major = _major; |
2400 | md->disk->first_minor = minor; | |
2401 | md->disk->fops = &dm_blk_dops; | |
2402 | md->disk->queue = md->queue; | |
2403 | md->disk->private_data = md; | |
2404 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2405 | add_disk(md->disk); | |
7e51f257 | 2406 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2407 | |
670368a8 | 2408 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 2409 | if (!md->wq) |
0f20972f | 2410 | goto bad; |
304f3f6a | 2411 | |
32a926da MP |
2412 | md->bdev = bdget_disk(md->disk, 0); |
2413 | if (!md->bdev) | |
0f20972f | 2414 | goto bad; |
32a926da | 2415 | |
6a8736d1 TH |
2416 | bio_init(&md->flush_bio); |
2417 | md->flush_bio.bi_bdev = md->bdev; | |
e6047149 | 2418 | bio_set_op_attrs(&md->flush_bio, REQ_OP_WRITE, WRITE_FLUSH); |
6a8736d1 | 2419 | |
fd2ed4d2 MP |
2420 | dm_stats_init(&md->stats); |
2421 | ||
ba61fdd1 | 2422 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2423 | spin_lock(&_minor_lock); |
ba61fdd1 | 2424 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2425 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2426 | |
2427 | BUG_ON(old_md != MINOR_ALLOCED); | |
2428 | ||
1da177e4 LT |
2429 | return md; |
2430 | ||
0f20972f MS |
2431 | bad: |
2432 | cleanup_mapped_device(md); | |
83d5e5b0 | 2433 | bad_io_barrier: |
1da177e4 | 2434 | free_minor(minor); |
6ed7ade8 | 2435 | bad_minor: |
10da4f79 | 2436 | module_put(THIS_MODULE); |
6ed7ade8 | 2437 | bad_module_get: |
1da177e4 LT |
2438 | kfree(md); |
2439 | return NULL; | |
2440 | } | |
2441 | ||
ae9da83f JN |
2442 | static void unlock_fs(struct mapped_device *md); |
2443 | ||
1da177e4 LT |
2444 | static void free_dev(struct mapped_device *md) |
2445 | { | |
f331c029 | 2446 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2447 | |
32a926da | 2448 | unlock_fs(md); |
2eb6e1e3 | 2449 | |
0f20972f | 2450 | cleanup_mapped_device(md); |
1c357a1e MS |
2451 | if (md->tag_set) { |
2452 | blk_mq_free_tag_set(md->tag_set); | |
2453 | kfree(md->tag_set); | |
2454 | } | |
63a4f065 | 2455 | |
86f1152b | 2456 | free_table_devices(&md->table_devices); |
63a4f065 | 2457 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2458 | free_minor(minor); |
2459 | ||
10da4f79 | 2460 | module_put(THIS_MODULE); |
1da177e4 LT |
2461 | kfree(md); |
2462 | } | |
2463 | ||
e6ee8c0b KU |
2464 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2465 | { | |
c0820cf5 | 2466 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2467 | |
4e6e36c3 MS |
2468 | if (md->bs) { |
2469 | /* The md already has necessary mempools. */ | |
2470 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
16245bdc JN |
2471 | /* |
2472 | * Reload bioset because front_pad may have changed | |
2473 | * because a different table was loaded. | |
2474 | */ | |
2475 | bioset_free(md->bs); | |
2476 | md->bs = p->bs; | |
2477 | p->bs = NULL; | |
16245bdc | 2478 | } |
4e6e36c3 MS |
2479 | /* |
2480 | * There's no need to reload with request-based dm | |
2481 | * because the size of front_pad doesn't change. | |
2482 | * Note for future: If you are to reload bioset, | |
2483 | * prep-ed requests in the queue may refer | |
2484 | * to bio from the old bioset, so you must walk | |
2485 | * through the queue to unprep. | |
2486 | */ | |
2487 | goto out; | |
c0820cf5 | 2488 | } |
e6ee8c0b | 2489 | |
cbc4e3c1 MS |
2490 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
2491 | ||
e6ee8c0b KU |
2492 | md->io_pool = p->io_pool; |
2493 | p->io_pool = NULL; | |
1ae49ea2 MS |
2494 | md->rq_pool = p->rq_pool; |
2495 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2496 | md->bs = p->bs; |
2497 | p->bs = NULL; | |
4e6e36c3 | 2498 | |
e6ee8c0b | 2499 | out: |
02233342 | 2500 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
2501 | dm_table_free_md_mempools(t); |
2502 | } | |
2503 | ||
1da177e4 LT |
2504 | /* |
2505 | * Bind a table to the device. | |
2506 | */ | |
2507 | static void event_callback(void *context) | |
2508 | { | |
7a8c3d3b MA |
2509 | unsigned long flags; |
2510 | LIST_HEAD(uevents); | |
1da177e4 LT |
2511 | struct mapped_device *md = (struct mapped_device *) context; |
2512 | ||
7a8c3d3b MA |
2513 | spin_lock_irqsave(&md->uevent_lock, flags); |
2514 | list_splice_init(&md->uevent_list, &uevents); | |
2515 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2516 | ||
ed9e1982 | 2517 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2518 | |
1da177e4 LT |
2519 | atomic_inc(&md->event_nr); |
2520 | wake_up(&md->eventq); | |
2521 | } | |
2522 | ||
c217649b MS |
2523 | /* |
2524 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2525 | */ | |
4e90188b | 2526 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2527 | { |
4e90188b | 2528 | set_capacity(md->disk, size); |
1da177e4 | 2529 | |
db8fef4f | 2530 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2531 | } |
2532 | ||
042d2a9b AK |
2533 | /* |
2534 | * Returns old map, which caller must destroy. | |
2535 | */ | |
2536 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2537 | struct queue_limits *limits) | |
1da177e4 | 2538 | { |
042d2a9b | 2539 | struct dm_table *old_map; |
165125e1 | 2540 | struct request_queue *q = md->queue; |
1da177e4 LT |
2541 | sector_t size; |
2542 | ||
2543 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2544 | |
2545 | /* | |
2546 | * Wipe any geometry if the size of the table changed. | |
2547 | */ | |
fd2ed4d2 | 2548 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2549 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2550 | ||
32a926da | 2551 | __set_size(md, size); |
d5816876 | 2552 | |
2ca3310e AK |
2553 | dm_table_event_callback(t, event_callback, md); |
2554 | ||
e6ee8c0b KU |
2555 | /* |
2556 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2557 | * for request-based during suspension. So stop it to prevent | |
2558 | * I/O mapping before resume. | |
2559 | * This must be done before setting the queue restrictions, | |
2560 | * because request-based dm may be run just after the setting. | |
2561 | */ | |
16f12266 | 2562 | if (dm_table_request_based(t)) { |
