rjw | 1f88458 | 2022-01-06 17:20:42 +0800 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. |
| 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
| 4 | * |
| 5 | * This file is released under the GPL. |
| 6 | */ |
| 7 | |
| 8 | #include "dm-core.h" |
| 9 | #include "dm-rq.h" |
| 10 | #include "dm-uevent.h" |
| 11 | |
| 12 | #include <linux/init.h> |
| 13 | #include <linux/module.h> |
| 14 | #include <linux/mutex.h> |
| 15 | #include <linux/sched/mm.h> |
| 16 | #include <linux/sched/signal.h> |
| 17 | #include <linux/blkpg.h> |
| 18 | #include <linux/bio.h> |
| 19 | #include <linux/mempool.h> |
| 20 | #include <linux/dax.h> |
| 21 | #include <linux/slab.h> |
| 22 | #include <linux/idr.h> |
| 23 | #include <linux/uio.h> |
| 24 | #include <linux/hdreg.h> |
| 25 | #include <linux/delay.h> |
| 26 | #include <linux/wait.h> |
| 27 | #include <linux/pr.h> |
| 28 | |
| 29 | #define DM_MSG_PREFIX "core" |
| 30 | |
| 31 | /* |
| 32 | * Cookies are numeric values sent with CHANGE and REMOVE |
| 33 | * uevents while resuming, removing or renaming the device. |
| 34 | */ |
| 35 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" |
| 36 | #define DM_COOKIE_LENGTH 24 |
| 37 | |
| 38 | static const char *_name = DM_NAME; |
| 39 | |
| 40 | static unsigned int major = 0; |
| 41 | static unsigned int _major = 0; |
| 42 | |
| 43 | static DEFINE_IDR(_minor_idr); |
| 44 | |
| 45 | static DEFINE_SPINLOCK(_minor_lock); |
| 46 | |
| 47 | static void do_deferred_remove(struct work_struct *w); |
| 48 | |
| 49 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); |
| 50 | |
| 51 | static struct workqueue_struct *deferred_remove_workqueue; |
| 52 | |
| 53 | atomic_t dm_global_event_nr = ATOMIC_INIT(0); |
| 54 | DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq); |
| 55 | |
| 56 | void dm_issue_global_event(void) |
| 57 | { |
| 58 | atomic_inc(&dm_global_event_nr); |
| 59 | wake_up(&dm_global_eventq); |
| 60 | } |
| 61 | |
| 62 | /* |
| 63 | * One of these is allocated per bio. |
| 64 | */ |
| 65 | struct dm_io { |
| 66 | struct mapped_device *md; |
| 67 | blk_status_t status; |
| 68 | atomic_t io_count; |
| 69 | struct bio *bio; |
| 70 | unsigned long start_time; |
| 71 | spinlock_t endio_lock; |
| 72 | struct dm_stats_aux stats_aux; |
| 73 | }; |
| 74 | |
| 75 | #define MINOR_ALLOCED ((void *)-1) |
| 76 | |
| 77 | /* |
| 78 | * Bits for the md->flags field. |
| 79 | */ |
| 80 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
| 81 | #define DMF_SUSPENDED 1 |
| 82 | #define DMF_FROZEN 2 |
| 83 | #define DMF_FREEING 3 |
| 84 | #define DMF_DELETING 4 |
| 85 | #define DMF_NOFLUSH_SUSPENDING 5 |
| 86 | #define DMF_DEFERRED_REMOVE 6 |
| 87 | #define DMF_SUSPENDED_INTERNALLY 7 |
| 88 | |
| 89 | #define DM_NUMA_NODE NUMA_NO_NODE |
| 90 | static int dm_numa_node = DM_NUMA_NODE; |
| 91 | |
| 92 | /* |
| 93 | * For mempools pre-allocation at the table loading time. |
| 94 | */ |
| 95 | struct dm_md_mempools { |
| 96 | mempool_t *io_pool; |
| 97 | struct bio_set *bs; |
| 98 | }; |
| 99 | |
| 100 | struct table_device { |
| 101 | struct list_head list; |
| 102 | atomic_t count; |
| 103 | struct dm_dev dm_dev; |
| 104 | }; |
| 105 | |
| 106 | static struct kmem_cache *_io_cache; |
| 107 | static struct kmem_cache *_rq_tio_cache; |
| 108 | static struct kmem_cache *_rq_cache; |
| 109 | |
| 110 | /* |
| 111 | * Bio-based DM's mempools' reserved IOs set by the user. |
| 112 | */ |
| 113 | #define RESERVED_BIO_BASED_IOS 16 |
| 114 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; |
| 115 | |
| 116 | static int __dm_get_module_param_int(int *module_param, int min, int max) |
| 117 | { |
| 118 | int param = ACCESS_ONCE(*module_param); |
| 119 | int modified_param = 0; |
| 120 | bool modified = true; |
| 121 | |
| 122 | if (param < min) |
| 123 | modified_param = min; |
| 124 | else if (param > max) |
| 125 | modified_param = max; |
| 126 | else |
| 127 | modified = false; |
| 128 | |
| 129 | if (modified) { |
| 130 | (void)cmpxchg(module_param, param, modified_param); |
| 131 | param = modified_param; |
| 132 | } |
| 133 | |
| 134 | return param; |
| 135 | } |
| 136 | |
| 137 | unsigned __dm_get_module_param(unsigned *module_param, |
| 138 | unsigned def, unsigned max) |
| 139 | { |
| 140 | unsigned param = ACCESS_ONCE(*module_param); |
| 141 | unsigned modified_param = 0; |
| 142 | |
| 143 | if (!param) |
| 144 | modified_param = def; |
| 145 | else if (param > max) |
| 146 | modified_param = max; |
| 147 | |
| 148 | if (modified_param) { |
| 149 | (void)cmpxchg(module_param, param, modified_param); |
| 150 | param = modified_param; |
| 151 | } |
| 152 | |
| 153 | return param; |
| 154 | } |
| 155 | |
| 156 | unsigned dm_get_reserved_bio_based_ios(void) |
| 157 | { |
| 158 | return __dm_get_module_param(&reserved_bio_based_ios, |
| 159 | RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS); |
| 160 | } |
| 161 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); |
| 162 | |
| 163 | static unsigned dm_get_numa_node(void) |
| 164 | { |
| 165 | return __dm_get_module_param_int(&dm_numa_node, |
| 166 | DM_NUMA_NODE, num_online_nodes() - 1); |
| 167 | } |
| 168 | |
| 169 | static int __init local_init(void) |
| 170 | { |
| 171 | int r = -ENOMEM; |
| 172 | |
| 173 | /* allocate a slab for the dm_ios */ |
| 174 | _io_cache = KMEM_CACHE(dm_io, 0); |
| 175 | if (!_io_cache) |
| 176 | return r; |
| 177 | |
| 178 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
| 179 | if (!_rq_tio_cache) |
| 180 | goto out_free_io_cache; |
| 181 | |
| 182 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
| 183 | __alignof__(struct request), 0, NULL); |
| 184 | if (!_rq_cache) |
| 185 | goto out_free_rq_tio_cache; |
| 186 | |
| 187 | r = dm_uevent_init(); |
| 188 | if (r) |
| 189 | goto out_free_rq_cache; |
| 190 | |
| 191 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
| 192 | if (!deferred_remove_workqueue) { |
| 193 | r = -ENOMEM; |
| 194 | goto out_uevent_exit; |
| 195 | } |
| 196 | |
| 197 | _major = major; |
| 198 | r = register_blkdev(_major, _name); |
| 199 | if (r < 0) |
| 200 | goto out_free_workqueue; |
| 201 | |
| 202 | if (!_major) |
| 203 | _major = r; |
| 204 | |
| 205 | return 0; |
| 206 | |
| 207 | out_free_workqueue: |
| 208 | destroy_workqueue(deferred_remove_workqueue); |
| 209 | out_uevent_exit: |
| 210 | dm_uevent_exit(); |
| 211 | out_free_rq_cache: |
| 212 | kmem_cache_destroy(_rq_cache); |
| 213 | out_free_rq_tio_cache: |
| 214 | kmem_cache_destroy(_rq_tio_cache); |
| 215 | out_free_io_cache: |
| 216 | kmem_cache_destroy(_io_cache); |
| 217 | |
| 218 | return r; |
| 219 | } |
| 220 | |
| 221 | static void local_exit(void) |
| 222 | { |
| 223 | flush_scheduled_work(); |
| 224 | destroy_workqueue(deferred_remove_workqueue); |
| 225 | |
| 226 | kmem_cache_destroy(_rq_cache); |
| 227 | kmem_cache_destroy(_rq_tio_cache); |
| 228 | kmem_cache_destroy(_io_cache); |
| 229 | unregister_blkdev(_major, _name); |
| 230 | dm_uevent_exit(); |
| 231 | |
| 232 | _major = 0; |
| 233 | |
| 234 | DMINFO("cleaned up"); |
| 235 | } |
| 236 | |
| 237 | static int (*_inits[])(void) __initdata = { |
| 238 | local_init, |
| 239 | dm_target_init, |
| 240 | dm_linear_init, |
| 241 | dm_stripe_init, |
| 242 | dm_io_init, |
| 243 | dm_kcopyd_init, |
| 244 | dm_interface_init, |
| 245 | dm_statistics_init, |
| 246 | }; |
| 247 | |
| 248 | static void (*_exits[])(void) = { |
| 249 | local_exit, |
| 250 | dm_target_exit, |
| 251 | dm_linear_exit, |
| 252 | dm_stripe_exit, |
| 253 | dm_io_exit, |
| 254 | dm_kcopyd_exit, |
| 255 | dm_interface_exit, |
| 256 | dm_statistics_exit, |
| 257 | }; |
| 258 | |
| 259 | static int __init dm_init(void) |
| 260 | { |
| 261 | const int count = ARRAY_SIZE(_inits); |
| 262 | |
| 263 | int r, i; |
| 264 | |
| 265 | for (i = 0; i < count; i++) { |
| 266 | r = _inits[i](); |
| 267 | if (r) |
| 268 | goto bad; |
| 269 | } |
| 270 | |
| 271 | return 0; |
| 272 | |
| 273 | bad: |
| 274 | while (i--) |
| 275 | _exits[i](); |
| 276 | |
| 277 | return r; |
| 278 | } |
| 279 | |
| 280 | static void __exit dm_exit(void) |
| 281 | { |
| 282 | int i = ARRAY_SIZE(_exits); |
| 283 | |
| 284 | while (i--) |
| 285 | _exits[i](); |
| 286 | |
| 287 | /* |
| 288 | * Should be empty by this point. |
| 289 | */ |
| 290 | idr_destroy(&_minor_idr); |
| 291 | } |
| 292 | |
| 293 | /* |
| 294 | * Block device functions |
| 295 | */ |
| 296 | int dm_deleting_md(struct mapped_device *md) |
| 297 | { |
| 298 | return test_bit(DMF_DELETING, &md->flags); |
| 299 | } |
| 300 | |
| 301 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
| 302 | { |
| 303 | struct mapped_device *md; |
| 304 | |
| 305 | spin_lock(&_minor_lock); |
| 306 | |
| 307 | md = bdev->bd_disk->private_data; |
| 308 | if (!md) |
| 309 | goto out; |
| 310 | |
| 311 | if (test_bit(DMF_FREEING, &md->flags) || |
| 312 | dm_deleting_md(md)) { |
| 313 | md = NULL; |
| 314 | goto out; |
| 315 | } |
| 316 | |
| 317 | dm_get(md); |
| 318 | atomic_inc(&md->open_count); |
| 319 | out: |
| 320 | spin_unlock(&_minor_lock); |
| 321 | |
| 322 | return md ? 0 : -ENXIO; |
| 323 | } |
| 324 | |
| 325 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
| 326 | { |
| 327 | struct mapped_device *md; |
| 328 | |
| 329 | spin_lock(&_minor_lock); |
| 330 | |
| 331 | md = disk->private_data; |
| 332 | if (WARN_ON(!md)) |
| 333 | goto out; |
| 334 | |
| 335 | if (atomic_dec_and_test(&md->open_count) && |
| 336 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) |
| 337 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
| 338 | |
| 339 | dm_put(md); |
| 340 | out: |
| 341 | spin_unlock(&_minor_lock); |
| 342 | } |
| 343 | |
| 344 | int dm_open_count(struct mapped_device *md) |
| 345 | { |
| 346 | return atomic_read(&md->open_count); |
| 347 | } |
| 348 | |
| 349 | /* |
| 350 | * Guarantees nothing is using the device before it's deleted. |
| 351 | */ |
| 352 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
| 353 | { |
| 354 | int r = 0; |
| 355 | |
| 356 | spin_lock(&_minor_lock); |
| 357 | |
| 358 | if (dm_open_count(md)) { |
| 359 | r = -EBUSY; |
| 360 | if (mark_deferred) |
| 361 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); |
| 362 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) |
| 363 | r = -EEXIST; |
| 364 | else |
| 365 | set_bit(DMF_DELETING, &md->flags); |
| 366 | |
| 367 | spin_unlock(&_minor_lock); |
| 368 | |
| 369 | return r; |
| 370 | } |
| 371 | |
| 372 | int dm_cancel_deferred_remove(struct mapped_device *md) |
| 373 | { |
| 374 | int r = 0; |
| 375 | |
| 376 | spin_lock(&_minor_lock); |
| 377 | |
| 378 | if (test_bit(DMF_DELETING, &md->flags)) |
| 379 | r = -EBUSY; |
| 380 | else |
| 381 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); |
| 382 | |
| 383 | spin_unlock(&_minor_lock); |
| 384 | |
| 385 | return r; |
| 386 | } |
| 387 | |
| 388 | static void do_deferred_remove(struct work_struct *w) |
| 389 | { |
| 390 | dm_deferred_remove(); |
| 391 | } |
| 392 | |
| 393 | sector_t dm_get_size(struct mapped_device *md) |
| 394 | { |
| 395 | return get_capacity(md->disk); |
| 396 | } |
| 397 | |
| 398 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
| 399 | { |
| 400 | return md->queue; |
| 401 | } |
| 402 | |
| 403 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
| 404 | { |
| 405 | return &md->stats; |
| 406 | } |
| 407 | |
| 408 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
| 409 | { |
| 410 | struct mapped_device *md = bdev->bd_disk->private_data; |
| 411 | |
| 412 | return dm_get_geometry(md, geo); |
| 413 | } |
| 414 | |
| 415 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
| 416 | struct block_device **bdev, |
| 417 | fmode_t *mode) |
| 418 | { |
| 419 | struct dm_target *tgt; |
| 420 | struct dm_table *map; |
| 421 | int srcu_idx, r; |
| 422 | |
| 423 | retry: |
| 424 | r = -ENOTTY; |
| 425 | map = dm_get_live_table(md, &srcu_idx); |
| 426 | if (!map || !dm_table_get_size(map)) |
| 427 | goto out; |
| 428 | |
| 429 | /* We only support devices that have a single target */ |
| 430 | if (dm_table_get_num_targets(map) != 1) |
| 431 | goto out; |
| 432 | |
| 433 | tgt = dm_table_get_target(map, 0); |
| 434 | if (!tgt->type->prepare_ioctl) |
