1// SPDX-License-Identifier: GPL-2.0-only 2/* 3 * Copyright (C) 2003 Sistina Software (UK) Limited. 4 * Copyright (C) 2004, 2010-2011 Red Hat, Inc. All rights reserved. 5 * 6 * This file is released under the GPL. 7 */ 8 9#include <linux/device-mapper.h> 10 11#include <linux/module.h> 12#include <linux/init.h> 13#include <linux/blkdev.h> 14#include <linux/bio.h> 15#include <linux/slab.h> 16 17#define DM_MSG_PREFIX "flakey" 18 19#define PROBABILITY_BASE 1000000000 20 21#define all_corrupt_bio_flags_match(bio, fc) \ 22 (((bio)->bi_opf & (fc)->corrupt_bio_flags) == (fc)->corrupt_bio_flags) 23 24/* 25 * Flakey: Used for testing only, simulates intermittent, 26 * catastrophic device failure. 27 */ 28struct flakey_c { 29 struct dm_dev *dev; 30 unsigned long start_time; 31 sector_t start; 32 unsigned int up_interval; 33 unsigned int down_interval; 34 unsigned long flags; 35 unsigned int corrupt_bio_byte; 36 unsigned int corrupt_bio_rw; 37 unsigned int corrupt_bio_value; 38 blk_opf_t corrupt_bio_flags; 39 unsigned int random_read_corrupt; 40 unsigned int random_write_corrupt; 41}; 42 43enum feature_flag_bits { 44 ERROR_READS, 45 DROP_WRITES, 46 ERROR_WRITES 47}; 48 49struct per_bio_data { 50 bool bio_submitted; 51}; 52 53static int parse_features(struct dm_arg_set *as, struct flakey_c *fc, 54 struct dm_target *ti) 55{ 56 int r; 57 unsigned int argc; 58 const char *arg_name; 59 60 static const struct dm_arg _args[] = { 61 {0, 11, "Invalid number of feature args"}, 62 {1, UINT_MAX, "Invalid corrupt bio byte"}, 63 {0, 255, "Invalid corrupt value to write into bio byte (0-255)"}, 64 {0, UINT_MAX, "Invalid corrupt bio flags mask"}, 65 {0, PROBABILITY_BASE, "Invalid random corrupt argument"}, 66 }; 67 68 /* No feature arguments supplied. */ 69 if (!as->argc) 70 return 0; 71 72 r = dm_read_arg_group(_args, as, &argc, &ti->error); 73 if (r) 74 return r; 75 76 while (argc) { 77 arg_name = dm_shift_arg(as); 78 argc--; 79 80 if (!arg_name) { 81 ti->error = "Insufficient feature arguments"; 82 return -EINVAL; 83 } 84 85 /* 86 * error_reads 87 */ 88 if (!strcasecmp(arg_name, "error_reads")) { 89 if (test_and_set_bit(ERROR_READS, &fc->flags)) { 90 ti->error = "Feature error_reads duplicated"; 91 return -EINVAL; 92 } 93 continue; 94 } 95 96 /* 97 * drop_writes 98 */ 99 if (!strcasecmp(arg_name, "drop_writes")) { 100 if (test_and_set_bit(DROP_WRITES, &fc->flags)) { 101 ti->error = "Feature drop_writes duplicated"; 102 return -EINVAL; 103 } else if (test_bit(ERROR_WRITES, &fc->flags)) { 104 ti->error = "Feature drop_writes conflicts with feature error_writes"; 105 return -EINVAL; 106 } 107 108 continue; 109 } 110 111 /* 112 * error_writes 113 */ 114 if (!strcasecmp(arg_name, "error_writes")) { 115 if (test_and_set_bit(ERROR_WRITES, &fc->flags)) { 116 ti->error = "Feature error_writes duplicated"; 117 return -EINVAL; 118 119 } else if (test_bit(DROP_WRITES, &fc->flags)) { 120 ti->error = "Feature error_writes conflicts with feature drop_writes"; 121 return -EINVAL; 122 } 123 124 continue; 125 } 126 127 /* 128 * corrupt_bio_byte <Nth_byte> <direction> <value> <bio_flags> 129 */ 130 if (!strcasecmp(arg_name, "corrupt_bio_byte")) { 131 if (!argc) { 132 ti->error = "Feature corrupt_bio_byte requires parameters"; 133 return -EINVAL; 134 } 135 136 r = dm_read_arg(_args + 1, as, &fc->corrupt_bio_byte, &ti->error); 137 if (r) 138 return r; 139 argc--; 140 141 /* 142 * Direction r or w? 143 */ 144 arg_name = dm_shift_arg(as); 