eca7ee6d | 2563 | dm_stop_queue(q); |
16f12266 MS |
2564 | /* |
2565 | * Leverage the fact that request-based DM targets are | |
2566 | * immutable singletons and establish md->immutable_target | |
2567 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
2568 | */ | |
2569 | md->immutable_target = dm_table_get_immutable_target(t); | |
2570 | } | |
e6ee8c0b KU |
2571 | |
2572 | __bind_mempools(md, t); | |
2573 | ||
a12f5d48 | 2574 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2575 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2576 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2577 | ||
754c5fc7 | 2578 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2579 | if (old_map) |
2580 | dm_sync_table(md); | |
1da177e4 | 2581 | |
042d2a9b | 2582 | return old_map; |
1da177e4 LT |
2583 | } |
2584 | ||
a7940155 AK |
2585 | /* |
2586 | * Returns unbound table for the caller to free. | |
2587 | */ | |
2588 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2589 | { |
a12f5d48 | 2590 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2591 | |
2592 | if (!map) | |
a7940155 | 2593 | return NULL; |
1da177e4 LT |
2594 | |
2595 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2596 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2597 | dm_sync_table(md); |
a7940155 AK |
2598 | |
2599 | return map; | |
1da177e4 LT |
2600 | } |
2601 | ||
2602 | /* | |
2603 | * Constructor for a new device. | |
2604 | */ | |
2b06cfff | 2605 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2606 | { |
2607 | struct mapped_device *md; | |
2608 | ||
2b06cfff | 2609 | md = alloc_dev(minor); |
1da177e4 LT |
2610 | if (!md) |
2611 | return -ENXIO; | |
2612 | ||
784aae73 MB |
2613 | dm_sysfs_init(md); |
2614 | ||
1da177e4 LT |
2615 | *result = md; |
2616 | return 0; | |
2617 | } | |
2618 | ||
a5664dad MS |
2619 | /* |
2620 | * Functions to manage md->type. | |
2621 | * All are required to hold md->type_lock. | |
2622 | */ | |
2623 | void dm_lock_md_type(struct mapped_device *md) | |
2624 | { | |
2625 | mutex_lock(&md->type_lock); | |
2626 | } | |
2627 | ||
2628 | void dm_unlock_md_type(struct mapped_device *md) | |
2629 | { | |
2630 | mutex_unlock(&md->type_lock); | |
2631 | } | |
2632 | ||
2633 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2634 | { | |
00c4fc3b | 2635 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2636 | md->type = type; |
2637 | } | |
2638 | ||
2639 | unsigned dm_get_md_type(struct mapped_device *md) | |
2640 | { | |
2641 | return md->type; | |
2642 | } | |
2643 | ||
36a0456f AK |
2644 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2645 | { | |
2646 | return md->immutable_target_type; | |
2647 | } | |
2648 | ||
f84cb8a4 MS |
2649 | /* |
2650 | * The queue_limits are only valid as long as you have a reference | |
2651 | * count on 'md'. | |
2652 | */ | |
2653 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2654 | { | |
2655 | BUG_ON(!atomic_read(&md->holders)); | |
2656 | return &md->queue->limits; | |
2657 | } | |
2658 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2659 | ||
eca7ee6d | 2660 | static void dm_old_init_rq_based_worker_thread(struct mapped_device *md) |
bfebd1cd MS |
2661 | { |
2662 | /* Initialize the request-based DM worker thread */ | |
2663 | init_kthread_worker(&md->kworker); | |
2664 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2665 | "kdmwork-%s", dm_device_name(md)); | |
2666 | } | |
2667 | ||
4a0b4ddf | 2668 | /* |
eca7ee6d | 2669 | * Fully initialize a .request_fn request-based queue. |
4a0b4ddf | 2670 | */ |
eca7ee6d | 2671 | static int dm_old_init_request_queue(struct mapped_device *md) |
4a0b4ddf | 2672 | { |
4a0b4ddf | 2673 | /* Fully initialize the queue */ |
e233d800 | 2674 | if (!blk_init_allocated_queue(md->queue, dm_request_fn, NULL)) |
bfebd1cd | 2675 | return -EINVAL; |
4a0b4ddf | 2676 | |
0ce65797 MS |
2677 | /* disable dm_request_fn's merge heuristic by default */ |
2678 | md->seq_rq_merge_deadline_usecs = 0; | |
2679 | ||
eca7ee6d | 2680 | dm_init_normal_md_queue(md); |
4a0b4ddf | 2681 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
eca7ee6d | 2682 | blk_queue_prep_rq(md->queue, dm_old_prep_fn); |
4a0b4ddf | 2683 | |
eca7ee6d | 2684 | dm_old_init_rq_based_worker_thread(md); |
2eb6e1e3 | 2685 | |
4a0b4ddf MS |
2686 | elv_register_queue(md->queue); |
2687 | ||
bfebd1cd MS |
2688 | return 0; |
2689 | } | |
2690 | ||
2691 | static int dm_mq_init_request(void *data, struct request *rq, | |
2692 | unsigned int hctx_idx, unsigned int request_idx, | |
2693 | unsigned int numa_node) | |
2694 | { | |
2695 | struct mapped_device *md = data; | |
2696 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2697 | ||
2698 | /* | |
2699 | * Must initialize md member of tio, otherwise it won't | |
2700 | * be available in dm_mq_queue_rq. | |
2701 | */ | |
2702 | tio->md = md; | |
2703 | ||
591ddcfc MS |
2704 | if (md->init_tio_pdu) { |
2705 | /* target-specific per-io data is immediately after the tio */ | |
2706 | tio->info.ptr = tio + 1; | |
2707 | } | |
2708 | ||
bfebd1cd MS |
2709 | return 0; |
2710 | } | |
2711 | ||
2712 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
2713 | const struct blk_mq_queue_data *bd) | |
2714 | { | |
2715 | struct request *rq = bd->rq; | |
2716 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2717 | struct mapped_device *md = tio->md; | |
16f12266 | 2718 | struct dm_target *ti = md->immutable_target; |
bfebd1cd | 2719 | |
16f12266 MS |
2720 | if (unlikely(!ti)) { |
2721 | int srcu_idx; | |
2722 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
bfebd1cd | 2723 | |
16f12266 | 2724 | ti = dm_table_find_target(map, 0); |
bfebd1cd | 2725 | dm_put_live_table(md, srcu_idx); |
bfebd1cd | 2726 | } |
bfebd1cd MS |
2727 | |
2728 | if (ti->type->busy && ti->type->busy(ti)) | |
2729 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2730 | ||
2731 | dm_start_request(md, rq); | |
2732 | ||
2733 | /* Init tio using md established in .init_request */ | |
2734 | init_tio(tio, rq, md); | |
2735 | ||
02233342 MS |
2736 | /* |
