]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
4cc96131 MS |
8 | #include "dm-core.h" |
9 | #include "dm-rq.h" | |
51e5b2bd | 10 | #include "dm-uevent.h" |
1da177e4 LT |
11 | |
12 | #include <linux/init.h> | |
13 | #include <linux/module.h> | |
48c9c27b | 14 | #include <linux/mutex.h> |
6958c1c6 | 15 | #include <linux/sched/mm.h> |
174cd4b1 | 16 | #include <linux/sched/signal.h> |
1da177e4 LT |
17 | #include <linux/blkpg.h> |
18 | #include <linux/bio.h> | |
1da177e4 | 19 | #include <linux/mempool.h> |
f26c5719 | 20 | #include <linux/dax.h> |
1da177e4 LT |
21 | #include <linux/slab.h> |
22 | #include <linux/idr.h> | |
7e026c8c | 23 | #include <linux/uio.h> |
3ac51e74 | 24 | #include <linux/hdreg.h> |
3f77316d | 25 | #include <linux/delay.h> |
ffcc3936 | 26 | #include <linux/wait.h> |
71cdb697 | 27 | #include <linux/pr.h> |
b0b4d7c6 | 28 | #include <linux/refcount.h> |
c6a564ff | 29 | #include <linux/part_stat.h> |
a892c8d5 | 30 | #include <linux/blk-crypto.h> |
55782138 | 31 | |
72d94861 AK |
32 | #define DM_MSG_PREFIX "core" |
33 | ||
60935eb2 MB |
34 | /* |
35 | * Cookies are numeric values sent with CHANGE and REMOVE | |
36 | * uevents while resuming, removing or renaming the device. | |
37 | */ | |
38 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
39 | #define DM_COOKIE_LENGTH 24 | |
40 | ||
1da177e4 LT |
41 | static const char *_name = DM_NAME; |
42 | ||
43 | static unsigned int major = 0; | |
44 | static unsigned int _major = 0; | |
45 | ||
d15b774c AK |
46 | static DEFINE_IDR(_minor_idr); |
47 | ||
f32c10b0 | 48 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
49 | |
50 | static void do_deferred_remove(struct work_struct *w); | |
51 | ||
52 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
53 | ||
acfe0ad7 MP |
54 | static struct workqueue_struct *deferred_remove_workqueue; |
55 | ||
93e6442c MP |
56 | atomic_t dm_global_event_nr = ATOMIC_INIT(0); |
57 | DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq); | |
58 | ||
62e08243 MP |
59 | void dm_issue_global_event(void) |
60 | { | |
61 | atomic_inc(&dm_global_event_nr); | |
62 | wake_up(&dm_global_eventq); | |
63 | } | |
64 | ||
1da177e4 | 65 | /* |
64f52b0e | 66 | * One of these is allocated (on-stack) per original bio. |
1da177e4 | 67 | */ |
64f52b0e | 68 | struct clone_info { |
64f52b0e MS |
69 | struct dm_table *map; |
70 | struct bio *bio; | |
71 | struct dm_io *io; | |
72 | sector_t sector; | |
73 | unsigned sector_count; | |
74 | }; | |
75 | ||
76 | /* | |
77 | * One of these is allocated per clone bio. | |
78 | */ | |
79 | #define DM_TIO_MAGIC 7282014 | |
80 | struct dm_target_io { | |
81 | unsigned magic; | |
82 | struct dm_io *io; | |
83 | struct dm_target *ti; | |
84 | unsigned target_bio_nr; | |
85 | unsigned *len_ptr; | |
86 | bool inside_dm_io; | |
87 | struct bio clone; | |
88 | }; | |
89 | ||
1da177e4 | 90 | /* |
745dc570 | 91 | * One of these is allocated per original bio. |
64f52b0e | 92 | * It contains the first clone used for that original. |
1da177e4 | 93 | */ |
64f52b0e | 94 | #define DM_IO_MAGIC 5191977 |
1da177e4 | 95 | struct dm_io { |
64f52b0e | 96 | unsigned magic; |
1da177e4 | 97 | struct mapped_device *md; |
4e4cbee9 | 98 | blk_status_t status; |
1da177e4 | 99 | atomic_t io_count; |
745dc570 | 100 | struct bio *orig_bio; |
3eaf840e | 101 | unsigned long start_time; |
f88fb981 | 102 | spinlock_t endio_lock; |
fd2ed4d2 | 103 | struct dm_stats_aux stats_aux; |
64f52b0e MS |
104 | /* last member of dm_target_io is 'struct bio' */ |
105 | struct dm_target_io tio; | |
1da177e4 LT |
106 | }; |
107 | ||
64f52b0e MS |
108 | void *dm_per_bio_data(struct bio *bio, size_t data_size) |
109 | { | |
110 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
111 | if (!tio->inside_dm_io) | |
112 | return (char *)bio - offsetof(struct dm_target_io, clone) - data_size; | |
113 | return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size; | |
114 | } | |
115 | EXPORT_SYMBOL_GPL(dm_per_bio_data); | |
116 | ||
117 | struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size) | |
118 | { | |
119 | struct dm_io *io = (struct dm_io *)((char *)data + data_size); | |
120 | if (io->magic == DM_IO_MAGIC) | |
121 | return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone)); | |
122 | BUG_ON(io->magic != DM_TIO_MAGIC); | |
123 | return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone)); | |
124 | } | |
125 | EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data); | |
126 | ||
127 | unsigned dm_bio_get_target_bio_nr(const struct bio *bio) | |
128 | { | |
129 | return container_of(bio, struct dm_target_io, clone)->target_bio_nr; | |
130 | } | |
131 | EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr); | |
132 | ||
ba61fdd1 JM |
133 | #define MINOR_ALLOCED ((void *)-1) |
134 | ||
1da177e4 LT |
135 | /* |
136 | * Bits for the md->flags field. | |
137 | */ | |
1eb787ec | 138 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 139 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 140 | #define DMF_FROZEN 2 |
fba9f90e | 141 | #define DMF_FREEING 3 |
5c6bd75d | 142 | #define DMF_DELETING 4 |
2e93ccc1 | 143 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
144 | #define DMF_DEFERRED_REMOVE 6 |
145 | #define DMF_SUSPENDED_INTERNALLY 7 | |
5df96f2b | 146 | #define DMF_POST_SUSPENDING 8 |
1da177e4 | 147 | |
115485e8 | 148 | #define DM_NUMA_NODE NUMA_NO_NODE |
115485e8 | 149 | static int dm_numa_node = DM_NUMA_NODE; |
faad87df | 150 | |
e6ee8c0b KU |
151 | /* |
152 | * For mempools pre-allocation at the table loading time. | |
153 | */ | |
154 | struct dm_md_mempools { | |
6f1c819c KO |
155 | struct bio_set bs; |
156 | struct bio_set io_bs; | |
e6ee8c0b KU |
157 | }; |
158 | ||
86f1152b BM |
159 | struct table_device { |
160 | struct list_head list; | |
b0b4d7c6 | 161 | refcount_t count; |
86f1152b BM |
162 | struct dm_dev dm_dev; |
163 | }; | |
164 | ||
e8603136 MS |
165 | /* |
166 | * Bio-based DM's mempools' reserved IOs set by the user. | |
167 | */ | |
4cc96131 | 168 | #define RESERVED_BIO_BASED_IOS 16 |
e8603136 MS |
169 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
170 | ||
115485e8 MS |
171 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
172 | { | |
6aa7de05 | 173 | int param = READ_ONCE(*module_param); |
115485e8 MS |
174 | int modified_param = 0; |
175 | bool modified = true; | |
176 | ||
177 | if (param < min) | |
178 | modified_param = min; | |
179 | else if (param > max) | |
180 | modified_param = max; | |
181 | else | |
182 | modified = false; | |
183 | ||
184 | if (modified) { | |
185 | (void)cmpxchg(module_param, param, modified_param); | |
186 | param = modified_param; | |
187 | } | |
188 | ||
189 | return param; | |
190 | } | |
191 | ||
4cc96131 MS |
192 | unsigned __dm_get_module_param(unsigned *module_param, |
193 | unsigned def, unsigned max) | |
f4790826 | 194 | { |
6aa7de05 | 195 | unsigned param = READ_ONCE(*module_param); |
09c2d531 | 196 | unsigned modified_param = 0; |
f4790826 | 197 | |
09c2d531 MS |
198 | if (!param) |
199 | modified_param = def; | |
200 | else if (param > max) | |
201 | modified_param = max; | |
f4790826 | 202 | |
09c2d531 MS |
203 | if (modified_param) { |
204 | (void)cmpxchg(module_param, param, modified_param); | |
205 | param = modified_param; | |
f4790826 MS |
206 | } |
207 | ||
09c2d531 | 208 | return param; |
f4790826 MS |
209 | } |
210 | ||
e8603136 MS |
211 | unsigned dm_get_reserved_bio_based_ios(void) |
212 | { | |
09c2d531 | 213 | return __dm_get_module_param(&reserved_bio_based_ios, |
4cc96131 | 214 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
e8603136 MS |
215 | } |
216 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
217 | ||
115485e8 MS |
218 | static unsigned dm_get_numa_node(void) |
219 | { | |
220 | return __dm_get_module_param_int(&dm_numa_node, | |
221 | DM_NUMA_NODE, num_online_nodes() - 1); | |
222 | } | |
223 | ||
1da177e4 LT |
224 | static int __init local_init(void) |
225 | { | |
e689fbab | 226 | int r; |
1ae49ea2 | 227 | |
51e5b2bd | 228 | r = dm_uevent_init(); |
51157b4a | 229 | if (r) |
e689fbab | 230 | return r; |
51e5b2bd | 231 | |
acfe0ad7 MP |
232 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
233 | if (!deferred_remove_workqueue) { | |
234 | r = -ENOMEM; | |
235 | goto out_uevent_exit; | |
236 | } | |
237 | ||
1da177e4 LT |
238 | _major = major; |
239 | r = register_blkdev(_major, _name); | |
51157b4a | 240 | if (r < 0) |
acfe0ad7 | 241 | goto out_free_workqueue; |
1da177e4 LT |
242 | |
243 | if (!_major) | |
244 | _major = r; | |
245 | ||
246 | return 0; | |
51157b4a | 247 | |
acfe0ad7 MP |
248 | out_free_workqueue: |
249 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
250 | out_uevent_exit: |
251 | dm_uevent_exit(); | |
51157b4a KU |
252 | |
253 | return r; | |
1da177e4 LT |
254 | } |
255 | ||
256 | static void local_exit(void) | |
257 | { | |
2c140a24 | 258 | flush_scheduled_work(); |
acfe0ad7 | 259 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 260 | |
00d59405 | 261 | unregister_blkdev(_major, _name); |
51e5b2bd | 262 | dm_uevent_exit(); |
1da177e4 LT |
263 | |
264 | _major = 0; | |
265 | ||
266 | DMINFO("cleaned up"); | |
267 | } | |
268 | ||
b9249e55 | 269 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
270 | local_init, |
271 | dm_target_init, | |
272 | dm_linear_init, | |
273 | dm_stripe_init, | |
952b3557 | 274 | dm_io_init, |
945fa4d2 | 275 | dm_kcopyd_init, |
1da177e4 | 276 | dm_interface_init, |
fd2ed4d2 | 277 | dm_statistics_init, |
1da177e4 LT |
278 | }; |
279 | ||
b9249e55 | 280 | static void (*_exits[])(void) = { |
1da177e4 LT |
281 | local_exit, |
282 | dm_target_exit, | |
283 | dm_linear_exit, | |
284 | dm_stripe_exit, | |
952b3557 | 285 | dm_io_exit, |
945fa4d2 | 286 | dm_kcopyd_exit, |
1da177e4 | 287 | dm_interface_exit, |
fd2ed4d2 | 288 | dm_statistics_exit, |
1da177e4 LT |
289 | }; |
290 | ||
291 | static int __init dm_init(void) | |
292 | { | |
293 | const int count = ARRAY_SIZE(_inits); | |
294 | ||
295 | int r, i; | |
296 | ||
297 | for (i = 0; i < count; i++) { | |
298 | r = _inits[i](); | |
299 | if (r) | |
300 | goto bad; | |
301 | } | |
302 | ||
303 | return 0; | |
304 | ||
305 | bad: | |
306 | while (i--) | |
307 | _exits[i](); | |
308 | ||
309 | return r; | |
310 | } | |
311 | ||
312 | static void __exit dm_exit(void) | |
313 | { | |
314 | int i = ARRAY_SIZE(_exits); | |
315 | ||
316 | while (i--) | |
317 | _exits[i](); | |
d15b774c AK |
318 | |
319 | /* | |
320 | * Should be empty by this point. | |
321 | */ | |
d15b774c | 322 | idr_destroy(&_minor_idr); |
1da177e4 LT |
323 | } |
324 | ||
325 | /* | |
326 | * Block device functions | |
327 | */ | |
432a212c MA |
328 | int dm_deleting_md(struct mapped_device *md) |
329 | { | |
330 | return test_bit(DMF_DELETING, &md->flags); | |
331 | } | |
332 | ||
fe5f9f2c | 333 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
334 | { |
335 | struct mapped_device *md; | |
336 | ||
fba9f90e JM |
337 | spin_lock(&_minor_lock); |
338 | ||
fe5f9f2c | 339 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
340 | if (!md) |
341 | goto out; | |
342 | ||
5c6bd75d | 343 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 344 | dm_deleting_md(md)) { |
fba9f90e JM |
345 | md = NULL; |
346 | goto out; | |
347 | } | |
348 | ||
1da177e4 | 349 | dm_get(md); |
5c6bd75d | 350 | atomic_inc(&md->open_count); |
fba9f90e JM |
351 | out: |
352 | spin_unlock(&_minor_lock); | |
353 | ||
354 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
355 | } |
356 | ||
db2a144b | 357 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 358 | { |
63a4f065 | 359 | struct mapped_device *md; |
6e9624b8 | 360 | |
4a1aeb98 MB |
361 | spin_lock(&_minor_lock); |
362 | ||
63a4f065 MS |
363 | md = disk->private_data; |
364 | if (WARN_ON(!md)) | |
365 | goto out; | |
366 | ||
2c140a24 MP |
367 | if (atomic_dec_and_test(&md->open_count) && |
368 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 369 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 370 | |
1da177e4 | 371 | dm_put(md); |
63a4f065 | 372 | out: |
4a1aeb98 | 373 | spin_unlock(&_minor_lock); |
1da177e4 LT |
374 | } |
375 | ||
5c6bd75d AK |
376 | int dm_open_count(struct mapped_device *md) |
377 | { | |
378 | return atomic_read(&md->open_count); | |
379 | } | |
380 | ||
381 | /* | |
382 | * Guarantees nothing is using the device before it's deleted. | |
383 | */ | |
2c140a24 | 384 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
385 | { |
386 | int r = 0; | |
387 | ||
388 | spin_lock(&_minor_lock); | |
389 | ||
2c140a24 | 390 | if (dm_open_count(md)) { |
5c6bd75d | 391 | r = -EBUSY; |
2c140a24 MP |
392 | if (mark_deferred) |
393 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
394 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
395 | r = -EEXIST; | |
5c6bd75d AK |
396 | else |
397 | set_bit(DMF_DELETING, &md->flags); | |
398 | ||
399 | spin_unlock(&_minor_lock); | |
400 | ||
401 | return r; | |
402 | } | |
403 | ||
2c140a24 MP |
404 | int dm_cancel_deferred_remove(struct mapped_device *md) |
405 | { | |
406 | int r = 0; | |
407 | ||
408 | spin_lock(&_minor_lock); | |
409 | ||
410 | if (test_bit(DMF_DELETING, &md->flags)) | |
411 | r = -EBUSY; | |
412 | else | |
413 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
414 | ||
415 | spin_unlock(&_minor_lock); | |
416 | ||
417 | return r; | |
418 | } | |
419 | ||
420 | static void do_deferred_remove(struct work_struct *w) | |
421 | { | |
422 | dm_deferred_remove(); | |
423 | } | |
424 | ||
fd2ed4d2 MP |
425 | sector_t dm_get_size(struct mapped_device *md) |
426 | { | |
427 | return get_capacity(md->disk); | |
428 | } | |
429 | ||
9974fa2c MS |
430 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
431 | { | |
432 | return md->queue; | |
433 | } | |
434 | ||
fd2ed4d2 MP |
435 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
436 | { | |
437 | return &md->stats; | |
438 | } | |
439 | ||
3ac51e74 DW |
440 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
441 | { | |
442 | struct mapped_device *md = bdev->bd_disk->private_data; | |
443 | ||
444 | return dm_get_geometry(md, geo); | |
445 | } | |
446 | ||
d4100351 CH |
447 | #ifdef CONFIG_BLK_DEV_ZONED |
448 | int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data) | |
449 | { | |
450 | struct dm_report_zones_args *args = data; | |
451 | sector_t sector_diff = args->tgt->begin - args->start; | |
452 | ||
453 | /* | |
454 | * Ignore zones beyond the target range. | |
455 | */ | |
456 | if (zone->start >= args->start + args->tgt->len) | |
457 | return 0; | |
458 | ||
459 | /* | |