145 if (arg_name && !strcasecmp(arg_name, "w")) 146 fc->corrupt_bio_rw = WRITE; 147 else if (arg_name && !strcasecmp(arg_name, "r")) 148 fc->corrupt_bio_rw = READ; 149 else { 150 ti->error = "Invalid corrupt bio direction (r or w)"; 151 return -EINVAL; 152 } 153 argc--; 154 155 /* 156 * Value of byte (0-255) to write in place of correct one. 157 */ 158 r = dm_read_arg(_args + 2, as, &fc->corrupt_bio_value, &ti->error); 159 if (r) 160 return r; 161 argc--; 162 163 /* 164 * Only corrupt bios with these flags set. 165 */ 166 BUILD_BUG_ON(sizeof(fc->corrupt_bio_flags) != 167 sizeof(unsigned int)); 168 r = dm_read_arg(_args + 3, as, 169 (__force unsigned int *)&fc->corrupt_bio_flags, 170 &ti->error); 171 if (r) 172 return r; 173 argc--; 174 175 continue; 176 } 177 178 if (!strcasecmp(arg_name, "random_read_corrupt")) { 179 if (!argc) { 180 ti->error = "Feature random_read_corrupt requires a parameter"; 181 return -EINVAL; 182 } 183 r = dm_read_arg(_args + 4, as, &fc->random_read_corrupt, &ti->error); 184 if (r) 185 return r; 186 argc--; 187 188 continue; 189 } 190 191 if (!strcasecmp(arg_name, "random_write_corrupt")) { 192 if (!argc) { 193 ti->error = "Feature random_write_corrupt requires a parameter"; 194 return -EINVAL; 195 } 196 r = dm_read_arg(_args + 4, as, &fc->random_write_corrupt, &ti->error); 197 if (r) 198 return r; 199 argc--; 200 201 continue; 202 } 203 204 ti->error = "Unrecognised flakey feature requested"; 205 return -EINVAL; 206 } 207 208 if (test_bit(DROP_WRITES, &fc->flags) && (fc->corrupt_bio_rw == WRITE)) { 209 ti->error = "drop_writes is incompatible with corrupt_bio_byte with the WRITE flag set"; 210 return -EINVAL; 211 212 } else if (test_bit(ERROR_WRITES, &fc->flags) && (fc->corrupt_bio_rw == WRITE)) { 213 ti->error = "error_writes is incompatible with corrupt_bio_byte with the WRITE flag set"; 214 return -EINVAL; 215 } 216 217 if (!fc->corrupt_bio_byte && !test_bit(ERROR_READS, &fc->flags) && 218 !test_bit(DROP_WRITES, &fc->flags) && !test_bit(ERROR_WRITES, &fc->flags) && 219 !fc->random_read_corrupt && !fc->random_write_corrupt) { 220 set_bit(ERROR_WRITES, &fc->flags); 221 set_bit(ERROR_READS, &fc->flags); 222 } 223 224 return 0; 225} 226 227/* 228 * Construct a flakey mapping: 229 * <dev_path> <offset> <up interval> <down interval> [<#feature args> [<arg>]*] 230 * 231 * Feature args: 232 * [drop_writes] 233 * [corrupt_bio_byte <Nth_byte> <direction> <value> <bio_flags>] 234 * 235 * Nth_byte starts from 1 for the first byte. 236 * Direction is r for READ or w for WRITE. 237 * bio_flags is ignored if 0. 238 */ 239static int flakey_ctr(struct dm_target *ti, unsigned int argc, char **argv) 240{ 241 static const struct dm_arg _args[] = { 242 {0, UINT_MAX, "Invalid up interval"}, 243 {0, UINT_MAX, "Invalid down interval"}, 244 }; 245 246 int r; 247 struct flakey_c *fc; 248 unsigned long long tmpll; 249 struct dm_arg_set as; 250 const char *devname; 251 char dummy; 252 253 as.argc = argc; 254 as.argv = argv; 255 256 if (argc < 4) { 257 ti->error = "Invalid argument count"; 258 return -EINVAL; 259 } 260 261 fc = kzalloc(sizeof(*fc), GFP_KERNEL); 262 if (!fc) { 263 ti->error = "Cannot allocate context"; 264 return -ENOMEM; 265 } 266 fc->start_time = jiffies; 267 268 devname = dm_shift_arg(&as); 269 270 r = -EINVAL; 271 if (sscanf(dm_shift_arg(&as), "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) { 272 ti->error = "Invalid device sector"; 273 goto bad; 274 } 275 fc->start = tmpll; 276 277 r = dm_read_arg(_args, &as, &fc->up_interval, &ti->error); 278 if (r) 279 goto bad; 280 281 r = dm_read_arg(_args, &as, &fc->down_interval, &ti->error); 282 if (r) 283 goto bad; 284 285 if (!