2737 | * Establish tio->ti before queuing work (map_tio_request) | |
2738 | * or making direct call to map_request(). | |
2739 | */ | |
bfebd1cd | 2740 | tio->ti = ti; |
02233342 | 2741 | |
c5248f79 MS |
2742 | /* Direct call is fine since .queue_rq allows allocations */ |
2743 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) { | |
2744 | /* Undo dm_start_request() before requeuing */ | |
2745 | rq_end_stats(md, rq); | |
2746 | rq_completed(md, rq_data_dir(rq), false); | |
2747 | return BLK_MQ_RQ_QUEUE_BUSY; | |
02233342 | 2748 | } |
bfebd1cd MS |
2749 | |
2750 | return BLK_MQ_RQ_QUEUE_OK; | |
2751 | } | |
2752 | ||
2753 | static struct blk_mq_ops dm_mq_ops = { | |
2754 | .queue_rq = dm_mq_queue_rq, | |
2755 | .map_queue = blk_mq_map_queue, | |
2756 | .complete = dm_softirq_done, | |
2757 | .init_request = dm_mq_init_request, | |
2758 | }; | |
2759 | ||
591ddcfc MS |
2760 | static int dm_mq_init_request_queue(struct mapped_device *md, |
2761 | struct dm_target *immutable_tgt) | |
bfebd1cd MS |
2762 | { |
2763 | struct request_queue *q; | |
2764 | int err; | |
2765 | ||
c5248f79 MS |
2766 | if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) { |
2767 | DMERR("request-based dm-mq may only be stacked on blk-mq device(s)"); | |
2768 | return -EINVAL; | |
2769 | } | |
2770 | ||
115485e8 | 2771 | md->tag_set = kzalloc_node(sizeof(struct blk_mq_tag_set), GFP_KERNEL, md->numa_node_id); |
1c357a1e MS |
2772 | if (!md->tag_set) |
2773 | return -ENOMEM; | |
2774 | ||
2775 | md->tag_set->ops = &dm_mq_ops; | |
2776 | md->tag_set->queue_depth = dm_get_blk_mq_queue_depth(); | |
115485e8 | 2777 | md->tag_set->numa_node = md->numa_node_id; |
1c357a1e MS |
2778 | md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; |
2779 | md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues(); | |
2780 | md->tag_set->driver_data = md; | |
2781 | ||
2782 | md->tag_set->cmd_size = sizeof(struct dm_rq_target_io); | |
591ddcfc MS |
2783 | if (immutable_tgt && immutable_tgt->per_io_data_size) { |
2784 | /* any target-specific per-io data is immediately after the tio */ | |
2785 | md->tag_set->cmd_size += immutable_tgt->per_io_data_size; | |
2786 | md->init_tio_pdu = true; | |
2787 | } | |
bfebd1cd | 2788 | |
1c357a1e | 2789 | err = blk_mq_alloc_tag_set(md->tag_set); |
bfebd1cd | 2790 | if (err) |
1c357a1e | 2791 | goto out_kfree_tag_set; |
bfebd1cd | 2792 | |
1c357a1e | 2793 | q = blk_mq_init_allocated_queue(md->tag_set, md->queue); |
bfebd1cd MS |
2794 | if (IS_ERR(q)) { |
2795 | err = PTR_ERR(q); | |
2796 | goto out_tag_set; | |
2797 | } | |
bfebd1cd MS |
2798 | dm_init_md_queue(md); |
2799 | ||
2800 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2801 | blk_mq_register_disk(md->disk); | |
2802 | ||
bfebd1cd MS |
2803 | return 0; |
2804 | ||
2805 | out_tag_set: | |
1c357a1e MS |
2806 | blk_mq_free_tag_set(md->tag_set); |
2807 | out_kfree_tag_set: | |
2808 | kfree(md->tag_set); | |
2809 | ||
bfebd1cd | 2810 | return err; |
4a0b4ddf MS |
2811 | } |
2812 | ||
4e6e36c3 MS |
2813 | static unsigned filter_md_type(unsigned type, struct mapped_device *md) |
2814 | { | |
2815 | if (type == DM_TYPE_BIO_BASED) | |
2816 | return type; | |
2817 | ||
2818 | return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED; | |
2819 | } | |
2820 | ||
4a0b4ddf MS |
2821 | /* |
2822 | * Setup the DM device's queue based on md's type | |
2823 | */ | |
591ddcfc | 2824 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2825 | { |
bfebd1cd | 2826 | int r; |
17e149b8 | 2827 | unsigned md_type = filter_md_type(dm_get_md_type(md), md); |
bfebd1cd MS |
2828 | |
2829 | switch (md_type) { | |
2830 | case DM_TYPE_REQUEST_BASED: | |
eca7ee6d | 2831 | r = dm_old_init_request_queue(md); |
bfebd1cd | 2832 | if (r) { |
eca7ee6d | 2833 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2834 | return r; |
ff36ab34 | 2835 | } |
bfebd1cd MS |
2836 | break; |
2837 | case DM_TYPE_MQ_REQUEST_BASED: | |
591ddcfc | 2838 | r = dm_mq_init_request_queue(md, dm_table_get_immutable_target(t)); |
bfebd1cd | 2839 | if (r) { |
eca7ee6d | 2840 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2841 | return r; |
2842 | } | |
2843 | break; | |
2844 | case DM_TYPE_BIO_BASED: | |
eca7ee6d | 2845 | dm_init_normal_md_queue(md); |
ff36ab34 | 2846 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
2847 | /* |
2848 | * DM handles splitting bios as needed. Free the bio_split bioset | |
2849 | * since it won't be used (saves 1 process per bio-based DM device). | |
2850 | */ | |
2851 | bioset_free(md->queue->bio_split); | |
2852 | md->queue->bio_split = NULL; | |
bfebd1cd | 2853 | break; |
4a0b4ddf MS |
2854 | } |
2855 | ||
2856 | return 0; | |
2857 | } | |
2858 | ||
2bec1f4a | 2859 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2860 | { |
2861 | struct mapped_device *md; | |
1da177e4 LT |
2862 | unsigned minor = MINOR(dev); |
2863 | ||
2864 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2865 | return NULL; | |
2866 | ||
f32c10b0 | 2867 | spin_lock(&_minor_lock); |
1da177e4 LT |
2868 | |
2869 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2870 | if (md) { |
2871 | if ((md == MINOR_ALLOCED || | |
2872 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2873 | dm_deleting_md(md) || | |
2874 | test_bit(DMF_FREEING, &md->flags))) { | |
2875 | md = NULL; | |
2876 | goto out; | |
2877 | } | |
2878 | dm_get(md); | |
fba9f90e | 2879 | } |
1da177e4 | 2880 | |
fba9f90e | 2881 | out: |
f32c10b0 | 2882 | spin_unlock(&_minor_lock); |
1da177e4 | 2883 | |
637842cf DT |
2884 | return md; |
2885 | } | |
3cf2e4ba | 2886 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2887 | |
9ade92a9 | 2888 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2889 | { |
9ade92a9 | 2890 | return md->interface_ptr; |
1da177e4 LT |
2891 | } |
2892 | ||
2893 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2894 | { | |
2895 | md->interface_ptr = ptr; | |
2896 | } | |
2897 | ||
2898 | void dm_get(struct mapped_device *md) | |
2899 | { | |
2900 | atomic_inc(&md->holders); | |
3f77316d | 2901 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2902 | } |