460 | * Remap the start sector and write pointer position of the zone | |
461 | * to match its position in the target range. | |
462 | */ | |
463 | zone->start += sector_diff; | |
464 | if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { | |
465 | if (zone->cond == BLK_ZONE_COND_FULL) | |
466 | zone->wp = zone->start + zone->len; | |
467 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
468 | zone->wp = zone->start; | |
469 | else | |
470 | zone->wp += sector_diff; | |
471 | } | |
472 | ||
473 | args->next_sector = zone->start + zone->len; | |
474 | return args->orig_cb(zone, args->zone_idx++, args->orig_data); | |
475 | } | |
476 | EXPORT_SYMBOL_GPL(dm_report_zones_cb); | |
477 | ||
e76239a3 | 478 | static int dm_blk_report_zones(struct gendisk *disk, sector_t sector, |
d4100351 | 479 | unsigned int nr_zones, report_zones_cb cb, void *data) |
e76239a3 | 480 | { |
e76239a3 | 481 | struct mapped_device *md = disk->private_data; |
e76239a3 CH |
482 | struct dm_table *map; |
483 | int srcu_idx, ret; | |
d4100351 CH |
484 | struct dm_report_zones_args args = { |
485 | .next_sector = sector, | |
486 | .orig_data = data, | |
487 | .orig_cb = cb, | |
488 | }; | |
e76239a3 CH |
489 | |
490 | if (dm_suspended_md(md)) | |
491 | return -EAGAIN; | |
492 | ||
493 | map = dm_get_live_table(md, &srcu_idx); | |
494 | if (!map) | |
495 | return -EIO; | |
496 | ||
d4100351 CH |
497 | do { |
498 | struct dm_target *tgt; | |
e76239a3 | 499 | |
d4100351 CH |
500 | tgt = dm_table_find_target(map, args.next_sector); |
501 | if (WARN_ON_ONCE(!tgt->type->report_zones)) { | |
502 | ret = -EIO; | |
503 | goto out; | |
504 | } | |
e76239a3 | 505 | |
d4100351 | 506 | args.tgt = tgt; |
a9cb9f41 JT |
507 | ret = tgt->type->report_zones(tgt, &args, |
508 | nr_zones - args.zone_idx); | |
d4100351 CH |
509 | if (ret < 0) |
510 | goto out; | |
511 | } while (args.zone_idx < nr_zones && | |
512 | args.next_sector < get_capacity(disk)); | |
e76239a3 | 513 | |
d4100351 | 514 | ret = args.zone_idx; |
e76239a3 CH |
515 | out: |
516 | dm_put_live_table(md, srcu_idx); | |
517 | return ret; | |
e76239a3 | 518 | } |
d4100351 CH |
519 | #else |
520 | #define dm_blk_report_zones NULL | |
521 | #endif /* CONFIG_BLK_DEV_ZONED */ | |
e76239a3 | 522 | |
971888c4 | 523 | static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, |
5bd5e8d8 | 524 | struct block_device **bdev) |
971888c4 | 525 | __acquires(md->io_barrier) |
aa129a22 | 526 | { |
66482026 | 527 | struct dm_target *tgt; |
6c182cd8 | 528 | struct dm_table *map; |
971888c4 | 529 | int r; |
aa129a22 | 530 | |
6c182cd8 | 531 | retry: |
e56f81e0 | 532 | r = -ENOTTY; |
971888c4 | 533 | map = dm_get_live_table(md, srcu_idx); |
aa129a22 | 534 | if (!map || !dm_table_get_size(map)) |
971888c4 | 535 | return r; |
aa129a22 MB |
536 | |
537 | /* We only support devices that have a single target */ | |
538 | if (dm_table_get_num_targets(map) != 1) | |
971888c4 | 539 | return r; |
aa129a22 | 540 | |
66482026 MS |
541 | tgt = dm_table_get_target(map, 0); |
542 | if (!tgt->type->prepare_ioctl) | |
971888c4 | 543 | return r; |
519049af | 544 | |
971888c4 MS |
545 | if (dm_suspended_md(md)) |
546 | return -EAGAIN; | |
aa129a22 | 547 | |
5bd5e8d8 | 548 | r = tgt->type->prepare_ioctl(tgt, bdev); |
5bbbfdf6 | 549 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
971888c4 | 550 | dm_put_live_table(md, *srcu_idx); |
6c182cd8 HR |
551 | msleep(10); |
552 | goto retry; | |
553 | } | |
971888c4 | 554 | |
e56f81e0 CH |
555 | return r; |
556 | } | |
557 | ||
971888c4 MS |
558 | static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) |
559 | __releases(md->io_barrier) | |
560 | { | |
561 | dm_put_live_table(md, srcu_idx); | |
562 | } | |
563 | ||
e56f81e0 CH |
564 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
565 | unsigned int cmd, unsigned long arg) | |
566 | { | |
567 | struct mapped_device *md = bdev->bd_disk->private_data; | |
971888c4 | 568 | int r, srcu_idx; |
e56f81e0 | 569 | |
5bd5e8d8 | 570 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
e56f81e0 | 571 | if (r < 0) |
971888c4 | 572 | goto out; |
6c182cd8 | 573 | |
e56f81e0 CH |
574 | if (r > 0) { |
575 | /* | |
e980f623 CH |
576 | * Target determined this ioctl is being issued against a |
577 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 578 | */ |
e980f623 CH |
579 | if (!capable(CAP_SYS_RAWIO)) { |
580 | DMWARN_LIMIT( | |
581 | "%s: sending ioctl %x to DM device without required privilege.", | |
582 | current->comm, cmd); | |
583 | r = -ENOIOCTLCMD; | |
e56f81e0 | 584 | goto out; |
e980f623 | 585 | } |
e56f81e0 | 586 | } |
6c182cd8 | 587 | |
66482026 | 588 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 589 | out: |
971888c4 | 590 | dm_unprepare_ioctl(md, srcu_idx); |
aa129a22 MB |
591 | return r; |
592 | } | |
593 | ||
978e51ba MS |
594 | static void start_io_acct(struct dm_io *io); |
595 | ||
596 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) | |
1da177e4 | 597 | { |
64f52b0e MS |
598 | struct dm_io *io; |
599 | struct dm_target_io *tio; | |
600 | struct bio *clone; | |
601 | ||
6f1c819c | 602 | clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs); |
64f52b0e MS |
603 | if (!clone) |
604 | return NULL; | |
605 | ||
606 | tio = container_of(clone, struct dm_target_io, clone); | |
607 | tio->inside_dm_io = true; | |
608 | tio->io = NULL; | |
609 | ||
610 | io = container_of(tio, struct dm_io, tio); | |
611 | io->magic = DM_IO_MAGIC; | |
978e51ba MS |
612 | io->status = 0; |
613 | atomic_set(&io->io_count, 1); | |
614 | io->orig_bio = bio; | |
615 | io->md = md; | |
616 | spin_lock_init(&io->endio_lock); | |
617 | ||
618 | start_io_acct(io); | |
64f52b0e MS |
619 | |
620 | return io; | |
1da177e4 LT |
621 | } |
622 | ||
028867ac | 623 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 | 624 | { |
64f52b0e MS |
625 | bio_put(&io->tio.clone); |
626 | } | |
627 | ||
628 | static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, | |
629 | unsigned target_bio_nr, gfp_t gfp_mask) | |
630 | { | |
631 | struct dm_target_io *tio; | |
632 | ||
633 | if (!ci->io->tio.io) { | |
634 | /* the dm_target_io embedded in ci->io is available */ | |
635 | tio = &ci->io->tio; | |
636 | } else { | |
6f1c819c | 637 | struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs); |
64f52b0e MS |
638 | if (!clone) |
639 | return NULL; | |
640 | ||
641 | tio = container_of(clone, struct dm_target_io, clone); | |
642 | tio->inside_dm_io = false; | |
643 | } | |
644 | ||
645 | tio->magic = DM_TIO_MAGIC; | |
646 | tio->io = ci->io; | |
647 | tio->ti = ti; | |
648 | tio->target_bio_nr = target_bio_nr; | |
649 | ||
650 | return tio; | |
1da177e4 LT |
651 | } |
652 | ||
cfae7529 | 653 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 654 | { |
64f52b0e MS |
655 | if (tio->inside_dm_io) |
656 | return; | |
dba14160 | 657 | bio_put(&tio->clone); |
1da177e4 LT |
658 | } |
659 | ||
087615bf GKB |
660 | u64 dm_start_time_ns_from_clone(struct bio *bio) |
661 | { | |
662 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
663 | struct dm_io *io = tio->io; | |
664 | ||
665 | return jiffies_to_nsecs(io->start_time); | |
666 | } | |
667 | EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone); | |
668 | ||
3eaf840e JNN |
669 | static void start_io_acct(struct dm_io *io) |
670 | { | |
671 | struct mapped_device *md = io->md; | |
745dc570 | 672 | struct bio *bio = io->orig_bio; |
3eaf840e | 673 | |
86240d5b | 674 | io->start_time = bio_start_io_acct(bio); |
fd2ed4d2 | 675 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
676 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
677 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
678 | false, 0, &io->stats_aux); | |
3eaf840e JNN |
679 | } |
680 | ||
d221d2e7 | 681 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
682 | { |
683 | struct mapped_device *md = io->md; | |
745dc570 | 684 | struct bio *bio = io->orig_bio; |
3eaf840e | 685 | unsigned long duration = jiffies - io->start_time; |
3eaf840e | 686 | |
86240d5b | 687 | bio_end_io_acct(bio, io->start_time); |
3eaf840e | 688 | |
fd2ed4d2 | 689 | if (unlikely(dm_stats_used(&md->stats))) |
528ec5ab MC |
690 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
691 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
692 | true, duration, &io->stats_aux); | |
fd2ed4d2 | 693 | |
d221d2e7 | 694 | /* nudge anyone waiting on suspend queue */ |
645efa84 | 695 | if (unlikely(wq_has_sleeper(&md->wait))) |
d221d2e7 | 696 | wake_up(&md->wait); |
3eaf840e JNN |
697 | } |
698 | ||
1da177e4 LT |
699 | /* |
700 | * Add the bio to the list of deferred io. | |
701 | */ | |
92c63902 | 702 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 703 | { |
05447420 | 704 | unsigned long flags; |
1da177e4 | 705 | |
05447420 | 706 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 707 | bio_list_add(&md->deferred, bio); |
05447420 | 708 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 709 | queue_work(md->wq, &md->work); |
1da177e4 LT |
710 | } |
711 | ||
712 | /* | |
713 | * Everyone (including functions in this file), should use this | |
714 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 715 | * dm_put_live_table() when finished. |
1da177e4 | 716 | */ |
83d5e5b0 | 717 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 718 | { |
83d5e5b0 MP |
719 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
720 | ||
721 | return srcu_dereference(md->map, &md->io_barrier); | |
722 | } | |
1da177e4 | 723 | |
83d5e5b0 MP |
724 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
725 | { | |
726 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
727 | } | |
728 | ||
729 | void dm_sync_table(struct mapped_device *md) | |
730 | { | |
731 | synchronize_srcu(&md->io_barrier); | |
732 | synchronize_rcu_expedited(); | |
733 | } | |
734 | ||
735 | /* | |
736 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
737 | * The caller must not block between these two functions. | |
738 | */ | |
739 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
740 | { | |
741 | rcu_read_lock(); | |
742 | return rcu_dereference(md->map); | |
743 | } | |
1da177e4 | 744 | |
83d5e5b0 MP |
745 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
746 | { | |
747 | rcu_read_unlock(); | |
1da177e4 LT |
748 | } |
749 | ||
971888c4 MS |
750 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
751 | ||
86f1152b BM |
752 | /* |
753 | * Open a table device so we can use it as a map destination. | |
754 | */ | |
755 | static int open_table_device(struct table_device *td, dev_t dev, | |
756 | struct mapped_device *md) | |
757 | { | |
86f1152b BM |
758 | struct block_device *bdev; |
759 | ||
760 | int r; | |
761 | ||
762 | BUG_ON(td->dm_dev.bdev); | |
763 | ||
519049af | 764 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr); |
86f1152b BM |
765 | if (IS_ERR(bdev)) |
766 | return PTR_ERR(bdev); | |
767 | ||
768 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
769 | if (r) { | |
770 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
771 | return r; | |
772 | } | |
773 | ||
774 | td->dm_dev.bdev = bdev; | |
817bf402 | 775 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
776 | return 0; |
777 | } | |
778 | ||
779 | /* | |
780 | * Close a table device that we've been using. | |
781 | */ | |
782 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
783 | { | |
784 | if (!td->dm_dev.bdev) | |
785 | return; | |
786 | ||
787 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
788 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 789 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 790 | td->dm_dev.bdev = NULL; |
817bf402 | 791 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
792 | } |
793 | ||
794 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
8454fca4 SS |
795 | fmode_t mode) |
796 | { | |
86f1152b BM |
797 | struct table_device *td; |
798 | ||
799 | list_for_each_entry(td, l, list) | |
800 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
801 | return td; | |
802 | ||
803 | return NULL; | |
804 | } | |
805 | ||
806 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
8454fca4 SS |
807 | struct dm_dev **result) |
808 | { | |
86f1152b BM |
809 | int r; |
810 | struct table_device *td; | |
811 | ||
812 | mutex_lock(&md->table_devices_lock); | |
813 | td = find_table_device(&md->table_devices, dev, mode); | |
814 | if (!td) { | |
115485e8 | 815 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
816 | if (!td) { |
817 | mutex_unlock(&md->table_devices_lock); | |
818 | return -ENOMEM; | |
819 | } | |
820 | ||
821 | td->dm_dev.mode = mode; | |
822 | td->dm_dev.bdev = NULL; | |
823 | ||
824 | if ((r = open_table_device(td, dev, md))) { | |
825 | mutex_unlock(&md->table_devices_lock); | |
826 | kfree(td); | |
827 | return r; | |
828 | } | |
829 | ||
830 | format_dev_t(td->dm_dev.name, dev); | |
831 | ||
b0b4d7c6 | 832 | refcount_set(&td->count, 1); |
86f1152b | 833 | list_add(&td->list, &md->table_devices); |
b0b4d7c6 ER |
834 | } else { |
835 | refcount_inc(&td->count); | |
86f1152b | 836 | } |
86f1152b BM |
837 | mutex_unlock(&md->table_devices_lock); |
838 | ||
839 | *result = &td->dm_dev; | |
840 | return 0; | |
841 | } | |
842 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
843 | ||
844 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
845 | { | |
846 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
847 | ||
848 | mutex_lock(&md->table_devices_lock); | |
b0b4d7c6 | 849 | if (refcount_dec_and_test(&td->count)) { |
86f1152b BM |
850 | close_table_device(td, md); |
851 | list_del(&td->list); | |
852 | kfree(td); | |
853 | } | |
854 | mutex_unlock(&md->table_devices_lock); | |
855 | } | |
856 | EXPORT_SYMBOL(dm_put_table_device); | |
857 | ||
858 | static void free_table_devices(struct list_head *devices) | |
859 | { | |
860 | struct list_head *tmp, *next; | |
861 | ||
862 | list_for_each_safe(tmp, next, devices) { | |
863 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
864 | ||
865 | DMWARN("dm_destroy: %s still exists with %d references", | |
b0b4d7c6 | 866 | td->dm_dev.name, refcount_read(&td->count)); |
86f1152b BM |
867 | kfree(td); |
868 | } | |
869 | } | |
870 | ||
3ac51e74 DW |
871 | /* |
872 | * Get the geometry associated with a dm device | |
873 | */ | |
874 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
875 | { | |
876 | *geo = md->geometry; | |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
881 | /* | |
882 | * Set the geometry of a device. | |
883 | */ | |
884 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
885 | { | |
886 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