| 435 | goto out; |
| 436 | |
| 437 | if (dm_suspended_md(md)) { |
| 438 | r = -EAGAIN; |
| 439 | goto out; |
| 440 | } |
| 441 | |
| 442 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
| 443 | if (r < 0) |
| 444 | goto out; |
| 445 | |
| 446 | bdgrab(*bdev); |
| 447 | dm_put_live_table(md, srcu_idx); |
| 448 | return r; |
| 449 | |
| 450 | out: |
| 451 | dm_put_live_table(md, srcu_idx); |
| 452 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
| 453 | msleep(10); |
| 454 | goto retry; |
| 455 | } |
| 456 | return r; |
| 457 | } |
| 458 | |
| 459 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
| 460 | unsigned int cmd, unsigned long arg) |
| 461 | { |
| 462 | struct mapped_device *md = bdev->bd_disk->private_data; |
| 463 | int r; |
| 464 | |
| 465 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
| 466 | if (r < 0) |
| 467 | return r; |
| 468 | |
| 469 | if (r > 0) { |
| 470 | /* |
| 471 | * Target determined this ioctl is being issued against a |
| 472 | * subset of the parent bdev; require extra privileges. |
| 473 | */ |
| 474 | if (!capable(CAP_SYS_RAWIO)) { |
| 475 | DMWARN_LIMIT( |
| 476 | "%s: sending ioctl %x to DM device without required privilege.", |
| 477 | current->comm, cmd); |
| 478 | r = -ENOIOCTLCMD; |
| 479 | goto out; |
| 480 | } |
| 481 | } |
| 482 | |
| 483 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
| 484 | out: |
| 485 | bdput(bdev); |
| 486 | return r; |
| 487 | } |
| 488 | |
| 489 | static struct dm_io *alloc_io(struct mapped_device *md) |
| 490 | { |
| 491 | return mempool_alloc(md->io_pool, GFP_NOIO); |
| 492 | } |
| 493 | |
| 494 | static void free_io(struct mapped_device *md, struct dm_io *io) |
| 495 | { |
| 496 | mempool_free(io, md->io_pool); |
| 497 | } |
| 498 | |
| 499 | static void free_tio(struct dm_target_io *tio) |
| 500 | { |
| 501 | bio_put(&tio->clone); |
| 502 | } |
| 503 | |
| 504 | int md_in_flight(struct mapped_device *md) |
| 505 | { |
| 506 | return atomic_read(&md->pending[READ]) + |
| 507 | atomic_read(&md->pending[WRITE]); |
| 508 | } |
| 509 | |
| 510 | static void start_io_acct(struct dm_io *io) |
| 511 | { |
| 512 | struct mapped_device *md = io->md; |
| 513 | struct bio *bio = io->bio; |
| 514 | int cpu; |
| 515 | int rw = bio_data_dir(bio); |
| 516 | |
| 517 | io->start_time = jiffies; |
| 518 | |
| 519 | cpu = part_stat_lock(); |
| 520 | part_round_stats(md->queue, cpu, &dm_disk(md)->part0); |
| 521 | part_stat_unlock(); |
| 522 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
| 523 | atomic_inc_return(&md->pending[rw])); |
| 524 | |
| 525 | if (unlikely(dm_stats_used(&md->stats))) |
| 526 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
| 527 | bio->bi_iter.bi_sector, bio_sectors(bio), |
| 528 | false, 0, &io->stats_aux); |
| 529 | } |
| 530 | |
| 531 | static void end_io_acct(struct dm_io *io) |
| 532 | { |
| 533 | struct mapped_device *md = io->md; |
| 534 | struct bio *bio = io->bio; |
| 535 | unsigned long duration = jiffies - io->start_time; |
| 536 | int pending; |
| 537 | int rw = bio_data_dir(bio); |
| 538 | |
| 539 | generic_end_io_acct(md->queue, rw, &dm_disk(md)->part0, io->start_time); |
| 540 | |
| 541 | if (unlikely(dm_stats_used(&md->stats))) |
| 542 | dm_stats_account_io(&md->stats, bio_data_dir(bio), |
| 543 | bio->bi_iter.bi_sector, bio_sectors(bio), |
| 544 | true, duration, &io->stats_aux); |
| 545 | |
| 546 | /* |
| 547 | * After this is decremented the bio must not be touched if it is |
| 548 | * a flush. |
| 549 | */ |
| 550 | pending = atomic_dec_return(&md->pending[rw]); |
| 551 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); |
| 552 | pending += atomic_read(&md->pending[rw^0x1]); |
| 553 | |
| 554 | /* nudge anyone waiting on suspend queue */ |
| 555 | if (!pending) |
| 556 | wake_up(&md->wait); |
| 557 | } |
| 558 | |
| 559 | /* |
| 560 | * Add the bio to the list of deferred io. |
| 561 | */ |
| 562 | static void queue_io(struct mapped_device *md, struct bio *bio) |
| 563 | { |
| 564 | unsigned long flags; |
| 565 | |
| 566 | spin_lock_irqsave(&md->deferred_lock, flags); |
| 567 | bio_list_add(&md->deferred, bio); |
| 568 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
| 569 | queue_work(md->wq, &md->work); |
| 570 | } |
| 571 | |
| 572 | /* |
| 573 | * Everyone (including functions in this file), should use this |
| 574 | * function to access the md->map field, and make sure they call |
| 575 | * dm_put_live_table() when finished. |
| 576 | */ |
| 577 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
| 578 | { |
| 579 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
| 580 | |
| 581 | return srcu_dereference(md->map, &md->io_barrier); |
| 582 | } |
| 583 | |
| 584 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
| 585 | { |
| 586 | srcu_read_unlock(&md->io_barrier, srcu_idx); |
| 587 | } |
| 588 | |
| 589 | void dm_sync_table(struct mapped_device *md) |
| 590 | { |
| 591 | synchronize_srcu(&md->io_barrier); |
| 592 | synchronize_rcu_expedited(); |
| 593 | } |
| 594 | |
| 595 | /* |
| 596 | * A fast alternative to dm_get_live_table/dm_put_live_table. |
| 597 | * The caller must not block between these two functions. |
| 598 | */ |
| 599 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) |
| 600 | { |
| 601 | rcu_read_lock(); |
| 602 | return rcu_dereference(md->map); |
| 603 | } |
| 604 | |
| 605 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
| 606 | { |
| 607 | rcu_read_unlock(); |
| 608 | } |
| 609 | |
| 610 | /* |
| 611 | * Open a table device so we can use it as a map destination. |
| 612 | */ |
| 613 | static int open_table_device(struct table_device *td, dev_t dev, |
| 614 | struct mapped_device *md) |
| 615 | { |
| 616 | static char *_claim_ptr = "I belong to device-mapper"; |
| 617 | struct block_device *bdev; |
| 618 | |
| 619 | int r; |
| 620 | |
| 621 | BUG_ON(td->dm_dev.bdev); |
| 622 | |
| 623 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); |
| 624 | if (IS_ERR(bdev)) |
| 625 | return PTR_ERR(bdev); |
| 626 | |
| 627 | r = bd_link_disk_holder(bdev, dm_disk(md)); |
| 628 | if (r) { |
| 629 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); |
| 630 | return r; |
| 631 | } |
| 632 | |
| 633 | td->dm_dev.bdev = bdev; |
| 634 | td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name); |
| 635 | return 0; |
| 636 | } |
| 637 | |
| 638 | /* |
| 639 | * Close a table device that we've been using. |
| 640 | */ |
| 641 | static void close_table_device(struct table_device *td, struct mapped_device *md) |
| 642 | { |
| 643 | if (!td->dm_dev.bdev) |
| 644 | return; |
| 645 | |
| 646 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); |
| 647 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); |
| 648 | put_dax(td->dm_dev.dax_dev); |
| 649 | td->dm_dev.bdev = NULL; |
| 650 | td->dm_dev.dax_dev = NULL; |
| 651 | } |
| 652 | |
| 653 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, |
| 654 | fmode_t mode) { |
| 655 | struct table_device *td; |
| 656 | |
| 657 | list_for_each_entry(td, l, list) |
| 658 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) |
| 659 | return td; |
| 660 | |
| 661 | return NULL; |
| 662 | } |
| 663 | |
| 664 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, |
| 665 | struct dm_dev **result) { |
| 666 | int r; |
| 667 | struct table_device *td; |
| 668 | |
| 669 | mutex_lock(&md->table_devices_lock); |
| 670 | td = find_table_device(&md->table_devices, dev, mode); |
| 671 | if (!td) { |
| 672 | td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id); |
| 673 | if (!td) { |
| 674 | mutex_unlock(&md->table_devices_lock); |
| 675 | return -ENOMEM; |
| 676 | } |
| 677 | |
| 678 | td->dm_dev.mode = mode; |
| 679 | td->dm_dev.bdev = NULL; |
| 680 | |
| 681 | if ((r = open_table_device(td, dev, md))) { |
| 682 | mutex_unlock(&md->table_devices_lock); |
| 683 | kfree(td); |
| 684 | return r; |
| 685 | } |
| 686 | |
| 687 | format_dev_t(td->dm_dev.name, dev); |
| 688 | |
| 689 | atomic_set(&td->count, 0); |
| 690 | list_add(&td->list, &md->table_devices); |
| 691 | } |
| 692 | atomic_inc(&td->count); |
| 693 | mutex_unlock(&md->table_devices_lock); |
| 694 | |
| 695 | *result = &td->dm_dev; |
| 696 | return 0; |
| 697 | } |
| 698 | EXPORT_SYMBOL_GPL(dm_get_table_device); |
| 699 | |
| 700 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) |
| 701 | { |
| 702 | struct table_device *td = container_of(d, struct table_device, dm_dev); |
| 703 | |
| 704 | mutex_lock(&md->table_devices_lock); |
| 705 | if (atomic_dec_and_test(&td->count)) { |
| 706 | close_table_device(td, md); |
| 707 | list_del(&td->list); |
| 708 | kfree(td); |
| 709 | } |
| 710 | mutex_unlock(&md->table_devices_lock); |
| 711 | } |
| 712 | EXPORT_SYMBOL(dm_put_table_device); |
| 713 | |
| 714 | static void free_table_devices(struct list_head *devices) |
| 715 | { |
| 716 | struct list_head *tmp, *next; |
| 717 | |
| 718 | list_for_each_safe(tmp, next, devices) { |
| 719 | struct table_device *td = list_entry(tmp, struct table_device, list); |
| 720 | |
| 721 | DMWARN("dm_destroy: %s still exists with %d references", |
| 722 | td->dm_dev.name, atomic_read(&td->count)); |
| 723 | kfree(td); |
| 724 | } |
| 725 | } |
| 726 | |
| 727 | /* |
| 728 | * Get the geometry associated with a dm device |
| 729 | */ |
| 730 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) |
| 731 | { |
| 732 | *geo = md->geometry; |
| 733 | |
| 734 | return 0; |
| 735 | } |
| 736 | |
| 737 | /* |
| 738 | * Set the geometry of a device. |
| 739 | */ |
| 740 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) |
| 741 | { |
| 742 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; |
| 743 | |
| 744 | if (geo->start > sz) { |
| 745 | DMWARN("Start sector is beyond the geometry limits."); |
| 746 | return -EINVAL; |
| 747 | } |
| 748 | |
| 749 | md->geometry = *geo; |
| 750 | |
| 751 | return 0; |
| 752 | } |
| 753 | |
| 754 | /*----------------------------------------------------------------- |
| 755 | * CRUD START: |
| 756 | * A more elegant soln is in the works that uses the queue |
| 757 | * merge fn, unfortunately there are a couple of changes to |
| 758 | * the block layer that I want to make for this. So in the |
| 759 | * interests of getting something for people to use I give |
| 760 | * you this clearly demarcated crap. |
| 761 | *---------------------------------------------------------------*/ |
| 762 | |
| 763 | static int __noflush_suspending(struct mapped_device *md) |
| 764 | { |
| 765 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
| 766 | } |
| 767 | |
| 768 | /* |
| 769 | * Decrements the number of outstanding ios that a bio has been |
| 770 | * cloned into, completing the original io if necc. |
| 771 | */ |
| 772 | static void dec_pending(struct dm_io *io, blk_status_t error) |
| 773 | { |
| 774 | unsigned long flags; |
| 775 | blk_status_t io_error; |
| 776 | struct bio *bio; |
| 777 | struct mapped_device *md = io->md; |
| 778 | |
| 779 | /* Push-back supersedes any I/O errors */ |
| 780 | if (unlikely(error)) { |
| 781 | spin_lock_irqsave(&io->endio_lock, flags); |
| 782 | if (!(io->status == BLK_STS_DM_REQUEUE && |
| 783 | __noflush_suspending(md))) |
| 784 | io->status = error; |
| 785 | spin_unlock_irqrestore(&io->endio_lock, flags); |
| 786 | } |
| 787 | |
| 788 | if (atomic_dec_and_test(&io->io_count)) { |
| 789 | if (io->status == BLK_STS_DM_REQUEUE) { |
| 790 | /* |
| 791 | * Target requested pushing back the I/O. |
| 792 | */ |
| 793 | spin_lock_irqsave(&md->deferred_lock, flags); |
| 794 | if (__noflush_suspending(md)) |
| 795 | bio_list_add_head(&md->deferred, io->bio); |
| 796 | else |
| 797 | /* noflush suspend was interrupted. */ |
| 798 | io->status = BLK_STS_IOERR; |
| 799 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
| 800 | } |
| 801 | |
| 802 | io_error = io->status; |
| 803 | bio = io->bio; |
| 804 | end_io_acct(io); |
| 805 | free_io(md, io); |
| 806 | |
| 807 | if (io_error == BLK_STS_DM_REQUEUE) |
| 808 | return; |
| 809 | |
| 810 | if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) { |
| 811 | /* |
| 812 | * Preflush done for flush with data, reissue |
| 813 | * without REQ_PREFLUSH. |