2903 | ||
09ee96b2 MP |
2904 | int dm_hold(struct mapped_device *md) |
2905 | { | |
2906 | spin_lock(&_minor_lock); | |
2907 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2908 | spin_unlock(&_minor_lock); | |
2909 | return -EBUSY; | |
2910 | } | |
2911 | dm_get(md); | |
2912 | spin_unlock(&_minor_lock); | |
2913 | return 0; | |
2914 | } | |
2915 | EXPORT_SYMBOL_GPL(dm_hold); | |
2916 | ||
72d94861 AK |
2917 | const char *dm_device_name(struct mapped_device *md) |
2918 | { | |
2919 | return md->name; | |
2920 | } | |
2921 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2922 | ||
3f77316d | 2923 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2924 | { |
1134e5ae | 2925 | struct dm_table *map; |
83d5e5b0 | 2926 | int srcu_idx; |
1da177e4 | 2927 | |
3f77316d | 2928 | might_sleep(); |
fba9f90e | 2929 | |
63a4f065 | 2930 | spin_lock(&_minor_lock); |
3f77316d KU |
2931 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2932 | set_bit(DMF_FREEING, &md->flags); | |
2933 | spin_unlock(&_minor_lock); | |
2934 | ||
02233342 | 2935 | if (dm_request_based(md) && md->kworker_task) |
2eb6e1e3 KB |
2936 | flush_kthread_worker(&md->kworker); |
2937 | ||
ab7c7bb6 MP |
2938 | /* |
2939 | * Take suspend_lock so that presuspend and postsuspend methods | |
2940 | * do not race with internal suspend. | |
2941 | */ | |
2942 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2943 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2944 | if (!dm_suspended_md(md)) { |
2945 | dm_table_presuspend_targets(map); | |
2946 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2947 | } |
83d5e5b0 MP |
2948 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2949 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2950 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2951 | |
3f77316d KU |
2952 | /* |
2953 | * Rare, but there may be I/O requests still going to complete, | |
2954 | * for example. Wait for all references to disappear. | |
2955 | * No one should increment the reference count of the mapped_device, | |
2956 | * after the mapped_device state becomes DMF_FREEING. | |
2957 | */ | |
2958 | if (wait) | |
2959 | while (atomic_read(&md->holders)) | |
2960 | msleep(1); | |
2961 | else if (atomic_read(&md->holders)) | |
2962 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2963 | dm_device_name(md), atomic_read(&md->holders)); | |
2964 | ||
2965 | dm_sysfs_exit(md); | |
3f77316d KU |
2966 | dm_table_destroy(__unbind(md)); |
2967 | free_dev(md); | |
2968 | } | |
2969 | ||
2970 | void dm_destroy(struct mapped_device *md) | |
2971 | { | |
2972 | __dm_destroy(md, true); | |
2973 | } | |
2974 | ||
2975 | void dm_destroy_immediate(struct mapped_device *md) | |
2976 | { | |
2977 | __dm_destroy(md, false); | |
2978 | } | |
2979 | ||
2980 | void dm_put(struct mapped_device *md) | |
2981 | { | |
2982 | atomic_dec(&md->holders); | |
1da177e4 | 2983 | } |
79eb885c | 2984 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2985 | |
401600df | 2986 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2987 | { |
2988 | int r = 0; | |
b44ebeb0 MP |
2989 | DECLARE_WAITQUEUE(wait, current); |
2990 | ||
b44ebeb0 | 2991 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2992 | |
2993 | while (1) { | |
401600df | 2994 | set_current_state(interruptible); |
46125c1c | 2995 | |
b4324fee | 2996 | if (!md_in_flight(md)) |
46125c1c MB |
2997 | break; |
2998 | ||
401600df MP |
2999 | if (interruptible == TASK_INTERRUPTIBLE && |
3000 | signal_pending(current)) { | |
46125c1c MB |
3001 | r = -EINTR; |
3002 | break; | |
3003 | } | |
3004 | ||
3005 | io_schedule(); | |
3006 | } | |
3007 | set_current_state(TASK_RUNNING); | |
3008 | ||
b44ebeb0 MP |
3009 | remove_wait_queue(&md->wait, &wait); |
3010 | ||
46125c1c MB |
3011 | return r; |
3012 | } | |
3013 | ||
1da177e4 LT |
3014 | /* |
3015 | * Process the deferred bios | |
3016 | */ | |
ef208587 | 3017 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 3018 | { |
ef208587 MP |
3019 | struct mapped_device *md = container_of(work, struct mapped_device, |
3020 | work); | |
6d6f10df | 3021 | struct bio *c; |
83d5e5b0 MP |
3022 | int srcu_idx; |
3023 | struct dm_table *map; | |
1da177e4 | 3024 | |
83d5e5b0 | 3025 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 3026 | |
3b00b203 | 3027 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
3028 | spin_lock_irq(&md->deferred_lock); |
3029 | c = bio_list_pop(&md->deferred); | |
3030 | spin_unlock_irq(&md->deferred_lock); | |
3031 | ||
6a8736d1 | 3032 | if (!c) |
df12ee99 | 3033 | break; |
022c2611 | 3034 | |
e6ee8c0b KU |
3035 | if (dm_request_based(md)) |
3036 | generic_make_request(c); | |
6a8736d1 | 3037 | else |
83d5e5b0 | 3038 | __split_and_process_bio(md, map, c); |
022c2611 | 3039 | } |
73d410c0 | 3040 | |
83d5e5b0 | 3041 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
3042 | } |
3043 | ||
9a1fb464 | 3044 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 3045 | { |
3b00b203 | 3046 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 3047 | smp_mb__after_atomic(); |
53d5914f | 3048 | queue_work(md->wq, &md->work); |
304f3f6a MB |
3049 | } |
3050 | ||
1da177e4 | 3051 | /* |
042d2a9b | 3052 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 3053 | */ |
042d2a9b | 3054 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 3055 | { |
87eb5b21 | 3056 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 3057 | struct queue_limits limits; |
042d2a9b | 3058 | int r; |
1da177e4 | 3059 | |
e61290a4 | 3060 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
3061 | |
3062 | /* device must be suspended */ | |
4f186f8b | 3063 | if (!dm_suspended_md(md)) |
93c534ae | 3064 | goto out; |
1da177e4 | 3065 | |
3ae70656 MS |
3066 | /* |
3067 | * If the new table has no data devices, retain the existing limits. | |
3068 | * This helps multipath with queue_if_no_path if all paths disappear, | |
3069 | * then new I/O is queued based on these limits, and then some paths | |
3070 | * reappear. | |
3071 | */ | |
3072 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 3073 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