887 | ||
888 | if (geo->start > sz) { | |
889 | DMWARN("Start sector is beyond the geometry limits."); | |
890 | return -EINVAL; | |
891 | } | |
892 | ||
893 | md->geometry = *geo; | |
894 | ||
895 | return 0; | |
896 | } | |
897 | ||
2e93ccc1 KU |
898 | static int __noflush_suspending(struct mapped_device *md) |
899 | { | |
900 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
901 | } | |
902 | ||
1da177e4 LT |
903 | /* |
904 | * Decrements the number of outstanding ios that a bio has been | |
905 | * cloned into, completing the original io if necc. | |
906 | */ | |
4e4cbee9 | 907 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 908 | { |
2e93ccc1 | 909 | unsigned long flags; |
4e4cbee9 | 910 | blk_status_t io_error; |
b35f8caa MB |
911 | struct bio *bio; |
912 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
913 | |
914 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
915 | if (unlikely(error)) { |
916 | spin_lock_irqsave(&io->endio_lock, flags); | |
745dc570 | 917 | if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) |
4e4cbee9 | 918 | io->status = error; |
f88fb981 KU |
919 | spin_unlock_irqrestore(&io->endio_lock, flags); |
920 | } | |
1da177e4 LT |
921 | |
922 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 923 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
924 | /* |
925 | * Target requested pushing back the I/O. | |
2e93ccc1 | 926 | */ |
022c2611 | 927 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 | 928 | if (__noflush_suspending(md)) |
745dc570 MS |
929 | /* NOTE early return due to BLK_STS_DM_REQUEUE below */ |
930 | bio_list_add_head(&md->deferred, io->orig_bio); | |
6a8736d1 | 931 | else |
2e93ccc1 | 932 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 933 | io->status = BLK_STS_IOERR; |
022c2611 | 934 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
935 | } |
936 | ||
4e4cbee9 | 937 | io_error = io->status; |
745dc570 | 938 | bio = io->orig_bio; |
6a8736d1 TH |
939 | end_io_acct(io); |
940 | free_io(md, io); | |
941 | ||
4e4cbee9 | 942 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 943 | return; |
2e93ccc1 | 944 | |
1eff9d32 | 945 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 946 | /* |
6a8736d1 | 947 | * Preflush done for flush with data, reissue |
28a8f0d3 | 948 | * without REQ_PREFLUSH. |
af7e466a | 949 | */ |
1eff9d32 | 950 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 951 | queue_io(md, bio); |
af7e466a | 952 | } else { |
b372d360 | 953 | /* done with normal IO or empty flush */ |
8dd601fa N |
954 | if (io_error) |
955 | bio->bi_status = io_error; | |
4246a0b6 | 956 | bio_endio(bio); |
b35f8caa | 957 | } |
1da177e4 LT |
958 | } |
959 | } | |
960 | ||
bcb44433 MS |
961 | void disable_discard(struct mapped_device *md) |
962 | { | |
963 | struct queue_limits *limits = dm_get_queue_limits(md); | |
964 | ||
965 | /* device doesn't really support DISCARD, disable it */ | |
966 | limits->max_discard_sectors = 0; | |
967 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue); | |
968 | } | |
969 | ||
4cc96131 | 970 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
971 | { |
972 | struct queue_limits *limits = dm_get_queue_limits(md); | |
973 | ||
974 | /* device doesn't really support WRITE SAME, disable it */ | |
975 | limits->max_write_same_sectors = 0; | |
976 | } | |
977 | ||
ac62d620 CH |
978 | void disable_write_zeroes(struct mapped_device *md) |
979 | { | |
980 | struct queue_limits *limits = dm_get_queue_limits(md); | |
981 | ||
982 | /* device doesn't really support WRITE ZEROES, disable it */ | |
983 | limits->max_write_zeroes_sectors = 0; | |
984 | } | |
985 | ||
4246a0b6 | 986 | static void clone_endio(struct bio *bio) |
1da177e4 | 987 | { |
4e4cbee9 | 988 | blk_status_t error = bio->bi_status; |
bfc6d41c | 989 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 990 | struct dm_io *io = tio->io; |
9faf400f | 991 | struct mapped_device *md = tio->io->md; |
1da177e4 | 992 | dm_endio_fn endio = tio->ti->type->end_io; |
415c79e1 | 993 | struct bio *orig_bio = io->orig_bio; |
1da177e4 | 994 | |
978e51ba | 995 | if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) { |
bcb44433 MS |
996 | if (bio_op(bio) == REQ_OP_DISCARD && |
997 | !bio->bi_disk->queue->limits.max_discard_sectors) | |
998 | disable_discard(md); | |
999 | else if (bio_op(bio) == REQ_OP_WRITE_SAME && | |
1000 | !bio->bi_disk->queue->limits.max_write_same_sectors) | |
ac62d620 | 1001 | disable_write_same(md); |
bcb44433 MS |
1002 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES && |
1003 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) | |
ac62d620 CH |
1004 | disable_write_zeroes(md); |
1005 | } | |
7eee4ae2 | 1006 | |
415c79e1 JT |
1007 | /* |
1008 | * For zone-append bios get offset in zone of the written | |
1009 | * sector and add that to the original bio sector pos. | |
1010 | */ | |
1011 | if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) { | |
1012 | sector_t written_sector = bio->bi_iter.bi_sector; | |
1013 | struct request_queue *q = orig_bio->bi_disk->queue; | |
1014 | u64 mask = (u64)blk_queue_zone_sectors(q) - 1; | |
1015 | ||
1016 | orig_bio->bi_iter.bi_sector += written_sector & mask; | |
1017 | } | |
1018 | ||
1be56909 | 1019 | if (endio) { |
4e4cbee9 | 1020 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
1021 | switch (r) { |
1022 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 1023 | error = BLK_STS_DM_REQUEUE; |
df561f66 | 1024 | fallthrough; |
1be56909 CH |
1025 | case DM_ENDIO_DONE: |
1026 | break; | |
1027 | case DM_ENDIO_INCOMPLETE: | |
1028 | /* The target will handle the io */ | |
1029 | return; | |
1030 | default: | |
1031 | DMWARN("unimplemented target endio return value: %d", r); | |
1032 | BUG(); | |
1033 | } | |
1034 | } | |
1035 | ||
cfae7529 | 1036 | free_tio(tio); |
b35f8caa | 1037 | dec_pending(io, error); |
1da177e4 LT |
1038 | } |
1039 | ||
56a67df7 MS |
1040 | /* |
1041 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1042 | * target boundary. | |
1043 | */ | |
1044 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1045 | { | |
1046 | sector_t target_offset = dm_target_offset(ti, sector); | |
1047 | ||
1048 | return ti->len - target_offset; | |
1049 | } | |
1050 | ||
1051 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1052 | { |
56a67df7 | 1053 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1054 | sector_t offset, max_len; |
1da177e4 LT |
1055 | |
1056 | /* | |
542f9038 | 1057 | * Does the target need to split even further? |
1da177e4 | 1058 | */ |
542f9038 MS |
1059 | if (ti->max_io_len) { |
1060 | offset = dm_target_offset(ti, sector); | |
1061 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1062 | max_len = sector_div(offset, ti->max_io_len); | |
1063 | else | |
1064 | max_len = offset & (ti->max_io_len - 1); | |
1065 | max_len = ti->max_io_len - max_len; | |
1066 | ||
1067 | if (len > max_len) | |
1068 | len = max_len; | |
1da177e4 LT |
1069 | } |
1070 | ||
1071 | return len; | |
1072 | } | |
1073 | ||
542f9038 MS |
1074 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1075 | { | |
1076 | if (len > UINT_MAX) { | |
1077 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1078 | (unsigned long long)len, UINT_MAX); | |
1079 | ti->error = "Maximum size of target IO is too large"; | |
1080 | return -EINVAL; | |
1081 | } | |
1082 | ||
75ae1936 | 1083 | ti->max_io_len = (uint32_t) len; |
542f9038 MS |
1084 | |
1085 | return 0; | |
1086 | } | |
1087 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1088 | ||
f26c5719 | 1089 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
3d97c829 MS |
1090 | sector_t sector, int *srcu_idx) |
1091 | __acquires(md->io_barrier) | |
545ed20e | 1092 | { |
545ed20e TK |
1093 | struct dm_table *map; |
1094 | struct dm_target *ti; | |
545ed20e | 1095 | |
f26c5719 | 1096 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1097 | if (!map) |
f26c5719 | 1098 | return NULL; |
545ed20e TK |
1099 | |
1100 | ti = dm_table_find_target(map, sector); | |
123d87d5 | 1101 | if (!ti) |
f26c5719 | 1102 | return NULL; |
545ed20e | 1103 | |
f26c5719 DW |
1104 | return ti; |
1105 | } | |
545ed20e | 1106 | |
f26c5719 | 1107 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1108 | long nr_pages, void **kaddr, pfn_t *pfn) |
f26c5719 DW |
1109 | { |
1110 | struct mapped_device *md = dax_get_private(dax_dev); | |
1111 | sector_t sector = pgoff * PAGE_SECTORS; | |
1112 | struct dm_target *ti; | |
1113 | long len, ret = -EIO; | |
1114 | int srcu_idx; | |
545ed20e | 1115 | |
f26c5719 | 1116 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1117 | |
f26c5719 DW |
1118 | if (!ti) |
1119 | goto out; | |
1120 | if (!ti->type->direct_access) | |
1121 | goto out; | |
1122 | len = max_io_len(sector, ti) / PAGE_SECTORS; | |
1123 | if (len < 1) | |
1124 | goto out; | |
1125 | nr_pages = min(len, nr_pages); | |
dbc62659 | 1126 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
817bf402 | 1127 | |
f26c5719 | 1128 | out: |
545ed20e | 1129 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1130 | |
1131 | return ret; | |
545ed20e TK |
1132 | } |
1133 | ||
7bf7eac8 DW |
1134 | static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev, |
1135 | int blocksize, sector_t start, sector_t len) | |
1136 | { | |
1137 | struct mapped_device *md = dax_get_private(dax_dev); | |
1138 | struct dm_table *map; | |
1139 | int srcu_idx; | |
1140 | bool ret; | |
1141 | ||
1142 | map = dm_get_live_table(md, &srcu_idx); | |
1143 | if (!map) | |
1144 | return false; | |
1145 | ||
2e9ee095 | 1146 | ret = dm_table_supports_dax(map, device_supports_dax, &blocksize); |
7bf7eac8 DW |
1147 | |
1148 | dm_put_live_table(md, srcu_idx); | |
1149 | ||
1150 | return ret; | |
1151 | } | |
1152 | ||
7e026c8c | 1153 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1154 | void *addr, size_t bytes, struct iov_iter *i) |
7e026c8c DW |
1155 | { |
1156 | struct mapped_device *md = dax_get_private(dax_dev); | |
1157 | sector_t sector = pgoff * PAGE_SECTORS; | |
1158 | struct dm_target *ti; | |
1159 | long ret = 0; | |
1160 | int srcu_idx; | |
1161 | ||
1162 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1163 | ||
1164 | if (!ti) | |
1165 | goto out; | |
1166 | if (!ti->type->dax_copy_from_iter) { | |
1167 | ret = copy_from_iter(addr, bytes, i); | |
1168 | goto out; | |
1169 | } | |
1170 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
1171 | out: | |
1172 | dm_put_live_table(md, srcu_idx); | |
1173 | ||
1174 | return ret; | |
1175 | } | |
1176 | ||
b3a9a0c3 DW |
1177 | static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
1178 | void *addr, size_t bytes, struct iov_iter *i) | |
1179 | { | |
1180 | struct mapped_device *md = dax_get_private(dax_dev); | |
1181 | sector_t sector = pgoff * PAGE_SECTORS; | |
1182 | struct dm_target *ti; | |
1183 | long ret = 0; | |
1184 | int srcu_idx; | |
1185 | ||
1186 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1187 | ||
1188 | if (!ti) | |
1189 | goto out; | |
1190 | if (!ti->type->dax_copy_to_iter) { | |
1191 | ret = copy_to_iter(addr, bytes, i); | |
1192 | goto out; | |
1193 | } | |
1194 | ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i); | |
1195 | out: | |
1196 | dm_put_live_table(md, srcu_idx); | |
1197 | ||
1198 | return ret; | |
1199 | } | |
1200 | ||
cdf6cdcd VG |
1201 | static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, |
1202 | size_t nr_pages) | |
1203 | { | |
1204 | struct mapped_device *md = dax_get_private(dax_dev); | |
1205 | sector_t sector = pgoff * PAGE_SECTORS; | |
1206 | struct dm_target *ti; | |
1207 | int ret = -EIO; | |
1208 | int srcu_idx; | |
1209 | ||
1210 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1211 | ||
1212 | if (!ti) | |
1213 | goto out; | |
1214 | if (WARN_ON(!ti->type->dax_zero_page_range)) { | |
1215 | /* | |
1216 | * ->zero_page_range() is mandatory dax operation. If we are | |
1217 | * here, something is wrong. | |
1218 | */ | |
1219 | dm_put_live_table(md, srcu_idx); | |
1220 | goto out; | |
1221 | } | |
1222 | ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages); | |
1223 | ||
1224 | out: | |
1225 | dm_put_live_table(md, srcu_idx); | |
1226 | ||
1227 | return ret; | |
1228 | } | |
1229 | ||
1dd40c3e MP |
1230 | /* |
1231 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
2e2d6f7e AJ |
1232 | * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET, |
1233 | * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH. | |
1dd40c3e MP |
1234 | * |
1235 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1236 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1237 | * sent in a next bio. | |
1238 | * | |
1239 | * A diagram that explains the arithmetics: | |
1240 | * +--------------------+---------------+-------+ | |
1241 | * | 1 | 2 | 3 | | |
1242 | * +--------------------+---------------+-------+ | |
1243 | * | |
1244 | * <-------------- *tio->len_ptr ---------------> | |
1245 | * <------- bi_size -------> | |
1246 | * <-- n_sectors --> | |
1247 | * | |
1248 | * Region 1 was already iterated over with bio_advance or similar function. | |
1249 | * (it may be empty if the target doesn't use bio_advance) | |
1250 | * Region 2 is the remaining bio size that the target wants to process. | |
1251 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1252 | * to make it empty) | |
1253 | * The target requires that region 3 is to be sent in the next bio. | |
1254 | * | |
1255 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1256 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1257 | * copies of the bio. | |
1258 | */ | |
1259 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1260 | { | |
1261 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1262 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1263 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1264 | BUG_ON(bi_size > *tio->len_ptr); |
1265 | BUG_ON(n_sectors > bi_size); | |
1266 | *tio->len_ptr -= bi_size - n_sectors; | |
1267 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1268 | } | |
1269 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1270 | ||
978e51ba | 1271 | static blk_qc_t __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1272 | { |
1273 | int r; | |
2056a782 | 1274 | sector_t sector; |
dba14160 | 1275 | struct bio *clone = &tio->clone; |
64f52b0e | 1276 | struct dm_io *io = tio->io; |
bd2a49b8 | 1277 | struct dm_target *ti = tio->ti; |
978e51ba | 1278 | blk_qc_t ret = BLK_QC_T_NONE; |
1da177e4 | 1279 | |
1da177e4 | 1280 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1281 | |
1282 | /* | |
