]>
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 | ||
3ac51e74 DW |
425 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
426 | { | |
427 | struct mapped_device *md = bdev->bd_disk->private_data; | |
428 | ||
429 | return dm_get_geometry(md, geo); | |
430 | } | |
431 | ||
d4100351 CH |
432 | #ifdef CONFIG_BLK_DEV_ZONED |
433 | int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data) | |
434 | { | |
435 | struct dm_report_zones_args *args = data; | |
436 | sector_t sector_diff = args->tgt->begin - args->start; | |
437 | ||
438 | /* | |
439 | * Ignore zones beyond the target range. | |
440 | */ | |
441 | if (zone->start >= args->start + args->tgt->len) | |
442 | return 0; | |
443 | ||
444 | /* | |
445 | * Remap the start sector and write pointer position of the zone | |
446 | * to match its position in the target range. | |
447 | */ | |
448 | zone->start += sector_diff; | |
449 | if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { | |
450 | if (zone->cond == BLK_ZONE_COND_FULL) | |
451 | zone->wp = zone->start + zone->len; | |
452 | else if (zone->cond == BLK_ZONE_COND_EMPTY) | |
453 | zone->wp = zone->start; | |
454 | else | |
455 | zone->wp += sector_diff; | |
456 | } | |
457 | ||
458 | args->next_sector = zone->start + zone->len; | |
459 | return args->orig_cb(zone, args->zone_idx++, args->orig_data); | |
460 | } | |
461 | EXPORT_SYMBOL_GPL(dm_report_zones_cb); | |
462 | ||
e76239a3 | 463 | static int dm_blk_report_zones(struct gendisk *disk, sector_t sector, |
d4100351 | 464 | unsigned int nr_zones, report_zones_cb cb, void *data) |
e76239a3 | 465 | { |
e76239a3 | 466 | struct mapped_device *md = disk->private_data; |
e76239a3 CH |
467 | struct dm_table *map; |
468 | int srcu_idx, ret; | |
d4100351 CH |
469 | struct dm_report_zones_args args = { |
470 | .next_sector = sector, | |
471 | .orig_data = data, | |
472 | .orig_cb = cb, | |
473 | }; | |
e76239a3 CH |
474 | |
475 | if (dm_suspended_md(md)) | |
476 | return -EAGAIN; | |
477 | ||
478 | map = dm_get_live_table(md, &srcu_idx); | |
479 | if (!map) | |
480 | return -EIO; | |
481 | ||
d4100351 CH |
482 | do { |
483 | struct dm_target *tgt; | |
e76239a3 | 484 | |
d4100351 CH |
485 | tgt = dm_table_find_target(map, args.next_sector); |
486 | if (WARN_ON_ONCE(!tgt->type->report_zones)) { | |
487 | ret = -EIO; | |
488 | goto out; | |
489 | } | |
e76239a3 | 490 | |
d4100351 | 491 | args.tgt = tgt; |
a9cb9f41 JT |
492 | ret = tgt->type->report_zones(tgt, &args, |
493 | nr_zones - args.zone_idx); | |
d4100351 CH |
494 | if (ret < 0) |
495 | goto out; | |
496 | } while (args.zone_idx < nr_zones && | |
497 | args.next_sector < get_capacity(disk)); | |
e76239a3 | 498 | |
d4100351 | 499 | ret = args.zone_idx; |
e76239a3 CH |
500 | out: |
501 | dm_put_live_table(md, srcu_idx); | |
502 | return ret; | |
e76239a3 | 503 | } |
d4100351 CH |
504 | #else |
505 | #define dm_blk_report_zones NULL | |
506 | #endif /* CONFIG_BLK_DEV_ZONED */ | |
e76239a3 | 507 | |
971888c4 | 508 | static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx, |
5bd5e8d8 | 509 | struct block_device **bdev) |
971888c4 | 510 | __acquires(md->io_barrier) |
aa129a22 | 511 | { |
66482026 | 512 | struct dm_target *tgt; |
6c182cd8 | 513 | struct dm_table *map; |
971888c4 | 514 | int r; |
aa129a22 | 515 | |
6c182cd8 | 516 | retry: |
e56f81e0 | 517 | r = -ENOTTY; |
971888c4 | 518 | map = dm_get_live_table(md, srcu_idx); |
aa129a22 | 519 | if (!map || !dm_table_get_size(map)) |
971888c4 | 520 | return r; |
aa129a22 MB |
521 | |
522 | /* We only support devices that have a single target */ | |
523 | if (dm_table_get_num_targets(map) != 1) | |
971888c4 | 524 | return r; |
aa129a22 | 525 | |
66482026 MS |
526 | tgt = dm_table_get_target(map, 0); |
527 | if (!tgt->type->prepare_ioctl) | |
971888c4 | 528 | return r; |
519049af | 529 | |
971888c4 MS |
530 | if (dm_suspended_md(md)) |
531 | return -EAGAIN; | |
aa129a22 | 532 | |
5bd5e8d8 | 533 | r = tgt->type->prepare_ioctl(tgt, bdev); |
5bbbfdf6 | 534 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
971888c4 | 535 | dm_put_live_table(md, *srcu_idx); |
6c182cd8 HR |
536 | msleep(10); |
537 | goto retry; | |
538 | } | |
971888c4 | 539 | |
e56f81e0 CH |
540 | return r; |
541 | } | |
542 | ||
971888c4 MS |
543 | static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx) |
544 | __releases(md->io_barrier) | |
545 | { | |
546 | dm_put_live_table(md, srcu_idx); | |
547 | } | |
548 | ||
e56f81e0 CH |
549 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
550 | unsigned int cmd, unsigned long arg) | |
551 | { | |
552 | struct mapped_device *md = bdev->bd_disk->private_data; | |
971888c4 | 553 | int r, srcu_idx; |
e56f81e0 | 554 | |
5bd5e8d8 | 555 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
e56f81e0 | 556 | if (r < 0) |
971888c4 | 557 | goto out; |
6c182cd8 | 558 | |
e56f81e0 CH |
559 | if (r > 0) { |
560 | /* | |
e980f623 CH |
561 | * Target determined this ioctl is being issued against a |
562 | * subset of the parent bdev; require extra privileges. | |
e56f81e0 | 563 | */ |
e980f623 CH |
564 | if (!capable(CAP_SYS_RAWIO)) { |
565 | DMWARN_LIMIT( | |
566 | "%s: sending ioctl %x to DM device without required privilege.", | |
567 | current->comm, cmd); | |
568 | r = -ENOIOCTLCMD; | |
e56f81e0 | 569 | goto out; |
e980f623 | 570 | } |
e56f81e0 | 571 | } |
6c182cd8 | 572 | |
66482026 | 573 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 574 | out: |
971888c4 | 575 | dm_unprepare_ioctl(md, srcu_idx); |
aa129a22 MB |
576 | return r; |
577 | } | |
578 | ||
7465d7ac MS |
579 | u64 dm_start_time_ns_from_clone(struct bio *bio) |
580 | { | |
581 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
582 | struct dm_io *io = tio->io; | |
583 | ||
584 | return jiffies_to_nsecs(io->start_time); | |
585 | } | |
586 | EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone); | |
587 | ||
588 | static void start_io_acct(struct dm_io *io) | |
589 | { | |
590 | struct mapped_device *md = io->md; | |
591 | struct bio *bio = io->orig_bio; | |
592 | ||
593 | io->start_time = bio_start_io_acct(bio); | |
594 | if (unlikely(dm_stats_used(&md->stats))) | |
595 | dm_stats_account_io(&md->stats, bio_data_dir(bio), | |
596 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
597 | false, 0, &io->stats_aux); | |
598 | } | |
599 | ||
600 | static void end_io_acct(struct dm_io *io) | |
601 | { | |
602 | struct mapped_device *md = io->md; | |
603 | struct bio *bio = io->orig_bio; | |
604 | unsigned long duration = jiffies - io->start_time; | |
605 | ||
606 | bio_end_io_acct(bio, io->start_time); | |
607 | ||
608 | if (unlikely(dm_stats_used(&md->stats))) | |
609 | dm_stats_account_io(&md->stats, bio_data_dir(bio), | |
610 | bio->bi_iter.bi_sector, bio_sectors(bio), | |
611 | true, duration, &io->stats_aux); | |
612 | ||
613 | /* nudge anyone waiting on suspend queue */ | |
614 | if (unlikely(wq_has_sleeper(&md->wait))) | |
615 | wake_up(&md->wait); | |
616 | } | |
978e51ba MS |
617 | |
618 | static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio) | |
1da177e4 | 619 | { |
64f52b0e MS |
620 | struct dm_io *io; |
621 | struct dm_target_io *tio; | |
622 | struct bio *clone; | |
623 | ||
6f1c819c | 624 | clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs); |
64f52b0e MS |
625 | if (!clone) |
626 | return NULL; | |
627 | ||
628 | tio = container_of(clone, struct dm_target_io, clone); | |
629 | tio->inside_dm_io = true; | |
630 | tio->io = NULL; | |
631 | ||
632 | io = container_of(tio, struct dm_io, tio); | |
633 | io->magic = DM_IO_MAGIC; | |
978e51ba MS |
634 | io->status = 0; |
635 | atomic_set(&io->io_count, 1); | |
636 | io->orig_bio = bio; | |
637 | io->md = md; | |
638 | spin_lock_init(&io->endio_lock); | |
639 | ||
640 | start_io_acct(io); | |
64f52b0e MS |
641 | |
642 | return io; | |
1da177e4 LT |
643 | } |
644 | ||
028867ac | 645 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 | 646 | { |
64f52b0e MS |
647 | bio_put(&io->tio.clone); |
648 | } | |
649 | ||
650 | static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti, | |
651 | unsigned target_bio_nr, gfp_t gfp_mask) | |
652 | { | |
653 | struct dm_target_io *tio; | |
654 | ||
655 | if (!ci->io->tio.io) { | |
656 | /* the dm_target_io embedded in ci->io is available */ | |
657 | tio = &ci->io->tio; | |
658 | } else { | |
6f1c819c | 659 | struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs); |
64f52b0e MS |
660 | if (!clone) |
661 | return NULL; | |
662 | ||
663 | tio = container_of(clone, struct dm_target_io, clone); | |
664 | tio->inside_dm_io = false; | |
665 | } | |
666 | ||
667 | tio->magic = DM_TIO_MAGIC; | |
668 | tio->io = ci->io; | |
669 | tio->ti = ti; | |
670 | tio->target_bio_nr = target_bio_nr; | |
671 | ||
672 | return tio; | |
1da177e4 LT |
673 | } |
674 | ||
cfae7529 | 675 | static void free_tio(struct dm_target_io *tio) |
1da177e4 | 676 | { |
64f52b0e MS |
677 | if (tio->inside_dm_io) |
678 | return; | |
dba14160 | 679 | bio_put(&tio->clone); |
1da177e4 LT |
680 | } |
681 | ||
682 | /* | |
683 | * Add the bio to the list of deferred io. | |
684 | */ | |
92c63902 | 685 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 686 | { |
05447420 | 687 | unsigned long flags; |
1da177e4 | 688 | |
05447420 | 689 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 690 | bio_list_add(&md->deferred, bio); |
05447420 | 691 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 692 | queue_work(md->wq, &md->work); |
1da177e4 LT |
693 | } |
694 | ||
695 | /* | |
696 | * Everyone (including functions in this file), should use this | |
697 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 698 | * dm_put_live_table() when finished. |
1da177e4 | 699 | */ |
83d5e5b0 | 700 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 701 | { |
83d5e5b0 MP |
702 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
703 | ||
704 | return srcu_dereference(md->map, &md->io_barrier); | |
705 | } | |
1da177e4 | 706 | |
83d5e5b0 MP |
707 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
708 | { | |
709 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
710 | } | |
711 | ||
712 | void dm_sync_table(struct mapped_device *md) | |
713 | { | |
714 | synchronize_srcu(&md->io_barrier); | |
715 | synchronize_rcu_expedited(); | |
716 | } | |
717 | ||
718 | /* | |
719 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
720 | * The caller must not block between these two functions. | |
721 | */ | |
722 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
723 | { | |
724 | rcu_read_lock(); | |
725 | return rcu_dereference(md->map); | |
726 | } | |
1da177e4 | 727 | |
83d5e5b0 MP |
728 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
729 | { | |
730 | rcu_read_unlock(); | |
1da177e4 LT |
731 | } |
732 | ||
971888c4 MS |
733 | static char *_dm_claim_ptr = "I belong to device-mapper"; |
734 | ||
86f1152b BM |
735 | /* |
736 | * Open a table device so we can use it as a map destination. | |
737 | */ | |
738 | static int open_table_device(struct table_device *td, dev_t dev, | |
739 | struct mapped_device *md) | |
740 | { | |
86f1152b BM |
741 | struct block_device *bdev; |
742 | ||
743 | int r; | |
744 | ||
745 | BUG_ON(td->dm_dev.bdev); | |
746 | ||
519049af | 747 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr); |
86f1152b BM |
748 | if (IS_ERR(bdev)) |
749 | return PTR_ERR(bdev); | |
750 | ||
751 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
752 | if (r) { | |
753 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
754 | return r; | |
755 | } | |
756 | ||
757 | td->dm_dev.bdev = bdev; | |
817bf402 | 758 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
86f1152b BM |
759 | return 0; |
760 | } | |
761 | ||
762 | /* | |
763 | * Close a table device that we've been using. | |
764 | */ | |
765 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
766 | { | |
767 | if (!td->dm_dev.bdev) | |
768 | return; | |
769 | ||
770 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
771 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
817bf402 | 772 | put_dax(td->dm_dev.dax_dev); |
86f1152b | 773 | td->dm_dev.bdev = NULL; |
817bf402 | 774 | td->dm_dev.dax_dev = NULL; |
86f1152b BM |
775 | } |
776 | ||
777 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
8454fca4 SS |
778 | fmode_t mode) |
779 | { | |
86f1152b BM |
780 | struct table_device *td; |
781 | ||
782 | list_for_each_entry(td, l, list) | |
783 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
784 | return td; | |
785 | ||