3074 | if (live_map) |
3075 | limits = md->queue->limits; | |
83d5e5b0 | 3076 | dm_put_live_table_fast(md); |
3ae70656 MS |
3077 | } |
3078 | ||
87eb5b21 MC |
3079 | if (!live_map) { |
3080 | r = dm_calculate_queue_limits(table, &limits); | |
3081 | if (r) { | |
3082 | map = ERR_PTR(r); | |
3083 | goto out; | |
3084 | } | |
042d2a9b | 3085 | } |
754c5fc7 | 3086 | |
042d2a9b | 3087 | map = __bind(md, table, &limits); |
1da177e4 | 3088 | |
93c534ae | 3089 | out: |
e61290a4 | 3090 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3091 | return map; |
1da177e4 LT |
3092 | } |
3093 | ||
3094 | /* | |
3095 | * Functions to lock and unlock any filesystem running on the | |
3096 | * device. | |
3097 | */ | |
2ca3310e | 3098 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3099 | { |
e39e2e95 | 3100 | int r; |
1da177e4 LT |
3101 | |
3102 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3103 | |
db8fef4f | 3104 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3105 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3106 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3107 | md->frozen_sb = NULL; |
3108 | return r; | |
dfbe03f6 AK |
3109 | } |
3110 | ||
aa8d7c2f AK |
3111 | set_bit(DMF_FROZEN, &md->flags); |
3112 | ||
1da177e4 LT |
3113 | return 0; |
3114 | } | |
3115 | ||
2ca3310e | 3116 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3117 | { |
aa8d7c2f AK |
3118 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3119 | return; | |
3120 | ||
db8fef4f | 3121 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3122 | md->frozen_sb = NULL; |
aa8d7c2f | 3123 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3124 | } |
3125 | ||
3126 | /* | |
ffcc3936 MS |
3127 | * If __dm_suspend returns 0, the device is completely quiescent |
3128 | * now. There is no request-processing activity. All new requests | |
3129 | * are being added to md->deferred list. | |
cec47e3d | 3130 | * |
ffcc3936 | 3131 | * Caller must hold md->suspend_lock |
cec47e3d | 3132 | */ |
ffcc3936 MS |
3133 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3134 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3135 | { |
ffcc3936 MS |
3136 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3137 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3138 | int r; | |
1da177e4 | 3139 | |
2e93ccc1 KU |
3140 | /* |
3141 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3142 | * This flag is cleared before dm_suspend returns. | |
3143 | */ | |
3144 | if (noflush) | |
3145 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3146 | ||
d67ee213 MS |
3147 | /* |
3148 | * This gets reverted if there's an error later and the targets | |
3149 | * provide the .presuspend_undo hook. | |
3150 | */ | |
cf222b37 AK |
3151 | dm_table_presuspend_targets(map); |
3152 | ||
32a926da | 3153 | /* |
9f518b27 KU |
3154 | * Flush I/O to the device. |
3155 | * Any I/O submitted after lock_fs() may not be flushed. | |
3156 | * noflush takes precedence over do_lockfs. | |
3157 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3158 | */ |
3159 | if (!noflush && do_lockfs) { | |
3160 | r = lock_fs(md); | |
d67ee213 MS |
3161 | if (r) { |
3162 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3163 | return r; |
d67ee213 | 3164 | } |
aa8d7c2f | 3165 | } |
1da177e4 LT |
3166 | |
3167 | /* | |
3b00b203 MP |
3168 | * Here we must make sure that no processes are submitting requests |
3169 | * to target drivers i.e. no one may be executing | |
3170 | * __split_and_process_bio. This is called from dm_request and | |
3171 | * dm_wq_work. | |
3172 | * | |
3173 | * To get all processes out of __split_and_process_bio in dm_request, | |
3174 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3175 | * __split_and_process_bio from dm_request and quiesce the thread |
3176 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3177 | * flush_workqueue(md->wq). | |
1da177e4 | 3178 | */ |
1eb787ec | 3179 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3180 | if (map) |
3181 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3182 | |
d0bcb878 | 3183 | /* |
29e4013d TH |
3184 | * Stop md->queue before flushing md->wq in case request-based |
3185 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3186 | */ |
2eb6e1e3 | 3187 | if (dm_request_based(md)) { |
eca7ee6d | 3188 | dm_stop_queue(md->queue); |
02233342 MS |
3189 | if (md->kworker_task) |
3190 | flush_kthread_worker(&md->kworker); | |
2eb6e1e3 | 3191 | } |
cec47e3d | 3192 | |
d0bcb878 KU |
3193 | flush_workqueue(md->wq); |
3194 | ||
1da177e4 | 3195 | /* |
3b00b203 MP |
3196 | * At this point no more requests are entering target request routines. |
3197 | * We call dm_wait_for_completion to wait for all existing requests | |
3198 | * to finish. | |
1da177e4 | 3199 | */ |
ffcc3936 | 3200 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3201 | |
6d6f10df | 3202 | if (noflush) |
022c2611 | 3203 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3204 | if (map) |
3205 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3206 | |
1da177e4 | 3207 | /* were we interrupted ? */ |
46125c1c | 3208 | if (r < 0) { |
9a1fb464 | 3209 | dm_queue_flush(md); |
73d410c0 | 3210 | |
cec47e3d | 3211 | if (dm_request_based(md)) |
eca7ee6d | 3212 | dm_start_queue(md->queue); |
cec47e3d | 3213 | |
2ca3310e | 3214 | unlock_fs(md); |
d67ee213 | 3215 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3216 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3217 | } |
1da177e4 | 3218 | |
ffcc3936 MS |
3219 | return r; |
3220 | } | |
3221 | ||
3222 | /* | |
3223 | * We need to be able to change a mapping table under a mounted | |
3224 | * filesystem. For example we might want to move some data in | |
3225 | * the background. Before the table can be swapped with | |
3226 | * dm_bind_table, dm_suspend must be called to flush any in | |
3227 | * flight bios and ensure that any further io gets deferred. | |
3228 | */ | |
3229 | /* | |
3230 | * Suspend mechanism in request-based dm. | |
3231 | * | |