1283 | * Map the clone. If r == 0 we don't need to do | |
1284 | * anything, the target has assumed ownership of | |
1285 | * this io. | |
1286 | */ | |
64f52b0e | 1287 | atomic_inc(&io->io_count); |
4f024f37 | 1288 | sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1289 | |
7de3ee57 | 1290 | r = ti->type->map(ti, clone); |
846785e6 CH |
1291 | switch (r) { |
1292 | case DM_MAPIO_SUBMITTED: | |
1293 | break; | |
1294 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1295 | /* the bio has been remapped so dispatch it */ |
74d46992 | 1296 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
64f52b0e | 1297 | bio_dev(io->orig_bio), sector); |
5a6c35f9 | 1298 | ret = submit_bio_noacct(clone); |
846785e6 CH |
1299 | break; |
1300 | case DM_MAPIO_KILL: | |
4e4cbee9 | 1301 | free_tio(tio); |
64f52b0e | 1302 | dec_pending(io, BLK_STS_IOERR); |
4e4cbee9 | 1303 | break; |
846785e6 | 1304 | case DM_MAPIO_REQUEUE: |
cfae7529 | 1305 | free_tio(tio); |
64f52b0e | 1306 | dec_pending(io, BLK_STS_DM_REQUEUE); |
846785e6 CH |
1307 | break; |
1308 | default: | |
45cbcd79 KU |
1309 | DMWARN("unimplemented target map return value: %d", r); |
1310 | BUG(); | |
1da177e4 | 1311 | } |
1da177e4 | 1312 | |
978e51ba | 1313 | return ret; |
1da177e4 | 1314 | } |
1da177e4 | 1315 | |
e0d6609a | 1316 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1317 | { |
4f024f37 KO |
1318 | bio->bi_iter.bi_sector = sector; |
1319 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1320 | } |
1321 | ||
1322 | /* | |
1323 | * Creates a bio that consists of range of complete bvecs. | |
1324 | */ | |
c80914e8 MS |
1325 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1326 | sector_t sector, unsigned len) | |
1da177e4 | 1327 | { |
dba14160 | 1328 | struct bio *clone = &tio->clone; |
1da177e4 | 1329 | |
1c3b13e6 KO |
1330 | __bio_clone_fast(clone, bio); |
1331 | ||
a892c8d5 ST |
1332 | bio_crypt_clone(clone, bio, GFP_NOIO); |
1333 | ||
57c36519 | 1334 | if (bio_integrity(bio)) { |
e2460f2a MP |
1335 | int r; |
1336 | ||
1337 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1338 | !dm_target_passes_integrity(tio->ti->type))) { | |
1339 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1340 | dm_device_name(tio->io->md), | |
1341 | tio->ti->type->name); | |
1342 | return -EIO; | |
1343 | } | |
1344 | ||
1345 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1346 | if (r < 0) |
1347 | return r; | |
1348 | } | |
bd2a49b8 | 1349 | |
fa8db494 MS |
1350 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1351 | clone->bi_iter.bi_size = to_bytes(len); | |
1352 | ||
1353 | if (bio_integrity(bio)) | |
1354 | bio_integrity_trim(clone); | |
c80914e8 MS |
1355 | |
1356 | return 0; | |
1da177e4 LT |
1357 | } |
1358 | ||
318716dd MS |
1359 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
1360 | struct dm_target *ti, unsigned num_bios) | |
f9ab94ce | 1361 | { |
dba14160 | 1362 | struct dm_target_io *tio; |
318716dd | 1363 | int try; |
dba14160 | 1364 | |
318716dd MS |
1365 | if (!num_bios) |
1366 | return; | |
f9ab94ce | 1367 | |
318716dd MS |
1368 | if (num_bios == 1) { |
1369 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); | |
1370 | bio_list_add(blist, &tio->clone); | |
1371 | return; | |
1372 | } | |
9015df24 | 1373 | |
318716dd MS |
1374 | for (try = 0; try < 2; try++) { |
1375 | int bio_nr; | |
1376 | struct bio *bio; | |
1377 | ||
1378 | if (try) | |
bc02cdbe | 1379 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd MS |
1380 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
1381 | tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); | |
1382 | if (!tio) | |
1383 | break; | |
1384 | ||
1385 | bio_list_add(blist, &tio->clone); | |
1386 | } | |
1387 | if (try) | |
bc02cdbe | 1388 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1389 | if (bio_nr == num_bios) |
1390 | return; | |
1391 | ||
1392 | while ((bio = bio_list_pop(blist))) { | |
1393 | tio = container_of(bio, struct dm_target_io, clone); | |
1394 | free_tio(tio); | |
1395 | } | |
1396 | } | |
9015df24 AK |
1397 | } |
1398 | ||
978e51ba MS |
1399 | static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci, |
1400 | struct dm_target_io *tio, unsigned *len) | |
9015df24 | 1401 | { |
dba14160 | 1402 | struct bio *clone = &tio->clone; |
9015df24 | 1403 | |
1dd40c3e MP |
1404 | tio->len_ptr = len; |
1405 | ||
99778273 | 1406 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1407 | if (len) |
1dd40c3e | 1408 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1409 | |
978e51ba | 1410 | return __map_bio(tio); |
f9ab94ce MP |
1411 | } |
1412 | ||
14fe594d | 1413 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1414 | unsigned num_bios, unsigned *len) |
06a426ce | 1415 | { |
318716dd MS |
1416 | struct bio_list blist = BIO_EMPTY_LIST; |
1417 | struct bio *bio; | |
1418 | struct dm_target_io *tio; | |
1419 | ||
1420 | alloc_multiple_bios(&blist, ci, ti, num_bios); | |
06a426ce | 1421 | |
318716dd MS |
1422 | while ((bio = bio_list_pop(&blist))) { |
1423 | tio = container_of(bio, struct dm_target_io, clone); | |
978e51ba | 1424 | (void) __clone_and_map_simple_bio(ci, tio, len); |
318716dd | 1425 | } |
06a426ce MS |
1426 | } |
1427 | ||
14fe594d | 1428 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1429 | { |
06a426ce | 1430 | unsigned target_nr = 0; |
f9ab94ce MP |
1431 | struct dm_target *ti; |
1432 | ||
892ad71f | 1433 | /* |
dbe3ece1 JA |
1434 | * Empty flush uses a statically initialized bio, as the base for |
1435 | * cloning. However, blkg association requires that a bdev is | |
1436 | * associated with a gendisk, which doesn't happen until the bdev is | |
1437 | * opened. So, blkg association is done at issue time of the flush | |
1438 | * rather than when the device is created in alloc_dev(). | |
892ad71f DZ |
1439 | */ |
1440 | bio_set_dev(ci->bio, ci->io->md->bdev); | |
1441 | ||
b372d360 | 1442 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1443 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1444 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce MP |
1445 | return 0; |
1446 | } | |
1447 | ||
c80914e8 | 1448 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
f31c21e4 | 1449 | sector_t sector, unsigned *len) |
5ae89a87 | 1450 | { |
dba14160 | 1451 | struct bio *bio = ci->bio; |
5ae89a87 | 1452 | struct dm_target_io *tio; |
f31c21e4 | 1453 | int r; |
5ae89a87 | 1454 | |
318716dd | 1455 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); |
f31c21e4 N |
1456 | tio->len_ptr = len; |
1457 | r = clone_bio(tio, bio, sector, *len); | |
1458 | if (r < 0) { | |
1459 | free_tio(tio); | |
1460 | return r; | |
b0d8ed4d | 1461 | } |
978e51ba | 1462 | (void) __map_bio(tio); |
c80914e8 | 1463 | |
f31c21e4 | 1464 | return 0; |
5ae89a87 MS |
1465 | } |
1466 | ||
55a62eef | 1467 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1468 | |
55a62eef | 1469 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1470 | { |
55a62eef | 1471 | return ti->num_discard_bios; |
23508a96 MS |
1472 | } |
1473 | ||
00716545 DS |
1474 | static unsigned get_num_secure_erase_bios(struct dm_target *ti) |
1475 | { | |
1476 | return ti->num_secure_erase_bios; | |
1477 | } | |
1478 | ||
55a62eef | 1479 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1480 | { |
55a62eef | 1481 | return ti->num_write_same_bios; |
23508a96 MS |
1482 | } |
1483 | ||
ac62d620 CH |
1484 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
1485 | { | |
1486 | return ti->num_write_zeroes_bios; | |
1487 | } | |
1488 | ||
3d7f4562 | 1489 | static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
61697a6a | 1490 | unsigned num_bios) |
ba1cbad9 | 1491 | { |
51b86f9a | 1492 | unsigned len; |
ba1cbad9 | 1493 | |
3d7f4562 MS |
1494 | /* |
1495 | * Even though the device advertised support for this type of | |
1496 | * request, that does not mean every target supports it, and | |
1497 | * reconfiguration might also have changed that since the | |
1498 | * check was performed. | |
1499 | */ | |
3d7f4562 MS |
1500 | if (!num_bios) |
1501 | return -EOPNOTSUPP; | |
ba1cbad9 | 1502 | |
51b86f9a ML |
1503 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
1504 | ||
3d7f4562 | 1505 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1506 | |
3d7f4562 MS |
1507 | ci->sector += len; |
1508 | ci->sector_count -= len; | |
5ae89a87 MS |
1509 | |
1510 | return 0; | |
ba1cbad9 MS |
1511 | } |
1512 | ||
3d7f4562 | 1513 | static int __send_discard(struct clone_info *ci, struct dm_target *ti) |
23508a96 | 1514 | { |
61697a6a | 1515 | return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti)); |
23508a96 | 1516 | } |
0ce65797 | 1517 | |
00716545 DS |
1518 | static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti) |
1519 | { | |
61697a6a | 1520 | return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti)); |
00716545 DS |
1521 | } |
1522 | ||
3d7f4562 | 1523 | static int __send_write_same(struct clone_info *ci, struct dm_target *ti) |
0ce65797 | 1524 | { |
61697a6a | 1525 | return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti)); |
0ce65797 MS |
1526 | } |
1527 | ||
3d7f4562 | 1528 | static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti) |
ac62d620 | 1529 | { |
61697a6a | 1530 | return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti)); |
ac62d620 CH |
1531 | } |
1532 | ||
568c73a3 MS |
1533 | static bool is_abnormal_io(struct bio *bio) |
1534 | { | |
1535 | bool r = false; | |
1536 | ||
1537 | switch (bio_op(bio)) { | |
1538 | case REQ_OP_DISCARD: | |
1539 | case REQ_OP_SECURE_ERASE: | |
1540 | case REQ_OP_WRITE_SAME: | |
1541 | case REQ_OP_WRITE_ZEROES: | |
1542 | r = true; | |
1543 | break; | |
1544 | } | |
1545 | ||
1546 | return r; | |
1547 | } | |
1548 | ||
0519c71e MS |
1549 | static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti, |
1550 | int *result) | |
1551 | { | |
1552 | struct bio *bio = ci->bio; | |
1553 | ||
1554 | if (bio_op(bio) == REQ_OP_DISCARD) | |
1555 | *result = __send_discard(ci, ti); | |
00716545 DS |
1556 | else if (bio_op(bio) == REQ_OP_SECURE_ERASE) |
1557 | *result = __send_secure_erase(ci, ti); | |
0519c71e MS |
1558 | else if (bio_op(bio) == REQ_OP_WRITE_SAME) |
1559 | *result = __send_write_same(ci, ti); | |
1560 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES) | |
1561 | *result = __send_write_zeroes(ci, ti); | |
1562 | else | |
1563 | return false; | |
1564 | ||
1565 | return true; | |
1566 | } | |
1567 | ||
e4c93811 AK |
1568 | /* |
1569 | * Select the correct strategy for processing a non-flush bio. | |
1570 | */ | |
14fe594d | 1571 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1572 | { |
512875bd | 1573 | struct dm_target *ti; |
1c3b13e6 | 1574 | unsigned len; |
c80914e8 | 1575 | int r; |
0ce65797 | 1576 | |
512875bd | 1577 | ti = dm_table_find_target(ci->map, ci->sector); |
123d87d5 | 1578 | if (!ti) |
512875bd JN |
1579 | return -EIO; |
1580 | ||
568c73a3 | 1581 | if (__process_abnormal_io(ci, ti, &r)) |
0519c71e | 1582 | return r; |
3d7f4562 | 1583 | |
e76239a3 | 1584 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
0ce65797 | 1585 | |
c80914e8 MS |
1586 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1587 | if (r < 0) | |
1588 | return r; | |
0ce65797 | 1589 | |
1c3b13e6 KO |
1590 | ci->sector += len; |
1591 | ci->sector_count -= len; | |
0ce65797 | 1592 | |
1c3b13e6 | 1593 | return 0; |
0ce65797 MS |
1594 | } |
1595 | ||
978e51ba MS |
1596 | static void init_clone_info(struct clone_info *ci, struct mapped_device *md, |
1597 | struct dm_table *map, struct bio *bio) | |
1598 | { | |
1599 | ci->map = map; | |
1600 | ci->io = alloc_io(md, bio); | |
1601 | ci->sector = bio->bi_iter.bi_sector; | |
1602 | } | |
1603 | ||
a1e1cb72 MS |
1604 | #define __dm_part_stat_sub(part, field, subnd) \ |
1605 | (part_stat_get(part, field) -= (subnd)) | |
1606 | ||
1da177e4 | 1607 | /* |
14fe594d | 1608 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1609 | */ |
978e51ba MS |
1610 | static blk_qc_t __split_and_process_bio(struct mapped_device *md, |
1611 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1612 | { |
1da177e4 | 1613 | struct clone_info ci; |
978e51ba | 1614 | blk_qc_t ret = BLK_QC_T_NONE; |
512875bd | 1615 | int error = 0; |
1da177e4 | 1616 | |
978e51ba | 1617 | init_clone_info(&ci, md, map, bio); |
0ce65797 | 1618 | |
1eff9d32 | 1619 | if (bio->bi_opf & REQ_PREFLUSH) { |
dbe3ece1 JA |
1620 | struct bio flush_bio; |
1621 | ||
1622 | /* | |
1623 | * Use an on-stack bio for this, it's safe since we don't | |
1624 | * need to reference it after submit. It's just used as | |
1625 | * the basis for the clone(s). | |
1626 | */ | |
1627 | bio_init(&flush_bio, NULL, 0); | |
1628 | flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; | |
1629 | ci.bio = &flush_bio; | |
b372d360 | 1630 | ci.sector_count = 0; |
14fe594d | 1631 | error = __send_empty_flush(&ci); |
382761dc | 1632 | bio_uninit(ci.bio); |
b372d360 | 1633 | /* dec_pending submits any data associated with flush */ |
2e2d6f7e | 1634 | } else if (op_is_zone_mgmt(bio_op(bio))) { |
a4aa5e56 DLM |
1635 | ci.bio = bio; |
1636 | ci.sector_count = 0; | |
1637 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1638 | } else { |
6a8736d1 | 1639 | ci.bio = bio; |
d87f4c14 | 1640 | ci.sector_count = bio_sectors(bio); |
18a25da8 | 1641 | while (ci.sector_count && !error) { |
14fe594d | 1642 | error = __split_and_process_non_flush(&ci); |
18a25da8 N |
1643 | if (current->bio_list && ci.sector_count && !error) { |
1644 | /* | |
ed00aabd | 1645 | * Remainder must be passed to submit_bio_noacct() |
18a25da8 N |
1646 | * so that it gets handled *after* bios already submitted |
1647 | * have been completely processed. | |
1648 | * We take a clone of the original to store in | |
745dc570 | 1649 | * ci.io->orig_bio to be used by end_io_acct() and |
18a25da8 | 1650 | * for dec_pending to use for completion handling. |
18a25da8 | 1651 | */ |
f21c601a MS |
1652 | struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count, |
1653 | GFP_NOIO, &md->queue->bio_split); | |
745dc570 | 1654 | ci.io->orig_bio = b; |
a1e1cb72 MS |
1655 | |
1656 | /* | |
1657 | * Adjust IO stats for each split, otherwise upon queue | |
1658 | * reentry there will be redundant IO accounting. | |
1659 | * NOTE: this is a stop-gap fix, a proper fix involves | |
1660 | * significant refactoring of DM core's bio splitting | |
1661 | * (by eliminating DM's splitting and just using bio_split) | |
1662 | */ | |
1663 | part_stat_lock(); | |
1664 | __dm_part_stat_sub(&dm_disk(md)->part0, | |
1665 | sectors[op_stat_group(bio_op(bio))], ci.sector_count); | |
1666 | part_stat_unlock(); | |