786 | return NULL; | |
787 | } | |
788 | ||
789 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
8454fca4 SS |
790 | struct dm_dev **result) |
791 | { | |
86f1152b BM |
792 | int r; |
793 | struct table_device *td; | |
794 | ||
795 | mutex_lock(&md->table_devices_lock); | |
796 | td = find_table_device(&md->table_devices, dev, mode); | |
797 | if (!td) { | |
115485e8 | 798 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
86f1152b BM |
799 | if (!td) { |
800 | mutex_unlock(&md->table_devices_lock); | |
801 | return -ENOMEM; | |
802 | } | |
803 | ||
804 | td->dm_dev.mode = mode; | |
805 | td->dm_dev.bdev = NULL; | |
806 | ||
807 | if ((r = open_table_device(td, dev, md))) { | |
808 | mutex_unlock(&md->table_devices_lock); | |
809 | kfree(td); | |
810 | return r; | |
811 | } | |
812 | ||
813 | format_dev_t(td->dm_dev.name, dev); | |
814 | ||
b0b4d7c6 | 815 | refcount_set(&td->count, 1); |
86f1152b | 816 | list_add(&td->list, &md->table_devices); |
b0b4d7c6 ER |
817 | } else { |
818 | refcount_inc(&td->count); | |
86f1152b | 819 | } |
86f1152b BM |
820 | mutex_unlock(&md->table_devices_lock); |
821 | ||
822 | *result = &td->dm_dev; | |
823 | return 0; | |
824 | } | |
825 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
826 | ||
827 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
828 | { | |
829 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
830 | ||
831 | mutex_lock(&md->table_devices_lock); | |
b0b4d7c6 | 832 | if (refcount_dec_and_test(&td->count)) { |
86f1152b BM |
833 | close_table_device(td, md); |
834 | list_del(&td->list); | |
835 | kfree(td); | |
836 | } | |
837 | mutex_unlock(&md->table_devices_lock); | |
838 | } | |
839 | EXPORT_SYMBOL(dm_put_table_device); | |
840 | ||
841 | static void free_table_devices(struct list_head *devices) | |
842 | { | |
843 | struct list_head *tmp, *next; | |
844 | ||
845 | list_for_each_safe(tmp, next, devices) { | |
846 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
847 | ||
848 | DMWARN("dm_destroy: %s still exists with %d references", | |
b0b4d7c6 | 849 | td->dm_dev.name, refcount_read(&td->count)); |
86f1152b BM |
850 | kfree(td); |
851 | } | |
852 | } | |
853 | ||
3ac51e74 DW |
854 | /* |
855 | * Get the geometry associated with a dm device | |
856 | */ | |
857 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
858 | { | |
859 | *geo = md->geometry; | |
860 | ||
861 | return 0; | |
862 | } | |
863 | ||
864 | /* | |
865 | * Set the geometry of a device. | |
866 | */ | |
867 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
868 | { | |
869 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
870 | ||
871 | if (geo->start > sz) { | |
872 | DMWARN("Start sector is beyond the geometry limits."); | |
873 | return -EINVAL; | |
874 | } | |
875 | ||
876 | md->geometry = *geo; | |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
2e93ccc1 KU |
881 | static int __noflush_suspending(struct mapped_device *md) |
882 | { | |
883 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
884 | } | |
885 | ||
1da177e4 LT |
886 | /* |
887 | * Decrements the number of outstanding ios that a bio has been | |
888 | * cloned into, completing the original io if necc. | |
889 | */ | |
4e4cbee9 | 890 | static void dec_pending(struct dm_io *io, blk_status_t error) |
1da177e4 | 891 | { |
2e93ccc1 | 892 | unsigned long flags; |
4e4cbee9 | 893 | blk_status_t io_error; |
b35f8caa MB |
894 | struct bio *bio; |
895 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
896 | |
897 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
898 | if (unlikely(error)) { |
899 | spin_lock_irqsave(&io->endio_lock, flags); | |
745dc570 | 900 | if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md))) |
4e4cbee9 | 901 | io->status = error; |
f88fb981 KU |
902 | spin_unlock_irqrestore(&io->endio_lock, flags); |
903 | } | |
1da177e4 LT |
904 | |
905 | if (atomic_dec_and_test(&io->io_count)) { | |
4e4cbee9 | 906 | if (io->status == BLK_STS_DM_REQUEUE) { |
2e93ccc1 KU |
907 | /* |
908 | * Target requested pushing back the I/O. | |
2e93ccc1 | 909 | */ |
022c2611 | 910 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 | 911 | if (__noflush_suspending(md)) |
745dc570 MS |
912 | /* NOTE early return due to BLK_STS_DM_REQUEUE below */ |
913 | bio_list_add_head(&md->deferred, io->orig_bio); | |
6a8736d1 | 914 | else |
2e93ccc1 | 915 | /* noflush suspend was interrupted. */ |
4e4cbee9 | 916 | io->status = BLK_STS_IOERR; |
022c2611 | 917 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
918 | } |
919 | ||
4e4cbee9 | 920 | io_error = io->status; |
745dc570 | 921 | bio = io->orig_bio; |
6a8736d1 TH |
922 | end_io_acct(io); |
923 | free_io(md, io); | |
924 | ||
4e4cbee9 | 925 | if (io_error == BLK_STS_DM_REQUEUE) |
6a8736d1 | 926 | return; |
2e93ccc1 | 927 | |
1eff9d32 | 928 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 929 | /* |
6a8736d1 | 930 | * Preflush done for flush with data, reissue |
28a8f0d3 | 931 | * without REQ_PREFLUSH. |
af7e466a | 932 | */ |
1eff9d32 | 933 | bio->bi_opf &= ~REQ_PREFLUSH; |
6a8736d1 | 934 | queue_io(md, bio); |
af7e466a | 935 | } else { |
b372d360 | 936 | /* done with normal IO or empty flush */ |
8dd601fa N |
937 | if (io_error) |
938 | bio->bi_status = io_error; | |
4246a0b6 | 939 | bio_endio(bio); |
b35f8caa | 940 | } |
1da177e4 LT |
941 | } |
942 | } | |
943 | ||
bcb44433 MS |
944 | void disable_discard(struct mapped_device *md) |
945 | { | |
946 | struct queue_limits *limits = dm_get_queue_limits(md); | |
947 | ||
948 | /* device doesn't really support DISCARD, disable it */ | |
949 | limits->max_discard_sectors = 0; | |
950 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue); | |
951 | } | |
952 | ||
4cc96131 | 953 | void disable_write_same(struct mapped_device *md) |
7eee4ae2 MS |
954 | { |
955 | struct queue_limits *limits = dm_get_queue_limits(md); | |
956 | ||
957 | /* device doesn't really support WRITE SAME, disable it */ | |
958 | limits->max_write_same_sectors = 0; | |
959 | } | |
960 | ||
ac62d620 CH |
961 | void disable_write_zeroes(struct mapped_device *md) |
962 | { | |
963 | struct queue_limits *limits = dm_get_queue_limits(md); | |
964 | ||
965 | /* device doesn't really support WRITE ZEROES, disable it */ | |
966 | limits->max_write_zeroes_sectors = 0; | |
967 | } | |
968 | ||
4246a0b6 | 969 | static void clone_endio(struct bio *bio) |
1da177e4 | 970 | { |
4e4cbee9 | 971 | blk_status_t error = bio->bi_status; |
bfc6d41c | 972 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 973 | struct dm_io *io = tio->io; |
9faf400f | 974 | struct mapped_device *md = tio->io->md; |
1da177e4 | 975 | dm_endio_fn endio = tio->ti->type->end_io; |
415c79e1 | 976 | struct bio *orig_bio = io->orig_bio; |
1da177e4 | 977 | |
978e51ba | 978 | if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) { |
bcb44433 MS |
979 | if (bio_op(bio) == REQ_OP_DISCARD && |
980 | !bio->bi_disk->queue->limits.max_discard_sectors) | |
981 | disable_discard(md); | |
982 | else if (bio_op(bio) == REQ_OP_WRITE_SAME && | |
983 | !bio->bi_disk->queue->limits.max_write_same_sectors) | |
ac62d620 | 984 | disable_write_same(md); |
bcb44433 MS |
985 | else if (bio_op(bio) == REQ_OP_WRITE_ZEROES && |
986 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) | |
ac62d620 CH |
987 | disable_write_zeroes(md); |
988 | } | |
7eee4ae2 | 989 | |
415c79e1 JT |
990 | /* |
991 | * For zone-append bios get offset in zone of the written | |
992 | * sector and add that to the original bio sector pos. | |
993 | */ | |
994 | if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) { | |
995 | sector_t written_sector = bio->bi_iter.bi_sector; | |
996 | struct request_queue *q = orig_bio->bi_disk->queue; | |
997 | u64 mask = (u64)blk_queue_zone_sectors(q) - 1; | |
998 | ||
999 | orig_bio->bi_iter.bi_sector += written_sector & mask; | |
1000 | } | |
1001 | ||
1be56909 | 1002 | if (endio) { |
4e4cbee9 | 1003 | int r = endio(tio->ti, bio, &error); |
1be56909 CH |
1004 | switch (r) { |
1005 | case DM_ENDIO_REQUEUE: | |
4e4cbee9 | 1006 | error = BLK_STS_DM_REQUEUE; |
df561f66 | 1007 | fallthrough; |
1be56909 CH |
1008 | case DM_ENDIO_DONE: |
1009 | break; | |
1010 | case DM_ENDIO_INCOMPLETE: | |
1011 | /* The target will handle the io */ | |
1012 | return; | |
1013 | default: | |
1014 | DMWARN("unimplemented target endio return value: %d", r); | |
1015 | BUG(); | |
1016 | } | |
1017 | } | |
1018 | ||
cfae7529 | 1019 | free_tio(tio); |
b35f8caa | 1020 | dec_pending(io, error); |
1da177e4 LT |
1021 | } |
1022 | ||
56a67df7 MS |
1023 | /* |
1024 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1025 | * target boundary. | |
1026 | */ | |
3720281d MS |
1027 | static inline sector_t max_io_len_target_boundary(struct dm_target *ti, |
1028 | sector_t target_offset) | |
56a67df7 | 1029 | { |
56a67df7 MS |
1030 | return ti->len - target_offset; |
1031 | } | |
1032 | ||
3720281d | 1033 | static sector_t max_io_len(struct dm_target *ti, sector_t sector) |
1da177e4 | 1034 | { |
3720281d MS |
1035 | sector_t target_offset = dm_target_offset(ti, sector); |
1036 | sector_t len = max_io_len_target_boundary(ti, target_offset); | |
5091cdec | 1037 | sector_t max_len; |
1da177e4 LT |
1038 | |
1039 | /* | |
542f9038 | 1040 | * Does the target need to split even further? |
5091cdec MS |
1041 | * - q->limits.chunk_sectors reflects ti->max_io_len so |
1042 | * blk_max_size_offset() provides required splitting. | |
1043 | * - blk_max_size_offset() also respects q->limits.max_sectors | |
1da177e4 | 1044 | */ |
33bd6f06 | 1045 | max_len = blk_max_size_offset(ti->table->md->queue, |
3720281d | 1046 | target_offset); |
5091cdec MS |
1047 | if (len > max_len) |
1048 | len = max_len; | |
1da177e4 LT |
1049 | |
1050 | return len; | |
1051 | } | |
1052 | ||
542f9038 MS |
1053 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1054 | { | |
1055 | if (len > UINT_MAX) { | |
1056 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1057 | (unsigned long long)len, UINT_MAX); | |
1058 | ti->error = "Maximum size of target IO is too large"; | |
1059 | return -EINVAL; | |
1060 | } | |
1061 | ||
75ae1936 | 1062 | ti->max_io_len = (uint32_t) len; |
542f9038 MS |
1063 | |
1064 | return 0; | |
1065 | } | |
1066 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1067 | ||
f26c5719 | 1068 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
3d97c829 MS |
1069 | sector_t sector, int *srcu_idx) |
1070 | __acquires(md->io_barrier) | |
545ed20e | 1071 | { |
545ed20e TK |
1072 | struct dm_table *map; |
1073 | struct dm_target *ti; | |
545ed20e | 1074 | |
f26c5719 | 1075 | map = dm_get_live_table(md, srcu_idx); |
545ed20e | 1076 | if (!map) |
f26c5719 | 1077 | return NULL; |
545ed20e TK |
1078 | |
1079 | ti = dm_table_find_target(map, sector); | |
123d87d5 | 1080 | if (!ti) |
f26c5719 | 1081 | return NULL; |
545ed20e | 1082 | |
f26c5719 DW |
1083 | return ti; |
1084 | } | |
545ed20e | 1085 | |
f26c5719 | 1086 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1087 | long nr_pages, void **kaddr, pfn_t *pfn) |
f26c5719 DW |
1088 | { |
1089 | struct mapped_device *md = dax_get_private(dax_dev); | |
1090 | sector_t sector = pgoff * PAGE_SECTORS; | |
1091 | struct dm_target *ti; | |
1092 | long len, ret = -EIO; | |
1093 | int srcu_idx; | |
545ed20e | 1094 | |
f26c5719 | 1095 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
545ed20e | 1096 | |
f26c5719 DW |
1097 | if (!ti) |
1098 | goto out; | |
1099 | if (!ti->type->direct_access) | |
1100 | goto out; | |
3720281d | 1101 | len = max_io_len(ti, sector) / PAGE_SECTORS; |
f26c5719 DW |
1102 | if (len < 1) |
1103 | goto out; | |
1104 | nr_pages = min(len, nr_pages); | |
dbc62659 | 1105 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
817bf402 | 1106 | |
f26c5719 | 1107 | out: |
545ed20e | 1108 | dm_put_live_table(md, srcu_idx); |
f26c5719 DW |
1109 | |
1110 | return ret; | |
545ed20e TK |
1111 | } |
1112 | ||
7bf7eac8 DW |
1113 | static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev, |
1114 | int blocksize, sector_t start, sector_t len) | |
1115 | { | |
1116 | struct mapped_device *md = dax_get_private(dax_dev); | |
1117 | struct dm_table *map; | |
02186d88 | 1118 | bool ret = false; |
7bf7eac8 | 1119 | int srcu_idx; |
7bf7eac8 DW |
1120 | |
1121 | map = dm_get_live_table(md, &srcu_idx); | |
1122 | if (!map) | |
02186d88 | 1123 | goto out; |
7bf7eac8 | 1124 | |
2e9ee095 | 1125 | ret = dm_table_supports_dax(map, device_supports_dax, &blocksize); |
7bf7eac8 | 1126 | |
02186d88 | 1127 | out: |