3232 | * 1. Flush all I/Os by lock_fs() if needed. | |
3233 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3234 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3235 | * | |
3236 | * To abort suspend, start the request_queue. | |
3237 | */ | |
3238 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3239 | { | |
3240 | struct dm_table *map = NULL; | |
3241 | int r = 0; | |
3242 | ||
3243 | retry: | |
3244 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3245 | ||
3246 | if (dm_suspended_md(md)) { | |
3247 | r = -EINVAL; | |
3248 | goto out_unlock; | |
3249 | } | |
3250 | ||
3251 | if (dm_suspended_internally_md(md)) { | |
3252 | /* already internally suspended, wait for internal resume */ | |
3253 | mutex_unlock(&md->suspend_lock); | |
3254 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3255 | if (r) | |
3256 | return r; | |
3257 | goto retry; | |
3258 | } | |
3259 | ||
a12f5d48 | 3260 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3261 | |
3262 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3263 | if (r) | |
3264 | goto out_unlock; | |
3b00b203 | 3265 | |
2ca3310e | 3266 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3267 | |
4d4471cb KU |
3268 | dm_table_postsuspend_targets(map); |
3269 | ||
d287483d | 3270 | out_unlock: |
e61290a4 | 3271 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3272 | return r; |
1da177e4 LT |
3273 | } |
3274 | ||
ffcc3936 MS |
3275 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3276 | { | |
3277 | if (map) { | |
3278 | int r = dm_table_resume_targets(map); | |
3279 | if (r) | |
3280 | return r; | |
3281 | } | |
3282 | ||
3283 | dm_queue_flush(md); | |
3284 | ||
3285 | /* | |
3286 | * Flushing deferred I/Os must be done after targets are resumed | |
3287 | * so that mapping of targets can work correctly. | |
3288 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3289 | */ | |
3290 | if (dm_request_based(md)) | |
eca7ee6d | 3291 | dm_start_queue(md->queue); |
ffcc3936 MS |
3292 | |
3293 | unlock_fs(md); | |
3294 | ||
3295 | return 0; | |
3296 | } | |
3297 | ||
1da177e4 LT |
3298 | int dm_resume(struct mapped_device *md) |
3299 | { | |
cf222b37 | 3300 | int r = -EINVAL; |
cf222b37 | 3301 | struct dm_table *map = NULL; |
1da177e4 | 3302 | |
ffcc3936 MS |
3303 | retry: |
3304 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3305 | ||
4f186f8b | 3306 | if (!dm_suspended_md(md)) |
cf222b37 | 3307 | goto out; |
cf222b37 | 3308 | |
ffcc3936 MS |
3309 | if (dm_suspended_internally_md(md)) { |
3310 | /* already internally suspended, wait for internal resume */ | |
3311 | mutex_unlock(&md->suspend_lock); | |
3312 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3313 | if (r) | |
3314 | return r; | |
3315 | goto retry; | |
3316 | } | |
3317 | ||
a12f5d48 | 3318 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3319 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3320 | goto out; |
1da177e4 | 3321 | |
ffcc3936 | 3322 | r = __dm_resume(md, map); |
8757b776 MB |
3323 | if (r) |
3324 | goto out; | |
2ca3310e | 3325 | |
2ca3310e AK |
3326 | clear_bit(DMF_SUSPENDED, &md->flags); |
3327 | ||
cf222b37 AK |
3328 | r = 0; |
3329 | out: | |
e61290a4 | 3330 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3331 | |
cf222b37 | 3332 | return r; |
1da177e4 LT |
3333 | } |
3334 | ||
fd2ed4d2 MP |
3335 | /* |
3336 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3337 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3338 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3339 | */ |
3340 | ||
ffcc3936 | 3341 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3342 | { |
ffcc3936 MS |
3343 | struct dm_table *map = NULL; |
3344 | ||
96b26c8c | 3345 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3346 | return; /* nested internal suspend */ |
3347 | ||
3348 | if (dm_suspended_md(md)) { | |
3349 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3350 | return; /* nest suspend */ | |
3351 | } | |
3352 | ||
a12f5d48 | 3353 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3354 | |
3355 | /* | |
3356 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3357 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3358 | * would require changing .presuspend to return an error -- avoid this | |
3359 | * until there is a need for more elaborate variants of internal suspend. | |
3360 | */ | |
3361 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3362 | ||
3363 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3364 | ||
3365 | dm_table_postsuspend_targets(map); | |
3366 | } | |
3367 | ||
3368 | static void __dm_internal_resume(struct mapped_device *md) | |
3369 | { | |
96b26c8c MP |
3370 | BUG_ON(!md->internal_suspend_count); |
3371 | ||
3372 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3373 | return; /* resume from nested internal suspend */ |
3374 | ||
fd2ed4d2 | 3375 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3376 | goto done; /* resume from nested suspend */ |
3377 | ||
3378 | /* | |
3379 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3380 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3381 | */ | |
3382 | (void) __dm_resume(md, NULL); | |
3383 | ||
3384 | done: | |
3385 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3386 | smp_mb__after_atomic(); | |
3387 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3388 | } | |
3389 | ||
3390 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3391 | { | |
3392 | mutex_lock(&md->suspend_lock); | |
3393 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3394 | mutex_unlock(&md->suspend_lock); | |
3395 | } | |
3396 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3397 | ||
3398 | void dm_internal_resume(struct mapped_device *md) | |
3399 | { | |
3400 | mutex_lock(&md->suspend_lock); | |
3401 | __dm_internal_resume(md); | |
3402 | mutex_unlock(&md->suspend_lock); | |
3403 | } | |
3404 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3405 | ||
3406 | /* | |
3407 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3408 | * which prevents interaction with userspace-driven suspend. | |
3409 | */ | |
3410 | ||