1667 | ||
18a25da8 | 1668 | bio_chain(b, bio); |
075c18c3 | 1669 | trace_block_split(md->queue, b, bio->bi_iter.bi_sector); |
ed00aabd | 1670 | ret = submit_bio_noacct(bio); |
18a25da8 N |
1671 | break; |
1672 | } | |
1673 | } | |
d87f4c14 | 1674 | } |
0ce65797 | 1675 | |
1da177e4 | 1676 | /* drop the extra reference count */ |
54385bf7 | 1677 | dec_pending(ci.io, errno_to_blk_status(error)); |
978e51ba | 1678 | return ret; |
0ce65797 MS |
1679 | } |
1680 | ||
cec47e3d | 1681 | /* |
978e51ba MS |
1682 | * Optimized variant of __split_and_process_bio that leverages the |
1683 | * fact that targets that use it do _not_ have a need to split bios. | |
cec47e3d | 1684 | */ |
568c73a3 MS |
1685 | static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map, |
1686 | struct bio *bio, struct dm_target *ti) | |
978e51ba MS |
1687 | { |
1688 | struct clone_info ci; | |
1689 | blk_qc_t ret = BLK_QC_T_NONE; | |
1690 | int error = 0; | |
1691 | ||
978e51ba MS |
1692 | init_clone_info(&ci, md, map, bio); |
1693 | ||
1694 | if (bio->bi_opf & REQ_PREFLUSH) { | |
dbe3ece1 JA |
1695 | struct bio flush_bio; |
1696 | ||
1697 | /* | |
1698 | * Use an on-stack bio for this, it's safe since we don't | |
1699 | * need to reference it after submit. It's just used as | |
1700 | * the basis for the clone(s). | |
1701 | */ | |
1702 | bio_init(&flush_bio, NULL, 0); | |
1703 | flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; | |
1704 | ci.bio = &flush_bio; | |
978e51ba MS |
1705 | ci.sector_count = 0; |
1706 | error = __send_empty_flush(&ci); | |
382761dc | 1707 | bio_uninit(ci.bio); |
978e51ba MS |
1708 | /* dec_pending submits any data associated with flush */ |
1709 | } else { | |
978e51ba MS |
1710 | struct dm_target_io *tio; |
1711 | ||
978e51ba MS |
1712 | ci.bio = bio; |
1713 | ci.sector_count = bio_sectors(bio); | |
568c73a3 | 1714 | if (__process_abnormal_io(&ci, ti, &error)) |
0519c71e MS |
1715 | goto out; |
1716 | ||
1717 | tio = alloc_tio(&ci, ti, 0, GFP_NOIO); | |
978e51ba MS |
1718 | ret = __clone_and_map_simple_bio(&ci, tio, NULL); |
1719 | } | |
1720 | out: | |
1721 | /* drop the extra reference count */ | |
1722 | dec_pending(ci.io, errno_to_blk_status(error)); | |
1723 | return ret; | |
1724 | } | |
1725 | ||
568c73a3 MS |
1726 | static void dm_queue_split(struct mapped_device *md, struct dm_target *ti, struct bio **bio) |
1727 | { | |
1728 | unsigned len, sector_count; | |
1729 | ||
1730 | sector_count = bio_sectors(*bio); | |
1731 | len = min_t(sector_t, max_io_len((*bio)->bi_iter.bi_sector, ti), sector_count); | |
1732 | ||
1733 | if (sector_count > len) { | |
1734 | struct bio *split = bio_split(*bio, len, GFP_NOIO, &md->queue->bio_split); | |
1735 | ||
1736 | bio_chain(split, *bio); | |
1737 | trace_block_split(md->queue, split, (*bio)->bi_iter.bi_sector); | |
ed00aabd | 1738 | submit_bio_noacct(*bio); |
568c73a3 MS |
1739 | *bio = split; |
1740 | } | |
1741 | } | |
1742 | ||
6548c7c5 MS |
1743 | static blk_qc_t dm_process_bio(struct mapped_device *md, |
1744 | struct dm_table *map, struct bio *bio) | |
1745 | { | |
568c73a3 MS |
1746 | blk_qc_t ret = BLK_QC_T_NONE; |
1747 | struct dm_target *ti = md->immutable_target; | |
1748 | ||
1749 | if (unlikely(!map)) { | |
1750 | bio_io_error(bio); | |
1751 | return ret; | |
1752 | } | |
1753 | ||
1754 | if (!ti) { | |
1755 | ti = dm_table_find_target(map, bio->bi_iter.bi_sector); | |
123d87d5 | 1756 | if (unlikely(!ti)) { |
568c73a3 MS |
1757 | bio_io_error(bio); |
1758 | return ret; | |
1759 | } | |
1760 | } | |
1761 | ||
1762 | /* | |
c62b37d9 | 1763 | * If in ->queue_bio we need to use blk_queue_split(), otherwise |
568c73a3 MS |
1764 | * queue_limits for abnormal requests (e.g. discard, writesame, etc) |
1765 | * won't be imposed. | |
1766 | */ | |
1767 | if (current->bio_list) { | |
120c9257 | 1768 | if (is_abnormal_io(bio)) |
f695ca38 | 1769 | blk_queue_split(&bio); |
120c9257 | 1770 | else |
568c73a3 MS |
1771 | dm_queue_split(md, ti, &bio); |
1772 | } | |
1773 | ||
6548c7c5 | 1774 | if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED) |
568c73a3 | 1775 | return __process_bio(md, map, bio, ti); |
6548c7c5 MS |
1776 | else |
1777 | return __split_and_process_bio(md, map, bio); | |
1778 | } | |
1779 | ||
c62b37d9 | 1780 | static blk_qc_t dm_submit_bio(struct bio *bio) |
cec47e3d | 1781 | { |
c4a59c4e | 1782 | struct mapped_device *md = bio->bi_disk->private_data; |
978e51ba | 1783 | blk_qc_t ret = BLK_QC_T_NONE; |
83d5e5b0 MP |
1784 | int srcu_idx; |
1785 | struct dm_table *map; | |
cec47e3d | 1786 | |
ac7c5675 CH |
1787 | if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) { |
1788 | /* | |
1789 | * We are called with a live reference on q_usage_counter, but | |
1790 | * that one will be released as soon as we return. Grab an | |
c62b37d9 CH |
1791 | * extra one as blk_mq_submit_bio expects to be able to consume |
1792 | * a reference (which lives until the request is freed in case a | |
1793 | * request is allocated). | |
ac7c5675 | 1794 | */ |
c62b37d9 CH |
1795 | percpu_ref_get(&bio->bi_disk->queue->q_usage_counter); |
1796 | return blk_mq_submit_bio(bio); | |
ac7c5675 | 1797 | } |
8cf7961d | 1798 | |
83d5e5b0 | 1799 | map = dm_get_live_table(md, &srcu_idx); |
29e4013d | 1800 | |
6a8736d1 TH |
1801 | /* if we're suspended, we have to queue this io for later */ |
1802 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1803 | dm_put_live_table(md, srcu_idx); |
9eef87da | 1804 | |
6abc4946 KK |
1805 | if (bio->bi_opf & REQ_NOWAIT) |
1806 | bio_wouldblock_error(bio); | |
1807 | else if (!(bio->bi_opf & REQ_RAHEAD)) | |
6a8736d1 TH |
1808 | queue_io(md, bio); |
1809 | else | |
54d9a1b4 | 1810 | bio_io_error(bio); |
978e51ba | 1811 | return ret; |
cec47e3d | 1812 | } |
1da177e4 | 1813 | |
6548c7c5 | 1814 | ret = dm_process_bio(md, map, bio); |
978e51ba | 1815 | |
83d5e5b0 | 1816 | dm_put_live_table(md, srcu_idx); |
978e51ba MS |
1817 | return ret; |
1818 | } | |
1819 | ||
1da177e4 LT |
1820 | /*----------------------------------------------------------------- |
1821 | * An IDR is used to keep track of allocated minor numbers. | |
1822 | *---------------------------------------------------------------*/ | |
2b06cfff | 1823 | static void free_minor(int minor) |
1da177e4 | 1824 | { |
f32c10b0 | 1825 | spin_lock(&_minor_lock); |
1da177e4 | 1826 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1827 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1828 | } |
1829 | ||
1830 | /* | |
1831 | * See if the device with a specific minor # is free. | |
1832 | */ | |
cf13ab8e | 1833 | static int specific_minor(int minor) |
1da177e4 | 1834 | { |
c9d76be6 | 1835 | int r; |
1da177e4 LT |
1836 | |
1837 | if (minor >= (1 << MINORBITS)) | |
1838 | return -EINVAL; | |
1839 | ||
c9d76be6 | 1840 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1841 | spin_lock(&_minor_lock); |
1da177e4 | 1842 | |
c9d76be6 | 1843 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1844 | |
f32c10b0 | 1845 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1846 | idr_preload_end(); |
1847 | if (r < 0) | |
1848 | return r == -ENOSPC ? -EBUSY : r; | |
1849 | return 0; | |
1da177e4 LT |
1850 | } |
1851 | ||
cf13ab8e | 1852 | static int next_free_minor(int *minor) |
1da177e4 | 1853 | { |
c9d76be6 | 1854 | int r; |
62f75c2f | 1855 | |
c9d76be6 | 1856 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1857 | spin_lock(&_minor_lock); |
1da177e4 | 1858 | |
c9d76be6 | 1859 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1860 | |
f32c10b0 | 1861 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1862 | idr_preload_end(); |
1863 | if (r < 0) | |
1864 | return r; | |
1865 | *minor = r; | |
1866 | return 0; | |
1da177e4 LT |
1867 | } |
1868 | ||
83d5cde4 | 1869 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1870 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1871 | |
53d5914f MP |
1872 | static void dm_wq_work(struct work_struct *work); |
1873 | ||
0f20972f MS |
1874 | static void cleanup_mapped_device(struct mapped_device *md) |
1875 | { | |
0f20972f MS |
1876 | if (md->wq) |
1877 | destroy_workqueue(md->wq); | |
6f1c819c KO |
1878 | bioset_exit(&md->bs); |
1879 | bioset_exit(&md->io_bs); | |
0f20972f | 1880 | |
f26c5719 DW |
1881 | if (md->dax_dev) { |
1882 | kill_dax(md->dax_dev); | |
1883 | put_dax(md->dax_dev); | |
1884 | md->dax_dev = NULL; | |
1885 | } | |
1886 | ||
0f20972f MS |
1887 | if (md->disk) { |
1888 | spin_lock(&_minor_lock); | |
1889 | md->disk->private_data = NULL; | |
1890 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1891 | del_gendisk(md->disk); |
1892 | put_disk(md->disk); | |
1893 | } | |
1894 | ||
1895 | if (md->queue) | |
1896 | blk_cleanup_queue(md->queue); | |
1897 | ||
d09960b0 TE |
1898 | cleanup_srcu_struct(&md->io_barrier); |
1899 | ||
0f20972f MS |
1900 | if (md->bdev) { |
1901 | bdput(md->bdev); | |
1902 | md->bdev = NULL; | |
1903 | } | |
4cc96131 | 1904 | |
d5ffebdd MS |
1905 | mutex_destroy(&md->suspend_lock); |
1906 | mutex_destroy(&md->type_lock); | |
1907 | mutex_destroy(&md->table_devices_lock); | |
1908 | ||
4cc96131 | 1909 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
1910 | } |
1911 | ||
1da177e4 LT |
1912 | /* |
1913 | * Allocate and initialise a blank device with a given minor. | |
1914 | */ | |
2b06cfff | 1915 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1916 | { |
115485e8 MS |
1917 | int r, numa_node_id = dm_get_numa_node(); |
1918 | struct mapped_device *md; | |
ba61fdd1 | 1919 | void *old_md; |
1da177e4 | 1920 | |
856eb091 | 1921 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1922 | if (!md) { |
1923 | DMWARN("unable to allocate device, out of memory."); | |
1924 | return NULL; | |
1925 | } | |
1926 | ||
10da4f79 | 1927 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1928 | goto bad_module_get; |
10da4f79 | 1929 | |
1da177e4 | 1930 | /* get a minor number for the dev */ |
2b06cfff | 1931 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1932 | r = next_free_minor(&minor); |
2b06cfff | 1933 | else |
cf13ab8e | 1934 | r = specific_minor(minor); |
1da177e4 | 1935 | if (r < 0) |
6ed7ade8 | 1936 | goto bad_minor; |
1da177e4 | 1937 | |
83d5e5b0 MP |
1938 | r = init_srcu_struct(&md->io_barrier); |
1939 | if (r < 0) | |
1940 | goto bad_io_barrier; | |
1941 | ||
115485e8 | 1942 | md->numa_node_id = numa_node_id; |
591ddcfc | 1943 | md->init_tio_pdu = false; |
a5664dad | 1944 | md->type = DM_TYPE_NONE; |
e61290a4 | 1945 | mutex_init(&md->suspend_lock); |
a5664dad | 1946 | mutex_init(&md->type_lock); |
86f1152b | 1947 | mutex_init(&md->table_devices_lock); |
022c2611 | 1948 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1949 | atomic_set(&md->holders, 1); |
5c6bd75d | 1950 | atomic_set(&md->open_count, 0); |
1da177e4 | 1951 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1952 | atomic_set(&md->uevent_seq, 0); |
1953 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1954 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1955 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1956 | |
47ace7e0 | 1957 | /* |
c62b37d9 CH |
1958 | * default to bio-based until DM table is loaded and md->type |
1959 | * established. If request-based table is loaded: blk-mq will | |
1960 | * override accordingly. | |
47ace7e0 | 1961 | */ |
c62b37d9 | 1962 | md->queue = blk_alloc_queue(numa_node_id); |
3d745ea5 CH |
1963 | if (!md->queue) |
1964 | goto bad; | |
1da177e4 | 1965 | |
c12c9a3c | 1966 | md->disk = alloc_disk_node(1, md->numa_node_id); |
1da177e4 | 1967 | if (!md->disk) |
0f20972f | 1968 | goto bad; |
1da177e4 | 1969 | |
f0b04115 | 1970 | init_waitqueue_head(&md->wait); |
53d5914f | 1971 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1972 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1973 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 1974 | |
1da177e4 LT |
1975 | md->disk->major = _major; |
1976 | md->disk->first_minor = minor; | |
1977 | md->disk->fops = &dm_blk_dops; | |
1978 | md->disk->queue = md->queue; | |
1979 | md->disk->private_data = md; | |
1980 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 | 1981 | |
976431b0 | 1982 | if (IS_ENABLED(CONFIG_DAX_DRIVER)) { |
fefc1d97 PG |
1983 | md->dax_dev = alloc_dax(md, md->disk->disk_name, |
1984 | &dm_dax_ops, 0); | |
4e4ced93 | 1985 | if (IS_ERR(md->dax_dev)) |
976431b0 DW |
1986 | goto bad; |
1987 | } | |
f26c5719 | 1988 | |
c100ec49 | 1989 | add_disk_no_queue_reg(md->disk); |
7e51f257 | 1990 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1991 | |
670368a8 | 1992 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1993 | if (!md->wq) |
0f20972f | 1994 | goto bad; |
304f3f6a | 1995 | |
32a926da MP |
1996 | md->bdev = bdget_disk(md->disk, 0); |
1997 | if (!md->bdev) | |
0f20972f | 1998 | goto bad; |
32a926da | 1999 | |
fd2ed4d2 MP |
2000 | dm_stats_init(&md->stats); |
2001 | ||
ba61fdd1 | 2002 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2003 | spin_lock(&_minor_lock); |
ba61fdd1 | 2004 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2005 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2006 | |
2007 | BUG_ON(old_md != MINOR_ALLOCED); | |
2008 | ||
1da177e4 LT |
2009 | return md; |
2010 | ||
0f20972f MS |
2011 | bad: |
2012 | cleanup_mapped_device(md); | |
83d5e5b0 | 2013 | bad_io_barrier: |
1da177e4 | 2014 | free_minor(minor); |
6ed7ade8 | 2015 | bad_minor: |
10da4f79 | 2016 | module_put(THIS_MODULE); |
6ed7ade8 | 2017 | bad_module_get: |
856eb091 | 2018 | kvfree(md); |
1da177e4 LT |
2019 | return NULL; |
2020 | } | |
2021 | ||
ae9da83f JN |
2022 | static void unlock_fs(struct mapped_device *md); |
2023 | ||
1da177e4 LT |
2024 | static void free_dev(struct mapped_device *md) |
2025 | { | |
f331c029 | 2026 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2027 | |
32a926da | 2028 | unlock_fs(md); |
2eb6e1e3 | 2029 | |
0f20972f | 2030 | cleanup_mapped_device(md); |
63a4f065 | 2031 | |
86f1152b | 2032 | free_table_devices(&md->table_devices); |
63a4f065 | 2033 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2034 | free_minor(minor); |
2035 | ||
10da4f79 | 2036 | module_put(THIS_MODULE); |
856eb091 | 2037 | kvfree(md); |
1da177e4 LT |
2038 | } |
2039 | ||
2a2a4c51 | 2040 | static int __bind_mempools(struct mapped_device *md, struct dm_table *t) |
e6ee8c0b | 2041 | { |
c0820cf5 | 2042 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
2a2a4c51 | 2043 | int ret = 0; |
e6ee8c0b | 2044 | |
0776aa0e | 2045 | if (dm_table_bio_based(t)) { |
64f52b0e MS |
2046 | /* |
2047 | * The md may already have mempools that need changing. | |