7bf7eac8 DW |
1128 | dm_put_live_table(md, srcu_idx); |
1129 | ||
1130 | return ret; | |
1131 | } | |
1132 | ||
7e026c8c | 1133 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
3d97c829 | 1134 | void *addr, size_t bytes, struct iov_iter *i) |
7e026c8c DW |
1135 | { |
1136 | struct mapped_device *md = dax_get_private(dax_dev); | |
1137 | sector_t sector = pgoff * PAGE_SECTORS; | |
1138 | struct dm_target *ti; | |
1139 | long ret = 0; | |
1140 | int srcu_idx; | |
1141 | ||
1142 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1143 | ||
1144 | if (!ti) | |
1145 | goto out; | |
1146 | if (!ti->type->dax_copy_from_iter) { | |
1147 | ret = copy_from_iter(addr, bytes, i); | |
1148 | goto out; | |
1149 | } | |
1150 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); | |
1151 | out: | |
1152 | dm_put_live_table(md, srcu_idx); | |
1153 | ||
1154 | return ret; | |
1155 | } | |
1156 | ||
b3a9a0c3 DW |
1157 | static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
1158 | void *addr, size_t bytes, struct iov_iter *i) | |
1159 | { | |
1160 | struct mapped_device *md = dax_get_private(dax_dev); | |
1161 | sector_t sector = pgoff * PAGE_SECTORS; | |
1162 | struct dm_target *ti; | |
1163 | long ret = 0; | |
1164 | int srcu_idx; | |
1165 | ||
1166 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1167 | ||
1168 | if (!ti) | |
1169 | goto out; | |
1170 | if (!ti->type->dax_copy_to_iter) { | |
1171 | ret = copy_to_iter(addr, bytes, i); | |
1172 | goto out; | |
1173 | } | |
1174 | ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i); | |
1175 | out: | |
1176 | dm_put_live_table(md, srcu_idx); | |
1177 | ||
1178 | return ret; | |
1179 | } | |
1180 | ||
cdf6cdcd VG |
1181 | static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff, |
1182 | size_t nr_pages) | |
1183 | { | |
1184 | struct mapped_device *md = dax_get_private(dax_dev); | |
1185 | sector_t sector = pgoff * PAGE_SECTORS; | |
1186 | struct dm_target *ti; | |
1187 | int ret = -EIO; | |
1188 | int srcu_idx; | |
1189 | ||
1190 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); | |
1191 | ||
1192 | if (!ti) | |
1193 | goto out; | |
1194 | if (WARN_ON(!ti->type->dax_zero_page_range)) { | |
1195 | /* | |
1196 | * ->zero_page_range() is mandatory dax operation. If we are | |
1197 | * here, something is wrong. | |
1198 | */ | |
1199 | dm_put_live_table(md, srcu_idx); | |
1200 | goto out; | |
1201 | } | |
1202 | ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages); | |
1203 | ||
1204 | out: | |
1205 | dm_put_live_table(md, srcu_idx); | |
1206 | ||
1207 | return ret; | |
1208 | } | |
1209 | ||
1dd40c3e MP |
1210 | /* |
1211 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
2e2d6f7e AJ |
1212 | * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_RESET, |
1213 | * REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH. | |
1dd40c3e MP |
1214 | * |
1215 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1216 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1217 | * sent in a next bio. | |
1218 | * | |
1219 | * A diagram that explains the arithmetics: | |
1220 | * +--------------------+---------------+-------+ | |
1221 | * | 1 | 2 | 3 | | |
1222 | * +--------------------+---------------+-------+ | |
1223 | * | |
1224 | * <-------------- *tio->len_ptr ---------------> | |
1225 | * <------- bi_size -------> | |
1226 | * <-- n_sectors --> | |
1227 | * | |
1228 | * Region 1 was already iterated over with bio_advance or similar function. | |
1229 | * (it may be empty if the target doesn't use bio_advance) | |
1230 | * Region 2 is the remaining bio size that the target wants to process. | |
1231 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1232 | * to make it empty) | |
1233 | * The target requires that region 3 is to be sent in the next bio. | |
1234 | * | |
1235 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1236 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1237 | * copies of the bio. | |
1238 | */ | |
1239 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1240 | { | |
1241 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1242 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1eff9d32 | 1243 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
1dd40c3e MP |
1244 | BUG_ON(bi_size > *tio->len_ptr); |
1245 | BUG_ON(n_sectors > bi_size); | |
1246 | *tio->len_ptr -= bi_size - n_sectors; | |
1247 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1248 | } | |
1249 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1250 | ||
978e51ba | 1251 | static blk_qc_t __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1252 | { |
1253 | int r; | |
2056a782 | 1254 | sector_t sector; |
dba14160 | 1255 | struct bio *clone = &tio->clone; |
64f52b0e | 1256 | struct dm_io *io = tio->io; |
bd2a49b8 | 1257 | struct dm_target *ti = tio->ti; |
978e51ba | 1258 | blk_qc_t ret = BLK_QC_T_NONE; |
1da177e4 | 1259 | |
1da177e4 | 1260 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1261 | |
1262 | /* | |
1263 | * Map the clone. If r == 0 we don't need to do | |
1264 | * anything, the target has assumed ownership of | |
1265 | * this io. | |
1266 | */ | |
64f52b0e | 1267 | atomic_inc(&io->io_count); |
4f024f37 | 1268 | sector = clone->bi_iter.bi_sector; |
d67a5f4b | 1269 | |
7de3ee57 | 1270 | r = ti->type->map(ti, clone); |
846785e6 CH |
1271 | switch (r) { |
1272 | case DM_MAPIO_SUBMITTED: | |
1273 | break; | |
1274 | case DM_MAPIO_REMAPPED: | |
1da177e4 | 1275 | /* the bio has been remapped so dispatch it */ |
74d46992 | 1276 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
64f52b0e | 1277 | bio_dev(io->orig_bio), sector); |
5a6c35f9 | 1278 | ret = submit_bio_noacct(clone); |
846785e6 CH |
1279 | break; |
1280 | case DM_MAPIO_KILL: | |
4e4cbee9 | 1281 | free_tio(tio); |
64f52b0e | 1282 | dec_pending(io, BLK_STS_IOERR); |
4e4cbee9 | 1283 | break; |
846785e6 | 1284 | case DM_MAPIO_REQUEUE: |
cfae7529 | 1285 | free_tio(tio); |
64f52b0e | 1286 | dec_pending(io, BLK_STS_DM_REQUEUE); |
846785e6 CH |
1287 | break; |
1288 | default: | |
45cbcd79 KU |
1289 | DMWARN("unimplemented target map return value: %d", r); |
1290 | BUG(); | |
1da177e4 | 1291 | } |
1da177e4 | 1292 | |
978e51ba | 1293 | return ret; |
1da177e4 | 1294 | } |
1da177e4 | 1295 | |
e0d6609a | 1296 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1297 | { |
4f024f37 KO |
1298 | bio->bi_iter.bi_sector = sector; |
1299 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1300 | } |
1301 | ||
1302 | /* | |
1303 | * Creates a bio that consists of range of complete bvecs. | |
1304 | */ | |
c80914e8 MS |
1305 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
1306 | sector_t sector, unsigned len) | |
1da177e4 | 1307 | { |
dba14160 | 1308 | struct bio *clone = &tio->clone; |
1da177e4 | 1309 | |
1c3b13e6 KO |
1310 | __bio_clone_fast(clone, bio); |
1311 | ||
a892c8d5 ST |
1312 | bio_crypt_clone(clone, bio, GFP_NOIO); |
1313 | ||
57c36519 | 1314 | if (bio_integrity(bio)) { |
e2460f2a MP |
1315 | int r; |
1316 | ||
1317 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && | |
1318 | !dm_target_passes_integrity(tio->ti->type))) { | |
1319 | DMWARN("%s: the target %s doesn't support integrity data.", | |
1320 | dm_device_name(tio->io->md), | |
1321 | tio->ti->type->name); | |
1322 | return -EIO; | |
1323 | } | |
1324 | ||
1325 | r = bio_integrity_clone(clone, bio, GFP_NOIO); | |
c80914e8 MS |
1326 | if (r < 0) |
1327 | return r; | |
1328 | } | |
bd2a49b8 | 1329 | |
fa8db494 MS |
1330 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1331 | clone->bi_iter.bi_size = to_bytes(len); | |
1332 | ||
1333 | if (bio_integrity(bio)) | |
1334 | bio_integrity_trim(clone); | |
c80914e8 MS |
1335 | |
1336 | return 0; | |
1da177e4 LT |
1337 | } |
1338 | ||
318716dd MS |
1339 | static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci, |
1340 | struct dm_target *ti, unsigned num_bios) | |
f9ab94ce | 1341 | { |
dba14160 | 1342 | struct dm_target_io *tio; |
318716dd | 1343 | int try; |
dba14160 | 1344 | |
318716dd MS |
1345 | if (!num_bios) |
1346 | return; | |
f9ab94ce | 1347 | |
318716dd MS |
1348 | if (num_bios == 1) { |
1349 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); | |
1350 | bio_list_add(blist, &tio->clone); | |
1351 | return; | |
1352 | } | |
9015df24 | 1353 | |
318716dd MS |
1354 | for (try = 0; try < 2; try++) { |
1355 | int bio_nr; | |
1356 | struct bio *bio; | |
1357 | ||
1358 | if (try) | |
bc02cdbe | 1359 | mutex_lock(&ci->io->md->table_devices_lock); |
318716dd MS |
1360 | for (bio_nr = 0; bio_nr < num_bios; bio_nr++) { |
1361 | tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT); | |
1362 | if (!tio) | |
1363 | break; | |
1364 | ||
1365 | bio_list_add(blist, &tio->clone); | |
1366 | } | |
1367 | if (try) | |
bc02cdbe | 1368 | mutex_unlock(&ci->io->md->table_devices_lock); |
318716dd MS |
1369 | if (bio_nr == num_bios) |
1370 | return; | |
1371 | ||
1372 | while ((bio = bio_list_pop(blist))) { | |
1373 | tio = container_of(bio, struct dm_target_io, clone); | |
1374 | free_tio(tio); | |
1375 | } | |
1376 | } | |
9015df24 AK |
1377 | } |
1378 | ||
978e51ba MS |
1379 | static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci, |
1380 | struct dm_target_io *tio, unsigned *len) | |
9015df24 | 1381 | { |
dba14160 | 1382 | struct bio *clone = &tio->clone; |
9015df24 | 1383 | |
1dd40c3e MP |
1384 | tio->len_ptr = len; |
1385 | ||
99778273 | 1386 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1387 | if (len) |
1dd40c3e | 1388 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1389 | |
978e51ba | 1390 | return __map_bio(tio); |
f9ab94ce MP |
1391 | } |
1392 | ||
14fe594d | 1393 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1394 | unsigned num_bios, unsigned *len) |
06a426ce | 1395 | { |
318716dd MS |
1396 | struct bio_list blist = BIO_EMPTY_LIST; |
1397 | struct bio *bio; | |
1398 | struct dm_target_io *tio; | |
1399 | ||
1400 | alloc_multiple_bios(&blist, ci, ti, num_bios); | |
06a426ce | 1401 | |
318716dd MS |
1402 | while ((bio = bio_list_pop(&blist))) { |
1403 | tio = container_of(bio, struct dm_target_io, clone); | |
978e51ba | 1404 | (void) __clone_and_map_simple_bio(ci, tio, len); |
318716dd | 1405 | } |
06a426ce MS |
1406 | } |
1407 | ||
14fe594d | 1408 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1409 | { |
06a426ce | 1410 | unsigned target_nr = 0; |
f9ab94ce | 1411 | struct dm_target *ti; |
828678b8 MS |
1412 | struct bio flush_bio; |
1413 | ||
1414 | /* | |
1415 | * Use an on-stack bio for this, it's safe since we don't | |
1416 | * need to reference it after submit. It's just used as | |
1417 | * the basis for the clone(s). | |
1418 | */ | |
1419 | bio_init(&flush_bio, NULL, 0); | |
1420 | flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; | |
1421 | ci->bio = &flush_bio; | |
1422 | ci->sector_count = 0; | |
f9ab94ce | 1423 | |
892ad71f | 1424 | /* |
dbe3ece1 JA |
1425 | * Empty flush uses a statically initialized bio, as the base for |
1426 | * cloning. However, blkg association requires that a bdev is | |
1427 | * associated with a gendisk, which doesn't happen until the bdev is | |
1428 | * opened. So, blkg association is done at issue time of the flush | |
1429 | * rather than when the device is created in alloc_dev(). | |
892ad71f DZ |
1430 | */ |
1431 | bio_set_dev(ci->bio, ci->io->md->bdev); | |
1432 | ||
b372d360 | 1433 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1434 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1435 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
828678b8 MS |
1436 | |
1437 | bio_uninit(ci->bio); | |
f9ab94ce MP |
1438 | return 0; |
1439 | } | |
1440 | ||
c80914e8 | 1441 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
f31c21e4 | 1442 | sector_t sector, unsigned *len) |
5ae89a87 | 1443 | { |
dba14160 | 1444 | struct bio *bio = ci->bio; |
5ae89a87 | 1445 | struct dm_target_io *tio; |
f31c21e4 | 1446 | int r; |
5ae89a87 | 1447 | |
318716dd | 1448 | tio = alloc_tio(ci, ti, 0, GFP_NOIO); |
f31c21e4 N |
1449 | tio->len_ptr = len; |
1450 | r = clone_bio(tio, bio, sector, *len); | |
1451 | if (r < 0) { | |
1452 | free_tio(tio); | |
1453 | return r; | |
b0d8ed4d | 1454 | } |
978e51ba | 1455 | (void) __map_bio(tio); |
c80914e8 | 1456 | |
f31c21e4 | 1457 | return 0; |
5ae89a87 MS |
1458 | } |
1459 | ||
3d7f4562 | 1460 | static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, |
61697a6a | 1461 | unsigned num_bios) |
ba1cbad9 | 1462 | { |
51b86f9a | 1463 | unsigned len; |
ba1cbad9 | 1464 | |