| 814 | */ |
| 815 | bio->bi_opf &= ~REQ_PREFLUSH; |
| 816 | queue_io(md, bio); |
| 817 | } else { |
| 818 | /* done with normal IO or empty flush */ |
| 819 | if (io_error) |
| 820 | bio->bi_status = io_error; |
| 821 | bio_endio(bio); |
| 822 | } |
| 823 | } |
| 824 | } |
| 825 | |
| 826 | void disable_write_same(struct mapped_device *md) |
| 827 | { |
| 828 | struct queue_limits *limits = dm_get_queue_limits(md); |
| 829 | |
| 830 | /* device doesn't really support WRITE SAME, disable it */ |
| 831 | limits->max_write_same_sectors = 0; |
| 832 | } |
| 833 | |
| 834 | void disable_write_zeroes(struct mapped_device *md) |
| 835 | { |
| 836 | struct queue_limits *limits = dm_get_queue_limits(md); |
| 837 | |
| 838 | /* device doesn't really support WRITE ZEROES, disable it */ |
| 839 | limits->max_write_zeroes_sectors = 0; |
| 840 | } |
| 841 | |
| 842 | static void clone_endio(struct bio *bio) |
| 843 | { |
| 844 | blk_status_t error = bio->bi_status; |
| 845 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
| 846 | struct dm_io *io = tio->io; |
| 847 | struct mapped_device *md = tio->io->md; |
| 848 | dm_endio_fn endio = tio->ti->type->end_io; |
| 849 | |
| 850 | if (unlikely(error == BLK_STS_TARGET)) { |
| 851 | if (bio_op(bio) == REQ_OP_WRITE_SAME && |
| 852 | !bio->bi_disk->queue->limits.max_write_same_sectors) |
| 853 | disable_write_same(md); |
| 854 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && |
| 855 | !bio->bi_disk->queue->limits.max_write_zeroes_sectors) |
| 856 | disable_write_zeroes(md); |
| 857 | } |
| 858 | |
| 859 | if (endio) { |
| 860 | int r = endio(tio->ti, bio, &error); |
| 861 | switch (r) { |
| 862 | case DM_ENDIO_REQUEUE: |
| 863 | error = BLK_STS_DM_REQUEUE; |
| 864 | /*FALLTHRU*/ |
| 865 | case DM_ENDIO_DONE: |
| 866 | break; |
| 867 | case DM_ENDIO_INCOMPLETE: |
| 868 | /* The target will handle the io */ |
| 869 | return; |
| 870 | default: |
| 871 | DMWARN("unimplemented target endio return value: %d", r); |
| 872 | BUG(); |
| 873 | } |
| 874 | } |
| 875 | |
| 876 | free_tio(tio); |
| 877 | dec_pending(io, error); |
| 878 | } |
| 879 | |
| 880 | /* |
| 881 | * Return maximum size of I/O possible at the supplied sector up to the current |
| 882 | * target boundary. |
| 883 | */ |
| 884 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) |
| 885 | { |
| 886 | sector_t target_offset = dm_target_offset(ti, sector); |
| 887 | |
| 888 | return ti->len - target_offset; |
| 889 | } |
| 890 | |
| 891 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) |
| 892 | { |
| 893 | sector_t len = max_io_len_target_boundary(sector, ti); |
| 894 | sector_t offset, max_len; |
| 895 | |
| 896 | /* |
| 897 | * Does the target need to split even further? |
| 898 | */ |
| 899 | if (ti->max_io_len) { |
| 900 | offset = dm_target_offset(ti, sector); |
| 901 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) |
| 902 | max_len = sector_div(offset, ti->max_io_len); |
| 903 | else |
| 904 | max_len = offset & (ti->max_io_len - 1); |
| 905 | max_len = ti->max_io_len - max_len; |
| 906 | |
| 907 | if (len > max_len) |
| 908 | len = max_len; |
| 909 | } |
| 910 | |
| 911 | return len; |
| 912 | } |
| 913 | |
| 914 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
| 915 | { |
| 916 | if (len > UINT_MAX) { |
| 917 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", |
| 918 | (unsigned long long)len, UINT_MAX); |
| 919 | ti->error = "Maximum size of target IO is too large"; |
| 920 | return -EINVAL; |
| 921 | } |
| 922 | |
| 923 | ti->max_io_len = (uint32_t) len; |
| 924 | |
| 925 | return 0; |
| 926 | } |
| 927 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); |
| 928 | |
| 929 | static struct dm_target *dm_dax_get_live_target(struct mapped_device *md, |
| 930 | sector_t sector, int *srcu_idx) |
| 931 | { |
| 932 | struct dm_table *map; |
| 933 | struct dm_target *ti; |
| 934 | |
| 935 | map = dm_get_live_table(md, srcu_idx); |
| 936 | if (!map) |
| 937 | return NULL; |
| 938 | |
| 939 | ti = dm_table_find_target(map, sector); |
| 940 | if (!dm_target_is_valid(ti)) |
| 941 | return NULL; |
| 942 | |
| 943 | return ti; |
| 944 | } |
| 945 | |
| 946 | static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, |
| 947 | long nr_pages, void **kaddr, pfn_t *pfn) |
| 948 | { |
| 949 | struct mapped_device *md = dax_get_private(dax_dev); |
| 950 | sector_t sector = pgoff * PAGE_SECTORS; |
| 951 | struct dm_target *ti; |
| 952 | long len, ret = -EIO; |
| 953 | int srcu_idx; |
| 954 | |
| 955 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
| 956 | |
| 957 | if (!ti) |
| 958 | goto out; |
| 959 | if (!ti->type->direct_access) |
| 960 | goto out; |
| 961 | len = max_io_len(sector, ti) / PAGE_SECTORS; |
| 962 | if (len < 1) |
| 963 | goto out; |
| 964 | nr_pages = min(len, nr_pages); |
| 965 | ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn); |
| 966 | |
| 967 | out: |
| 968 | dm_put_live_table(md, srcu_idx); |
| 969 | |
| 970 | return ret; |
| 971 | } |
| 972 | |
| 973 | static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, |
| 974 | void *addr, size_t bytes, struct iov_iter *i) |
| 975 | { |
| 976 | struct mapped_device *md = dax_get_private(dax_dev); |
| 977 | sector_t sector = pgoff * PAGE_SECTORS; |
| 978 | struct dm_target *ti; |
| 979 | long ret = 0; |
| 980 | int srcu_idx; |
| 981 | |
| 982 | ti = dm_dax_get_live_target(md, sector, &srcu_idx); |
| 983 | |
| 984 | if (!ti) |
| 985 | goto out; |
| 986 | if (!ti->type->dax_copy_from_iter) { |
| 987 | ret = copy_from_iter(addr, bytes, i); |
| 988 | goto out; |
| 989 | } |
| 990 | ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); |
| 991 | out: |
| 992 | dm_put_live_table(md, srcu_idx); |
| 993 | |
| 994 | return ret; |
| 995 | } |
| 996 | |
| 997 | /* |
| 998 | * A target may call dm_accept_partial_bio only from the map routine. It is |
| 999 | * allowed for all bio types except REQ_PREFLUSH. |
| 1000 | * |
| 1001 | * dm_accept_partial_bio informs the dm that the target only wants to process |
| 1002 | * additional n_sectors sectors of the bio and the rest of the data should be |
| 1003 | * sent in a next bio. |
| 1004 | * |
| 1005 | * A diagram that explains the arithmetics: |
| 1006 | * +--------------------+---------------+-------+ |
| 1007 | * | 1 | 2 | 3 | |
| 1008 | * +--------------------+---------------+-------+ |
| 1009 | * |
| 1010 | * <-------------- *tio->len_ptr ---------------> |
| 1011 | * <------- bi_size -------> |
| 1012 | * <-- n_sectors --> |
| 1013 | * |
| 1014 | * Region 1 was already iterated over with bio_advance or similar function. |
| 1015 | * (it may be empty if the target doesn't use bio_advance) |
| 1016 | * Region 2 is the remaining bio size that the target wants to process. |
| 1017 | * (it may be empty if region 1 is non-empty, although there is no reason |
| 1018 | * to make it empty) |
| 1019 | * The target requires that region 3 is to be sent in the next bio. |
| 1020 | * |
| 1021 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), |
| 1022 | * the partially processed part (the sum of regions 1+2) must be the same for all |
| 1023 | * copies of the bio. |
| 1024 | */ |
| 1025 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) |
| 1026 | { |
| 1027 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
| 1028 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; |
| 1029 | BUG_ON(bio->bi_opf & REQ_PREFLUSH); |
| 1030 | BUG_ON(bi_size > *tio->len_ptr); |
| 1031 | BUG_ON(n_sectors > bi_size); |
| 1032 | *tio->len_ptr -= bi_size - n_sectors; |
| 1033 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; |
| 1034 | } |
| 1035 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); |
| 1036 | |
| 1037 | /* |
| 1038 | * The zone descriptors obtained with a zone report indicate zone positions |
| 1039 | * within the target backing device, regardless of that device is a partition |
| 1040 | * and regardless of the target mapping start sector on the device or partition. |
| 1041 | * The zone descriptors start sector and write pointer position must be adjusted |
| 1042 | * to match their relative position within the dm device. |
| 1043 | * A target may call dm_remap_zone_report() after completion of a |
| 1044 | * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained from the |
| 1045 | * backing device. |
| 1046 | */ |
| 1047 | void dm_remap_zone_report(struct dm_target *ti, struct bio *bio, sector_t start) |
| 1048 | { |
| 1049 | #ifdef CONFIG_BLK_DEV_ZONED |
| 1050 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
| 1051 | struct bio *report_bio = tio->io->bio; |
| 1052 | struct blk_zone_report_hdr *hdr = NULL; |
| 1053 | struct blk_zone *zone; |
| 1054 | unsigned int nr_rep = 0; |
| 1055 | unsigned int ofst; |
| 1056 | sector_t part_offset; |
| 1057 | struct bio_vec bvec; |
| 1058 | struct bvec_iter iter; |
| 1059 | void *addr; |
| 1060 | |
| 1061 | if (bio->bi_status) |
| 1062 | return; |
| 1063 | |
| 1064 | /* |
| 1065 | * bio sector was incremented by the request size on completion. Taking |
| 1066 | * into account the original request sector, the target start offset on |
| 1067 | * the backing device and the target mapping offset (ti->begin), the |
| 1068 | * start sector of the backing device. The partition offset is always 0 |
| 1069 | * if the target uses a whole device. |
| 1070 | */ |
| 1071 | part_offset = bio->bi_iter.bi_sector + ti->begin - (start + bio_end_sector(report_bio)); |
| 1072 | |
| 1073 | /* |
| 1074 | * Remap the start sector of the reported zones. For sequential zones, |
| 1075 | * also remap the write pointer position. |
| 1076 | */ |
| 1077 | bio_for_each_segment(bvec, report_bio, iter) { |
| 1078 | addr = kmap_atomic(bvec.bv_page); |
| 1079 | |
| 1080 | /* Remember the report header in the first page */ |
| 1081 | if (!hdr) { |
| 1082 | hdr = addr; |
| 1083 | ofst = sizeof(struct blk_zone_report_hdr); |
| 1084 | } else |
| 1085 | ofst = 0; |
| 1086 | |
| 1087 | /* Set zones start sector */ |
| 1088 | while (hdr->nr_zones && ofst < bvec.bv_len) { |
| 1089 | zone = addr + ofst; |
| 1090 | zone->start -= part_offset; |
| 1091 | if (zone->start >= start + ti->len) { |
| 1092 | hdr->nr_zones = 0; |
| 1093 | break; |
| 1094 | } |
| 1095 | zone->start = zone->start + ti->begin - start; |
| 1096 | if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) { |
| 1097 | if (zone->cond == BLK_ZONE_COND_FULL) |
| 1098 | zone->wp = zone->start + zone->len; |
| 1099 | else if (zone->cond == BLK_ZONE_COND_EMPTY) |
| 1100 | zone->wp = zone->start; |
| 1101 | else |
| 1102 | zone->wp = zone->wp + ti->begin - start - part_offset; |
| 1103 | } |
| 1104 | ofst += sizeof(struct blk_zone); |
| 1105 | hdr->nr_zones--; |
| 1106 | nr_rep++; |
| 1107 | } |
| 1108 | |
| 1109 | if (addr != hdr) |
| 1110 | kunmap_atomic(addr); |
| 1111 | |
| 1112 | if (!hdr->nr_zones) |
| 1113 | break; |
| 1114 | } |
| 1115 | |
| 1116 | if (hdr) { |
| 1117 | hdr->nr_zones = nr_rep; |
| 1118 | kunmap_atomic(hdr); |
| 1119 | } |
| 1120 | |
| 1121 | bio_advance(report_bio, report_bio->bi_iter.bi_size); |
| 1122 | |
| 1123 | #else /* !CONFIG_BLK_DEV_ZONED */ |
| 1124 | bio->bi_status = BLK_STS_NOTSUPP; |
| 1125 | #endif |
| 1126 | } |
| 1127 | EXPORT_SYMBOL_GPL(dm_remap_zone_report); |
| 1128 | |
| 1129 | /* |
| 1130 | * Flush current->bio_list when the target map method blocks. |
| 1131 | * This fixes deadlocks in snapshot and possibly in other targets. |
| 1132 | */ |
| 1133 | struct dm_offload { |
| 1134 | struct blk_plug plug; |
| 1135 | struct blk_plug_cb cb; |
| 1136 | }; |
| 1137 | |
| 1138 | static void flush_current_bio_list(struct blk_plug_cb *cb, bool from_schedule) |
| 1139 | { |
| 1140 | struct dm_offload *o = container_of(cb, struct dm_offload, cb); |
| 1141 | struct bio_list list; |
| 1142 | struct bio *bio; |
| 1143 | int i; |
| 1144 | |
| 1145 | INIT_LIST_HEAD(&o->cb.list); |
| 1146 | |
| 1147 | if (unlikely(!current->bio_list)) |
| 1148 | return; |
| 1149 | |
| 1150 | for (i = 0; i < 2; i++) { |
| 1151 | list = current->bio_list[i]; |
| 1152 | bio_list_init(¤t->bio_list[i]); |
| 1153 | |
| 1154 | while ((bio = bio_list_pop(&list))) { |
| 1155 | struct bio_set *bs = bio->bi_pool; |