(fc->up_interval + fc->down_interval)) { 286 ti->error = "Total (up + down) interval is zero"; 287 r = -EINVAL; 288 goto bad; 289 } 290 291 if (fc->up_interval + fc->down_interval < fc->up_interval) { 292 ti->error = "Interval overflow"; 293 r = -EINVAL; 294 goto bad; 295 } 296 297 r = parse_features(&as, fc, ti); 298 if (r) 299 goto bad; 300 301 r = dm_get_device(ti, devname, dm_table_get_mode(ti->table), &fc->dev); 302 if (r) { 303 ti->error = "Device lookup failed"; 304 goto bad; 305 } 306 307 ti->num_flush_bios = 1; 308 ti->num_discard_bios = 1; 309 ti->per_io_data_size = sizeof(struct per_bio_data); 310 ti->private = fc; 311 return 0; 312 313bad: 314 kfree(fc); 315 return r; 316} 317 318static void flakey_dtr(struct dm_target *ti) 319{ 320 struct flakey_c *fc = ti->private; 321 322 dm_put_device(ti, fc->dev); 323 kfree(fc); 324} 325 326static sector_t flakey_map_sector(struct dm_target *ti, sector_t bi_sector) 327{ 328 struct flakey_c *fc = ti->private; 329 330 return fc->start + dm_target_offset(ti, bi_sector); 331} 332 333static void flakey_map_bio(struct dm_target *ti, struct bio *bio) 334{ 335 struct flakey_c *fc = ti->private; 336 337 bio_set_dev(bio, fc->dev->bdev); 338 bio->bi_iter.bi_sector = flakey_map_sector(ti, bio->bi_iter.bi_sector); 339} 340 341static void corrupt_bio_common(struct bio *bio, unsigned int corrupt_bio_byte, 342 unsigned char corrupt_bio_value) 343{ 344 struct bvec_iter iter; 345 struct bio_vec bvec; 346 347 /* 348 * Overwrite the Nth byte of the bio's data, on whichever page 349 * it falls. 350 */ 351 bio_for_each_segment(bvec, bio, iter) { 352 if (bio_iter_len(bio, iter) > corrupt_bio_byte) { 353 unsigned char *segment = bvec_kmap_local(&bvec); 354 segment[corrupt_bio_byte] = corrupt_bio_value; 355 kunmap_local(segment); 356 DMDEBUG("Corrupting data bio=%p by writing %u to byte %u " 357 "(rw=%c bi_opf=%u bi_sector=%llu size=%u)\n", 358 bio, corrupt_bio_value, corrupt_bio_byte, 359 (bio_data_dir(bio) == WRITE) ? 'w' : 'r', bio->bi_opf, 360 (unsigned long long)bio->bi_iter.bi_sector, 361 bio->bi_iter.bi_size); 362 break; 363 } 364 corrupt_bio_byte -= bio_iter_len(bio, iter); 365 } 366} 367 368static void corrupt_bio_data(struct bio *bio, struct flakey_c *fc) 369{ 370 unsigned int corrupt_bio_byte = fc->corrupt_bio_byte - 1; 371 372 if (!bio_has_data(bio)) 373 return; 374 375 corrupt_bio_common(bio, corrupt_bio_byte, fc->corrupt_bio_value); 376} 377 378static void corrupt_bio_random(struct bio *bio) 379{ 380 unsigned int corrupt_byte; 381 unsigned char corrupt_value; 382 383 if (!bio_has_data(bio)) 384 return; 385 386 corrupt_byte = get_random_u32() % bio->bi_iter.bi_size; 387 corrupt_value = get_random_u8(); 388 389 corrupt_bio_common(bio, corrupt_byte, corrupt_value); 390} 391 392static void clone_free(struct bio *clone) 393{ 394 struct folio_iter fi; 395 396 if (clone->bi_vcnt > 0) { /* bio_for_each_folio_all crashes with an empty bio */ 397 bio_for_each_folio_all(fi, clone) 398 folio_put(fi.folio); 399 } 400 401 bio_uninit(clone); 402 kfree(clone); 403} 404 405static void clone_endio(struct bio *clone) 406{ 407 struct bio *bio = clone->bi_private; 408 bio->bi_status = clone->bi_status; 409 clone_free(clone); 410 bio_endio(bio); 411} 412 413static struct