3d7f4562 MS |
1465 | /* |
1466 | * Even though the device advertised support for this type of | |
1467 | * request, that does not mean every target supports it, and | |
1468 | * reconfiguration might also have changed that since the | |
1469 | * check was performed. | |
1470 | */ | |
3d7f4562 MS |
1471 | if (!num_bios) |
1472 | return -EOPNOTSUPP; | |
ba1cbad9 | 1473 | |
3720281d MS |
1474 | len = min_t(sector_t, ci->sector_count, |
1475 | max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector))); | |
51b86f9a | 1476 | |
3d7f4562 | 1477 | __send_duplicate_bios(ci, ti, num_bios, &len); |
e262f347 | 1478 | |
3d7f4562 MS |
1479 | ci->sector += len; |
1480 | ci->sector_count -= len; | |
5ae89a87 MS |
1481 | |
1482 | return 0; | |
ba1cbad9 MS |
1483 | } |
1484 | ||
568c73a3 MS |
1485 | static bool is_abnormal_io(struct bio *bio) |
1486 | { | |
1487 | bool r = false; | |
1488 | ||
1489 | switch (bio_op(bio)) { | |
1490 | case REQ_OP_DISCARD: | |
1491 | case REQ_OP_SECURE_ERASE: | |
1492 | case REQ_OP_WRITE_SAME: | |
1493 | case REQ_OP_WRITE_ZEROES: | |
1494 | r = true; | |
1495 | break; | |
1496 | } | |
1497 | ||
1498 | return r; | |
1499 | } | |
1500 | ||
0519c71e MS |
1501 | static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti, |
1502 | int *result) | |
1503 | { | |
1504 | struct bio *bio = ci->bio; | |
9679b5a7 | 1505 | unsigned num_bios = 0; |
0519c71e | 1506 | |
9679b5a7 MS |
1507 | switch (bio_op(bio)) { |
1508 | case REQ_OP_DISCARD: | |
1509 | num_bios = ti->num_discard_bios; | |
1510 | break; | |
1511 | case REQ_OP_SECURE_ERASE: | |
1512 | num_bios = ti->num_secure_erase_bios; | |
1513 | break; | |
1514 | case REQ_OP_WRITE_SAME: | |
1515 | num_bios = ti->num_write_same_bios; | |
1516 | break; | |
1517 | case REQ_OP_WRITE_ZEROES: | |
1518 | num_bios = ti->num_write_zeroes_bios; | |
1519 | break; | |
1520 | default: | |
0519c71e | 1521 | return false; |
9679b5a7 | 1522 | } |
0519c71e | 1523 | |
9679b5a7 | 1524 | *result = __send_changing_extent_only(ci, ti, num_bios); |
0519c71e MS |
1525 | return true; |
1526 | } | |
1527 | ||
e4c93811 AK |
1528 | /* |
1529 | * Select the correct strategy for processing a non-flush bio. | |
1530 | */ | |
14fe594d | 1531 | static int __split_and_process_non_flush(struct clone_info *ci) |
0ce65797 | 1532 | { |
512875bd | 1533 | struct dm_target *ti; |
1c3b13e6 | 1534 | unsigned len; |
c80914e8 | 1535 | int r; |
0ce65797 | 1536 | |
512875bd | 1537 | ti = dm_table_find_target(ci->map, ci->sector); |
123d87d5 | 1538 | if (!ti) |
512875bd JN |
1539 | return -EIO; |
1540 | ||
568c73a3 | 1541 | if (__process_abnormal_io(ci, ti, &r)) |
0519c71e | 1542 | return r; |
3d7f4562 | 1543 | |
3720281d | 1544 | len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count); |
0ce65797 | 1545 | |
c80914e8 MS |
1546 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1547 | if (r < 0) | |
1548 | return r; | |
0ce65797 | 1549 | |
1c3b13e6 KO |
1550 | ci->sector += len; |
1551 | ci->sector_count -= len; | |
0ce65797 | 1552 | |
1c3b13e6 | 1553 | return 0; |
0ce65797 MS |
1554 | } |
1555 | ||
978e51ba MS |
1556 | static void init_clone_info(struct clone_info *ci, struct mapped_device *md, |
1557 | struct dm_table *map, struct bio *bio) | |
1558 | { | |
1559 | ci->map = map; | |
1560 | ci->io = alloc_io(md, bio); | |
1561 | ci->sector = bio->bi_iter.bi_sector; | |
1562 | } | |
1563 | ||
a1e1cb72 MS |
1564 | #define __dm_part_stat_sub(part, field, subnd) \ |
1565 | (part_stat_get(part, field) -= (subnd)) | |
1566 | ||
1da177e4 | 1567 | /* |
14fe594d | 1568 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1569 | */ |
978e51ba MS |
1570 | static blk_qc_t __split_and_process_bio(struct mapped_device *md, |
1571 | struct dm_table *map, struct bio *bio) | |
0ce65797 | 1572 | { |
1da177e4 | 1573 | struct clone_info ci; |
978e51ba | 1574 | blk_qc_t ret = BLK_QC_T_NONE; |
512875bd | 1575 | int error = 0; |
1da177e4 | 1576 | |
978e51ba | 1577 | init_clone_info(&ci, md, map, bio); |
0ce65797 | 1578 | |
1eff9d32 | 1579 | if (bio->bi_opf & REQ_PREFLUSH) { |
14fe594d | 1580 | error = __send_empty_flush(&ci); |
b372d360 | 1581 | /* dec_pending submits any data associated with flush */ |
2e2d6f7e | 1582 | } else if (op_is_zone_mgmt(bio_op(bio))) { |
a4aa5e56 DLM |
1583 | ci.bio = bio; |
1584 | ci.sector_count = 0; | |
1585 | error = __split_and_process_non_flush(&ci); | |
b372d360 | 1586 | } else { |
6a8736d1 | 1587 | ci.bio = bio; |
d87f4c14 | 1588 | ci.sector_count = bio_sectors(bio); |
18a25da8 | 1589 | while (ci.sector_count && !error) { |
14fe594d | 1590 | error = __split_and_process_non_flush(&ci); |
18a25da8 N |
1591 | if (current->bio_list && ci.sector_count && !error) { |
1592 | /* | |
ed00aabd | 1593 | * Remainder must be passed to submit_bio_noacct() |
18a25da8 N |
1594 | * so that it gets handled *after* bios already submitted |
1595 | * have been completely processed. | |
1596 | * We take a clone of the original to store in | |
745dc570 | 1597 | * ci.io->orig_bio to be used by end_io_acct() and |
18a25da8 | 1598 | * for dec_pending to use for completion handling. |
18a25da8 | 1599 | */ |
f21c601a MS |
1600 | struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count, |
1601 | GFP_NOIO, &md->queue->bio_split); | |
745dc570 | 1602 | ci.io->orig_bio = b; |
a1e1cb72 MS |
1603 | |
1604 | /* | |
1605 | * Adjust IO stats for each split, otherwise upon queue | |
1606 | * reentry there will be redundant IO accounting. | |
1607 | * NOTE: this is a stop-gap fix, a proper fix involves | |
1608 | * significant refactoring of DM core's bio splitting | |
1609 | * (by eliminating DM's splitting and just using bio_split) | |
1610 | */ | |
1611 | part_stat_lock(); | |
1612 | __dm_part_stat_sub(&dm_disk(md)->part0, | |
1613 | sectors[op_stat_group(bio_op(bio))], ci.sector_count); | |
1614 | part_stat_unlock(); | |
1615 | ||
18a25da8 | 1616 | bio_chain(b, bio); |
075c18c3 | 1617 | trace_block_split(md->queue, b, bio->bi_iter.bi_sector); |
ed00aabd | 1618 | ret = submit_bio_noacct(bio); |
18a25da8 N |
1619 | break; |
1620 | } | |
1621 | } | |
d87f4c14 | 1622 | } |
0ce65797 | 1623 | |
1da177e4 | 1624 | /* drop the extra reference count */ |
54385bf7 | 1625 | dec_pending(ci.io, errno_to_blk_status(error)); |
978e51ba | 1626 | return ret; |
0ce65797 MS |
1627 | } |
1628 | ||
cec47e3d | 1629 | /* |
978e51ba MS |
1630 | * Optimized variant of __split_and_process_bio that leverages the |
1631 | * fact that targets that use it do _not_ have a need to split bios. | |
cec47e3d | 1632 | */ |
568c73a3 | 1633 | static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map, |
094ee64d | 1634 | struct bio *bio) |
978e51ba MS |
1635 | { |
1636 | struct clone_info ci; | |
1637 | blk_qc_t ret = BLK_QC_T_NONE; | |
1638 | int error = 0; | |
1639 | ||
978e51ba MS |
1640 | init_clone_info(&ci, md, map, bio); |
1641 | ||
1642 | if (bio->bi_opf & REQ_PREFLUSH) { | |
978e51ba MS |
1643 | error = __send_empty_flush(&ci); |
1644 | /* dec_pending submits any data associated with flush */ | |
1645 | } else { | |
978e51ba | 1646 | struct dm_target_io *tio; |
094ee64d MS |
1647 | struct dm_target *ti = md->immutable_target; |
1648 | ||
1649 | if (WARN_ON_ONCE(!ti)) { | |
1650 | error = -EIO; | |
1651 | goto out; | |
1652 | } | |
978e51ba | 1653 | |
978e51ba MS |
1654 | ci.bio = bio; |
1655 | ci.sector_count = bio_sectors(bio); | |
568c73a3 | 1656 | if (__process_abnormal_io(&ci, ti, &error)) |
0519c71e MS |
1657 | goto out; |
1658 | ||
1659 | tio = alloc_tio(&ci, ti, 0, GFP_NOIO); | |
978e51ba MS |
1660 | ret = __clone_and_map_simple_bio(&ci, tio, NULL); |
1661 | } | |
1662 | out: | |
1663 | /* drop the extra reference count */ | |
1664 | dec_pending(ci.io, errno_to_blk_status(error)); | |
1665 | return ret; | |
1666 | } | |
1667 | ||
c62b37d9 | 1668 | static blk_qc_t dm_submit_bio(struct bio *bio) |
cec47e3d | 1669 | { |
c4a59c4e | 1670 | struct mapped_device *md = bio->bi_disk->private_data; |
978e51ba | 1671 | blk_qc_t ret = BLK_QC_T_NONE; |
83d5e5b0 MP |
1672 | int srcu_idx; |
1673 | struct dm_table *map; | |
cec47e3d | 1674 | |
ac7c5675 CH |
1675 | if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) { |
1676 | /* | |
1677 | * We are called with a live reference on q_usage_counter, but | |
1678 | * that one will be released as soon as we return. Grab an | |
c62b37d9 CH |
1679 | * extra one as blk_mq_submit_bio expects to be able to consume |
1680 | * a reference (which lives until the request is freed in case a | |
1681 | * request is allocated). | |
ac7c5675 | 1682 | */ |
c62b37d9 CH |
1683 | percpu_ref_get(&bio->bi_disk->queue->q_usage_counter); |
1684 | return blk_mq_submit_bio(bio); | |
ac7c5675 | 1685 | } |
8cf7961d | 1686 | |
83d5e5b0 | 1687 | map = dm_get_live_table(md, &srcu_idx); |
b2abdb1b MS |
1688 | if (unlikely(!map)) { |
1689 | DMERR_LIMIT("%s: mapping table unavailable, erroring io", | |
1690 | dm_device_name(md)); | |
1691 | bio_io_error(bio); | |
1692 | goto out; | |
1693 | } | |
29e4013d | 1694 | |
b2abdb1b | 1695 | /* If suspended, queue this IO for later */ |
6a8736d1 | 1696 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { |
6abc4946 KK |
1697 | if (bio->bi_opf & REQ_NOWAIT) |
1698 | bio_wouldblock_error(bio); | |
b2abdb1b | 1699 | else if (bio->bi_opf & REQ_RAHEAD) |
54d9a1b4 | 1700 | bio_io_error(bio); |
b2abdb1b MS |
1701 | else |
1702 | queue_io(md, bio); | |
1703 | goto out; | |
cec47e3d | 1704 | } |
1da177e4 | 1705 | |
b2abdb1b MS |
1706 | /* |
1707 | * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc) | |
1708 | * otherwise associated queue_limits won't be imposed. | |
1709 | */ | |
1710 | if (is_abnormal_io(bio)) | |
1711 | blk_queue_split(&bio); | |
978e51ba | 1712 | |
b2abdb1b MS |
1713 | if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED) |
1714 | ret = __process_bio(md, map, bio); | |
1715 | else | |
1716 | ret = __split_and_process_bio(md, map, bio); | |
1717 | out: | |
83d5e5b0 | 1718 | dm_put_live_table(md, srcu_idx); |
978e51ba MS |
1719 | return ret; |
1720 | } | |
1721 | ||
1da177e4 LT |
1722 | /*----------------------------------------------------------------- |
1723 | * An IDR is used to keep track of allocated minor numbers. | |
1724 | *---------------------------------------------------------------*/ | |
2b06cfff | 1725 | static void free_minor(int minor) |
1da177e4 | 1726 | { |
f32c10b0 | 1727 | spin_lock(&_minor_lock); |
1da177e4 | 1728 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1729 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1730 | } |
1731 | ||
1732 | /* | |
1733 | * See if the device with a specific minor # is free. | |
1734 | */ | |
cf13ab8e | 1735 | static int specific_minor(int minor) |
1da177e4 | 1736 | { |
c9d76be6 | 1737 | int r; |
1da177e4 LT |
1738 | |
1739 | if (minor >= (1 << MINORBITS)) | |
1740 | return -EINVAL; | |
1741 | ||
c9d76be6 | 1742 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1743 | spin_lock(&_minor_lock); |
1da177e4 | 1744 | |
c9d76be6 | 1745 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1746 | |
f32c10b0 | 1747 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1748 | idr_preload_end(); |
1749 | if (r < 0) | |
1750 | return r == -ENOSPC ? -EBUSY : r; | |
1751 | return 0; | |
1da177e4 LT |
1752 | } |
1753 | ||
cf13ab8e | 1754 | static int next_free_minor(int *minor) |
1da177e4 | 1755 | { |
c9d76be6 | 1756 | int r; |
62f75c2f | 1757 | |
c9d76be6 | 1758 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1759 | spin_lock(&_minor_lock); |
1da177e4 | 1760 | |
c9d76be6 | 1761 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1762 | |
f32c10b0 | 1763 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1764 | idr_preload_end(); |
1765 | if (r < 0) | |
1766 | return r; | |
1767 | *minor = r; | |
1768 | return 0; | |
1da177e4 LT |
1769 | } |
1770 | ||
83d5cde4 | 1771 | static const struct block_device_operations dm_blk_dops; |
f26c5719 | 1772 | static const struct dax_operations dm_dax_ops; |
1da177e4 | 1773 | |
53d5914f MP |
1774 | static void dm_wq_work(struct work_struct *work); |
1775 | ||
0f20972f MS |
1776 | static void cleanup_mapped_device(struct mapped_device *md) |
1777 | { | |
0f20972f MS |
1778 | if (md->wq) |
1779 | destroy_workqueue(md->wq); | |
6f1c819c KO |
1780 | bioset_exit(&md->bs); |
1781 | bioset_exit(&md->io_bs); | |
0f20972f | 1782 | |
f26c5719 DW |
1783 | if (md->dax_dev) { |
1784 | kill_dax(md->dax_dev); | |
1785 | put_dax(md->dax_dev); | |
1786 | md->dax_dev = NULL; | |
1787 | } | |
1788 | ||