| 1156 | if (unlikely(!bs) || bs == fs_bio_set || |
| 1157 | !bs->rescue_workqueue) { |
| 1158 | bio_list_add(¤t->bio_list[i], bio); |
| 1159 | continue; |
| 1160 | } |
| 1161 | |
| 1162 | spin_lock(&bs->rescue_lock); |
| 1163 | bio_list_add(&bs->rescue_list, bio); |
| 1164 | queue_work(bs->rescue_workqueue, &bs->rescue_work); |
| 1165 | spin_unlock(&bs->rescue_lock); |
| 1166 | } |
| 1167 | } |
| 1168 | } |
| 1169 | |
| 1170 | static void dm_offload_start(struct dm_offload *o) |
| 1171 | { |
| 1172 | blk_start_plug(&o->plug); |
| 1173 | o->cb.callback = flush_current_bio_list; |
| 1174 | list_add(&o->cb.list, ¤t->plug->cb_list); |
| 1175 | } |
| 1176 | |
| 1177 | static void dm_offload_end(struct dm_offload *o) |
| 1178 | { |
| 1179 | list_del(&o->cb.list); |
| 1180 | blk_finish_plug(&o->plug); |
| 1181 | } |
| 1182 | |
| 1183 | static void __map_bio(struct dm_target_io *tio) |
| 1184 | { |
| 1185 | int r; |
| 1186 | sector_t sector; |
| 1187 | struct dm_offload o; |
| 1188 | struct bio *clone = &tio->clone; |
| 1189 | struct dm_target *ti = tio->ti; |
| 1190 | |
| 1191 | clone->bi_end_io = clone_endio; |
| 1192 | |
| 1193 | /* |
| 1194 | * Map the clone. If r == 0 we don't need to do |
| 1195 | * anything, the target has assumed ownership of |
| 1196 | * this io. |
| 1197 | */ |
| 1198 | atomic_inc(&tio->io->io_count); |
| 1199 | sector = clone->bi_iter.bi_sector; |
| 1200 | |
| 1201 | dm_offload_start(&o); |
| 1202 | r = ti->type->map(ti, clone); |
| 1203 | dm_offload_end(&o); |
| 1204 | |
| 1205 | switch (r) { |
| 1206 | case DM_MAPIO_SUBMITTED: |
| 1207 | break; |
| 1208 | case DM_MAPIO_REMAPPED: |
| 1209 | /* the bio has been remapped so dispatch it */ |
| 1210 | trace_block_bio_remap(clone->bi_disk->queue, clone, |
| 1211 | bio_dev(tio->io->bio), sector); |
| 1212 | generic_make_request(clone); |
| 1213 | break; |
| 1214 | case DM_MAPIO_KILL: |
| 1215 | dec_pending(tio->io, BLK_STS_IOERR); |
| 1216 | free_tio(tio); |
| 1217 | break; |
| 1218 | case DM_MAPIO_REQUEUE: |
| 1219 | dec_pending(tio->io, BLK_STS_DM_REQUEUE); |
| 1220 | free_tio(tio); |
| 1221 | break; |
| 1222 | default: |
| 1223 | DMWARN("unimplemented target map return value: %d", r); |
| 1224 | BUG(); |
| 1225 | } |
| 1226 | } |
| 1227 | |
| 1228 | struct clone_info { |
| 1229 | struct mapped_device *md; |
| 1230 | struct dm_table *map; |
| 1231 | struct bio *bio; |
| 1232 | struct dm_io *io; |
| 1233 | sector_t sector; |
| 1234 | unsigned sector_count; |
| 1235 | }; |
| 1236 | |
| 1237 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
| 1238 | { |
| 1239 | bio->bi_iter.bi_sector = sector; |
| 1240 | bio->bi_iter.bi_size = to_bytes(len); |
| 1241 | } |
| 1242 | |
| 1243 | /* |
| 1244 | * Creates a bio that consists of range of complete bvecs. |
| 1245 | */ |
| 1246 | static int clone_bio(struct dm_target_io *tio, struct bio *bio, |
| 1247 | sector_t sector, unsigned len) |
| 1248 | { |
| 1249 | struct bio *clone = &tio->clone; |
| 1250 | |
| 1251 | __bio_clone_fast(clone, bio); |
| 1252 | |
| 1253 | if (unlikely(bio_integrity(bio) != NULL)) { |
| 1254 | int r; |
| 1255 | |
| 1256 | if (unlikely(!dm_target_has_integrity(tio->ti->type) && |
| 1257 | !dm_target_passes_integrity(tio->ti->type))) { |
| 1258 | DMWARN("%s: the target %s doesn't support integrity data.", |
| 1259 | dm_device_name(tio->io->md), |
| 1260 | tio->ti->type->name); |
| 1261 | return -EIO; |
| 1262 | } |
| 1263 | |
| 1264 | r = bio_integrity_clone(clone, bio, GFP_NOIO); |
| 1265 | if (r < 0) |
| 1266 | return r; |
| 1267 | } |
| 1268 | |
| 1269 | if (bio_op(bio) != REQ_OP_ZONE_REPORT) |
| 1270 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
| 1271 | clone->bi_iter.bi_size = to_bytes(len); |
| 1272 | |
| 1273 | if (unlikely(bio_integrity(bio) != NULL)) |
| 1274 | bio_integrity_trim(clone); |
| 1275 | |
| 1276 | return 0; |
| 1277 | } |
| 1278 | |
| 1279 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
| 1280 | struct dm_target *ti, |
| 1281 | unsigned target_bio_nr) |
| 1282 | { |
| 1283 | struct dm_target_io *tio; |
| 1284 | struct bio *clone; |
| 1285 | |
| 1286 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
| 1287 | tio = container_of(clone, struct dm_target_io, clone); |
| 1288 | |
| 1289 | tio->io = ci->io; |
| 1290 | tio->ti = ti; |
| 1291 | tio->target_bio_nr = target_bio_nr; |
| 1292 | |
| 1293 | return tio; |
| 1294 | } |
| 1295 | |
| 1296 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
| 1297 | struct dm_target *ti, |
| 1298 | unsigned target_bio_nr, unsigned *len) |
| 1299 | { |
| 1300 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
| 1301 | struct bio *clone = &tio->clone; |
| 1302 | |
| 1303 | tio->len_ptr = len; |
| 1304 | |
| 1305 | __bio_clone_fast(clone, ci->bio); |
| 1306 | if (len) |
| 1307 | bio_setup_sector(clone, ci->sector, *len); |
| 1308 | |
| 1309 | __map_bio(tio); |
| 1310 | } |
| 1311 | |
| 1312 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
| 1313 | unsigned num_bios, unsigned *len) |
| 1314 | { |
| 1315 | unsigned target_bio_nr; |
| 1316 | |
| 1317 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
| 1318 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
| 1319 | } |
| 1320 | |
| 1321 | static int __send_empty_flush(struct clone_info *ci) |
| 1322 | { |
| 1323 | unsigned target_nr = 0; |
| 1324 | struct dm_target *ti; |
| 1325 | |
| 1326 | BUG_ON(bio_has_data(ci->bio)); |
| 1327 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
| 1328 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
| 1329 | |
| 1330 | return 0; |
| 1331 | } |
| 1332 | |
| 1333 | static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
| 1334 | sector_t sector, unsigned *len) |
| 1335 | { |
| 1336 | struct bio *bio = ci->bio; |
| 1337 | struct dm_target_io *tio; |
| 1338 | unsigned target_bio_nr; |
| 1339 | unsigned num_target_bios = 1; |
| 1340 | int r = 0; |
| 1341 | |
| 1342 | /* |
| 1343 | * Does the target want to receive duplicate copies of the bio? |
| 1344 | */ |
| 1345 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) |
| 1346 | num_target_bios = ti->num_write_bios(ti, bio); |
| 1347 | |
| 1348 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
| 1349 | tio = alloc_tio(ci, ti, target_bio_nr); |
| 1350 | tio->len_ptr = len; |
| 1351 | r = clone_bio(tio, bio, sector, *len); |
| 1352 | if (r < 0) { |
| 1353 | free_tio(tio); |
| 1354 | break; |
| 1355 | } |
| 1356 | __map_bio(tio); |
| 1357 | } |
| 1358 | |
| 1359 | return r; |
| 1360 | } |
| 1361 | |
| 1362 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
| 1363 | |
| 1364 | static unsigned get_num_discard_bios(struct dm_target *ti) |
| 1365 | { |
| 1366 | return ti->num_discard_bios; |
| 1367 | } |
| 1368 | |
| 1369 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
| 1370 | { |
| 1371 | return ti->num_write_same_bios; |
| 1372 | } |
| 1373 | |
| 1374 | static unsigned get_num_write_zeroes_bios(struct dm_target *ti) |
| 1375 | { |
| 1376 | return ti->num_write_zeroes_bios; |
| 1377 | } |
| 1378 | |
| 1379 | typedef bool (*is_split_required_fn)(struct dm_target *ti); |
| 1380 | |
| 1381 | static bool is_split_required_for_discard(struct dm_target *ti) |
| 1382 | { |
| 1383 | return ti->split_discard_bios; |
| 1384 | } |
| 1385 | |
| 1386 | static int __send_changing_extent_only(struct clone_info *ci, |
| 1387 | get_num_bios_fn get_num_bios, |
| 1388 | is_split_required_fn is_split_required) |
| 1389 | { |
| 1390 | struct dm_target *ti; |
| 1391 | unsigned len; |
| 1392 | unsigned num_bios; |
| 1393 | |
| 1394 | do { |
| 1395 | ti = dm_table_find_target(ci->map, ci->sector); |
| 1396 | if (!dm_target_is_valid(ti)) |
| 1397 | return -EIO; |
| 1398 | |
| 1399 | /* |
| 1400 | * Even though the device advertised support for this type of |
| 1401 | * request, that does not mean every target supports it, and |
| 1402 | * reconfiguration might also have changed that since the |
| 1403 | * check was performed. |
| 1404 | */ |
| 1405 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
| 1406 | if (!num_bios) |
| 1407 | return -EOPNOTSUPP; |
| 1408 | |
| 1409 | if (is_split_required && !is_split_required(ti)) |
| 1410 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
| 1411 | else |
| 1412 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
| 1413 | |
| 1414 | __send_duplicate_bios(ci, ti, num_bios, &len); |
| 1415 | |
| 1416 | ci->sector += len; |
| 1417 | } while (ci->sector_count -= len); |
| 1418 | |
| 1419 | return 0; |
| 1420 | } |
| 1421 | |
| 1422 | static int __send_discard(struct clone_info *ci) |
| 1423 | { |
| 1424 | return __send_changing_extent_only(ci, get_num_discard_bios, |
| 1425 | is_split_required_for_discard); |
| 1426 | } |
| 1427 | |
| 1428 | static int __send_write_same(struct clone_info *ci) |
| 1429 | { |
| 1430 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
| 1431 | } |
| 1432 | |
| 1433 | static int __send_write_zeroes(struct clone_info *ci) |
| 1434 | { |
| 1435 | return __send_changing_extent_only(ci, get_num_write_zeroes_bios, NULL); |
| 1436 | } |
| 1437 | |
| 1438 | /* |
| 1439 | * Select the correct strategy for processing a non-flush bio. |
| 1440 | */ |
| 1441 | static int __split_and_process_non_flush(struct clone_info *ci) |
| 1442 | { |
| 1443 | struct bio *bio = ci->bio; |
| 1444 | struct dm_target *ti; |
| 1445 | unsigned len; |
| 1446 | int r; |
| 1447 | |
| 1448 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
| 1449 | return __send_discard(ci); |
| 1450 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
| 1451 | return __send_write_same(ci); |
| 1452 | else if (unlikely(bio_op(bio) == REQ_OP_WRITE_ZEROES)) |
| 1453 | return __send_write_zeroes(ci); |
| 1454 | |
| 1455 | ti = dm_table_find_target(ci->map, ci->sector); |
| 1456 | if (!dm_target_is_valid(ti)) |
| 1457 | return -EIO; |
| 1458 | |
| 1459 | if (bio_op(bio) == REQ_OP_ZONE_REPORT) |
| 1460 | len = ci->sector_count; |
| 1461 | else |
| 1462 | len = min_t(sector_t, max_io_len(ci->sector, ti), |
| 1463 | ci->sector_count); |
| 1464 | |
| 1465 | r = __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
| 1466 | if (r < 0) |
| 1467 | return r; |
| 1468 | |
| 1469 | ci->sector += len; |
| 1470 | ci->sector_count -= len; |
| 1471 | |
| 1472 | return 0; |
| 1473 | } |
| 1474 | |
| 1475 | /* |
| 1476 | * Entry point to split a bio into clones and submit them to the targets. |
| 1477 | */ |
| 1478 | static void __split_and_process_bio(struct mapped_device *md, |
| 1479 | struct dm_table *map, struct bio *bio) |
| 1480 | { |
| 1481 | struct clone_info ci; |
| 1482 | int error = 0; |
| 1483 | |
| 1484 | if (unlikely(!map)) { |
| 1485 | bio_io_error(bio); |
| 1486 | return; |
| 1487 | } |
| 1488 | |
| 1489 | ci.map = map; |
| 1490 | ci.md = md; |
| 1491 | ci.io = alloc_io(md); |
| 1492 | ci.io->status = 0; |
| 1493 | atomic_set(&ci.io->io_count, 1); |
| 1494 | ci.io->bio = bio; |
| 1495 | ci.io->md = md; |
| 1496 | spin_lock_init(&ci.io->endio_lock); |
| 1497 | ci.sector = bio->bi_iter.bi_sector; |
| 1498 | |
| 1499 | start_io_acct(ci.io); |
| 1500 | |
| 1501 | if (bio->bi_opf & REQ_PREFLUSH) { |
| 1502 | ci.bio = &ci.md->flush_bio; |
| 1503 | ci.sector_count = 0; |
| 1504 | error = __send_empty_flush(&ci); |
| 1505 | /* dec_pending submits any data associated with flush */ |
| 1506 | } else if (bio_op(bio) == REQ_OP_ZONE_RESET) { |
| 1507 | ci.bio = bio; |
| 1508 | ci.sector_count = 0; |
| 1509 | error = __split_and_process_non_flush(&ci); |
| 1510 | } else { |
| 1511 | ci.bio = bio; |
| 1512 | ci.sector_count = bio_sectors(bio); |
| 1513 | while (ci.sector_count && !error) |
| 1514 | error = __split_and_process_non_flush(&ci); |
| 1515 | } |
| 1516 | |
| 1517 | /* drop the extra reference count */ |
| 1518 | dec_pending(ci.io, errno_to_blk_status(error)); |
| 1519 | } |
| 1520 | /*----------------------------------------------------------------- |
| 1521 | * CRUD END |
| 1522 | *---------------------------------------------------------------*/ |
| 1523 | |
| 1524 | /* |
| 1525 | * The request function that just remaps the bio built up by |
| 1526 | * dm_merge_bvec. |
| 1527 | */ |
| 1528 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