2048 | * If so, reload bioset because front_pad may have changed | |
2049 | * because a different table was loaded. | |
2050 | */ | |
6f1c819c KO |
2051 | bioset_exit(&md->bs); |
2052 | bioset_exit(&md->io_bs); | |
0776aa0e | 2053 | |
6f1c819c | 2054 | } else if (bioset_initialized(&md->bs)) { |
4e6e36c3 MS |
2055 | /* |
2056 | * There's no need to reload with request-based dm | |
2057 | * because the size of front_pad doesn't change. | |
2058 | * Note for future: If you are to reload bioset, | |
2059 | * prep-ed requests in the queue may refer | |
2060 | * to bio from the old bioset, so you must walk | |
2061 | * through the queue to unprep. | |
2062 | */ | |
2063 | goto out; | |
c0820cf5 | 2064 | } |
e6ee8c0b | 2065 | |
6f1c819c KO |
2066 | BUG_ON(!p || |
2067 | bioset_initialized(&md->bs) || | |
2068 | bioset_initialized(&md->io_bs)); | |
cbc4e3c1 | 2069 | |
2a2a4c51 JA |
2070 | ret = bioset_init_from_src(&md->bs, &p->bs); |
2071 | if (ret) | |
2072 | goto out; | |
2073 | ret = bioset_init_from_src(&md->io_bs, &p->io_bs); | |
2074 | if (ret) | |
2075 | bioset_exit(&md->bs); | |
e6ee8c0b | 2076 | out: |
02233342 | 2077 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b | 2078 | dm_table_free_md_mempools(t); |
2a2a4c51 | 2079 | return ret; |
e6ee8c0b KU |
2080 | } |
2081 | ||
1da177e4 LT |
2082 | /* |
2083 | * Bind a table to the device. | |
2084 | */ | |
2085 | static void event_callback(void *context) | |
2086 | { | |
7a8c3d3b MA |
2087 | unsigned long flags; |
2088 | LIST_HEAD(uevents); | |
1da177e4 LT |
2089 | struct mapped_device *md = (struct mapped_device *) context; |
2090 | ||
7a8c3d3b MA |
2091 | spin_lock_irqsave(&md->uevent_lock, flags); |
2092 | list_splice_init(&md->uevent_list, &uevents); | |
2093 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2094 | ||
ed9e1982 | 2095 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2096 | |
1da177e4 LT |
2097 | atomic_inc(&md->event_nr); |
2098 | wake_up(&md->eventq); | |
62e08243 | 2099 | dm_issue_global_event(); |
1da177e4 LT |
2100 | } |
2101 | ||
042d2a9b AK |
2102 | /* |
2103 | * Returns old map, which caller must destroy. | |
2104 | */ | |
2105 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2106 | struct queue_limits *limits) | |
1da177e4 | 2107 | { |
042d2a9b | 2108 | struct dm_table *old_map; |
165125e1 | 2109 | struct request_queue *q = md->queue; |
978e51ba | 2110 | bool request_based = dm_table_request_based(t); |
1da177e4 | 2111 | sector_t size; |
2a2a4c51 | 2112 | int ret; |
1da177e4 | 2113 | |
5a8f1f80 BVA |
2114 | lockdep_assert_held(&md->suspend_lock); |
2115 | ||
1da177e4 | 2116 | size = dm_table_get_size(t); |
3ac51e74 DW |
2117 | |
2118 | /* | |
2119 | * Wipe any geometry if the size of the table changed. | |
2120 | */ | |
fd2ed4d2 | 2121 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2122 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2123 | ||
c2b4bb8c CH |
2124 | set_capacity(md->disk, size); |
2125 | bd_set_nr_sectors(md->bdev, size); | |
d5816876 | 2126 | |
2ca3310e AK |
2127 | dm_table_event_callback(t, event_callback, md); |
2128 | ||
e6ee8c0b KU |
2129 | /* |
2130 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2131 | * for request-based during suspension. So stop it to prevent | |
2132 | * I/O mapping before resume. | |
2133 | * This must be done before setting the queue restrictions, | |
2134 | * because request-based dm may be run just after the setting. | |
2135 | */ | |
978e51ba | 2136 | if (request_based) |
eca7ee6d | 2137 | dm_stop_queue(q); |
978e51ba MS |
2138 | |
2139 | if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) { | |
16f12266 | 2140 | /* |
978e51ba MS |
2141 | * Leverage the fact that request-based DM targets and |
2142 | * NVMe bio based targets are immutable singletons | |
2143 | * - used to optimize both dm_request_fn and dm_mq_queue_rq; | |
2144 | * and __process_bio. | |
16f12266 MS |
2145 | */ |
2146 | md->immutable_target = dm_table_get_immutable_target(t); | |
2147 | } | |
e6ee8c0b | 2148 | |
2a2a4c51 JA |
2149 | ret = __bind_mempools(md, t); |
2150 | if (ret) { | |
2151 | old_map = ERR_PTR(ret); | |
2152 | goto out; | |
2153 | } | |
e6ee8c0b | 2154 | |
a12f5d48 | 2155 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2156 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2157 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2158 | ||
754c5fc7 | 2159 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2160 | if (old_map) |
2161 | dm_sync_table(md); | |
1da177e4 | 2162 | |
2a2a4c51 | 2163 | out: |
042d2a9b | 2164 | return old_map; |
1da177e4 LT |
2165 | } |
2166 | ||
a7940155 AK |
2167 | /* |
2168 | * Returns unbound table for the caller to free. | |
2169 | */ | |
2170 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2171 | { |
a12f5d48 | 2172 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2173 | |
2174 | if (!map) | |
a7940155 | 2175 | return NULL; |
1da177e4 LT |
2176 | |
2177 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2178 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2179 | dm_sync_table(md); |
a7940155 AK |
2180 | |
2181 | return map; | |
1da177e4 LT |
2182 | } |
2183 | ||
2184 | /* | |
2185 | * Constructor for a new device. | |
2186 | */ | |
2b06cfff | 2187 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 | 2188 | { |
c12c9a3c | 2189 | int r; |
1da177e4 LT |
2190 | struct mapped_device *md; |
2191 | ||
2b06cfff | 2192 | md = alloc_dev(minor); |
1da177e4 LT |
2193 | if (!md) |
2194 | return -ENXIO; | |
2195 | ||
c12c9a3c MS |
2196 | r = dm_sysfs_init(md); |
2197 | if (r) { | |
2198 | free_dev(md); | |
2199 | return r; | |
2200 | } | |
784aae73 | 2201 | |
1da177e4 LT |
2202 | *result = md; |
2203 | return 0; | |
2204 | } | |
2205 | ||
a5664dad MS |
2206 | /* |
2207 | * Functions to manage md->type. | |
2208 | * All are required to hold md->type_lock. | |
2209 | */ | |
2210 | void dm_lock_md_type(struct mapped_device *md) | |
2211 | { | |
2212 | mutex_lock(&md->type_lock); | |
2213 | } | |
2214 | ||
2215 | void dm_unlock_md_type(struct mapped_device *md) | |
2216 | { | |
2217 | mutex_unlock(&md->type_lock); | |
2218 | } | |
2219 | ||
7e0d574f | 2220 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2221 | { |
00c4fc3b | 2222 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2223 | md->type = type; |
2224 | } | |
2225 | ||
7e0d574f | 2226 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2227 | { |
2228 | return md->type; | |
2229 | } | |
2230 | ||
36a0456f AK |
2231 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2232 | { | |
2233 | return md->immutable_target_type; | |
2234 | } | |
2235 | ||
f84cb8a4 MS |
2236 | /* |
2237 | * The queue_limits are only valid as long as you have a reference | |
2238 | * count on 'md'. | |
2239 | */ | |
2240 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2241 | { | |
2242 | BUG_ON(!atomic_read(&md->holders)); | |
2243 | return &md->queue->limits; | |
2244 | } | |
2245 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2246 | ||
4a0b4ddf MS |
2247 | /* |
2248 | * Setup the DM device's queue based on md's type | |
2249 | */ | |
591ddcfc | 2250 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2251 | { |
bfebd1cd | 2252 | int r; |
c100ec49 | 2253 | struct queue_limits limits; |
7e0d574f | 2254 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2255 | |
545ed20e | 2256 | switch (type) { |
bfebd1cd | 2257 | case DM_TYPE_REQUEST_BASED: |
e83068a5 | 2258 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2259 | if (r) { |
eca7ee6d | 2260 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2261 | return r; |
2262 | } | |
2263 | break; | |
2264 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2265 | case DM_TYPE_DAX_BIO_BASED: |
978e51ba | 2266 | case DM_TYPE_NVME_BIO_BASED: |
bfebd1cd | 2267 | break; |
7e0d574f BVA |
2268 | case DM_TYPE_NONE: |
2269 | WARN_ON_ONCE(true); | |
2270 | break; | |
4a0b4ddf MS |
2271 | } |
2272 | ||
c100ec49 MS |
2273 | r = dm_calculate_queue_limits(t, &limits); |
2274 | if (r) { | |
2275 | DMERR("Cannot calculate initial queue limits"); | |
2276 | return r; | |
2277 | } | |
2278 | dm_table_set_restrictions(t, md->queue, &limits); | |
2279 | blk_register_queue(md->disk); | |
2280 | ||
4a0b4ddf MS |
2281 | return 0; |
2282 | } | |
2283 | ||
2bec1f4a | 2284 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2285 | { |
2286 | struct mapped_device *md; | |
1da177e4 LT |
2287 | unsigned minor = MINOR(dev); |
2288 | ||
2289 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2290 | return NULL; | |
2291 | ||
f32c10b0 | 2292 | spin_lock(&_minor_lock); |
1da177e4 LT |
2293 | |
2294 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2295 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2296 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2297 | md = NULL; | |
2298 | goto out; | |
fba9f90e | 2299 | } |
49de5769 | 2300 | dm_get(md); |
fba9f90e | 2301 | out: |
f32c10b0 | 2302 | spin_unlock(&_minor_lock); |
1da177e4 | 2303 | |
637842cf DT |
2304 | return md; |
2305 | } | |
3cf2e4ba | 2306 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2307 | |
9ade92a9 | 2308 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2309 | { |
9ade92a9 | 2310 | return md->interface_ptr; |
1da177e4 LT |
2311 | } |
2312 | ||
2313 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2314 | { | |
2315 | md->interface_ptr = ptr; | |
2316 | } | |
2317 | ||
2318 | void dm_get(struct mapped_device *md) | |
2319 | { | |
2320 | atomic_inc(&md->holders); | |
3f77316d | 2321 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2322 | } |
2323 | ||
09ee96b2 MP |
2324 | int dm_hold(struct mapped_device *md) |
2325 | { | |
2326 | spin_lock(&_minor_lock); | |
2327 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2328 | spin_unlock(&_minor_lock); | |
2329 | return -EBUSY; | |
2330 | } | |
2331 | dm_get(md); | |
2332 | spin_unlock(&_minor_lock); | |
2333 | return 0; | |
2334 | } | |
2335 | EXPORT_SYMBOL_GPL(dm_hold); | |
2336 | ||
72d94861 AK |
2337 | const char *dm_device_name(struct mapped_device *md) |
2338 | { | |
2339 | return md->name; | |
2340 | } | |
2341 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2342 | ||
3f77316d | 2343 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2344 | { |
1134e5ae | 2345 | struct dm_table *map; |
83d5e5b0 | 2346 | int srcu_idx; |
1da177e4 | 2347 | |
3f77316d | 2348 | might_sleep(); |
fba9f90e | 2349 | |
63a4f065 | 2350 | spin_lock(&_minor_lock); |
3f77316d KU |
2351 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2352 | set_bit(DMF_FREEING, &md->flags); | |
2353 | spin_unlock(&_minor_lock); | |
3b785fbc | 2354 | |
c12c9a3c | 2355 | blk_set_queue_dying(md->queue); |
3f77316d | 2356 | |
ab7c7bb6 MP |
2357 | /* |
2358 | * Take suspend_lock so that presuspend and postsuspend methods | |
2359 | * do not race with internal suspend. | |
2360 | */ | |
2361 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2362 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2363 | if (!dm_suspended_md(md)) { |
2364 | dm_table_presuspend_targets(map); | |
adc0daad | 2365 | set_bit(DMF_SUSPENDED, &md->flags); |
5df96f2b | 2366 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
3f77316d | 2367 | dm_table_postsuspend_targets(map); |
1da177e4 | 2368 | } |
83d5e5b0 MP |
2369 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2370 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2371 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2372 | |
3f77316d KU |
2373 | /* |
2374 | * Rare, but there may be I/O requests still going to complete, | |
2375 | * for example. Wait for all references to disappear. | |
2376 | * No one should increment the reference count of the mapped_device, | |
2377 | * after the mapped_device state becomes DMF_FREEING. | |
2378 | */ | |
2379 | if (wait) | |
2380 | while (atomic_read(&md->holders)) | |
2381 | msleep(1); | |
2382 | else if (atomic_read(&md->holders)) | |
2383 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2384 | dm_device_name(md), atomic_read(&md->holders)); | |
2385 | ||
2386 | dm_sysfs_exit(md); | |
3f77316d KU |
2387 | dm_table_destroy(__unbind(md)); |
2388 | free_dev(md); | |
2389 | } | |
2390 | ||
2391 | void dm_destroy(struct mapped_device *md) | |
2392 | { | |
2393 | __dm_destroy(md, true); | |
2394 | } | |
2395 | ||
2396 | void dm_destroy_immediate(struct mapped_device *md) | |
2397 | { | |
2398 | __dm_destroy(md, false); | |
2399 | } | |
2400 | ||
2401 | void dm_put(struct mapped_device *md) | |
2402 | { | |
2403 | atomic_dec(&md->holders); | |
1da177e4 | 2404 | } |
79eb885c | 2405 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2406 | |
85067747 ML |
2407 | static bool md_in_flight_bios(struct mapped_device *md) |
2408 | { | |
2409 | int cpu; | |
2410 | struct hd_struct *part = &dm_disk(md)->part0; | |
2411 | long sum = 0; | |
2412 | ||
2413 | for_each_possible_cpu(cpu) { | |
2414 | sum += part_stat_local_read_cpu(part, in_flight[0], cpu); | |
2415 | sum += part_stat_local_read_cpu(part, in_flight[1], cpu); | |
2416 | } | |
2417 | ||
2418 | return sum != 0; | |
2419 | } | |
2420 | ||
2421 | static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state) | |
46125c1c MB |
2422 | { |
2423 | int r = 0; | |
9f4c3f87 | 2424 | DEFINE_WAIT(wait); |
46125c1c | 2425 | |
85067747 | 2426 | while (true) { |
9f4c3f87 | 2427 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2428 | |
85067747 | 2429 | if (!md_in_flight_bios(md)) |
46125c1c MB |
2430 | break; |
2431 | ||
e3fabdfd | 2432 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2433 | r = -EINTR; |
2434 | break; | |
2435 | } | |
2436 | ||
2437 | io_schedule(); | |
2438 | } | |
9f4c3f87 | 2439 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2440 | |
46125c1c MB |
2441 | return r; |
2442 | } | |
2443 | ||
85067747 ML |
2444 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
2445 | { | |
2446 | int r = 0; | |
2447 | ||
2448 | if (!queue_is_mq(md->queue)) | |
2449 | return dm_wait_for_bios_completion(md, task_state); | |
2450 | ||
2451 | while (true) { | |
2452 | if (!blk_mq_queue_inflight(md->queue)) | |
2453 | break; | |
2454 | ||
2455 | if (signal_pending_state(task_state, current)) { | |
2456 | r = -EINTR; | |
2457 | break; | |
2458 | } | |
2459 | ||
2460 | msleep(5); | |
2461 | } | |
2462 | ||
2463 | return r; | |
2464 | } | |
2465 | ||
1da177e4 LT |
2466 | /* |
2467 | * Process the deferred bios | |
2468 | */ | |
ef208587 | 2469 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2470 | { |
ef208587 MP |
2471 | struct mapped_device *md = container_of(work, struct mapped_device, |