0f20972f MS |
1789 | if (md->disk) { |
1790 | spin_lock(&_minor_lock); | |
1791 | md->disk->private_data = NULL; | |
1792 | spin_unlock(&_minor_lock); | |
0f20972f MS |
1793 | del_gendisk(md->disk); |
1794 | put_disk(md->disk); | |
1795 | } | |
1796 | ||
1797 | if (md->queue) | |
1798 | blk_cleanup_queue(md->queue); | |
1799 | ||
d09960b0 TE |
1800 | cleanup_srcu_struct(&md->io_barrier); |
1801 | ||
0f20972f MS |
1802 | if (md->bdev) { |
1803 | bdput(md->bdev); | |
1804 | md->bdev = NULL; | |
1805 | } | |
4cc96131 | 1806 | |
d5ffebdd MS |
1807 | mutex_destroy(&md->suspend_lock); |
1808 | mutex_destroy(&md->type_lock); | |
1809 | mutex_destroy(&md->table_devices_lock); | |
1810 | ||
4cc96131 | 1811 | dm_mq_cleanup_mapped_device(md); |
0f20972f MS |
1812 | } |
1813 | ||
1da177e4 LT |
1814 | /* |
1815 | * Allocate and initialise a blank device with a given minor. | |
1816 | */ | |
2b06cfff | 1817 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 | 1818 | { |
115485e8 MS |
1819 | int r, numa_node_id = dm_get_numa_node(); |
1820 | struct mapped_device *md; | |
ba61fdd1 | 1821 | void *old_md; |
1da177e4 | 1822 | |
856eb091 | 1823 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
1da177e4 LT |
1824 | if (!md) { |
1825 | DMWARN("unable to allocate device, out of memory."); | |
1826 | return NULL; | |
1827 | } | |
1828 | ||
10da4f79 | 1829 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1830 | goto bad_module_get; |
10da4f79 | 1831 | |
1da177e4 | 1832 | /* get a minor number for the dev */ |
2b06cfff | 1833 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1834 | r = next_free_minor(&minor); |
2b06cfff | 1835 | else |
cf13ab8e | 1836 | r = specific_minor(minor); |
1da177e4 | 1837 | if (r < 0) |
6ed7ade8 | 1838 | goto bad_minor; |
1da177e4 | 1839 | |
83d5e5b0 MP |
1840 | r = init_srcu_struct(&md->io_barrier); |
1841 | if (r < 0) | |
1842 | goto bad_io_barrier; | |
1843 | ||
115485e8 | 1844 | md->numa_node_id = numa_node_id; |
591ddcfc | 1845 | md->init_tio_pdu = false; |
a5664dad | 1846 | md->type = DM_TYPE_NONE; |
e61290a4 | 1847 | mutex_init(&md->suspend_lock); |
a5664dad | 1848 | mutex_init(&md->type_lock); |
86f1152b | 1849 | mutex_init(&md->table_devices_lock); |
022c2611 | 1850 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1851 | atomic_set(&md->holders, 1); |
5c6bd75d | 1852 | atomic_set(&md->open_count, 0); |
1da177e4 | 1853 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1854 | atomic_set(&md->uevent_seq, 0); |
1855 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 1856 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 1857 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 1858 | |
47ace7e0 | 1859 | /* |
c62b37d9 CH |
1860 | * default to bio-based until DM table is loaded and md->type |
1861 | * established. If request-based table is loaded: blk-mq will | |
1862 | * override accordingly. | |
47ace7e0 | 1863 | */ |
c62b37d9 | 1864 | md->queue = blk_alloc_queue(numa_node_id); |
3d745ea5 CH |
1865 | if (!md->queue) |
1866 | goto bad; | |
1da177e4 | 1867 | |
c12c9a3c | 1868 | md->disk = alloc_disk_node(1, md->numa_node_id); |
1da177e4 | 1869 | if (!md->disk) |
0f20972f | 1870 | goto bad; |
1da177e4 | 1871 | |
f0b04115 | 1872 | init_waitqueue_head(&md->wait); |
53d5914f | 1873 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1874 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1875 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 1876 | |
1da177e4 LT |
1877 | md->disk->major = _major; |
1878 | md->disk->first_minor = minor; | |
1879 | md->disk->fops = &dm_blk_dops; | |
1880 | md->disk->queue = md->queue; | |
1881 | md->disk->private_data = md; | |
1882 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
f26c5719 | 1883 | |
976431b0 | 1884 | if (IS_ENABLED(CONFIG_DAX_DRIVER)) { |
fefc1d97 PG |
1885 | md->dax_dev = alloc_dax(md, md->disk->disk_name, |
1886 | &dm_dax_ops, 0); | |
4e4ced93 | 1887 | if (IS_ERR(md->dax_dev)) |
976431b0 DW |
1888 | goto bad; |
1889 | } | |
f26c5719 | 1890 | |
c100ec49 | 1891 | add_disk_no_queue_reg(md->disk); |
7e51f257 | 1892 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1893 | |
670368a8 | 1894 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 1895 | if (!md->wq) |
0f20972f | 1896 | goto bad; |
304f3f6a | 1897 | |
32a926da MP |
1898 | md->bdev = bdget_disk(md->disk, 0); |
1899 | if (!md->bdev) | |
0f20972f | 1900 | goto bad; |
32a926da | 1901 | |
fd2ed4d2 MP |
1902 | dm_stats_init(&md->stats); |
1903 | ||
ba61fdd1 | 1904 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1905 | spin_lock(&_minor_lock); |
ba61fdd1 | 1906 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1907 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1908 | |
1909 | BUG_ON(old_md != MINOR_ALLOCED); | |
1910 | ||
1da177e4 LT |
1911 | return md; |
1912 | ||
0f20972f MS |
1913 | bad: |
1914 | cleanup_mapped_device(md); | |
83d5e5b0 | 1915 | bad_io_barrier: |
1da177e4 | 1916 | free_minor(minor); |
6ed7ade8 | 1917 | bad_minor: |
10da4f79 | 1918 | module_put(THIS_MODULE); |
6ed7ade8 | 1919 | bad_module_get: |
856eb091 | 1920 | kvfree(md); |
1da177e4 LT |
1921 | return NULL; |
1922 | } | |
1923 | ||
ae9da83f JN |
1924 | static void unlock_fs(struct mapped_device *md); |
1925 | ||
1da177e4 LT |
1926 | static void free_dev(struct mapped_device *md) |
1927 | { | |
f331c029 | 1928 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1929 | |
32a926da | 1930 | unlock_fs(md); |
2eb6e1e3 | 1931 | |
0f20972f | 1932 | cleanup_mapped_device(md); |
63a4f065 | 1933 | |
86f1152b | 1934 | free_table_devices(&md->table_devices); |
63a4f065 | 1935 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
1936 | free_minor(minor); |
1937 | ||
10da4f79 | 1938 | module_put(THIS_MODULE); |
856eb091 | 1939 | kvfree(md); |
1da177e4 LT |
1940 | } |
1941 | ||
2a2a4c51 | 1942 | static int __bind_mempools(struct mapped_device *md, struct dm_table *t) |
e6ee8c0b | 1943 | { |
c0820cf5 | 1944 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
2a2a4c51 | 1945 | int ret = 0; |
e6ee8c0b | 1946 | |
0776aa0e | 1947 | if (dm_table_bio_based(t)) { |
64f52b0e MS |
1948 | /* |
1949 | * The md may already have mempools that need changing. | |
1950 | * If so, reload bioset because front_pad may have changed | |
1951 | * because a different table was loaded. | |
1952 | */ | |
6f1c819c KO |
1953 | bioset_exit(&md->bs); |
1954 | bioset_exit(&md->io_bs); | |
0776aa0e | 1955 | |
6f1c819c | 1956 | } else if (bioset_initialized(&md->bs)) { |
4e6e36c3 MS |
1957 | /* |
1958 | * There's no need to reload with request-based dm | |
1959 | * because the size of front_pad doesn't change. | |
1960 | * Note for future: If you are to reload bioset, | |
1961 | * prep-ed requests in the queue may refer | |
1962 | * to bio from the old bioset, so you must walk | |
1963 | * through the queue to unprep. | |
1964 | */ | |
1965 | goto out; | |
c0820cf5 | 1966 | } |
e6ee8c0b | 1967 | |
6f1c819c KO |
1968 | BUG_ON(!p || |
1969 | bioset_initialized(&md->bs) || | |
1970 | bioset_initialized(&md->io_bs)); | |
cbc4e3c1 | 1971 | |
2a2a4c51 JA |
1972 | ret = bioset_init_from_src(&md->bs, &p->bs); |
1973 | if (ret) | |
1974 | goto out; | |
1975 | ret = bioset_init_from_src(&md->io_bs, &p->io_bs); | |
1976 | if (ret) | |
1977 | bioset_exit(&md->bs); | |
e6ee8c0b | 1978 | out: |
02233342 | 1979 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b | 1980 | dm_table_free_md_mempools(t); |
2a2a4c51 | 1981 | return ret; |
e6ee8c0b KU |
1982 | } |
1983 | ||
1da177e4 LT |
1984 | /* |
1985 | * Bind a table to the device. | |
1986 | */ | |
1987 | static void event_callback(void *context) | |
1988 | { | |
7a8c3d3b MA |
1989 | unsigned long flags; |
1990 | LIST_HEAD(uevents); | |
1da177e4 LT |
1991 | struct mapped_device *md = (struct mapped_device *) context; |
1992 | ||
7a8c3d3b MA |
1993 | spin_lock_irqsave(&md->uevent_lock, flags); |
1994 | list_splice_init(&md->uevent_list, &uevents); | |
1995 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1996 | ||
ed9e1982 | 1997 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1998 | |
1da177e4 LT |
1999 | atomic_inc(&md->event_nr); |
2000 | wake_up(&md->eventq); | |
62e08243 | 2001 | dm_issue_global_event(); |
1da177e4 LT |
2002 | } |
2003 | ||
042d2a9b AK |
2004 | /* |
2005 | * Returns old map, which caller must destroy. | |
2006 | */ | |
2007 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2008 | struct queue_limits *limits) | |
1da177e4 | 2009 | { |
042d2a9b | 2010 | struct dm_table *old_map; |
165125e1 | 2011 | struct request_queue *q = md->queue; |
978e51ba | 2012 | bool request_based = dm_table_request_based(t); |
1da177e4 | 2013 | sector_t size; |
2a2a4c51 | 2014 | int ret; |
1da177e4 | 2015 | |
5a8f1f80 BVA |
2016 | lockdep_assert_held(&md->suspend_lock); |
2017 | ||
1da177e4 | 2018 | size = dm_table_get_size(t); |
3ac51e74 DW |
2019 | |
2020 | /* | |
2021 | * Wipe any geometry if the size of the table changed. | |
2022 | */ | |
fd2ed4d2 | 2023 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2024 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2025 | ||
c2b4bb8c CH |
2026 | set_capacity(md->disk, size); |
2027 | bd_set_nr_sectors(md->bdev, size); | |
d5816876 | 2028 | |
2ca3310e AK |
2029 | dm_table_event_callback(t, event_callback, md); |
2030 | ||
e6ee8c0b KU |
2031 | /* |
2032 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2033 | * for request-based during suspension. So stop it to prevent | |
2034 | * I/O mapping before resume. | |
2035 | * This must be done before setting the queue restrictions, | |
2036 | * because request-based dm may be run just after the setting. | |
2037 | */ | |
978e51ba | 2038 | if (request_based) |
eca7ee6d | 2039 | dm_stop_queue(q); |
978e51ba MS |
2040 | |
2041 | if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) { | |
16f12266 | 2042 | /* |
978e51ba MS |
2043 | * Leverage the fact that request-based DM targets and |
2044 | * NVMe bio based targets are immutable singletons | |
094ee64d | 2045 | * - used to optimize both __process_bio and dm_mq_queue_rq |
16f12266 MS |
2046 | */ |
2047 | md->immutable_target = dm_table_get_immutable_target(t); | |
2048 | } | |
e6ee8c0b | 2049 | |
2a2a4c51 JA |
2050 | ret = __bind_mempools(md, t); |
2051 | if (ret) { | |
2052 | old_map = ERR_PTR(ret); | |
2053 | goto out; | |
2054 | } | |
e6ee8c0b | 2055 | |
a12f5d48 | 2056 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
1d3aa6f6 | 2057 | rcu_assign_pointer(md->map, (void *)t); |
36a0456f AK |
2058 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2059 | ||
754c5fc7 | 2060 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2061 | if (old_map) |
2062 | dm_sync_table(md); | |
1da177e4 | 2063 | |
2a2a4c51 | 2064 | out: |
042d2a9b | 2065 | return old_map; |
1da177e4 LT |
2066 | } |
2067 | ||
a7940155 AK |
2068 | /* |
2069 | * Returns unbound table for the caller to free. | |
2070 | */ | |
2071 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2072 | { |
a12f5d48 | 2073 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2074 | |
2075 | if (!map) | |
a7940155 | 2076 | return NULL; |
1da177e4 LT |
2077 | |
2078 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2079 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2080 | dm_sync_table(md); |
a7940155 AK |
2081 | |
2082 | return map; | |
1da177e4 LT |
2083 | } |
2084 | ||
2085 | /* | |
2086 | * Constructor for a new device. | |
2087 | */ | |
2b06cfff | 2088 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 | 2089 | { |
c12c9a3c | 2090 | int r; |
1da177e4 LT |
2091 | struct mapped_device *md; |
2092 | ||
2b06cfff | 2093 | md = alloc_dev(minor); |
1da177e4 LT |
2094 | if (!md) |
2095 | return -ENXIO; | |
2096 | ||
c12c9a3c MS |
2097 | r = dm_sysfs_init(md); |
2098 | if (r) { | |
2099 | free_dev(md); | |
2100 | return r; | |
2101 | } | |
784aae73 | 2102 | |
1da177e4 LT |
2103 | *result = md; |
2104 | return 0; | |
2105 | } | |
2106 | ||
a5664dad MS |
2107 | /* |
2108 | * Functions to manage md->type. | |
2109 | * All are required to hold md->type_lock. | |
2110 | */ | |
2111 | void dm_lock_md_type(struct mapped_device *md) | |
2112 | { | |
2113 | mutex_lock(&md->type_lock); | |
2114 | } | |
2115 | ||
2116 | void dm_unlock_md_type(struct mapped_device *md) | |
2117 | { | |
2118 | mutex_unlock(&md->type_lock); | |
2119 | } | |
2120 | ||
7e0d574f | 2121 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
a5664dad | 2122 | { |
00c4fc3b | 2123 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2124 | md->type = type; |