| 1529 | { |
| 1530 | int rw = bio_data_dir(bio); |
| 1531 | struct mapped_device *md = q->queuedata; |
| 1532 | int srcu_idx; |
| 1533 | struct dm_table *map; |
| 1534 | |
| 1535 | map = dm_get_live_table(md, &srcu_idx); |
| 1536 | |
| 1537 | generic_start_io_acct(q, rw, bio_sectors(bio), &dm_disk(md)->part0); |
| 1538 | |
| 1539 | /* if we're suspended, we have to queue this io for later */ |
| 1540 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { |
| 1541 | dm_put_live_table(md, srcu_idx); |
| 1542 | |
| 1543 | if (!(bio->bi_opf & REQ_RAHEAD)) |
| 1544 | queue_io(md, bio); |
| 1545 | else |
| 1546 | bio_io_error(bio); |
| 1547 | return BLK_QC_T_NONE; |
| 1548 | } |
| 1549 | |
| 1550 | __split_and_process_bio(md, map, bio); |
| 1551 | dm_put_live_table(md, srcu_idx); |
| 1552 | return BLK_QC_T_NONE; |
| 1553 | } |
| 1554 | |
| 1555 | static int dm_any_congested(void *congested_data, int bdi_bits) |
| 1556 | { |
| 1557 | int r = bdi_bits; |
| 1558 | struct mapped_device *md = congested_data; |
| 1559 | struct dm_table *map; |
| 1560 | |
| 1561 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
| 1562 | if (dm_request_based(md)) { |
| 1563 | /* |
| 1564 | * With request-based DM we only need to check the |
| 1565 | * top-level queue for congestion. |
| 1566 | */ |
| 1567 | r = md->queue->backing_dev_info->wb.state & bdi_bits; |
| 1568 | } else { |
| 1569 | map = dm_get_live_table_fast(md); |
| 1570 | if (map) |
| 1571 | r = dm_table_any_congested(map, bdi_bits); |
| 1572 | dm_put_live_table_fast(md); |
| 1573 | } |
| 1574 | } |
| 1575 | |
| 1576 | return r; |
| 1577 | } |
| 1578 | |
| 1579 | /*----------------------------------------------------------------- |
| 1580 | * An IDR is used to keep track of allocated minor numbers. |
| 1581 | *---------------------------------------------------------------*/ |
| 1582 | static void free_minor(int minor) |
| 1583 | { |
| 1584 | spin_lock(&_minor_lock); |
| 1585 | idr_remove(&_minor_idr, minor); |
| 1586 | spin_unlock(&_minor_lock); |
| 1587 | } |
| 1588 | |
| 1589 | /* |
| 1590 | * See if the device with a specific minor # is free. |
| 1591 | */ |
| 1592 | static int specific_minor(int minor) |
| 1593 | { |
| 1594 | int r; |
| 1595 | |
| 1596 | if (minor >= (1 << MINORBITS)) |
| 1597 | return -EINVAL; |
| 1598 | |
| 1599 | idr_preload(GFP_KERNEL); |
| 1600 | spin_lock(&_minor_lock); |
| 1601 | |
| 1602 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
| 1603 | |
| 1604 | spin_unlock(&_minor_lock); |
| 1605 | idr_preload_end(); |
| 1606 | if (r < 0) |
| 1607 | return r == -ENOSPC ? -EBUSY : r; |
| 1608 | return 0; |
| 1609 | } |
| 1610 | |
| 1611 | static int next_free_minor(int *minor) |
| 1612 | { |
| 1613 | int r; |
| 1614 | |
| 1615 | idr_preload(GFP_KERNEL); |
| 1616 | spin_lock(&_minor_lock); |
| 1617 | |
| 1618 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
| 1619 | |
| 1620 | spin_unlock(&_minor_lock); |
| 1621 | idr_preload_end(); |
| 1622 | if (r < 0) |
| 1623 | return r; |
| 1624 | *minor = r; |
| 1625 | return 0; |
| 1626 | } |
| 1627 | |
| 1628 | static const struct block_device_operations dm_blk_dops; |
| 1629 | static const struct dax_operations dm_dax_ops; |
| 1630 | |
| 1631 | static void dm_wq_work(struct work_struct *work); |
| 1632 | |
| 1633 | void dm_init_md_queue(struct mapped_device *md) |
| 1634 | { |
| 1635 | /* |
| 1636 | * Request-based dm devices cannot be stacked on top of bio-based dm |
| 1637 | * devices. The type of this dm device may not have been decided yet. |
| 1638 | * The type is decided at the first table loading time. |
| 1639 | * To prevent problematic device stacking, clear the queue flag |
| 1640 | * for request stacking support until then. |
| 1641 | * |
| 1642 | * This queue is new, so no concurrency on the queue_flags. |
| 1643 | */ |
| 1644 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); |
| 1645 | |
| 1646 | /* |
| 1647 | * Initialize data that will only be used by a non-blk-mq DM queue |
| 1648 | * - must do so here (in alloc_dev callchain) before queue is used |
| 1649 | */ |
| 1650 | md->queue->queuedata = md; |
| 1651 | } |
| 1652 | |
| 1653 | void dm_init_normal_md_queue(struct mapped_device *md) |
| 1654 | { |
| 1655 | md->use_blk_mq = false; |
| 1656 | dm_init_md_queue(md); |
| 1657 | |
| 1658 | /* |
| 1659 | * Initialize aspects of queue that aren't relevant for blk-mq |
| 1660 | */ |
| 1661 | md->queue->backing_dev_info->congested_data = md; |
| 1662 | md->queue->backing_dev_info->congested_fn = dm_any_congested; |
| 1663 | } |
| 1664 | |
| 1665 | static void cleanup_mapped_device(struct mapped_device *md) |
| 1666 | { |
| 1667 | if (md->wq) |
| 1668 | destroy_workqueue(md->wq); |
| 1669 | if (md->kworker_task) |
| 1670 | kthread_stop(md->kworker_task); |
| 1671 | mempool_destroy(md->io_pool); |
| 1672 | if (md->bs) |
| 1673 | bioset_free(md->bs); |
| 1674 | |
| 1675 | if (md->dax_dev) { |
| 1676 | kill_dax(md->dax_dev); |
| 1677 | put_dax(md->dax_dev); |
| 1678 | md->dax_dev = NULL; |
| 1679 | } |
| 1680 | |
| 1681 | if (md->disk) { |
| 1682 | spin_lock(&_minor_lock); |
| 1683 | md->disk->private_data = NULL; |
| 1684 | spin_unlock(&_minor_lock); |
| 1685 | del_gendisk(md->disk); |
| 1686 | put_disk(md->disk); |
| 1687 | } |
| 1688 | |
| 1689 | if (md->queue) |
| 1690 | blk_cleanup_queue(md->queue); |
| 1691 | |
| 1692 | cleanup_srcu_struct(&md->io_barrier); |
| 1693 | |
| 1694 | if (md->bdev) { |
| 1695 | bdput(md->bdev); |
| 1696 | md->bdev = NULL; |
| 1697 | } |
| 1698 | |
| 1699 | dm_mq_cleanup_mapped_device(md); |
| 1700 | } |
| 1701 | |
| 1702 | /* |
| 1703 | * Allocate and initialise a blank device with a given minor. |
| 1704 | */ |
| 1705 | static struct mapped_device *alloc_dev(int minor) |
| 1706 | { |
| 1707 | int r, numa_node_id = dm_get_numa_node(); |
| 1708 | struct dax_device *dax_dev; |
| 1709 | struct mapped_device *md; |
| 1710 | void *old_md; |
| 1711 | |
| 1712 | md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id); |
| 1713 | if (!md) { |
| 1714 | DMWARN("unable to allocate device, out of memory."); |
| 1715 | return NULL; |
| 1716 | } |
| 1717 | |
| 1718 | if (!try_module_get(THIS_MODULE)) |
| 1719 | goto bad_module_get; |
| 1720 | |
| 1721 | /* get a minor number for the dev */ |
| 1722 | if (minor == DM_ANY_MINOR) |
| 1723 | r = next_free_minor(&minor); |
| 1724 | else |
| 1725 | r = specific_minor(minor); |
| 1726 | if (r < 0) |
| 1727 | goto bad_minor; |
| 1728 | |
| 1729 | r = init_srcu_struct(&md->io_barrier); |
| 1730 | if (r < 0) |
| 1731 | goto bad_io_barrier; |
| 1732 | |
| 1733 | md->numa_node_id = numa_node_id; |
| 1734 | md->use_blk_mq = dm_use_blk_mq_default(); |
| 1735 | md->init_tio_pdu = false; |
| 1736 | md->type = DM_TYPE_NONE; |
| 1737 | mutex_init(&md->suspend_lock); |
| 1738 | mutex_init(&md->type_lock); |
| 1739 | mutex_init(&md->table_devices_lock); |
| 1740 | spin_lock_init(&md->deferred_lock); |
| 1741 | atomic_set(&md->holders, 1); |
| 1742 | atomic_set(&md->open_count, 0); |
| 1743 | atomic_set(&md->event_nr, 0); |
| 1744 | atomic_set(&md->uevent_seq, 0); |
| 1745 | INIT_LIST_HEAD(&md->uevent_list); |
| 1746 | INIT_LIST_HEAD(&md->table_devices); |
| 1747 | spin_lock_init(&md->uevent_lock); |
| 1748 | |
| 1749 | md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id); |
| 1750 | if (!md->queue) |
| 1751 | goto bad; |
| 1752 | |
| 1753 | dm_init_md_queue(md); |
| 1754 | /* |
| 1755 | * default to bio-based required ->make_request_fn until DM |
| 1756 | * table is loaded and md->type established. If request-based |
| 1757 | * table is loaded: blk-mq will override accordingly. |
| 1758 | */ |
| 1759 | blk_queue_make_request(md->queue, dm_make_request); |
| 1760 | |
| 1761 | md->disk = alloc_disk_node(1, numa_node_id); |
| 1762 | if (!md->disk) |
| 1763 | goto bad; |
| 1764 | |
| 1765 | atomic_set(&md->pending[0], 0); |
| 1766 | atomic_set(&md->pending[1], 0); |
| 1767 | init_waitqueue_head(&md->wait); |
| 1768 | INIT_WORK(&md->work, dm_wq_work); |
| 1769 | init_waitqueue_head(&md->eventq); |
| 1770 | init_completion(&md->kobj_holder.completion); |
| 1771 | md->kworker_task = NULL; |
| 1772 | |
| 1773 | md->disk->major = _major; |
| 1774 | md->disk->first_minor = minor; |
| 1775 | md->disk->fops = &dm_blk_dops; |
| 1776 | md->disk->queue = md->queue; |
| 1777 | md->disk->private_data = md; |
| 1778 | sprintf(md->disk->disk_name, "dm-%d", minor); |
| 1779 | |
| 1780 | dax_dev = alloc_dax(md, md->disk->disk_name, &dm_dax_ops); |
| 1781 | if (!dax_dev) |
| 1782 | goto bad; |
| 1783 | md->dax_dev = dax_dev; |
| 1784 | |
| 1785 | add_disk(md->disk); |
| 1786 | format_dev_t(md->name, MKDEV(_major, minor)); |
| 1787 | |
| 1788 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
| 1789 | if (!md->wq) |
| 1790 | goto bad; |
| 1791 | |
| 1792 | md->bdev = bdget_disk(md->disk, 0); |
| 1793 | if (!md->bdev) |
| 1794 | goto bad; |
| 1795 | |
| 1796 | bio_init(&md->flush_bio, NULL, 0); |
| 1797 | bio_set_dev(&md->flush_bio, md->bdev); |
| 1798 | md->flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC; |
| 1799 | |
| 1800 | dm_stats_init(&md->stats); |
| 1801 | |
| 1802 | /* Populate the mapping, nobody knows we exist yet */ |
| 1803 | spin_lock(&_minor_lock); |
| 1804 | old_md = idr_replace(&_minor_idr, md, minor); |
| 1805 | spin_unlock(&_minor_lock); |
| 1806 | |
| 1807 | BUG_ON(old_md != MINOR_ALLOCED); |
| 1808 | |
| 1809 | return md; |
| 1810 | |
| 1811 | bad: |
| 1812 | cleanup_mapped_device(md); |
| 1813 | bad_io_barrier: |
| 1814 | free_minor(minor); |
| 1815 | bad_minor: |
| 1816 | module_put(THIS_MODULE); |
| 1817 | bad_module_get: |
| 1818 | kvfree(md); |
| 1819 | return NULL; |
| 1820 | } |
| 1821 | |
| 1822 | static void unlock_fs(struct mapped_device *md); |
| 1823 | |
| 1824 | static void free_dev(struct mapped_device *md) |
| 1825 | { |
| 1826 | int minor = MINOR(disk_devt(md->disk)); |
| 1827 | |
| 1828 | unlock_fs(md); |
| 1829 | |
| 1830 | cleanup_mapped_device(md); |
| 1831 | |
| 1832 | free_table_devices(&md->table_devices); |
| 1833 | dm_stats_cleanup(&md->stats); |
| 1834 | free_minor(minor); |
| 1835 | |
| 1836 | module_put(THIS_MODULE); |
| 1837 | kvfree(md); |
| 1838 | } |
| 1839 | |
| 1840 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
| 1841 | { |
| 1842 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
| 1843 | |
| 1844 | if (md->bs) { |
| 1845 | /* The md already has necessary mempools. */ |
| 1846 | if (dm_table_bio_based(t)) { |
| 1847 | /* |
| 1848 | * Reload bioset because front_pad may have changed |
| 1849 | * because a different table was loaded. |
| 1850 | */ |
| 1851 | bioset_free(md->bs); |
| 1852 | md->bs = p->bs; |
| 1853 | p->bs = NULL; |
| 1854 | } |
| 1855 | /* |
| 1856 | * There's no need to reload with request-based dm |
| 1857 | * because the size of front_pad doesn't change. |
| 1858 | * Note for future: If you are to reload bioset, |
| 1859 | * prep-ed requests in the queue may refer |
| 1860 | * to bio from the old bioset, so you must walk |
| 1861 | * through the queue to unprep. |
| 1862 | */ |
| 1863 | goto out; |
| 1864 | } |
| 1865 | |
| 1866 | BUG_ON(!p || md->io_pool || md->bs); |
| 1867 | |
| 1868 | md->io_pool = p->io_pool; |
| 1869 | p->io_pool = NULL; |
| 1870 | md->bs = p->bs; |
| 1871 | p->bs = NULL; |
| 1872 | |
| 1873 | out: |
| 1874 | /* mempool bind completed, no longer need any mempools in the table */ |
| 1875 | dm_table_free_md_mempools(t); |
| 1876 | } |
| 1877 | |
| 1878 | /* |
| 1879 | * Bind a table to the device. |
| 1880 | */ |
| 1881 | static void event_callback(void *context) |
| 1882 | { |
| 1883 | unsigned long flags; |
| 1884 | LIST_HEAD(uevents); |
| 1885 | struct mapped_device *md = (struct mapped_device *) context; |
| 1886 | |
| 1887 | spin_lock_irqsave(&md->uevent_lock, flags); |
| 1888 | list_splice_init(&md->uevent_list, &uevents); |
| 1889 | spin_unlock_irqrestore(&md->uevent_lock, flags); |
| 1890 | |
| 1891 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
| 1892 | |
| 1893 | atomic_inc(&md->event_nr); |
| 1894 | wake_up(&md->eventq); |
| 1895 | dm_issue_global_event(); |
| 1896 | } |
| 1897 | |
| 1898 | /* |