2472 | work); | |
6d6f10df | 2473 | struct bio *c; |
83d5e5b0 MP |
2474 | int srcu_idx; |
2475 | struct dm_table *map; | |
1da177e4 | 2476 | |
83d5e5b0 | 2477 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2478 | |
3b00b203 | 2479 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2480 | spin_lock_irq(&md->deferred_lock); |
2481 | c = bio_list_pop(&md->deferred); | |
2482 | spin_unlock_irq(&md->deferred_lock); | |
2483 | ||
6a8736d1 | 2484 | if (!c) |
df12ee99 | 2485 | break; |
022c2611 | 2486 | |
e6ee8c0b | 2487 | if (dm_request_based(md)) |
ed00aabd | 2488 | (void) submit_bio_noacct(c); |
6a8736d1 | 2489 | else |
6548c7c5 | 2490 | (void) dm_process_bio(md, map, c); |
022c2611 | 2491 | } |
73d410c0 | 2492 | |
83d5e5b0 | 2493 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2494 | } |
2495 | ||
9a1fb464 | 2496 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2497 | { |
3b00b203 | 2498 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2499 | smp_mb__after_atomic(); |
53d5914f | 2500 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2501 | } |
2502 | ||
1da177e4 | 2503 | /* |
042d2a9b | 2504 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2505 | */ |
042d2a9b | 2506 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2507 | { |
87eb5b21 | 2508 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2509 | struct queue_limits limits; |
042d2a9b | 2510 | int r; |
1da177e4 | 2511 | |
e61290a4 | 2512 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2513 | |
2514 | /* device must be suspended */ | |
4f186f8b | 2515 | if (!dm_suspended_md(md)) |
93c534ae | 2516 | goto out; |
1da177e4 | 2517 | |
3ae70656 MS |
2518 | /* |
2519 | * If the new table has no data devices, retain the existing limits. | |
2520 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2521 | * then new I/O is queued based on these limits, and then some paths | |
2522 | * reappear. | |
2523 | */ | |
2524 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2525 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2526 | if (live_map) |
2527 | limits = md->queue->limits; | |
83d5e5b0 | 2528 | dm_put_live_table_fast(md); |
3ae70656 MS |
2529 | } |
2530 | ||
87eb5b21 MC |
2531 | if (!live_map) { |
2532 | r = dm_calculate_queue_limits(table, &limits); | |
2533 | if (r) { | |
2534 | map = ERR_PTR(r); | |
2535 | goto out; | |
2536 | } | |
042d2a9b | 2537 | } |
754c5fc7 | 2538 | |
042d2a9b | 2539 | map = __bind(md, table, &limits); |
62e08243 | 2540 | dm_issue_global_event(); |
1da177e4 | 2541 | |
93c534ae | 2542 | out: |
e61290a4 | 2543 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2544 | return map; |
1da177e4 LT |
2545 | } |
2546 | ||
2547 | /* | |
2548 | * Functions to lock and unlock any filesystem running on the | |
2549 | * device. | |
2550 | */ | |
2ca3310e | 2551 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2552 | { |
e39e2e95 | 2553 | int r; |
1da177e4 LT |
2554 | |
2555 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2556 | |
db8fef4f | 2557 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2558 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2559 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2560 | md->frozen_sb = NULL; |
2561 | return r; | |
dfbe03f6 AK |
2562 | } |
2563 | ||
aa8d7c2f AK |
2564 | set_bit(DMF_FROZEN, &md->flags); |
2565 | ||
1da177e4 LT |
2566 | return 0; |
2567 | } | |
2568 | ||
2ca3310e | 2569 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2570 | { |
aa8d7c2f AK |
2571 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2572 | return; | |
2573 | ||
db8fef4f | 2574 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2575 | md->frozen_sb = NULL; |
aa8d7c2f | 2576 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2577 | } |
2578 | ||
2579 | /* | |
b48633f8 BVA |
2580 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2581 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2582 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2583 | * | |
ffcc3936 MS |
2584 | * If __dm_suspend returns 0, the device is completely quiescent |
2585 | * now. There is no request-processing activity. All new requests | |
2586 | * are being added to md->deferred list. | |
cec47e3d | 2587 | */ |
ffcc3936 | 2588 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2589 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2590 | int dmf_suspended_flag) |
1da177e4 | 2591 | { |
ffcc3936 MS |
2592 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2593 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2594 | int r; | |
1da177e4 | 2595 | |
5a8f1f80 BVA |
2596 | lockdep_assert_held(&md->suspend_lock); |
2597 | ||
2e93ccc1 KU |
2598 | /* |
2599 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2600 | * This flag is cleared before dm_suspend returns. | |
2601 | */ | |
2602 | if (noflush) | |
2603 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 | 2604 | else |
ac75b09f | 2605 | DMDEBUG("%s: suspending with flush", dm_device_name(md)); |
2e93ccc1 | 2606 | |
d67ee213 MS |
2607 | /* |
2608 | * This gets reverted if there's an error later and the targets | |
2609 | * provide the .presuspend_undo hook. | |
2610 | */ | |
cf222b37 AK |
2611 | dm_table_presuspend_targets(map); |
2612 | ||
32a926da | 2613 | /* |
9f518b27 KU |
2614 | * Flush I/O to the device. |
2615 | * Any I/O submitted after lock_fs() may not be flushed. | |
2616 | * noflush takes precedence over do_lockfs. | |
2617 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2618 | */ |
2619 | if (!noflush && do_lockfs) { | |
2620 | r = lock_fs(md); | |
d67ee213 MS |
2621 | if (r) { |
2622 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2623 | return r; |
d67ee213 | 2624 | } |
aa8d7c2f | 2625 | } |
1da177e4 LT |
2626 | |
2627 | /* | |
3b00b203 MP |
2628 | * Here we must make sure that no processes are submitting requests |
2629 | * to target drivers i.e. no one may be executing | |
2630 | * __split_and_process_bio. This is called from dm_request and | |
2631 | * dm_wq_work. | |
2632 | * | |
2633 | * To get all processes out of __split_and_process_bio in dm_request, | |
2634 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2635 | * __split_and_process_bio from dm_request and quiesce the thread |
2636 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2637 | * flush_workqueue(md->wq). | |
1da177e4 | 2638 | */ |
1eb787ec | 2639 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2640 | if (map) |
2641 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2642 | |
d0bcb878 | 2643 | /* |
29e4013d TH |
2644 | * Stop md->queue before flushing md->wq in case request-based |
2645 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2646 | */ |
6a23e05c | 2647 | if (dm_request_based(md)) |
eca7ee6d | 2648 | dm_stop_queue(md->queue); |
cec47e3d | 2649 | |
d0bcb878 KU |
2650 | flush_workqueue(md->wq); |
2651 | ||
1da177e4 | 2652 | /* |
3b00b203 MP |
2653 | * At this point no more requests are entering target request routines. |
2654 | * We call dm_wait_for_completion to wait for all existing requests | |
2655 | * to finish. | |
1da177e4 | 2656 | */ |
b48633f8 | 2657 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2658 | if (!r) |
2659 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2660 | |
6d6f10df | 2661 | if (noflush) |
022c2611 | 2662 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2663 | if (map) |
2664 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2665 | |
1da177e4 | 2666 | /* were we interrupted ? */ |
46125c1c | 2667 | if (r < 0) { |
9a1fb464 | 2668 | dm_queue_flush(md); |
73d410c0 | 2669 | |
cec47e3d | 2670 | if (dm_request_based(md)) |
eca7ee6d | 2671 | dm_start_queue(md->queue); |
cec47e3d | 2672 | |
2ca3310e | 2673 | unlock_fs(md); |
d67ee213 | 2674 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2675 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2676 | } |
1da177e4 | 2677 | |
ffcc3936 MS |
2678 | return r; |
2679 | } | |
2680 | ||
2681 | /* | |
2682 | * We need to be able to change a mapping table under a mounted | |
2683 | * filesystem. For example we might want to move some data in | |
2684 | * the background. Before the table can be swapped with | |
2685 | * dm_bind_table, dm_suspend must be called to flush any in | |
2686 | * flight bios and ensure that any further io gets deferred. | |
2687 | */ | |
2688 | /* | |
2689 | * Suspend mechanism in request-based dm. | |
2690 | * | |
2691 | * 1. Flush all I/Os by lock_fs() if needed. | |
2692 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2693 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2694 | * | |
2695 | * To abort suspend, start the request_queue. | |
2696 | */ | |
2697 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2698 | { | |
2699 | struct dm_table *map = NULL; | |
2700 | int r = 0; | |
2701 | ||
2702 | retry: | |
2703 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2704 | ||
2705 | if (dm_suspended_md(md)) { | |
2706 | r = -EINVAL; | |
2707 | goto out_unlock; | |
2708 | } | |
2709 | ||
2710 | if (dm_suspended_internally_md(md)) { | |
2711 | /* already internally suspended, wait for internal resume */ | |
2712 | mutex_unlock(&md->suspend_lock); | |
2713 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2714 | if (r) | |
2715 | return r; | |
2716 | goto retry; | |
2717 | } | |
2718 | ||
a12f5d48 | 2719 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2720 | |
eaf9a736 | 2721 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2722 | if (r) |
2723 | goto out_unlock; | |
3b00b203 | 2724 | |
5df96f2b | 2725 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
4d4471cb | 2726 | dm_table_postsuspend_targets(map); |
5df96f2b | 2727 | clear_bit(DMF_POST_SUSPENDING, &md->flags); |
4d4471cb | 2728 | |
d287483d | 2729 | out_unlock: |
e61290a4 | 2730 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2731 | return r; |
1da177e4 LT |
2732 | } |
2733 | ||
ffcc3936 MS |
2734 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2735 | { | |
2736 | if (map) { | |
2737 | int r = dm_table_resume_targets(map); | |
2738 | if (r) | |
2739 | return r; | |
2740 | } | |
2741 | ||
2742 | dm_queue_flush(md); | |
2743 | ||
2744 | /* | |
2745 | * Flushing deferred I/Os must be done after targets are resumed | |
2746 | * so that mapping of targets can work correctly. | |
2747 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2748 | */ | |
2749 | if (dm_request_based(md)) | |
eca7ee6d | 2750 | dm_start_queue(md->queue); |
ffcc3936 MS |
2751 | |
2752 | unlock_fs(md); | |
2753 | ||
2754 | return 0; | |
2755 | } | |
2756 | ||
1da177e4 LT |
2757 | int dm_resume(struct mapped_device *md) |
2758 | { | |
8dc23658 | 2759 | int r; |
cf222b37 | 2760 | struct dm_table *map = NULL; |
1da177e4 | 2761 | |
ffcc3936 | 2762 | retry: |
8dc23658 | 2763 | r = -EINVAL; |
ffcc3936 MS |
2764 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2765 | ||
4f186f8b | 2766 | if (!dm_suspended_md(md)) |
cf222b37 | 2767 | goto out; |
cf222b37 | 2768 | |
ffcc3936 MS |
2769 | if (dm_suspended_internally_md(md)) { |
2770 | /* already internally suspended, wait for internal resume */ | |
2771 | mutex_unlock(&md->suspend_lock); | |
2772 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2773 | if (r) | |
2774 | return r; | |
2775 | goto retry; | |
2776 | } | |
2777 | ||
a12f5d48 | 2778 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2779 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2780 | goto out; |
1da177e4 | 2781 | |
ffcc3936 | 2782 | r = __dm_resume(md, map); |
8757b776 MB |
2783 | if (r) |
2784 | goto out; | |
2ca3310e | 2785 | |
2ca3310e | 2786 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2787 | out: |
e61290a4 | 2788 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2789 | |
cf222b37 | 2790 | return r; |
1da177e4 LT |
2791 | } |
2792 | ||
fd2ed4d2 MP |
2793 | /* |
2794 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2795 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2796 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2797 | */ |
2798 | ||
ffcc3936 | 2799 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2800 | { |
ffcc3936 MS |
2801 | struct dm_table *map = NULL; |
2802 | ||
1ea0654e BVA |
2803 | lockdep_assert_held(&md->suspend_lock); |
2804 | ||
96b26c8c | 2805 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2806 | return; /* nested internal suspend */ |
2807 | ||
2808 | if (dm_suspended_md(md)) { | |
2809 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2810 | return; /* nest suspend */ | |
2811 | } | |
2812 | ||
a12f5d48 | 2813 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2814 | |
2815 | /* | |
2816 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2817 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2818 | * would require changing .presuspend to return an error -- avoid this | |
2819 | * until there is a need for more elaborate variants of internal suspend. | |
2820 | */ | |
eaf9a736 MS |
2821 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2822 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 | 2823 | |
5df96f2b | 2824 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
ffcc3936 | 2825 | dm_table_postsuspend_targets(map); |
5df96f2b | 2826 | clear_bit(DMF_POST_SUSPENDING, &md->flags); |
ffcc3936 MS |
2827 | } |
2828 | ||
2829 | static void __dm_internal_resume(struct mapped_device *md) | |
2830 | { | |
96b26c8c MP |
2831 | BUG_ON(!md->internal_suspend_count); |
2832 | ||
2833 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2834 | return; /* resume from nested internal suspend */ |
2835 | ||
fd2ed4d2 | 2836 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2837 | goto done; /* resume from nested suspend */ |
2838 | ||
2839 | /* | |
2840 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2841 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2842 | */ | |
2843 | (void) __dm_resume(md, NULL); | |
2844 | ||
2845 | done: | |
2846 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2847 | smp_mb__after_atomic(); | |
2848 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2849 | } | |
2850 | ||
2851 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2852 | { | |
2853 | mutex_lock(&md->suspend_lock); | |
2854 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2855 | mutex_unlock(&md->suspend_lock); | |
2856 | } | |
2857 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2858 | ||
2859 | void dm_internal_resume(struct mapped_device *md) | |
2860 | { | |
2861 | mutex_lock(&md->suspend_lock); | |
2862 | __dm_internal_resume(md); | |
2863 | mutex_unlock(&md->suspend_lock); | |
2864 | } | |