2125 | } | |
2126 | ||
7e0d574f | 2127 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
a5664dad MS |
2128 | { |
2129 | return md->type; | |
2130 | } | |
2131 | ||
36a0456f AK |
2132 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2133 | { | |
2134 | return md->immutable_target_type; | |
2135 | } | |
2136 | ||
f84cb8a4 MS |
2137 | /* |
2138 | * The queue_limits are only valid as long as you have a reference | |
2139 | * count on 'md'. | |
2140 | */ | |
2141 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2142 | { | |
2143 | BUG_ON(!atomic_read(&md->holders)); | |
2144 | return &md->queue->limits; | |
2145 | } | |
2146 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2147 | ||
4a0b4ddf MS |
2148 | /* |
2149 | * Setup the DM device's queue based on md's type | |
2150 | */ | |
591ddcfc | 2151 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
4a0b4ddf | 2152 | { |
bfebd1cd | 2153 | int r; |
c100ec49 | 2154 | struct queue_limits limits; |
7e0d574f | 2155 | enum dm_queue_mode type = dm_get_md_type(md); |
bfebd1cd | 2156 | |
545ed20e | 2157 | switch (type) { |
bfebd1cd | 2158 | case DM_TYPE_REQUEST_BASED: |
e83068a5 | 2159 | r = dm_mq_init_request_queue(md, t); |
bfebd1cd | 2160 | if (r) { |
eca7ee6d | 2161 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2162 | return r; |
2163 | } | |
2164 | break; | |
2165 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2166 | case DM_TYPE_DAX_BIO_BASED: |
978e51ba | 2167 | case DM_TYPE_NVME_BIO_BASED: |
bfebd1cd | 2168 | break; |
7e0d574f BVA |
2169 | case DM_TYPE_NONE: |
2170 | WARN_ON_ONCE(true); | |
2171 | break; | |
4a0b4ddf MS |
2172 | } |
2173 | ||
c100ec49 MS |
2174 | r = dm_calculate_queue_limits(t, &limits); |
2175 | if (r) { | |
2176 | DMERR("Cannot calculate initial queue limits"); | |
2177 | return r; | |
2178 | } | |
2179 | dm_table_set_restrictions(t, md->queue, &limits); | |
2180 | blk_register_queue(md->disk); | |
2181 | ||
4a0b4ddf MS |
2182 | return 0; |
2183 | } | |
2184 | ||
2bec1f4a | 2185 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2186 | { |
2187 | struct mapped_device *md; | |
1da177e4 LT |
2188 | unsigned minor = MINOR(dev); |
2189 | ||
2190 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2191 | return NULL; | |
2192 | ||
f32c10b0 | 2193 | spin_lock(&_minor_lock); |
1da177e4 LT |
2194 | |
2195 | md = idr_find(&_minor_idr, minor); | |
49de5769 MS |
2196 | if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) || |
2197 | test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2198 | md = NULL; | |
2199 | goto out; | |
fba9f90e | 2200 | } |
49de5769 | 2201 | dm_get(md); |
fba9f90e | 2202 | out: |
f32c10b0 | 2203 | spin_unlock(&_minor_lock); |
1da177e4 | 2204 | |
637842cf DT |
2205 | return md; |
2206 | } | |
3cf2e4ba | 2207 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2208 | |
9ade92a9 | 2209 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2210 | { |
9ade92a9 | 2211 | return md->interface_ptr; |
1da177e4 LT |
2212 | } |
2213 | ||
2214 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2215 | { | |
2216 | md->interface_ptr = ptr; | |
2217 | } | |
2218 | ||
2219 | void dm_get(struct mapped_device *md) | |
2220 | { | |
2221 | atomic_inc(&md->holders); | |
3f77316d | 2222 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2223 | } |
2224 | ||
09ee96b2 MP |
2225 | int dm_hold(struct mapped_device *md) |
2226 | { | |
2227 | spin_lock(&_minor_lock); | |
2228 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2229 | spin_unlock(&_minor_lock); | |
2230 | return -EBUSY; | |
2231 | } | |
2232 | dm_get(md); | |
2233 | spin_unlock(&_minor_lock); | |
2234 | return 0; | |
2235 | } | |
2236 | EXPORT_SYMBOL_GPL(dm_hold); | |
2237 | ||
72d94861 AK |
2238 | const char *dm_device_name(struct mapped_device *md) |
2239 | { | |
2240 | return md->name; | |
2241 | } | |
2242 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2243 | ||
3f77316d | 2244 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2245 | { |
1134e5ae | 2246 | struct dm_table *map; |
83d5e5b0 | 2247 | int srcu_idx; |
1da177e4 | 2248 | |
3f77316d | 2249 | might_sleep(); |
fba9f90e | 2250 | |
63a4f065 | 2251 | spin_lock(&_minor_lock); |
3f77316d KU |
2252 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2253 | set_bit(DMF_FREEING, &md->flags); | |
2254 | spin_unlock(&_minor_lock); | |
3b785fbc | 2255 | |
c12c9a3c | 2256 | blk_set_queue_dying(md->queue); |
3f77316d | 2257 | |
ab7c7bb6 MP |
2258 | /* |
2259 | * Take suspend_lock so that presuspend and postsuspend methods | |
2260 | * do not race with internal suspend. | |
2261 | */ | |
2262 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2263 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2264 | if (!dm_suspended_md(md)) { |
2265 | dm_table_presuspend_targets(map); | |
adc0daad | 2266 | set_bit(DMF_SUSPENDED, &md->flags); |
5df96f2b | 2267 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
3f77316d | 2268 | dm_table_postsuspend_targets(map); |
1da177e4 | 2269 | } |
83d5e5b0 MP |
2270 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2271 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2272 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2273 | |
3f77316d KU |
2274 | /* |
2275 | * Rare, but there may be I/O requests still going to complete, | |
2276 | * for example. Wait for all references to disappear. | |
2277 | * No one should increment the reference count of the mapped_device, | |
2278 | * after the mapped_device state becomes DMF_FREEING. | |
2279 | */ | |
2280 | if (wait) | |
2281 | while (atomic_read(&md->holders)) | |
2282 | msleep(1); | |
2283 | else if (atomic_read(&md->holders)) | |
2284 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2285 | dm_device_name(md), atomic_read(&md->holders)); | |
2286 | ||
2287 | dm_sysfs_exit(md); | |
3f77316d KU |
2288 | dm_table_destroy(__unbind(md)); |
2289 | free_dev(md); | |
2290 | } | |
2291 | ||
2292 | void dm_destroy(struct mapped_device *md) | |
2293 | { | |
2294 | __dm_destroy(md, true); | |
2295 | } | |
2296 | ||
2297 | void dm_destroy_immediate(struct mapped_device *md) | |
2298 | { | |
2299 | __dm_destroy(md, false); | |
2300 | } | |
2301 | ||
2302 | void dm_put(struct mapped_device *md) | |
2303 | { | |
2304 | atomic_dec(&md->holders); | |
1da177e4 | 2305 | } |
79eb885c | 2306 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2307 | |
85067747 ML |
2308 | static bool md_in_flight_bios(struct mapped_device *md) |
2309 | { | |
2310 | int cpu; | |
2311 | struct hd_struct *part = &dm_disk(md)->part0; | |
2312 | long sum = 0; | |
2313 | ||
2314 | for_each_possible_cpu(cpu) { | |
2315 | sum += part_stat_local_read_cpu(part, in_flight[0], cpu); | |
2316 | sum += part_stat_local_read_cpu(part, in_flight[1], cpu); | |
2317 | } | |
2318 | ||
2319 | return sum != 0; | |
2320 | } | |
2321 | ||
2322 | static int dm_wait_for_bios_completion(struct mapped_device *md, long task_state) | |
46125c1c MB |
2323 | { |
2324 | int r = 0; | |
9f4c3f87 | 2325 | DEFINE_WAIT(wait); |
46125c1c | 2326 | |
85067747 | 2327 | while (true) { |
9f4c3f87 | 2328 | prepare_to_wait(&md->wait, &wait, task_state); |
46125c1c | 2329 | |
85067747 | 2330 | if (!md_in_flight_bios(md)) |
46125c1c MB |
2331 | break; |
2332 | ||
e3fabdfd | 2333 | if (signal_pending_state(task_state, current)) { |
46125c1c MB |
2334 | r = -EINTR; |
2335 | break; | |
2336 | } | |
2337 | ||
2338 | io_schedule(); | |
2339 | } | |
9f4c3f87 | 2340 | finish_wait(&md->wait, &wait); |
b44ebeb0 | 2341 | |
46125c1c MB |
2342 | return r; |
2343 | } | |
2344 | ||
85067747 ML |
2345 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
2346 | { | |
2347 | int r = 0; | |
2348 | ||
2349 | if (!queue_is_mq(md->queue)) | |
2350 | return dm_wait_for_bios_completion(md, task_state); | |
2351 | ||
2352 | while (true) { | |
2353 | if (!blk_mq_queue_inflight(md->queue)) | |
2354 | break; | |
2355 | ||
2356 | if (signal_pending_state(task_state, current)) { | |
2357 | r = -EINTR; | |
2358 | break; | |
2359 | } | |
2360 | ||
2361 | msleep(5); | |
2362 | } | |
2363 | ||
2364 | return r; | |
2365 | } | |
2366 | ||
1da177e4 LT |
2367 | /* |
2368 | * Process the deferred bios | |
2369 | */ | |
ef208587 | 2370 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2371 | { |
0c2915b8 MS |
2372 | struct mapped_device *md = container_of(work, struct mapped_device, work); |
2373 | struct bio *bio; | |
ef208587 | 2374 | |
3b00b203 | 2375 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 | 2376 | spin_lock_irq(&md->deferred_lock); |
0c2915b8 | 2377 | bio = bio_list_pop(&md->deferred); |
df12ee99 AK |
2378 | spin_unlock_irq(&md->deferred_lock); |
2379 | ||
0c2915b8 | 2380 | if (!bio) |
df12ee99 | 2381 | break; |
022c2611 | 2382 | |
0c2915b8 | 2383 | submit_bio_noacct(bio); |
022c2611 | 2384 | } |
1da177e4 LT |
2385 | } |
2386 | ||
9a1fb464 | 2387 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2388 | { |
3b00b203 | 2389 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2390 | smp_mb__after_atomic(); |
53d5914f | 2391 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2392 | } |
2393 | ||
1da177e4 | 2394 | /* |
042d2a9b | 2395 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2396 | */ |
042d2a9b | 2397 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2398 | { |
87eb5b21 | 2399 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2400 | struct queue_limits limits; |
042d2a9b | 2401 | int r; |
1da177e4 | 2402 | |
e61290a4 | 2403 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2404 | |
2405 | /* device must be suspended */ | |
4f186f8b | 2406 | if (!dm_suspended_md(md)) |
93c534ae | 2407 | goto out; |
1da177e4 | 2408 | |
3ae70656 MS |
2409 | /* |
2410 | * If the new table has no data devices, retain the existing limits. | |
2411 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2412 | * then new I/O is queued based on these limits, and then some paths | |
2413 | * reappear. | |
2414 | */ | |
2415 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2416 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2417 | if (live_map) |
2418 | limits = md->queue->limits; | |
83d5e5b0 | 2419 | dm_put_live_table_fast(md); |
3ae70656 MS |
2420 | } |
2421 | ||
87eb5b21 MC |
2422 | if (!live_map) { |
2423 | r = dm_calculate_queue_limits(table, &limits); | |
2424 | if (r) { | |
2425 | map = ERR_PTR(r); | |
2426 | goto out; | |
2427 | } | |
042d2a9b | 2428 | } |
754c5fc7 | 2429 | |
042d2a9b | 2430 | map = __bind(md, table, &limits); |
62e08243 | 2431 | dm_issue_global_event(); |
1da177e4 | 2432 | |
93c534ae | 2433 | out: |
e61290a4 | 2434 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2435 | return map; |
1da177e4 LT |
2436 | } |
2437 | ||
2438 | /* | |
2439 | * Functions to lock and unlock any filesystem running on the | |
2440 | * device. | |
2441 | */ | |
2ca3310e | 2442 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2443 | { |
e39e2e95 | 2444 | int r; |
1da177e4 LT |
2445 | |
2446 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2447 | |
db8fef4f | 2448 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2449 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2450 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2451 | md->frozen_sb = NULL; |
2452 | return r; | |
dfbe03f6 AK |
2453 | } |
2454 | ||
aa8d7c2f AK |
2455 | set_bit(DMF_FROZEN, &md->flags); |
2456 | ||
1da177e4 LT |
2457 | return 0; |
2458 | } | |
2459 | ||
2ca3310e | 2460 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2461 | { |
aa8d7c2f AK |
2462 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2463 | return; | |
2464 | ||
db8fef4f | 2465 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2466 | md->frozen_sb = NULL; |
aa8d7c2f | 2467 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2468 | } |
2469 | ||
2470 | /* | |
b48633f8 BVA |
2471 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
2472 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE | |
2473 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY | |
2474 | * | |
ffcc3936 MS |
2475 | * If __dm_suspend returns 0, the device is completely quiescent |
2476 | * now. There is no request-processing activity. All new requests | |
2477 | * are being added to md->deferred list. | |
cec47e3d | 2478 | */ |
ffcc3936 | 2479 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
b48633f8 | 2480 | unsigned suspend_flags, long task_state, |
eaf9a736 | 2481 | int dmf_suspended_flag) |
1da177e4 | 2482 | { |
ffcc3936 MS |