| 1899 | * Protected by md->suspend_lock obtained by dm_swap_table(). |
| 1900 | */ |
| 1901 | static void __set_size(struct mapped_device *md, sector_t size) |
| 1902 | { |
| 1903 | lockdep_assert_held(&md->suspend_lock); |
| 1904 | |
| 1905 | set_capacity(md->disk, size); |
| 1906 | |
| 1907 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
| 1908 | } |
| 1909 | |
| 1910 | /* |
| 1911 | * Returns old map, which caller must destroy. |
| 1912 | */ |
| 1913 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, |
| 1914 | struct queue_limits *limits) |
| 1915 | { |
| 1916 | struct dm_table *old_map; |
| 1917 | struct request_queue *q = md->queue; |
| 1918 | sector_t size; |
| 1919 | |
| 1920 | lockdep_assert_held(&md->suspend_lock); |
| 1921 | |
| 1922 | size = dm_table_get_size(t); |
| 1923 | |
| 1924 | /* |
| 1925 | * Wipe any geometry if the size of the table changed. |
| 1926 | */ |
| 1927 | if (size != dm_get_size(md)) |
| 1928 | memset(&md->geometry, 0, sizeof(md->geometry)); |
| 1929 | |
| 1930 | __set_size(md, size); |
| 1931 | |
| 1932 | dm_table_event_callback(t, event_callback, md); |
| 1933 | |
| 1934 | /* |
| 1935 | * The queue hasn't been stopped yet, if the old table type wasn't |
| 1936 | * for request-based during suspension. So stop it to prevent |
| 1937 | * I/O mapping before resume. |
| 1938 | * This must be done before setting the queue restrictions, |
| 1939 | * because request-based dm may be run just after the setting. |
| 1940 | */ |
| 1941 | if (dm_table_request_based(t)) { |
| 1942 | dm_stop_queue(q); |
| 1943 | /* |
| 1944 | * Leverage the fact that request-based DM targets are |
| 1945 | * immutable singletons and establish md->immutable_target |
| 1946 | * - used to optimize both dm_request_fn and dm_mq_queue_rq |
| 1947 | */ |
| 1948 | md->immutable_target = dm_table_get_immutable_target(t); |
| 1949 | } |
| 1950 | |
| 1951 | __bind_mempools(md, t); |
| 1952 | |
| 1953 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
| 1954 | rcu_assign_pointer(md->map, (void *)t); |
| 1955 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
| 1956 | |
| 1957 | dm_table_set_restrictions(t, q, limits); |
| 1958 | if (old_map) |
| 1959 | dm_sync_table(md); |
| 1960 | |
| 1961 | return old_map; |
| 1962 | } |
| 1963 | |
| 1964 | /* |
| 1965 | * Returns unbound table for the caller to free. |
| 1966 | */ |
| 1967 | static struct dm_table *__unbind(struct mapped_device *md) |
| 1968 | { |
| 1969 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
| 1970 | |
| 1971 | if (!map) |
| 1972 | return NULL; |
| 1973 | |
| 1974 | dm_table_event_callback(map, NULL, NULL); |
| 1975 | RCU_INIT_POINTER(md->map, NULL); |
| 1976 | dm_sync_table(md); |
| 1977 | |
| 1978 | return map; |
| 1979 | } |
| 1980 | |
| 1981 | /* |
| 1982 | * Constructor for a new device. |
| 1983 | */ |
| 1984 | int dm_create(int minor, struct mapped_device **result) |
| 1985 | { |
| 1986 | struct mapped_device *md; |
| 1987 | |
| 1988 | md = alloc_dev(minor); |
| 1989 | if (!md) |
| 1990 | return -ENXIO; |
| 1991 | |
| 1992 | dm_sysfs_init(md); |
| 1993 | |
| 1994 | *result = md; |
| 1995 | return 0; |
| 1996 | } |
| 1997 | |
| 1998 | /* |
| 1999 | * Functions to manage md->type. |
| 2000 | * All are required to hold md->type_lock. |
| 2001 | */ |
| 2002 | void dm_lock_md_type(struct mapped_device *md) |
| 2003 | { |
| 2004 | mutex_lock(&md->type_lock); |
| 2005 | } |
| 2006 | |
| 2007 | void dm_unlock_md_type(struct mapped_device *md) |
| 2008 | { |
| 2009 | mutex_unlock(&md->type_lock); |
| 2010 | } |
| 2011 | |
| 2012 | void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type) |
| 2013 | { |
| 2014 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
| 2015 | md->type = type; |
| 2016 | } |
| 2017 | |
| 2018 | enum dm_queue_mode dm_get_md_type(struct mapped_device *md) |
| 2019 | { |
| 2020 | return md->type; |
| 2021 | } |
| 2022 | |
| 2023 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
| 2024 | { |
| 2025 | return md->immutable_target_type; |
| 2026 | } |
| 2027 | |
| 2028 | /* |
| 2029 | * The queue_limits are only valid as long as you have a reference |
| 2030 | * count on 'md'. |
| 2031 | */ |
| 2032 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) |
| 2033 | { |
| 2034 | BUG_ON(!atomic_read(&md->holders)); |
| 2035 | return &md->queue->limits; |
| 2036 | } |
| 2037 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); |
| 2038 | |
| 2039 | /* |
| 2040 | * Setup the DM device's queue based on md's type |
| 2041 | */ |
| 2042 | int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t) |
| 2043 | { |
| 2044 | int r; |
| 2045 | enum dm_queue_mode type = dm_get_md_type(md); |
| 2046 | |
| 2047 | switch (type) { |
| 2048 | case DM_TYPE_REQUEST_BASED: |
| 2049 | r = dm_old_init_request_queue(md, t); |
| 2050 | if (r) { |
| 2051 | DMERR("Cannot initialize queue for request-based mapped device"); |
| 2052 | return r; |
| 2053 | } |
| 2054 | break; |
| 2055 | case DM_TYPE_MQ_REQUEST_BASED: |
| 2056 | r = dm_mq_init_request_queue(md, t); |
| 2057 | if (r) { |
| 2058 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
| 2059 | return r; |
| 2060 | } |
| 2061 | break; |
| 2062 | case DM_TYPE_BIO_BASED: |
| 2063 | case DM_TYPE_DAX_BIO_BASED: |
| 2064 | dm_init_normal_md_queue(md); |
| 2065 | /* |
| 2066 | * DM handles splitting bios as needed. Free the bio_split bioset |
| 2067 | * since it won't be used (saves 1 process per bio-based DM device). |
| 2068 | */ |
| 2069 | bioset_free(md->queue->bio_split); |
| 2070 | md->queue->bio_split = NULL; |
| 2071 | break; |
| 2072 | case DM_TYPE_NONE: |
| 2073 | WARN_ON_ONCE(true); |
| 2074 | break; |
| 2075 | } |
| 2076 | |
| 2077 | return 0; |
| 2078 | } |
| 2079 | |
| 2080 | struct mapped_device *dm_get_md(dev_t dev) |
| 2081 | { |
| 2082 | struct mapped_device *md; |
| 2083 | unsigned minor = MINOR(dev); |
| 2084 | |
| 2085 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) |
| 2086 | return NULL; |
| 2087 | |
| 2088 | spin_lock(&_minor_lock); |
| 2089 | |
| 2090 | md = idr_find(&_minor_idr, minor); |
| 2091 | if (md) { |
| 2092 | if ((md == MINOR_ALLOCED || |
| 2093 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
| 2094 | dm_deleting_md(md) || |
| 2095 | test_bit(DMF_FREEING, &md->flags))) { |
| 2096 | md = NULL; |
| 2097 | goto out; |
| 2098 | } |
| 2099 | dm_get(md); |
| 2100 | } |
| 2101 | |
| 2102 | out: |
| 2103 | spin_unlock(&_minor_lock); |
| 2104 | |
| 2105 | return md; |
| 2106 | } |
| 2107 | EXPORT_SYMBOL_GPL(dm_get_md); |
| 2108 | |
| 2109 | void *dm_get_mdptr(struct mapped_device *md) |
| 2110 | { |
| 2111 | return md->interface_ptr; |
| 2112 | } |
| 2113 | |
| 2114 | void dm_set_mdptr(struct mapped_device *md, void *ptr) |
| 2115 | { |
| 2116 | md->interface_ptr = ptr; |
| 2117 | } |
| 2118 | |
| 2119 | void dm_get(struct mapped_device *md) |
| 2120 | { |
| 2121 | atomic_inc(&md->holders); |
| 2122 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
| 2123 | } |
| 2124 | |
| 2125 | int dm_hold(struct mapped_device *md) |
| 2126 | { |
| 2127 | spin_lock(&_minor_lock); |
| 2128 | if (test_bit(DMF_FREEING, &md->flags)) { |
| 2129 | spin_unlock(&_minor_lock); |
| 2130 | return -EBUSY; |
| 2131 | } |
| 2132 | dm_get(md); |
| 2133 | spin_unlock(&_minor_lock); |
| 2134 | return 0; |
| 2135 | } |
| 2136 | EXPORT_SYMBOL_GPL(dm_hold); |
| 2137 | |
| 2138 | const char *dm_device_name(struct mapped_device *md) |
| 2139 | { |
| 2140 | return md->name; |
| 2141 | } |
| 2142 | EXPORT_SYMBOL_GPL(dm_device_name); |
| 2143 | |
| 2144 | static void __dm_destroy(struct mapped_device *md, bool wait) |
| 2145 | { |
| 2146 | struct request_queue *q = dm_get_md_queue(md); |
| 2147 | struct dm_table *map; |
| 2148 | int srcu_idx; |
| 2149 | |
| 2150 | might_sleep(); |
| 2151 | |
| 2152 | spin_lock(&_minor_lock); |
| 2153 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
| 2154 | set_bit(DMF_FREEING, &md->flags); |
| 2155 | spin_unlock(&_minor_lock); |
| 2156 | |
| 2157 | blk_set_queue_dying(q); |
| 2158 | |
| 2159 | if (dm_request_based(md) && md->kworker_task) |
| 2160 | kthread_flush_worker(&md->kworker); |
| 2161 | |
| 2162 | /* |
| 2163 | * Take suspend_lock so that presuspend and postsuspend methods |
| 2164 | * do not race with internal suspend. |
| 2165 | */ |
| 2166 | mutex_lock(&md->suspend_lock); |
| 2167 | map = dm_get_live_table(md, &srcu_idx); |
| 2168 | if (!dm_suspended_md(md)) { |
| 2169 | dm_table_presuspend_targets(map); |
| 2170 | dm_table_postsuspend_targets(map); |
| 2171 | } |
| 2172 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
| 2173 | dm_put_live_table(md, srcu_idx); |
| 2174 | mutex_unlock(&md->suspend_lock); |
| 2175 | |
| 2176 | /* |
| 2177 | * Rare, but there may be I/O requests still going to complete, |
| 2178 | * for example. Wait for all references to disappear. |
| 2179 | * No one should increment the reference count of the mapped_device, |
| 2180 | * after the mapped_device state becomes DMF_FREEING. |
| 2181 | */ |
| 2182 | if (wait) |
| 2183 | while (atomic_read(&md->holders)) |
| 2184 | msleep(1); |
| 2185 | else if (atomic_read(&md->holders)) |
| 2186 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", |
| 2187 | dm_device_name(md), atomic_read(&md->holders)); |
| 2188 | |
| 2189 | dm_sysfs_exit(md); |
| 2190 | dm_table_destroy(__unbind(md)); |
| 2191 | free_dev(md); |
| 2192 | } |
| 2193 | |
| 2194 | void dm_destroy(struct mapped_device *md) |
| 2195 | { |
| 2196 | __dm_destroy(md, true); |
| 2197 | } |
| 2198 | |
| 2199 | void dm_destroy_immediate(struct mapped_device *md) |
| 2200 | { |
| 2201 | __dm_destroy(md, false); |
| 2202 | } |
| 2203 | |
| 2204 | void dm_put(struct mapped_device *md) |
| 2205 | { |
| 2206 | atomic_dec(&md->holders); |
| 2207 | } |
| 2208 | EXPORT_SYMBOL_GPL(dm_put); |
| 2209 | |
| 2210 | static int dm_wait_for_completion(struct mapped_device *md, long task_state) |
| 2211 | { |
| 2212 | int r = 0; |
| 2213 | DEFINE_WAIT(wait); |
| 2214 | |
| 2215 | while (1) { |
| 2216 | prepare_to_wait(&md->wait, &wait, task_state); |
| 2217 | |
| 2218 | if (!md_in_flight(md)) |
| 2219 | break; |
| 2220 | |
| 2221 | if (signal_pending_state(task_state, current)) { |
| 2222 | r = -EINTR; |
| 2223 | break; |
| 2224 | } |
| 2225 | |
| 2226 | io_schedule(); |
| 2227 | } |
| 2228 | finish_wait(&md->wait, &wait); |
| 2229 | |
| 2230 | return r; |
| 2231 | } |
| 2232 | |
| 2233 | /* |
| 2234 | * Process the deferred bios |
| 2235 | */ |
| 2236 | static void dm_wq_work(struct work_struct *work) |
| 2237 | { |
| 2238 | struct mapped_device *md = container_of(work, struct mapped_device, |
| 2239 | work); |
| 2240 | struct bio *c; |
| 2241 | int srcu_idx; |
| 2242 | struct dm_table *map; |
| 2243 | |
| 2244 | map = dm_get_live_table(md, &srcu_idx); |
| 2245 | |
| 2246 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
| 2247 | spin_lock_irq(&md->deferred_lock); |
| 2248 | c = bio_list_pop(&md->deferred); |
| 2249 | spin_unlock_irq(&md->deferred_lock); |
| 2250 | |
| 2251 | if (!c) |
| 2252 | break; |
| 2253 | |
| 2254 | if (dm_request_based(md)) |
| 2255 | generic_make_request(c); |
| 2256 | else |
| 2257 | __split_and_process_bio(md, map, c); |
| 2258 | } |
| 2259 | |
| 2260 | dm_put_live_table(md, srcu_idx); |
| 2261 | } |
| 2262 | |
| 2263 | static void dm_queue_flush(struct mapped_device *md) |
| 2264 | { |
| 2265 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
| 2266 | smp_mb__after_atomic(); |
| 2267 | queue_work(md->wq, &md->work); |
| 2268 | } |
| 2269 | |
| 2270 | /* |
| 2271 | * Swap in a new table, returning the old one for the caller to destroy. |
| 2272 | */ |
| 2273 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
| 2274 | { |
| 2275 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
| 2276 | struct queue_limits limits; |
| 2277 | int r; |
| 2278 | |
| 2279 | mutex_lock(&md->suspend_lock); |
| 2280 | |
| 2281 | /* device must be suspended */ |
| 2282 | if (!dm_suspended_md(md)) |
| 2283 | goto out; |
| 2284 | |
| 2285 | /* |
| 2286 | * If the new table has no data devices, retain the existing limits. |
| 2287 | * This helps multipath with queue_if_no_path if all paths disappear, |