2865 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2866 | ||
2867 | /* | |
2868 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2869 | * which prevents interaction with userspace-driven suspend. | |
2870 | */ | |
2871 | ||
2872 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2873 | { | |
2874 | mutex_lock(&md->suspend_lock); | |
2875 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2876 | return; |
2877 | ||
2878 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2879 | synchronize_srcu(&md->io_barrier); | |
2880 | flush_workqueue(md->wq); | |
2881 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2882 | } | |
b735fede | 2883 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2884 | |
ffcc3936 | 2885 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2886 | { |
ffcc3936 | 2887 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2888 | goto done; |
2889 | ||
2890 | dm_queue_flush(md); | |
2891 | ||
2892 | done: | |
2893 | mutex_unlock(&md->suspend_lock); | |
2894 | } | |
b735fede | 2895 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2896 | |
1da177e4 LT |
2897 | /*----------------------------------------------------------------- |
2898 | * Event notification. | |
2899 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2900 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2901 | unsigned cookie) |
69267a30 | 2902 | { |
6958c1c6 MP |
2903 | int r; |
2904 | unsigned noio_flag; | |
60935eb2 MB |
2905 | char udev_cookie[DM_COOKIE_LENGTH]; |
2906 | char *envp[] = { udev_cookie, NULL }; | |
2907 | ||
6958c1c6 MP |
2908 | noio_flag = memalloc_noio_save(); |
2909 | ||
60935eb2 | 2910 | if (!cookie) |
6958c1c6 | 2911 | r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2912 | else { |
2913 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2914 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
6958c1c6 MP |
2915 | r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2916 | action, envp); | |
60935eb2 | 2917 | } |
6958c1c6 MP |
2918 | |
2919 | memalloc_noio_restore(noio_flag); | |
2920 | ||
2921 | return r; | |
69267a30 AK |
2922 | } |
2923 | ||
7a8c3d3b MA |
2924 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2925 | { | |
2926 | return atomic_add_return(1, &md->uevent_seq); | |
2927 | } | |
2928 | ||
1da177e4 LT |
2929 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2930 | { | |
2931 | return atomic_read(&md->event_nr); | |
2932 | } | |
2933 | ||
2934 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2935 | { | |
2936 | return wait_event_interruptible(md->eventq, | |
2937 | (event_nr != atomic_read(&md->event_nr))); | |
2938 | } | |
2939 | ||
7a8c3d3b MA |
2940 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2941 | { | |
2942 | unsigned long flags; | |
2943 | ||
2944 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2945 | list_add(elist, &md->uevent_list); | |
2946 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2947 | } | |
2948 | ||
1da177e4 LT |
2949 | /* |
2950 | * The gendisk is only valid as long as you have a reference | |
2951 | * count on 'md'. | |
2952 | */ | |
2953 | struct gendisk *dm_disk(struct mapped_device *md) | |
2954 | { | |
2955 | return md->disk; | |
2956 | } | |
65ff5b7d | 2957 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2958 | |
784aae73 MB |
2959 | struct kobject *dm_kobject(struct mapped_device *md) |
2960 | { | |
2995fa78 | 2961 | return &md->kobj_holder.kobj; |
784aae73 MB |
2962 | } |
2963 | ||
784aae73 MB |
2964 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2965 | { | |
2966 | struct mapped_device *md; | |
2967 | ||
2995fa78 | 2968 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2969 | |
b9a41d21 HT |
2970 | spin_lock(&_minor_lock); |
2971 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2972 | md = NULL; | |
2973 | goto out; | |
2974 | } | |
784aae73 | 2975 | dm_get(md); |
b9a41d21 HT |
2976 | out: |
2977 | spin_unlock(&_minor_lock); | |
2978 | ||
784aae73 MB |
2979 | return md; |
2980 | } | |
2981 | ||
4f186f8b | 2982 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2983 | { |
2984 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2985 | } | |
2986 | ||
5df96f2b MP |
2987 | static int dm_post_suspending_md(struct mapped_device *md) |
2988 | { | |
2989 | return test_bit(DMF_POST_SUSPENDING, &md->flags); | |
2990 | } | |
2991 | ||
ffcc3936 MS |
2992 | int dm_suspended_internally_md(struct mapped_device *md) |
2993 | { | |
2994 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2995 | } | |
2996 | ||
2c140a24 MP |
2997 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2998 | { | |
2999 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3000 | } | |
3001 | ||
64dbce58 KU |
3002 | int dm_suspended(struct dm_target *ti) |
3003 | { | |
ecdb2e25 | 3004 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3005 | } |
3006 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3007 | ||
5df96f2b MP |
3008 | int dm_post_suspending(struct dm_target *ti) |
3009 | { | |
3010 | return dm_post_suspending_md(dm_table_get_md(ti->table)); | |
3011 | } | |
3012 | EXPORT_SYMBOL_GPL(dm_post_suspending); | |
3013 | ||
2e93ccc1 KU |
3014 | int dm_noflush_suspending(struct dm_target *ti) |
3015 | { | |
ecdb2e25 | 3016 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3017 | } |
3018 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3019 | ||
7e0d574f | 3020 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
0776aa0e MS |
3021 | unsigned integrity, unsigned per_io_data_size, |
3022 | unsigned min_pool_size) | |
e6ee8c0b | 3023 | { |
115485e8 | 3024 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 3025 | unsigned int pool_size = 0; |
64f52b0e | 3026 | unsigned int front_pad, io_front_pad; |
6f1c819c | 3027 | int ret; |
e6ee8c0b KU |
3028 | |
3029 | if (!pools) | |
4e6e36c3 | 3030 | return NULL; |
e6ee8c0b | 3031 | |
78d8e58a MS |
3032 | switch (type) { |
3033 | case DM_TYPE_BIO_BASED: | |
545ed20e | 3034 | case DM_TYPE_DAX_BIO_BASED: |
22c11858 | 3035 | case DM_TYPE_NVME_BIO_BASED: |
0776aa0e | 3036 | pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); |
30187e1d | 3037 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
64f52b0e | 3038 | io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio); |
6f1c819c KO |
3039 | ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0); |
3040 | if (ret) | |
64f52b0e | 3041 | goto out; |
6f1c819c | 3042 | if (integrity && bioset_integrity_create(&pools->io_bs, pool_size)) |
eb8db831 | 3043 | goto out; |
78d8e58a MS |
3044 | break; |
3045 | case DM_TYPE_REQUEST_BASED: | |
0776aa0e | 3046 | pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); |
78d8e58a | 3047 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 3048 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
3049 | break; |
3050 | default: | |
3051 | BUG(); | |
3052 | } | |
3053 | ||
6f1c819c KO |
3054 | ret = bioset_init(&pools->bs, pool_size, front_pad, 0); |
3055 | if (ret) | |
5f015204 | 3056 | goto out; |
e6ee8c0b | 3057 | |
6f1c819c | 3058 | if (integrity && bioset_integrity_create(&pools->bs, pool_size)) |
5f015204 | 3059 | goto out; |
a91a2785 | 3060 | |
e6ee8c0b | 3061 | return pools; |
5f1b670d | 3062 | |
5f1b670d CH |
3063 | out: |
3064 | dm_free_md_mempools(pools); | |
78d8e58a | 3065 | |
4e6e36c3 | 3066 | return NULL; |
e6ee8c0b KU |
3067 | } |
3068 | ||
3069 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3070 | { | |
3071 | if (!pools) | |
3072 | return; | |
3073 | ||
6f1c819c KO |
3074 | bioset_exit(&pools->bs); |
3075 | bioset_exit(&pools->io_bs); | |
e6ee8c0b KU |
3076 | |
3077 | kfree(pools); | |
3078 | } | |
3079 | ||
9c72bad1 CH |
3080 | struct dm_pr { |
3081 | u64 old_key; | |
3082 | u64 new_key; | |
3083 | u32 flags; | |
3084 | bool fail_early; | |
3085 | }; | |
3086 | ||
3087 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
3088 | void *data) | |
71cdb697 CH |
3089 | { |
3090 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
3091 | struct dm_table *table; |
3092 | struct dm_target *ti; | |
3093 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 3094 | |
9c72bad1 CH |
3095 | table = dm_get_live_table(md, &srcu_idx); |
3096 | if (!table || !dm_table_get_size(table)) | |
3097 | goto out; | |
71cdb697 | 3098 | |
9c72bad1 CH |
3099 | /* We only support devices that have a single target */ |
3100 | if (dm_table_get_num_targets(table) != 1) | |
3101 | goto out; | |
3102 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 3103 | |
9c72bad1 CH |
3104 | ret = -EINVAL; |
3105 | if (!ti->type->iterate_devices) | |
3106 | goto out; | |
3107 | ||
3108 | ret = ti->type->iterate_devices(ti, fn, data); | |
3109 | out: | |
3110 | dm_put_live_table(md, srcu_idx); | |
3111 | return ret; | |
3112 | } | |
3113 | ||
3114 | /* | |
3115 | * For register / unregister we need to manually call out to every path. | |
3116 | */ | |
3117 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
3118 | sector_t start, sector_t len, void *data) | |
3119 | { | |
3120 | struct dm_pr *pr = data; | |
3121 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3122 | ||
3123 | if (!ops || !ops->pr_register) | |
3124 | return -EOPNOTSUPP; | |
3125 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
3126 | } | |
3127 | ||
3128 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
3129 | u32 flags) | |
3130 | { | |
3131 | struct dm_pr pr = { | |
3132 | .old_key = old_key, | |
3133 | .new_key = new_key, | |
3134 | .flags = flags, | |
3135 | .fail_early = true, | |
3136 | }; | |
3137 | int ret; | |
3138 | ||
3139 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
3140 | if (ret && new_key) { | |
3141 | /* unregister all paths if we failed to register any path */ | |
3142 | pr.old_key = new_key; | |
3143 | pr.new_key = 0; | |
3144 | pr.flags = 0; | |
3145 | pr.fail_early = false; | |
3146 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
3147 | } | |
3148 | ||
3149 | return ret; | |
71cdb697 CH |
3150 | } |
3151 | ||
3152 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3153 | u32 flags) |
71cdb697 CH |
3154 | { |
3155 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3156 | const struct pr_ops *ops; | |
971888c4 | 3157 | int r, srcu_idx; |
71cdb697 | 3158 | |
5bd5e8d8 | 3159 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3160 | if (r < 0) |
971888c4 | 3161 | goto out; |
71cdb697 CH |
3162 | |
3163 | ops = bdev->bd_disk->fops->pr_ops; | |
3164 | if (ops && ops->pr_reserve) | |
3165 | r = ops->pr_reserve(bdev, key, type, flags); | |
3166 | else | |
3167 | r = -EOPNOTSUPP; | |
971888c4 MS |
3168 | out: |
3169 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3170 | return r; |
3171 | } | |
3172 | ||
3173 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3174 | { | |
3175 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3176 | const struct pr_ops *ops; | |
971888c4 | 3177 | int r, srcu_idx; |
71cdb697 | 3178 | |
5bd5e8d8 | 3179 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3180 | if (r < 0) |
971888c4 | 3181 | goto out; |
71cdb697 CH |
3182 | |
3183 | ops = bdev->bd_disk->fops->pr_ops; | |
3184 | if (ops && ops->pr_release) | |
3185 | r = ops->pr_release(bdev, key, type); | |
3186 | else | |
3187 | r = -EOPNOTSUPP; | |
971888c4 MS |
3188 | out: |
3189 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3190 | return r; |
3191 | } | |
3192 | ||
3193 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3194 | enum pr_type type, bool abort) |
71cdb697 CH |
3195 | { |
3196 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3197 | const struct pr_ops *ops; | |
971888c4 | 3198 | int r, srcu_idx; |
71cdb697 | 3199 | |
5bd5e8d8 | 3200 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3201 | if (r < 0) |
971888c4 | 3202 | goto out; |
71cdb697 CH |
3203 | |
3204 | ops = bdev->bd_disk->fops->pr_ops; | |
3205 | if (ops && ops->pr_preempt) | |
3206 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3207 | else | |
3208 | r = -EOPNOTSUPP; | |
971888c4 MS |
3209 | out: |
3210 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3211 | return r; |
3212 | } | |
3213 | ||
3214 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3215 | { | |
3216 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3217 | const struct pr_ops *ops; | |
971888c4 | 3218 | int r, srcu_idx; |
71cdb697 | 3219 | |
5bd5e8d8 | 3220 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3221 | if (r < 0) |
971888c4 | 3222 | goto out; |
71cdb697 CH |
3223 | |
3224 | ops = bdev->bd_disk->fops->pr_ops; | |
3225 | if (ops && ops->pr_clear) | |
3226 | r = ops->pr_clear(bdev, key); | |
3227 | else | |
3228 | r = -EOPNOTSUPP; | |
971888c4 MS |
3229 | out: |
3230 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3231 | return r; |
3232 | } | |
3233 | ||
3234 | static const struct pr_ops dm_pr_ops = { | |
3235 | .pr_register = dm_pr_register, | |
3236 | .pr_reserve = dm_pr_reserve, | |
3237 | .pr_release = dm_pr_release, | |
3238 | .pr_preempt = dm_pr_preempt, | |
3239 | .pr_clear = dm_pr_clear, | |
3240 | }; | |
3241 | ||
83d5cde4 | 3242 | static const struct block_device_operations dm_blk_dops = { |
c62b37d9 | 3243 | .submit_bio = dm_submit_bio, |
1da177e4 LT |
3244 | .open = dm_blk_open, |
3245 | .release = dm_blk_close, | |
aa129a22 | 3246 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3247 | .getgeo = dm_blk_getgeo, |
e76239a3 | 3248 | .report_zones = dm_blk_report_zones, |
71cdb697 | 3249 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3250 | .owner = THIS_MODULE |
3251 | }; | |
3252 | ||
f26c5719 DW |
3253 | static const struct dax_operations dm_dax_ops = { |
3254 | .direct_access = dm_dax_direct_access, | |
7bf7eac8 | 3255 | .dax_supported = dm_dax_supported, |
7e026c8c | 3256 | .copy_from_iter = dm_dax_copy_from_iter, |
b3a9a0c3 | 3257 | .copy_to_iter = dm_dax_copy_to_iter, |
cdf6cdcd | 3258 | .zero_page_range = dm_dax_zero_page_range, |
f26c5719 DW |
3259 | }; |
3260 | ||
1da177e4 LT |
3261 | /* |
3262 | * module hooks | |
3263 | */ | |
3264 | module_init(dm_init); | |
3265 | module_exit(dm_exit); | |
3266 | ||
3267 | module_param(major, uint, 0); | |
3268 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3269 | |
e8603136 MS |
3270 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3271 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3272 | ||
115485e8 MS |
3273 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
3274 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
3275 | ||
1da177e4 LT |
3276 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3277 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3278 | MODULE_LICENSE("GPL"); |