3411 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3412 | { | |
3413 | mutex_lock(&md->suspend_lock); | |
3414 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3415 | return; |
3416 | ||
3417 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3418 | synchronize_srcu(&md->io_barrier); | |
3419 | flush_workqueue(md->wq); | |
3420 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3421 | } | |
b735fede | 3422 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3423 | |
ffcc3936 | 3424 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3425 | { |
ffcc3936 | 3426 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3427 | goto done; |
3428 | ||
3429 | dm_queue_flush(md); | |
3430 | ||
3431 | done: | |
3432 | mutex_unlock(&md->suspend_lock); | |
3433 | } | |
b735fede | 3434 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3435 | |
1da177e4 LT |
3436 | /*----------------------------------------------------------------- |
3437 | * Event notification. | |
3438 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3439 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3440 | unsigned cookie) |
69267a30 | 3441 | { |
60935eb2 MB |
3442 | char udev_cookie[DM_COOKIE_LENGTH]; |
3443 | char *envp[] = { udev_cookie, NULL }; | |
3444 | ||
3445 | if (!cookie) | |
3abf85b5 | 3446 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3447 | else { |
3448 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3449 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3450 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3451 | action, envp); | |
60935eb2 | 3452 | } |
69267a30 AK |
3453 | } |
3454 | ||
7a8c3d3b MA |
3455 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3456 | { | |
3457 | return atomic_add_return(1, &md->uevent_seq); | |
3458 | } | |
3459 | ||
1da177e4 LT |
3460 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3461 | { | |
3462 | return atomic_read(&md->event_nr); | |
3463 | } | |
3464 | ||
3465 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3466 | { | |
3467 | return wait_event_interruptible(md->eventq, | |
3468 | (event_nr != atomic_read(&md->event_nr))); | |
3469 | } | |
3470 | ||
7a8c3d3b MA |
3471 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3472 | { | |
3473 | unsigned long flags; | |
3474 | ||
3475 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3476 | list_add(elist, &md->uevent_list); | |
3477 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3478 | } | |
3479 | ||
1da177e4 LT |
3480 | /* |
3481 | * The gendisk is only valid as long as you have a reference | |
3482 | * count on 'md'. | |
3483 | */ | |
3484 | struct gendisk *dm_disk(struct mapped_device *md) | |
3485 | { | |
3486 | return md->disk; | |
3487 | } | |
65ff5b7d | 3488 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3489 | |
784aae73 MB |
3490 | struct kobject *dm_kobject(struct mapped_device *md) |
3491 | { | |
2995fa78 | 3492 | return &md->kobj_holder.kobj; |
784aae73 MB |
3493 | } |
3494 | ||
784aae73 MB |
3495 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3496 | { | |
3497 | struct mapped_device *md; | |
3498 | ||
2995fa78 | 3499 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3500 | |
4d89b7b4 | 3501 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3502 | dm_deleting_md(md)) |
4d89b7b4 MB |
3503 | return NULL; |
3504 | ||
784aae73 MB |
3505 | dm_get(md); |
3506 | return md; | |
3507 | } | |
3508 | ||
4f186f8b | 3509 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3510 | { |
3511 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3512 | } | |
3513 | ||
ffcc3936 MS |
3514 | int dm_suspended_internally_md(struct mapped_device *md) |
3515 | { | |
3516 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3517 | } | |
3518 | ||
2c140a24 MP |
3519 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3520 | { | |
3521 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3522 | } | |
3523 | ||
64dbce58 KU |
3524 | int dm_suspended(struct dm_target *ti) |
3525 | { | |
ecdb2e25 | 3526 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3527 | } |
3528 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3529 | ||
2e93ccc1 KU |
3530 | int dm_noflush_suspending(struct dm_target *ti) |
3531 | { | |
ecdb2e25 | 3532 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3533 | } |
3534 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3535 | ||
78d8e58a | 3536 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
30187e1d | 3537 | unsigned integrity, unsigned per_io_data_size) |
e6ee8c0b | 3538 | { |
115485e8 | 3539 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a MS |
3540 | struct kmem_cache *cachep = NULL; |
3541 | unsigned int pool_size = 0; | |
5f015204 | 3542 | unsigned int front_pad; |
e6ee8c0b KU |
3543 | |
3544 | if (!pools) | |
4e6e36c3 | 3545 | return NULL; |
e6ee8c0b | 3546 | |
78d8e58a | 3547 | type = filter_md_type(type, md); |
17e149b8 | 3548 | |
78d8e58a MS |
3549 | switch (type) { |
3550 | case DM_TYPE_BIO_BASED: | |
3551 | cachep = _io_cache; | |
3552 | pool_size = dm_get_reserved_bio_based_ios(); | |
30187e1d | 3553 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
78d8e58a MS |
3554 | break; |
3555 | case DM_TYPE_REQUEST_BASED: | |
3556 | cachep = _rq_tio_cache; | |
3557 | pool_size = dm_get_reserved_rq_based_ios(); | |
3558 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); | |
3559 | if (!pools->rq_pool) | |
3560 | goto out; | |
3561 | /* fall through to setup remaining rq-based pools */ | |
3562 | case DM_TYPE_MQ_REQUEST_BASED: | |
3563 | if (!pool_size) | |
3564 | pool_size = dm_get_reserved_rq_based_ios(); | |
3565 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); | |
591ddcfc | 3566 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
3567 | break; |
3568 | default: | |
3569 | BUG(); | |
3570 | } | |
3571 | ||
3572 | if (cachep) { | |
3573 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); | |
3574 | if (!pools->io_pool) | |
3575 | goto out; | |
3576 | } | |
e6ee8c0b | 3577 | |
3d8aab2d | 3578 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3579 | if (!pools->bs) |