| 2288 | * then new I/O is queued based on these limits, and then some paths |
| 2289 | * reappear. |
| 2290 | */ |
| 2291 | if (dm_table_has_no_data_devices(table)) { |
| 2292 | live_map = dm_get_live_table_fast(md); |
| 2293 | if (live_map) |
| 2294 | limits = md->queue->limits; |
| 2295 | dm_put_live_table_fast(md); |
| 2296 | } |
| 2297 | |
| 2298 | if (!live_map) { |
| 2299 | r = dm_calculate_queue_limits(table, &limits); |
| 2300 | if (r) { |
| 2301 | map = ERR_PTR(r); |
| 2302 | goto out; |
| 2303 | } |
| 2304 | } |
| 2305 | |
| 2306 | map = __bind(md, table, &limits); |
| 2307 | dm_issue_global_event(); |
| 2308 | |
| 2309 | out: |
| 2310 | mutex_unlock(&md->suspend_lock); |
| 2311 | return map; |
| 2312 | } |
| 2313 | |
| 2314 | /* |
| 2315 | * Functions to lock and unlock any filesystem running on the |
| 2316 | * device. |
| 2317 | */ |
| 2318 | static int lock_fs(struct mapped_device *md) |
| 2319 | { |
| 2320 | int r; |
| 2321 | |
| 2322 | WARN_ON(md->frozen_sb); |
| 2323 | |
| 2324 | md->frozen_sb = freeze_bdev(md->bdev); |
| 2325 | if (IS_ERR(md->frozen_sb)) { |
| 2326 | r = PTR_ERR(md->frozen_sb); |
| 2327 | md->frozen_sb = NULL; |
| 2328 | return r; |
| 2329 | } |
| 2330 | |
| 2331 | set_bit(DMF_FROZEN, &md->flags); |
| 2332 | |
| 2333 | return 0; |
| 2334 | } |
| 2335 | |
| 2336 | static void unlock_fs(struct mapped_device *md) |
| 2337 | { |
| 2338 | if (!test_bit(DMF_FROZEN, &md->flags)) |
| 2339 | return; |
| 2340 | |
| 2341 | thaw_bdev(md->bdev, md->frozen_sb); |
| 2342 | md->frozen_sb = NULL; |
| 2343 | clear_bit(DMF_FROZEN, &md->flags); |
| 2344 | } |
| 2345 | |
| 2346 | /* |
| 2347 | * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG |
| 2348 | * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE |
| 2349 | * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY |
| 2350 | * |
| 2351 | * If __dm_suspend returns 0, the device is completely quiescent |
| 2352 | * now. There is no request-processing activity. All new requests |
| 2353 | * are being added to md->deferred list. |
| 2354 | */ |
| 2355 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
| 2356 | unsigned suspend_flags, long task_state, |
| 2357 | int dmf_suspended_flag) |
| 2358 | { |
| 2359 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
| 2360 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; |
| 2361 | int r; |
| 2362 | |
| 2363 | lockdep_assert_held(&md->suspend_lock); |
| 2364 | |
| 2365 | /* |
| 2366 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. |
| 2367 | * This flag is cleared before dm_suspend returns. |
| 2368 | */ |
| 2369 | if (noflush) |
| 2370 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
| 2371 | else |
| 2372 | pr_debug("%s: suspending with flush\n", dm_device_name(md)); |
| 2373 | |
| 2374 | /* |
| 2375 | * This gets reverted if there's an error later and the targets |
| 2376 | * provide the .presuspend_undo hook. |
| 2377 | */ |
| 2378 | dm_table_presuspend_targets(map); |
| 2379 | |
| 2380 | /* |
| 2381 | * Flush I/O to the device. |
| 2382 | * Any I/O submitted after lock_fs() may not be flushed. |
| 2383 | * noflush takes precedence over do_lockfs. |
| 2384 | * (lock_fs() flushes I/Os and waits for them to complete.) |
| 2385 | */ |
| 2386 | if (!noflush && do_lockfs) { |
| 2387 | r = lock_fs(md); |
| 2388 | if (r) { |
| 2389 | dm_table_presuspend_undo_targets(map); |
| 2390 | return r; |
| 2391 | } |
| 2392 | } |
| 2393 | |
| 2394 | /* |
| 2395 | * Here we must make sure that no processes are submitting requests |
| 2396 | * to target drivers i.e. no one may be executing |
| 2397 | * __split_and_process_bio. This is called from dm_request and |
| 2398 | * dm_wq_work. |
| 2399 | * |
| 2400 | * To get all processes out of __split_and_process_bio in dm_request, |
| 2401 | * we take the write lock. To prevent any process from reentering |
| 2402 | * __split_and_process_bio from dm_request and quiesce the thread |
| 2403 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call |
| 2404 | * flush_workqueue(md->wq). |
| 2405 | */ |
| 2406 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
| 2407 | if (map) |
| 2408 | synchronize_srcu(&md->io_barrier); |
| 2409 | |
| 2410 | /* |
| 2411 | * Stop md->queue before flushing md->wq in case request-based |
| 2412 | * dm defers requests to md->wq from md->queue. |
| 2413 | */ |
| 2414 | if (dm_request_based(md)) { |
| 2415 | dm_stop_queue(md->queue); |
| 2416 | if (md->kworker_task) |
| 2417 | kthread_flush_worker(&md->kworker); |
| 2418 | } |
| 2419 | |
| 2420 | flush_workqueue(md->wq); |
| 2421 | |
| 2422 | /* |
| 2423 | * At this point no more requests are entering target request routines. |
| 2424 | * We call dm_wait_for_completion to wait for all existing requests |
| 2425 | * to finish. |
| 2426 | */ |
| 2427 | r = dm_wait_for_completion(md, task_state); |
| 2428 | if (!r) |
| 2429 | set_bit(dmf_suspended_flag, &md->flags); |
| 2430 | |
| 2431 | if (noflush) |
| 2432 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
| 2433 | if (map) |
| 2434 | synchronize_srcu(&md->io_barrier); |
| 2435 | |
| 2436 | /* were we interrupted ? */ |
| 2437 | if (r < 0) { |
| 2438 | dm_queue_flush(md); |
| 2439 | |
| 2440 | if (dm_request_based(md)) |
| 2441 | dm_start_queue(md->queue); |
| 2442 | |
| 2443 | unlock_fs(md); |
| 2444 | dm_table_presuspend_undo_targets(map); |
| 2445 | /* pushback list is already flushed, so skip flush */ |
| 2446 | } |
| 2447 | |
| 2448 | return r; |
| 2449 | } |
| 2450 | |
| 2451 | /* |
| 2452 | * We need to be able to change a mapping table under a mounted |
| 2453 | * filesystem. For example we might want to move some data in |
| 2454 | * the background. Before the table can be swapped with |
| 2455 | * dm_bind_table, dm_suspend must be called to flush any in |
| 2456 | * flight bios and ensure that any further io gets deferred. |
| 2457 | */ |
| 2458 | /* |
| 2459 | * Suspend mechanism in request-based dm. |
| 2460 | * |
| 2461 | * 1. Flush all I/Os by lock_fs() if needed. |
| 2462 | * 2. Stop dispatching any I/O by stopping the request_queue. |
| 2463 | * 3. Wait for all in-flight I/Os to be completed or requeued. |
| 2464 | * |
| 2465 | * To abort suspend, start the request_queue. |
| 2466 | */ |
| 2467 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
| 2468 | { |
| 2469 | struct dm_table *map = NULL; |
| 2470 | int r = 0; |
| 2471 | |
| 2472 | retry: |
| 2473 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
| 2474 | |
| 2475 | if (dm_suspended_md(md)) { |
| 2476 | r = -EINVAL; |
| 2477 | goto out_unlock; |
| 2478 | } |
| 2479 | |
| 2480 | if (dm_suspended_internally_md(md)) { |
| 2481 | /* already internally suspended, wait for internal resume */ |
| 2482 | mutex_unlock(&md->suspend_lock); |
| 2483 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); |
| 2484 | if (r) |
| 2485 | return r; |
| 2486 | goto retry; |
| 2487 | } |
| 2488 | |
| 2489 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
| 2490 | |
| 2491 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED); |
| 2492 | if (r) |
| 2493 | goto out_unlock; |
| 2494 | |
| 2495 | dm_table_postsuspend_targets(map); |
| 2496 | |
| 2497 | out_unlock: |
| 2498 | mutex_unlock(&md->suspend_lock); |
| 2499 | return r; |
| 2500 | } |
| 2501 | |
| 2502 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
| 2503 | { |
| 2504 | if (map) { |
| 2505 | int r = dm_table_resume_targets(map); |
| 2506 | if (r) |
| 2507 | return r; |
| 2508 | } |
| 2509 | |
| 2510 | dm_queue_flush(md); |
| 2511 | |
| 2512 | /* |
| 2513 | * Flushing deferred I/Os must be done after targets are resumed |
| 2514 | * so that mapping of targets can work correctly. |
| 2515 | * Request-based dm is queueing the deferred I/Os in its request_queue. |
| 2516 | */ |
| 2517 | if (dm_request_based(md)) |
| 2518 | dm_start_queue(md->queue); |
| 2519 | |
| 2520 | unlock_fs(md); |
| 2521 | |
| 2522 | return 0; |
| 2523 | } |
| 2524 | |
| 2525 | int dm_resume(struct mapped_device *md) |
| 2526 | { |
| 2527 | int r; |
| 2528 | struct dm_table *map = NULL; |
| 2529 | |
| 2530 | retry: |
| 2531 | r = -EINVAL; |
| 2532 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); |
| 2533 | |
| 2534 | if (!dm_suspended_md(md)) |
| 2535 | goto out; |
| 2536 | |
| 2537 | if (dm_suspended_internally_md(md)) { |
| 2538 | /* already internally suspended, wait for internal resume */ |
| 2539 | mutex_unlock(&md->suspend_lock); |
| 2540 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); |
| 2541 | if (r) |
| 2542 | return r; |
| 2543 | goto retry; |
| 2544 | } |
| 2545 | |
| 2546 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
| 2547 | if (!map || !dm_table_get_size(map)) |
| 2548 | goto out; |
| 2549 | |
| 2550 | r = __dm_resume(md, map); |
| 2551 | if (r) |
| 2552 | goto out; |
| 2553 | |
| 2554 | clear_bit(DMF_SUSPENDED, &md->flags); |
| 2555 | out: |
| 2556 | mutex_unlock(&md->suspend_lock); |
| 2557 | |
| 2558 | return r; |
| 2559 | } |
| 2560 | |
| 2561 | /* |
| 2562 | * Internal suspend/resume works like userspace-driven suspend. It waits |
| 2563 | * until all bios finish and prevents issuing new bios to the target drivers. |
| 2564 | * It may be used only from the kernel. |
| 2565 | */ |
| 2566 | |
| 2567 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
| 2568 | { |
| 2569 | struct dm_table *map = NULL; |
| 2570 | |
| 2571 | lockdep_assert_held(&md->suspend_lock); |
| 2572 | |
| 2573 | if (md->internal_suspend_count++) |
| 2574 | return; /* nested internal suspend */ |
| 2575 | |
| 2576 | if (dm_suspended_md(md)) { |
| 2577 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); |
| 2578 | return; /* nest suspend */ |
| 2579 | } |
| 2580 | |
| 2581 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
| 2582 | |
| 2583 | /* |
| 2584 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is |
| 2585 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend |
| 2586 | * would require changing .presuspend to return an error -- avoid this |
| 2587 | * until there is a need for more elaborate variants of internal suspend. |
| 2588 | */ |
| 2589 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE, |
| 2590 | DMF_SUSPENDED_INTERNALLY); |
| 2591 | |
| 2592 | dm_table_postsuspend_targets(map); |
| 2593 | } |
| 2594 | |
| 2595 | static void __dm_internal_resume(struct mapped_device *md) |
| 2596 | { |
| 2597 | BUG_ON(!md->internal_suspend_count); |
| 2598 | |
| 2599 | if (--md->internal_suspend_count) |
| 2600 | return; /* resume from nested internal suspend */ |
| 2601 | |
| 2602 | if (dm_suspended_md(md)) |
| 2603 | goto done; /* resume from nested suspend */ |
| 2604 | |
| 2605 | /* |
| 2606 | * NOTE: existing callers don't need to call dm_table_resume_targets |
| 2607 | * (which may fail -- so best to avoid it for now by passing NULL map) |
| 2608 | */ |
| 2609 | (void) __dm_resume(md, NULL); |
| 2610 | |
| 2611 | done: |
| 2612 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); |
| 2613 | smp_mb__after_atomic(); |
| 2614 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); |
| 2615 | } |
| 2616 | |
| 2617 | void dm_internal_suspend_noflush(struct mapped_device *md) |
| 2618 | { |
| 2619 | mutex_lock(&md->suspend_lock); |
| 2620 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); |
| 2621 | mutex_unlock(&md->suspend_lock); |
| 2622 | } |
| 2623 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); |
| 2624 | |
| 2625 | void dm_internal_resume(struct mapped_device *md) |
| 2626 | { |
| 2627 | mutex_lock(&md->suspend_lock); |
| 2628 | __dm_internal_resume(md); |
| 2629 | mutex_unlock(&md->suspend_lock); |
| 2630 | } |
| 2631 | EXPORT_SYMBOL_GPL(dm_internal_resume); |
| 2632 | |
| 2633 | /* |
| 2634 | * Fast variants of internal suspend/resume hold md->suspend_lock, |
| 2635 | * which prevents interaction with userspace-driven suspend. |
| 2636 | */ |
| 2637 | |
| 2638 | void dm_internal_suspend_fast(struct mapped_device *md) |
| 2639 | { |
| 2640 | mutex_lock(&md->suspend_lock); |
| 2641 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
| 2642 | return; |
| 2643 | |
| 2644 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
| 2645 | synchronize_srcu(&md->io_barrier); |