2483 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2484 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2485 | int r; | |
1da177e4 | 2486 | |
5a8f1f80 BVA |
2487 | lockdep_assert_held(&md->suspend_lock); |
2488 | ||
2e93ccc1 KU |
2489 | /* |
2490 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2491 | * This flag is cleared before dm_suspend returns. | |
2492 | */ | |
2493 | if (noflush) | |
2494 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
86331f39 | 2495 | else |
ac75b09f | 2496 | DMDEBUG("%s: suspending with flush", dm_device_name(md)); |
2e93ccc1 | 2497 | |
d67ee213 MS |
2498 | /* |
2499 | * This gets reverted if there's an error later and the targets | |
2500 | * provide the .presuspend_undo hook. | |
2501 | */ | |
cf222b37 AK |
2502 | dm_table_presuspend_targets(map); |
2503 | ||
32a926da | 2504 | /* |
9f518b27 KU |
2505 | * Flush I/O to the device. |
2506 | * Any I/O submitted after lock_fs() may not be flushed. | |
2507 | * noflush takes precedence over do_lockfs. | |
2508 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2509 | */ |
2510 | if (!noflush && do_lockfs) { | |
2511 | r = lock_fs(md); | |
d67ee213 MS |
2512 | if (r) { |
2513 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2514 | return r; |
d67ee213 | 2515 | } |
aa8d7c2f | 2516 | } |
1da177e4 LT |
2517 | |
2518 | /* | |
3b00b203 MP |
2519 | * Here we must make sure that no processes are submitting requests |
2520 | * to target drivers i.e. no one may be executing | |
0cede372 | 2521 | * __split_and_process_bio from dm_submit_bio. |
3b00b203 | 2522 | * |
0cede372 | 2523 | * To get all processes out of __split_and_process_bio in dm_submit_bio, |
3b00b203 | 2524 | * we take the write lock. To prevent any process from reentering |
0cede372 MS |
2525 | * __split_and_process_bio from dm_submit_bio and quiesce the thread |
2526 | * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call | |
6a8736d1 | 2527 | * flush_workqueue(md->wq). |
1da177e4 | 2528 | */ |
1eb787ec | 2529 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2530 | if (map) |
2531 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2532 | |
d0bcb878 | 2533 | /* |
29e4013d TH |
2534 | * Stop md->queue before flushing md->wq in case request-based |
2535 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2536 | */ |
6a23e05c | 2537 | if (dm_request_based(md)) |
eca7ee6d | 2538 | dm_stop_queue(md->queue); |
cec47e3d | 2539 | |
d0bcb878 KU |
2540 | flush_workqueue(md->wq); |
2541 | ||
1da177e4 | 2542 | /* |
3b00b203 MP |
2543 | * At this point no more requests are entering target request routines. |
2544 | * We call dm_wait_for_completion to wait for all existing requests | |
2545 | * to finish. | |
1da177e4 | 2546 | */ |
b48633f8 | 2547 | r = dm_wait_for_completion(md, task_state); |
eaf9a736 MS |
2548 | if (!r) |
2549 | set_bit(dmf_suspended_flag, &md->flags); | |
1da177e4 | 2550 | |
6d6f10df | 2551 | if (noflush) |
022c2611 | 2552 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2553 | if (map) |
2554 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2555 | |
1da177e4 | 2556 | /* were we interrupted ? */ |
46125c1c | 2557 | if (r < 0) { |
9a1fb464 | 2558 | dm_queue_flush(md); |
73d410c0 | 2559 | |
cec47e3d | 2560 | if (dm_request_based(md)) |
eca7ee6d | 2561 | dm_start_queue(md->queue); |
cec47e3d | 2562 | |
2ca3310e | 2563 | unlock_fs(md); |
d67ee213 | 2564 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2565 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2566 | } |
1da177e4 | 2567 | |
ffcc3936 MS |
2568 | return r; |
2569 | } | |
2570 | ||
2571 | /* | |
2572 | * We need to be able to change a mapping table under a mounted | |
2573 | * filesystem. For example we might want to move some data in | |
2574 | * the background. Before the table can be swapped with | |
2575 | * dm_bind_table, dm_suspend must be called to flush any in | |
2576 | * flight bios and ensure that any further io gets deferred. | |
2577 | */ | |
2578 | /* | |
2579 | * Suspend mechanism in request-based dm. | |
2580 | * | |
2581 | * 1. Flush all I/Os by lock_fs() if needed. | |
2582 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2583 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2584 | * | |
2585 | * To abort suspend, start the request_queue. | |
2586 | */ | |
2587 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2588 | { | |
2589 | struct dm_table *map = NULL; | |
2590 | int r = 0; | |
2591 | ||
2592 | retry: | |
2593 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2594 | ||
2595 | if (dm_suspended_md(md)) { | |
2596 | r = -EINVAL; | |
2597 | goto out_unlock; | |
2598 | } | |
2599 | ||
2600 | if (dm_suspended_internally_md(md)) { | |
2601 | /* already internally suspended, wait for internal resume */ | |
2602 | mutex_unlock(&md->suspend_lock); | |
2603 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2604 | if (r) | |
2605 | return r; | |
2606 | goto retry; | |
2607 | } | |
2608 | ||
a12f5d48 | 2609 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 | 2610 | |
eaf9a736 | 2611 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
ffcc3936 MS |
2612 | if (r) |
2613 | goto out_unlock; | |
3b00b203 | 2614 | |
5df96f2b | 2615 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
4d4471cb | 2616 | dm_table_postsuspend_targets(map); |
5df96f2b | 2617 | clear_bit(DMF_POST_SUSPENDING, &md->flags); |
4d4471cb | 2618 | |
d287483d | 2619 | out_unlock: |
e61290a4 | 2620 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2621 | return r; |
1da177e4 LT |
2622 | } |
2623 | ||
ffcc3936 MS |
2624 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2625 | { | |
2626 | if (map) { | |
2627 | int r = dm_table_resume_targets(map); | |
2628 | if (r) | |
2629 | return r; | |
2630 | } | |
2631 | ||
2632 | dm_queue_flush(md); | |
2633 | ||
2634 | /* | |
2635 | * Flushing deferred I/Os must be done after targets are resumed | |
2636 | * so that mapping of targets can work correctly. | |
2637 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2638 | */ | |
2639 | if (dm_request_based(md)) | |
eca7ee6d | 2640 | dm_start_queue(md->queue); |
ffcc3936 MS |
2641 | |
2642 | unlock_fs(md); | |
2643 | ||
2644 | return 0; | |
2645 | } | |
2646 | ||
1da177e4 LT |
2647 | int dm_resume(struct mapped_device *md) |
2648 | { | |
8dc23658 | 2649 | int r; |
cf222b37 | 2650 | struct dm_table *map = NULL; |
1da177e4 | 2651 | |
ffcc3936 | 2652 | retry: |
8dc23658 | 2653 | r = -EINVAL; |
ffcc3936 MS |
2654 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
2655 | ||
4f186f8b | 2656 | if (!dm_suspended_md(md)) |
cf222b37 | 2657 | goto out; |
cf222b37 | 2658 | |
ffcc3936 MS |
2659 | if (dm_suspended_internally_md(md)) { |
2660 | /* already internally suspended, wait for internal resume */ | |
2661 | mutex_unlock(&md->suspend_lock); | |
2662 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2663 | if (r) | |
2664 | return r; | |
2665 | goto retry; | |
2666 | } | |
2667 | ||
a12f5d48 | 2668 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2669 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2670 | goto out; |
1da177e4 | 2671 | |
ffcc3936 | 2672 | r = __dm_resume(md, map); |
8757b776 MB |
2673 | if (r) |
2674 | goto out; | |
2ca3310e | 2675 | |
2ca3310e | 2676 | clear_bit(DMF_SUSPENDED, &md->flags); |
cf222b37 | 2677 | out: |
e61290a4 | 2678 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2679 | |
cf222b37 | 2680 | return r; |
1da177e4 LT |
2681 | } |
2682 | ||
fd2ed4d2 MP |
2683 | /* |
2684 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2685 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2686 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2687 | */ |
2688 | ||
ffcc3936 | 2689 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2690 | { |
ffcc3936 MS |
2691 | struct dm_table *map = NULL; |
2692 | ||
1ea0654e BVA |
2693 | lockdep_assert_held(&md->suspend_lock); |
2694 | ||
96b26c8c | 2695 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2696 | return; /* nested internal suspend */ |
2697 | ||
2698 | if (dm_suspended_md(md)) { | |
2699 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2700 | return; /* nest suspend */ | |
2701 | } | |
2702 | ||
a12f5d48 | 2703 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2704 | |
2705 | /* | |
2706 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2707 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2708 | * would require changing .presuspend to return an error -- avoid this | |
2709 | * until there is a need for more elaborate variants of internal suspend. | |
2710 | */ | |
eaf9a736 MS |
2711 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
2712 | DMF_SUSPENDED_INTERNALLY); | |
ffcc3936 | 2713 | |
5df96f2b | 2714 | set_bit(DMF_POST_SUSPENDING, &md->flags); |
ffcc3936 | 2715 | dm_table_postsuspend_targets(map); |
5df96f2b | 2716 | clear_bit(DMF_POST_SUSPENDING, &md->flags); |
ffcc3936 MS |
2717 | } |
2718 | ||
2719 | static void __dm_internal_resume(struct mapped_device *md) | |
2720 | { | |
96b26c8c MP |
2721 | BUG_ON(!md->internal_suspend_count); |
2722 | ||
2723 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2724 | return; /* resume from nested internal suspend */ |
2725 | ||
fd2ed4d2 | 2726 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2727 | goto done; /* resume from nested suspend */ |
2728 | ||
2729 | /* | |
2730 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2731 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2732 | */ | |
2733 | (void) __dm_resume(md, NULL); | |
2734 | ||
2735 | done: | |
2736 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2737 | smp_mb__after_atomic(); | |
2738 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2739 | } | |
2740 | ||
2741 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2742 | { | |
2743 | mutex_lock(&md->suspend_lock); | |
2744 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2745 | mutex_unlock(&md->suspend_lock); | |
2746 | } | |
2747 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2748 | ||
2749 | void dm_internal_resume(struct mapped_device *md) | |
2750 | { | |
2751 | mutex_lock(&md->suspend_lock); | |
2752 | __dm_internal_resume(md); | |
2753 | mutex_unlock(&md->suspend_lock); | |
2754 | } | |
2755 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2756 | ||
2757 | /* | |
2758 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2759 | * which prevents interaction with userspace-driven suspend. | |
2760 | */ | |
2761 | ||
2762 | void dm_internal_suspend_fast(struct mapped_device *md) | |
2763 | { | |
2764 | mutex_lock(&md->suspend_lock); | |
2765 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
2766 | return; |
2767 | ||
2768 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2769 | synchronize_srcu(&md->io_barrier); | |
2770 | flush_workqueue(md->wq); | |
2771 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2772 | } | |
b735fede | 2773 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 2774 | |
ffcc3936 | 2775 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 2776 | { |
ffcc3936 | 2777 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
2778 | goto done; |
2779 | ||
2780 | dm_queue_flush(md); | |
2781 | ||
2782 | done: | |
2783 | mutex_unlock(&md->suspend_lock); | |
2784 | } | |
b735fede | 2785 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 2786 | |
1da177e4 LT |
2787 | /*----------------------------------------------------------------- |
2788 | * Event notification. | |
2789 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2790 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2791 | unsigned cookie) |
69267a30 | 2792 | { |
6958c1c6 MP |
2793 | int r; |
2794 | unsigned noio_flag; | |
60935eb2 MB |
2795 | char udev_cookie[DM_COOKIE_LENGTH]; |
2796 | char *envp[] = { udev_cookie, NULL }; | |
2797 | ||
6958c1c6 MP |
2798 | noio_flag = memalloc_noio_save(); |
2799 | ||
60935eb2 | 2800 | if (!cookie) |
6958c1c6 | 2801 | r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2802 | else { |
2803 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2804 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
6958c1c6 MP |
2805 | r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2806 | action, envp); | |
60935eb2 | 2807 | } |
6958c1c6 MP |
2808 | |
2809 | memalloc_noio_restore(noio_flag); | |
2810 | ||
2811 | return r; | |
69267a30 AK |
2812 | } |
2813 | ||
7a8c3d3b MA |
2814 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2815 | { | |
2816 | return atomic_add_return(1, &md->uevent_seq); | |