5f015204 | 3580 | goto out; |
e6ee8c0b | 3581 | |
a91a2785 | 3582 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3583 | goto out; |
a91a2785 | 3584 | |
e6ee8c0b | 3585 | return pools; |
5f1b670d | 3586 | |
5f1b670d CH |
3587 | out: |
3588 | dm_free_md_mempools(pools); | |
78d8e58a | 3589 | |
4e6e36c3 | 3590 | return NULL; |
e6ee8c0b KU |
3591 | } |
3592 | ||
3593 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3594 | { | |
3595 | if (!pools) | |
3596 | return; | |
3597 | ||
6f65985e JL |
3598 | mempool_destroy(pools->io_pool); |
3599 | mempool_destroy(pools->rq_pool); | |
1ae49ea2 | 3600 | |
e6ee8c0b KU |
3601 | if (pools->bs) |
3602 | bioset_free(pools->bs); | |
3603 | ||
3604 | kfree(pools); | |
3605 | } | |
3606 | ||
71cdb697 | 3607 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, |
956a4025 | 3608 | u32 flags) |
71cdb697 CH |
3609 | { |
3610 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3611 | const struct pr_ops *ops; | |
71cdb697 | 3612 | fmode_t mode; |
956a4025 | 3613 | int r; |
71cdb697 | 3614 | |
956a4025 | 3615 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3616 | if (r < 0) |
3617 | return r; | |
3618 | ||
3619 | ops = bdev->bd_disk->fops->pr_ops; | |
3620 | if (ops && ops->pr_register) | |
3621 | r = ops->pr_register(bdev, old_key, new_key, flags); | |
3622 | else | |
3623 | r = -EOPNOTSUPP; | |
3624 | ||
956a4025 | 3625 | bdput(bdev); |
71cdb697 CH |
3626 | return r; |
3627 | } | |
3628 | ||
3629 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3630 | u32 flags) |
71cdb697 CH |
3631 | { |
3632 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3633 | const struct pr_ops *ops; | |
71cdb697 | 3634 | fmode_t mode; |
956a4025 | 3635 | int r; |
71cdb697 | 3636 | |
956a4025 | 3637 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3638 | if (r < 0) |
3639 | return r; | |
3640 | ||
3641 | ops = bdev->bd_disk->fops->pr_ops; | |
3642 | if (ops && ops->pr_reserve) | |
3643 | r = ops->pr_reserve(bdev, key, type, flags); | |
3644 | else | |
3645 | r = -EOPNOTSUPP; | |
3646 | ||
956a4025 | 3647 | bdput(bdev); |
71cdb697 CH |
3648 | return r; |
3649 | } | |
3650 | ||
3651 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3652 | { | |
3653 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3654 | const struct pr_ops *ops; | |
71cdb697 | 3655 | fmode_t mode; |
956a4025 | 3656 | int r; |
71cdb697 | 3657 | |
956a4025 | 3658 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3659 | if (r < 0) |
3660 | return r; | |
3661 | ||
3662 | ops = bdev->bd_disk->fops->pr_ops; | |
3663 | if (ops && ops->pr_release) | |
3664 | r = ops->pr_release(bdev, key, type); | |
3665 | else | |
3666 | r = -EOPNOTSUPP; | |
3667 | ||
956a4025 | 3668 | bdput(bdev); |
71cdb697 CH |
3669 | return r; |
3670 | } | |
3671 | ||
3672 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3673 | enum pr_type type, bool abort) |
71cdb697 CH |
3674 | { |
3675 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3676 | const struct pr_ops *ops; | |
71cdb697 | 3677 | fmode_t mode; |
956a4025 | 3678 | int r; |
71cdb697 | 3679 | |
956a4025 | 3680 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3681 | if (r < 0) |
3682 | return r; | |
3683 | ||
3684 | ops = bdev->bd_disk->fops->pr_ops; | |
3685 | if (ops && ops->pr_preempt) | |
3686 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3687 | else | |
3688 | r = -EOPNOTSUPP; | |
3689 | ||
956a4025 | 3690 | bdput(bdev); |
71cdb697 CH |
3691 | return r; |
3692 | } | |
3693 | ||
3694 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3695 | { | |
3696 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3697 | const struct pr_ops *ops; | |
71cdb697 | 3698 | fmode_t mode; |
956a4025 | 3699 | int r; |
71cdb697 | 3700 | |
956a4025 | 3701 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3702 | if (r < 0) |
3703 | return r; | |
3704 | ||
3705 | ops = bdev->bd_disk->fops->pr_ops; | |
3706 | if (ops && ops->pr_clear) | |
3707 | r = ops->pr_clear(bdev, key); | |
3708 | else | |
3709 | r = -EOPNOTSUPP; | |
3710 | ||
956a4025 | 3711 | bdput(bdev); |
71cdb697 CH |
3712 | return r; |
3713 | } | |
3714 | ||
3715 | static const struct pr_ops dm_pr_ops = { | |
3716 | .pr_register = dm_pr_register, | |
3717 | .pr_reserve = dm_pr_reserve, | |
3718 | .pr_release = dm_pr_release, | |
3719 | .pr_preempt = dm_pr_preempt, | |
3720 | .pr_clear = dm_pr_clear, | |
3721 | }; | |
3722 | ||
83d5cde4 | 3723 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3724 | .open = dm_blk_open, |
3725 | .release = dm_blk_close, | |
aa129a22 | 3726 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3727 | .getgeo = dm_blk_getgeo, |
71cdb697 | 3728 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3729 | .owner = THIS_MODULE |
3730 | }; | |
3731 | ||
1da177e4 LT |
3732 | /* |
3733 | * module hooks | |
3734 | */ | |
3735 | module_init(dm_init); | |
3736 | module_exit(dm_exit); | |
3737 | ||
3738 | module_param(major, uint, 0); | |
3739 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3740 | |
e8603136 MS |
3741 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3742 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3743 | ||
f4790826 MS |
3744 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3745 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3746 | ||
17e149b8 MS |
3747 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
3748 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
3749 | ||
faad87df MS |
3750 | module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR); |
3751 | MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices"); | |
3752 | ||
3753 | module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR); | |
3754 | MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices"); | |
3755 | ||
115485e8 MS |
3756 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
3757 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
3758 | ||
1da177e4 LT |
3759 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3760 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3761 | MODULE_LICENSE("GPL"); |