| 2646 | flush_workqueue(md->wq); |
| 2647 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); |
| 2648 | } |
| 2649 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
| 2650 | |
| 2651 | void dm_internal_resume_fast(struct mapped_device *md) |
| 2652 | { |
| 2653 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
| 2654 | goto done; |
| 2655 | |
| 2656 | dm_queue_flush(md); |
| 2657 | |
| 2658 | done: |
| 2659 | mutex_unlock(&md->suspend_lock); |
| 2660 | } |
| 2661 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
| 2662 | |
| 2663 | /*----------------------------------------------------------------- |
| 2664 | * Event notification. |
| 2665 | *---------------------------------------------------------------*/ |
| 2666 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
| 2667 | unsigned cookie) |
| 2668 | { |
| 2669 | int r; |
| 2670 | unsigned noio_flag; |
| 2671 | char udev_cookie[DM_COOKIE_LENGTH]; |
| 2672 | char *envp[] = { udev_cookie, NULL }; |
| 2673 | |
| 2674 | noio_flag = memalloc_noio_save(); |
| 2675 | |
| 2676 | if (!cookie) |
| 2677 | r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
| 2678 | else { |
| 2679 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", |
| 2680 | DM_COOKIE_ENV_VAR_NAME, cookie); |
| 2681 | r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
| 2682 | action, envp); |
| 2683 | } |
| 2684 | |
| 2685 | memalloc_noio_restore(noio_flag); |
| 2686 | |
| 2687 | return r; |
| 2688 | } |
| 2689 | |
| 2690 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
| 2691 | { |
| 2692 | return atomic_add_return(1, &md->uevent_seq); |
| 2693 | } |
| 2694 | |
| 2695 | uint32_t dm_get_event_nr(struct mapped_device *md) |
| 2696 | { |
| 2697 | return atomic_read(&md->event_nr); |
| 2698 | } |
| 2699 | |
| 2700 | int dm_wait_event(struct mapped_device *md, int event_nr) |
| 2701 | { |
| 2702 | return wait_event_interruptible(md->eventq, |
| 2703 | (event_nr != atomic_read(&md->event_nr))); |
| 2704 | } |
| 2705 | |
| 2706 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
| 2707 | { |
| 2708 | unsigned long flags; |
| 2709 | |
| 2710 | spin_lock_irqsave(&md->uevent_lock, flags); |
| 2711 | list_add(elist, &md->uevent_list); |
| 2712 | spin_unlock_irqrestore(&md->uevent_lock, flags); |
| 2713 | } |
| 2714 | |
| 2715 | /* |
| 2716 | * The gendisk is only valid as long as you have a reference |
| 2717 | * count on 'md'. |
| 2718 | */ |
| 2719 | struct gendisk *dm_disk(struct mapped_device *md) |
| 2720 | { |
| 2721 | return md->disk; |
| 2722 | } |
| 2723 | EXPORT_SYMBOL_GPL(dm_disk); |
| 2724 | |
| 2725 | struct kobject *dm_kobject(struct mapped_device *md) |
| 2726 | { |
| 2727 | return &md->kobj_holder.kobj; |
| 2728 | } |
| 2729 | |
| 2730 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
| 2731 | { |
| 2732 | struct mapped_device *md; |
| 2733 | |
| 2734 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
| 2735 | |
| 2736 | spin_lock(&_minor_lock); |
| 2737 | if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) { |
| 2738 | md = NULL; |
| 2739 | goto out; |
| 2740 | } |
| 2741 | dm_get(md); |
| 2742 | out: |
| 2743 | spin_unlock(&_minor_lock); |
| 2744 | |
| 2745 | return md; |
| 2746 | } |
| 2747 | |
| 2748 | int dm_suspended_md(struct mapped_device *md) |
| 2749 | { |
| 2750 | return test_bit(DMF_SUSPENDED, &md->flags); |
| 2751 | } |
| 2752 | |
| 2753 | int dm_suspended_internally_md(struct mapped_device *md) |
| 2754 | { |
| 2755 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); |
| 2756 | } |
| 2757 | |
| 2758 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
| 2759 | { |
| 2760 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); |
| 2761 | } |
| 2762 | |
| 2763 | int dm_suspended(struct dm_target *ti) |
| 2764 | { |
| 2765 | return dm_suspended_md(dm_table_get_md(ti->table)); |
| 2766 | } |
| 2767 | EXPORT_SYMBOL_GPL(dm_suspended); |
| 2768 | |
| 2769 | int dm_noflush_suspending(struct dm_target *ti) |
| 2770 | { |
| 2771 | return __noflush_suspending(dm_table_get_md(ti->table)); |
| 2772 | } |
| 2773 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); |
| 2774 | |
| 2775 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type, |
| 2776 | unsigned integrity, unsigned per_io_data_size) |
| 2777 | { |
| 2778 | struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id); |
| 2779 | unsigned int pool_size = 0; |
| 2780 | unsigned int front_pad; |
| 2781 | |
| 2782 | if (!pools) |
| 2783 | return NULL; |
| 2784 | |
| 2785 | switch (type) { |
| 2786 | case DM_TYPE_BIO_BASED: |
| 2787 | case DM_TYPE_DAX_BIO_BASED: |
| 2788 | pool_size = dm_get_reserved_bio_based_ios(); |
| 2789 | front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
| 2790 | |
| 2791 | pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); |
| 2792 | if (!pools->io_pool) |
| 2793 | goto out; |
| 2794 | break; |
| 2795 | case DM_TYPE_REQUEST_BASED: |
| 2796 | case DM_TYPE_MQ_REQUEST_BASED: |
| 2797 | pool_size = dm_get_reserved_rq_based_ios(); |
| 2798 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
| 2799 | /* per_io_data_size is used for blk-mq pdu at queue allocation */ |
| 2800 | break; |
| 2801 | default: |
| 2802 | BUG(); |
| 2803 | } |
| 2804 | |
| 2805 | pools->bs = bioset_create(pool_size, front_pad, BIOSET_NEED_RESCUER); |
| 2806 | if (!pools->bs) |
| 2807 | goto out; |
| 2808 | |
| 2809 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
| 2810 | goto out; |
| 2811 | |
| 2812 | return pools; |
| 2813 | |
| 2814 | out: |
| 2815 | dm_free_md_mempools(pools); |
| 2816 | |
| 2817 | return NULL; |
| 2818 | } |
| 2819 | |
| 2820 | void dm_free_md_mempools(struct dm_md_mempools *pools) |
| 2821 | { |
| 2822 | if (!pools) |
| 2823 | return; |
| 2824 | |
| 2825 | mempool_destroy(pools->io_pool); |
| 2826 | |
| 2827 | if (pools->bs) |
| 2828 | bioset_free(pools->bs); |
| 2829 | |
| 2830 | kfree(pools); |
| 2831 | } |
| 2832 | |
| 2833 | struct dm_pr { |
| 2834 | u64 old_key; |
| 2835 | u64 new_key; |
| 2836 | u32 flags; |
| 2837 | bool fail_early; |
| 2838 | }; |
| 2839 | |
| 2840 | static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn, |
| 2841 | void *data) |
| 2842 | { |
| 2843 | struct mapped_device *md = bdev->bd_disk->private_data; |
| 2844 | struct dm_table *table; |
| 2845 | struct dm_target *ti; |
| 2846 | int ret = -ENOTTY, srcu_idx; |
| 2847 | |
| 2848 | table = dm_get_live_table(md, &srcu_idx); |
| 2849 | if (!table || !dm_table_get_size(table)) |
| 2850 | goto out; |
| 2851 | |
| 2852 | /* We only support devices that have a single target */ |
| 2853 | if (dm_table_get_num_targets(table) != 1) |
| 2854 | goto out; |
| 2855 | ti = dm_table_get_target(table, 0); |
| 2856 | |
| 2857 | ret = -EINVAL; |
| 2858 | if (!ti->type->iterate_devices) |
| 2859 | goto out; |
| 2860 | |
| 2861 | ret = ti->type->iterate_devices(ti, fn, data); |
| 2862 | out: |
| 2863 | dm_put_live_table(md, srcu_idx); |
| 2864 | return ret; |
| 2865 | } |
| 2866 | |
| 2867 | /* |
| 2868 | * For register / unregister we need to manually call out to every path. |
| 2869 | */ |
| 2870 | static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev, |
| 2871 | sector_t start, sector_t len, void *data) |
| 2872 | { |
| 2873 | struct dm_pr *pr = data; |
| 2874 | const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops; |
| 2875 | |
| 2876 | if (!ops || !ops->pr_register) |
| 2877 | return -EOPNOTSUPP; |
| 2878 | return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags); |
| 2879 | } |
| 2880 | |
| 2881 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, |
| 2882 | u32 flags) |
| 2883 | { |
| 2884 | struct dm_pr pr = { |
| 2885 | .old_key = old_key, |
| 2886 | .new_key = new_key, |
| 2887 | .flags = flags, |
| 2888 | .fail_early = true, |
| 2889 | }; |
| 2890 | int ret; |
| 2891 | |
| 2892 | ret = dm_call_pr(bdev, __dm_pr_register, &pr); |
| 2893 | if (ret && new_key) { |
| 2894 | /* unregister all paths if we failed to register any path */ |
| 2895 | pr.old_key = new_key; |
| 2896 | pr.new_key = 0; |
| 2897 | pr.flags = 0; |
| 2898 | pr.fail_early = false; |
| 2899 | dm_call_pr(bdev, __dm_pr_register, &pr); |
| 2900 | } |
| 2901 | |
| 2902 | return ret; |
| 2903 | } |
| 2904 | |
| 2905 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, |
| 2906 | u32 flags) |
| 2907 | { |
| 2908 | struct mapped_device *md = bdev->bd_disk->private_data; |
| 2909 | const struct pr_ops *ops; |
| 2910 | fmode_t mode; |
| 2911 | int r; |
| 2912 | |
| 2913 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
| 2914 | if (r < 0) |
| 2915 | return r; |
| 2916 | |
| 2917 | ops = bdev->bd_disk->fops->pr_ops; |
| 2918 | if (ops && ops->pr_reserve) |
| 2919 | r = ops->pr_reserve(bdev, key, type, flags); |
| 2920 | else |
| 2921 | r = -EOPNOTSUPP; |
| 2922 | |
| 2923 | bdput(bdev); |
| 2924 | return r; |
| 2925 | } |
| 2926 | |
| 2927 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) |
| 2928 | { |
| 2929 | struct mapped_device *md = bdev->bd_disk->private_data; |
| 2930 | const struct pr_ops *ops; |
| 2931 | fmode_t mode; |
| 2932 | int r; |
| 2933 | |
| 2934 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
| 2935 | if (r < 0) |
| 2936 | return r; |
| 2937 | |
| 2938 | ops = bdev->bd_disk->fops->pr_ops; |
| 2939 | if (ops && ops->pr_release) |
| 2940 | r = ops->pr_release(bdev, key, type); |
| 2941 | else |
| 2942 | r = -EOPNOTSUPP; |
| 2943 | |
| 2944 | bdput(bdev); |
| 2945 | return r; |
| 2946 | } |
| 2947 | |
| 2948 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, |
| 2949 | enum pr_type type, bool abort) |
| 2950 | { |
| 2951 | struct mapped_device *md = bdev->bd_disk->private_data; |
| 2952 | const struct pr_ops *ops; |
| 2953 | fmode_t mode; |
| 2954 | int r; |
| 2955 | |
| 2956 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
| 2957 | if (r < 0) |
| 2958 | return r; |
| 2959 | |
| 2960 | ops = bdev->bd_disk->fops->pr_ops; |
| 2961 | if (ops && ops->pr_preempt) |
| 2962 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); |
| 2963 | else |
| 2964 | r = -EOPNOTSUPP; |
| 2965 | |
| 2966 | bdput(bdev); |
| 2967 | return r; |
| 2968 | } |
| 2969 | |
| 2970 | static int dm_pr_clear(struct block_device *bdev, u64 key) |
| 2971 | { |
| 2972 | struct mapped_device *md = bdev->bd_disk->private_data; |
| 2973 | const struct pr_ops *ops; |
| 2974 | fmode_t mode; |
| 2975 | int r; |
| 2976 | |
| 2977 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
| 2978 | if (r < 0) |
| 2979 | return r; |
| 2980 | |
| 2981 | ops = bdev->bd_disk->fops->pr_ops; |
| 2982 | if (ops && ops->pr_clear) |
| 2983 | r = ops->pr_clear(bdev, key); |
| 2984 | else |
| 2985 | r = -EOPNOTSUPP; |
| 2986 | |
| 2987 | bdput(bdev); |
| 2988 | return r; |
| 2989 | } |
| 2990 | |
| 2991 | static const struct pr_ops dm_pr_ops = { |
| 2992 | .pr_register = dm_pr_register, |
| 2993 | .pr_reserve = dm_pr_reserve, |
| 2994 | .pr_release = dm_pr_release, |
| 2995 | .pr_preempt = dm_pr_preempt, |
| 2996 | .pr_clear = dm_pr_clear, |
| 2997 | }; |
| 2998 | |
| 2999 | static const struct block_device_operations dm_blk_dops = { |
| 3000 | .open = dm_blk_open, |
| 3001 | .release = dm_blk_close, |
| 3002 | .ioctl = dm_blk_ioctl, |
| 3003 | .getgeo = dm_blk_getgeo, |
| 3004 | .pr_ops = &dm_pr_ops, |
| 3005 | .owner = THIS_MODULE |
| 3006 | }; |
| 3007 | |
| 3008 | static const struct dax_operations dm_dax_ops = { |
| 3009 | .direct_access = dm_dax_direct_access, |
| 3010 | .copy_from_iter = dm_dax_copy_from_iter, |
| 3011 | }; |
| 3012 | |
| 3013 | /* |
| 3014 | * module hooks |
| 3015 | */ |
| 3016 | module_init(dm_init); |
| 3017 | module_exit(dm_exit); |
| 3018 | |
| 3019 | module_param(major, uint, 0); |
| 3020 | MODULE_PARM_DESC(major, "The major number of the device mapper"); |
| 3021 | |
| 3022 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
| 3023 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); |
| 3024 | |
| 3025 | module_param(dm_numa_node, int, S_IRUGO | S_IWUSR); |
| 3026 | MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations"); |
| 3027 | |
| 3028 | MODULE_DESCRIPTION(DM_NAME " driver"); |
| 3029 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); |
| 3030 | MODULE_LICENSE("GPL"); |