2817 | } | |
2818 | ||
1da177e4 LT |
2819 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2820 | { | |
2821 | return atomic_read(&md->event_nr); | |
2822 | } | |
2823 | ||
2824 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2825 | { | |
2826 | return wait_event_interruptible(md->eventq, | |
2827 | (event_nr != atomic_read(&md->event_nr))); | |
2828 | } | |
2829 | ||
7a8c3d3b MA |
2830 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2831 | { | |
2832 | unsigned long flags; | |
2833 | ||
2834 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2835 | list_add(elist, &md->uevent_list); | |
2836 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2837 | } | |
2838 | ||
1da177e4 LT |
2839 | /* |
2840 | * The gendisk is only valid as long as you have a reference | |
2841 | * count on 'md'. | |
2842 | */ | |
2843 | struct gendisk *dm_disk(struct mapped_device *md) | |
2844 | { | |
2845 | return md->disk; | |
2846 | } | |
65ff5b7d | 2847 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 2848 | |
784aae73 MB |
2849 | struct kobject *dm_kobject(struct mapped_device *md) |
2850 | { | |
2995fa78 | 2851 | return &md->kobj_holder.kobj; |
784aae73 MB |
2852 | } |
2853 | ||
784aae73 MB |
2854 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2855 | { | |
2856 | struct mapped_device *md; | |
2857 | ||
2995fa78 | 2858 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2859 | |
b9a41d21 HT |
2860 | spin_lock(&_minor_lock); |
2861 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { | |
2862 | md = NULL; | |
2863 | goto out; | |
2864 | } | |
784aae73 | 2865 | dm_get(md); |
b9a41d21 HT |
2866 | out: |
2867 | spin_unlock(&_minor_lock); | |
2868 | ||
784aae73 MB |
2869 | return md; |
2870 | } | |
2871 | ||
4f186f8b | 2872 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2873 | { |
2874 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2875 | } | |
2876 | ||
5df96f2b MP |
2877 | static int dm_post_suspending_md(struct mapped_device *md) |
2878 | { | |
2879 | return test_bit(DMF_POST_SUSPENDING, &md->flags); | |
2880 | } | |
2881 | ||
ffcc3936 MS |
2882 | int dm_suspended_internally_md(struct mapped_device *md) |
2883 | { | |
2884 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2885 | } | |
2886 | ||
2c140a24 MP |
2887 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2888 | { | |
2889 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2890 | } | |
2891 | ||
64dbce58 KU |
2892 | int dm_suspended(struct dm_target *ti) |
2893 | { | |
33bd6f06 | 2894 | return dm_suspended_md(ti->table->md); |
64dbce58 KU |
2895 | } |
2896 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2897 | ||
5df96f2b MP |
2898 | int dm_post_suspending(struct dm_target *ti) |
2899 | { | |
33bd6f06 | 2900 | return dm_post_suspending_md(ti->table->md); |
5df96f2b MP |
2901 | } |
2902 | EXPORT_SYMBOL_GPL(dm_post_suspending); | |
2903 | ||
2e93ccc1 KU |
2904 | int dm_noflush_suspending(struct dm_target *ti) |
2905 | { | |
33bd6f06 | 2906 | return __noflush_suspending(ti->table->md); |
2e93ccc1 KU |
2907 | } |
2908 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2909 | ||
7e0d574f | 2910 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
0776aa0e MS |
2911 | unsigned integrity, unsigned per_io_data_size, |
2912 | unsigned min_pool_size) | |
e6ee8c0b | 2913 | { |
115485e8 | 2914 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
78d8e58a | 2915 | unsigned int pool_size = 0; |
64f52b0e | 2916 | unsigned int front_pad, io_front_pad; |
6f1c819c | 2917 | int ret; |
e6ee8c0b KU |
2918 | |
2919 | if (!pools) | |
4e6e36c3 | 2920 | return NULL; |
e6ee8c0b | 2921 | |
78d8e58a MS |
2922 | switch (type) { |
2923 | case DM_TYPE_BIO_BASED: | |
545ed20e | 2924 | case DM_TYPE_DAX_BIO_BASED: |
22c11858 | 2925 | case DM_TYPE_NVME_BIO_BASED: |
0776aa0e | 2926 | pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size); |
30187e1d | 2927 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
64f52b0e | 2928 | io_front_pad = roundup(front_pad, __alignof__(struct dm_io)) + offsetof(struct dm_io, tio); |
6f1c819c KO |
2929 | ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0); |
2930 | if (ret) | |
64f52b0e | 2931 | goto out; |
6f1c819c | 2932 | if (integrity && bioset_integrity_create(&pools->io_bs, pool_size)) |
eb8db831 | 2933 | goto out; |
78d8e58a MS |
2934 | break; |
2935 | case DM_TYPE_REQUEST_BASED: | |
0776aa0e | 2936 | pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size); |
78d8e58a | 2937 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
591ddcfc | 2938 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
78d8e58a MS |
2939 | break; |
2940 | default: | |
2941 | BUG(); | |
2942 | } | |
2943 | ||
6f1c819c KO |
2944 | ret = bioset_init(&pools->bs, pool_size, front_pad, 0); |
2945 | if (ret) | |
5f015204 | 2946 | goto out; |
e6ee8c0b | 2947 | |
6f1c819c | 2948 | if (integrity && bioset_integrity_create(&pools->bs, pool_size)) |
5f015204 | 2949 | goto out; |
a91a2785 | 2950 | |
e6ee8c0b | 2951 | return pools; |
5f1b670d | 2952 | |
5f1b670d CH |
2953 | out: |
2954 | dm_free_md_mempools(pools); | |
78d8e58a | 2955 | |
4e6e36c3 | 2956 | return NULL; |
e6ee8c0b KU |
2957 | } |
2958 | ||
2959 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2960 | { | |
2961 | if (!pools) | |
2962 | return; | |
2963 | ||
6f1c819c KO |
2964 | bioset_exit(&pools->bs); |
2965 | bioset_exit(&pools->io_bs); | |
e6ee8c0b KU |
2966 | |
2967 | kfree(pools); | |
2968 | } | |
2969 | ||
9c72bad1 CH |
2970 | struct dm_pr { |
2971 | u64 old_key; | |
2972 | u64 new_key; | |
2973 | u32 flags; | |
2974 | bool fail_early; | |
2975 | }; | |
2976 | ||
2977 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, | |
2978 | void *data) | |
71cdb697 CH |
2979 | { |
2980 | struct mapped_device *md = bdev->bd_disk->private_data; | |
9c72bad1 CH |
2981 | struct dm_table *table; |
2982 | struct dm_target *ti; | |
2983 | int ret = -ENOTTY, srcu_idx; | |
71cdb697 | 2984 | |
9c72bad1 CH |
2985 | table = dm_get_live_table(md, &srcu_idx); |
2986 | if (!table || !dm_table_get_size(table)) | |
2987 | goto out; | |
71cdb697 | 2988 | |
9c72bad1 CH |
2989 | /* We only support devices that have a single target */ |
2990 | if (dm_table_get_num_targets(table) != 1) | |
2991 | goto out; | |
2992 | ti = dm_table_get_target(table, 0); | |
71cdb697 | 2993 | |
9c72bad1 CH |
2994 | ret = -EINVAL; |
2995 | if (!ti->type->iterate_devices) | |
2996 | goto out; | |
2997 | ||
2998 | ret = ti->type->iterate_devices(ti, fn, data); | |
2999 | out: | |
3000 | dm_put_live_table(md, srcu_idx); | |
3001 | return ret; | |
3002 | } | |
3003 | ||
3004 | /* | |
3005 | * For register / unregister we need to manually call out to every path. | |
3006 | */ | |
3007 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, | |
3008 | sector_t start, sector_t len, void *data) | |
3009 | { | |
3010 | struct dm_pr *pr = data; | |
3011 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; | |
3012 | ||
3013 | if (!ops || !ops->pr_register) | |
3014 | return -EOPNOTSUPP; | |
3015 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); | |
3016 | } | |
3017 | ||
3018 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, | |
3019 | u32 flags) | |
3020 | { | |
3021 | struct dm_pr pr = { | |
3022 | .old_key = old_key, | |
3023 | .new_key = new_key, | |
3024 | .flags = flags, | |
3025 | .fail_early = true, | |
3026 | }; | |
3027 | int ret; | |
3028 | ||
3029 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); | |
3030 | if (ret && new_key) { | |
3031 | /* unregister all paths if we failed to register any path */ | |
3032 | pr.old_key = new_key; | |
3033 | pr.new_key = 0; | |
3034 | pr.flags = 0; | |
3035 | pr.fail_early = false; | |
3036 | dm_call_pr(bdev, __dm_pr_register, &pr); | |
3037 | } | |
3038 | ||
3039 | return ret; | |
71cdb697 CH |
3040 | } |
3041 | ||
3042 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3043 | u32 flags) |
71cdb697 CH |
3044 | { |
3045 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3046 | const struct pr_ops *ops; | |
971888c4 | 3047 | int r, srcu_idx; |
71cdb697 | 3048 | |
5bd5e8d8 | 3049 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3050 | if (r < 0) |
971888c4 | 3051 | goto out; |
71cdb697 CH |
3052 | |
3053 | ops = bdev->bd_disk->fops->pr_ops; | |
3054 | if (ops && ops->pr_reserve) | |
3055 | r = ops->pr_reserve(bdev, key, type, flags); | |
3056 | else | |
3057 | r = -EOPNOTSUPP; | |
971888c4 MS |
3058 | out: |
3059 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3060 | return r; |
3061 | } | |
3062 | ||
3063 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3064 | { | |
3065 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3066 | const struct pr_ops *ops; | |
971888c4 | 3067 | int r, srcu_idx; |
71cdb697 | 3068 | |
5bd5e8d8 | 3069 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3070 | if (r < 0) |
971888c4 | 3071 | goto out; |
71cdb697 CH |
3072 | |
3073 | ops = bdev->bd_disk->fops->pr_ops; | |
3074 | if (ops && ops->pr_release) | |
3075 | r = ops->pr_release(bdev, key, type); | |
3076 | else | |
3077 | r = -EOPNOTSUPP; | |
971888c4 MS |
3078 | out: |
3079 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3080 | return r; |
3081 | } | |
3082 | ||
3083 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3084 | enum pr_type type, bool abort) |
71cdb697 CH |
3085 | { |
3086 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3087 | const struct pr_ops *ops; | |
971888c4 | 3088 | int r, srcu_idx; |
71cdb697 | 3089 | |
5bd5e8d8 | 3090 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3091 | if (r < 0) |
971888c4 | 3092 | goto out; |
71cdb697 CH |
3093 | |
3094 | ops = bdev->bd_disk->fops->pr_ops; | |
3095 | if (ops && ops->pr_preempt) | |
3096 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3097 | else | |
3098 | r = -EOPNOTSUPP; | |
971888c4 MS |
3099 | out: |
3100 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3101 | return r; |
3102 | } | |
3103 | ||
3104 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3105 | { | |
3106 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3107 | const struct pr_ops *ops; | |
971888c4 | 3108 | int r, srcu_idx; |
71cdb697 | 3109 | |
5bd5e8d8 | 3110 | r = dm_prepare_ioctl(md, &srcu_idx, &bdev); |
71cdb697 | 3111 | if (r < 0) |
971888c4 | 3112 | goto out; |
71cdb697 CH |
3113 | |
3114 | ops = bdev->bd_disk->fops->pr_ops; | |
3115 | if (ops && ops->pr_clear) | |
3116 | r = ops->pr_clear(bdev, key); | |
3117 | else | |
3118 | r = -EOPNOTSUPP; | |
971888c4 MS |
3119 | out: |
3120 | dm_unprepare_ioctl(md, srcu_idx); | |
71cdb697 CH |
3121 | return r; |
3122 | } | |
3123 | ||
3124 | static const struct pr_ops dm_pr_ops = { | |
3125 | .pr_register = dm_pr_register, | |
3126 | .pr_reserve = dm_pr_reserve, | |
3127 | .pr_release = dm_pr_release, | |
3128 | .pr_preempt = dm_pr_preempt, | |
3129 | .pr_clear = dm_pr_clear, | |
3130 | }; | |
3131 | ||
83d5cde4 | 3132 | static const struct block_device_operations dm_blk_dops = { |
c62b37d9 | 3133 | .submit_bio = dm_submit_bio, |
1da177e4 LT |
3134 | .open = dm_blk_open, |
3135 | .release = dm_blk_close, | |
aa129a22 | 3136 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3137 | .getgeo = dm_blk_getgeo, |
e76239a3 | 3138 | .report_zones = dm_blk_report_zones, |
71cdb697 | 3139 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3140 | .owner = THIS_MODULE |
3141 | }; | |
3142 | ||
f26c5719 DW |
3143 | static const struct dax_operations dm_dax_ops = { |
3144 | .direct_access = dm_dax_direct_access, | |
7bf7eac8 | 3145 | .dax_supported = dm_dax_supported, |
7e026c8c | 3146 | .copy_from_iter = dm_dax_copy_from_iter, |
b3a9a0c3 | 3147 | .copy_to_iter = dm_dax_copy_to_iter, |
cdf6cdcd | 3148 | .zero_page_range = dm_dax_zero_page_range, |
f26c5719 DW |
3149 | }; |
3150 | ||
1da177e4 LT |
3151 | /* |
3152 | * module hooks | |
3153 | */ | |
3154 | module_init(dm_init); | |
3155 | module_exit(dm_exit); | |
3156 | ||
3157 | module_param(major, uint, 0); | |
3158 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3159 | |
e8603136 MS |
3160 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3161 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3162 | ||
115485e8 MS |
3163 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
3164 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); | |
3165 | ||
1da177e4 LT |
3166 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3167 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3168 | MODULE_LICENSE("GPL"); |