bio *clone_bio(struct dm_target *ti, struct flakey_c *fc, struct bio *bio) 414{ 415 struct bio *clone; 416 unsigned size, remaining_size, nr_iovecs, order; 417 struct bvec_iter iter = bio->bi_iter; 418 419 if (unlikely(bio->bi_iter.bi_size > UIO_MAXIOV << PAGE_SHIFT)) 420 dm_accept_partial_bio(bio, UIO_MAXIOV << PAGE_SHIFT >> SECTOR_SHIFT); 421 422 size = bio->bi_iter.bi_size; 423 nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 424 425 clone = bio_kmalloc(nr_iovecs, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN); 426 if (!clone) 427 return NULL; 428 429 bio_init(clone, fc->dev->bdev, bio->bi_inline_vecs, nr_iovecs, bio->bi_opf); 430 431 clone->bi_iter.bi_sector = flakey_map_sector(ti, bio->bi_iter.bi_sector); 432 clone->bi_private = bio; 433 clone->bi_end_io = clone_endio; 434 435 remaining_size = size; 436 437 order = MAX_PAGE_ORDER; 438 while (remaining_size) { 439 struct page *pages; 440 unsigned size_to_add, to_copy; 441 unsigned char *virt; 442 unsigned remaining_order = __fls((remaining_size + PAGE_SIZE - 1) >> PAGE_SHIFT); 443 order = min(order, remaining_order); 444 445retry_alloc_pages: 446 pages = alloc_pages(GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN | __GFP_COMP, order); 447 if (unlikely(!pages)) { 448 if (order) { 449 order--; 450 goto retry_alloc_pages; 451 } 452 clone_free(clone); 453 return NULL; 454 } 455 size_to_add = min((unsigned)PAGE_SIZE << order, remaining_size); 456 457 virt = page_to_virt(pages); 458 to_copy = size_to_add; 459 do { 460 struct bio_vec bvec = bvec_iter_bvec(bio->bi_io_vec, iter); 461 unsigned this_step = min(bvec.bv_len, to_copy); 462 void *map = bvec_kmap_local(&bvec); 463 memcpy(virt, map, this_step); 464 kunmap_local(map); 465 466 bvec_iter_advance(bio->bi_io_vec, &iter, this_step); 467 to_copy -= this_step; 468 virt += this_step; 469 } while (to_copy); 470 471 __bio_add_page(clone, pages, size_to_add, 0); 472 remaining_size -= size_to_add; 473 } 474 475 return clone; 476} 477 478static int flakey_map(struct dm_target *ti, struct bio *bio) 479{ 480 struct flakey_c *fc = ti->private; 481 unsigned int elapsed; 482 struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); 483 484 pb->bio_submitted = false; 485 486 if (op_is_zone_mgmt(bio_op(bio))) 487 goto map_bio; 488 489 /* Are we alive ? */ 490 elapsed = (jiffies - fc->start_time) / HZ; 491 if (elapsed % (fc->up_interval + fc->down_interval) >= fc->up_interval) { 492 bool corrupt_fixed, corrupt_random; 493 /* 494 * Flag this bio as submitted while down. 495 */ 496 pb->bio_submitted = true; 497 498 /* 499 * Error reads if neither corrupt_bio_byte or drop_writes or error_writes are set. 500 * Otherwise, flakey_end_io() will decide if the reads should be modified. 501 */ 502 if (bio_data_dir(bio) == READ) { 503 if (test_bit(ERROR_READS, &fc->flags)) 504 return DM_MAPIO_KILL; 505 goto map_bio; 506 } 507 508 /* 509 * Drop or error writes? 510 */ 511 if (test_bit(DROP_WRITES, &fc->flags)) { 512 bio_endio(bio); 513 return DM_MAPIO_SUBMITTED; 514 } else if (test_bit(ERROR_WRITES, &fc->flags)) { 515 bio_io_error(bio); 516 return DM_MAPIO_SUBMITTED; 517 } 518 519 /* 520 * Corrupt matching writes. 521 */ 522 corrupt_fixed = false; 523 corrupt_random = false; 524 if (fc->corrupt_bio_byte && fc->corrupt_bio_rw == WRITE) { 525 if (all_corrupt_bio_flags_match(bio, fc)) 526 corrupt_fixed = true; 527 } 528 if (fc->random_write_corrupt) { 529 u64 rnd = get_random_u64(); 530 u32 rem = do_div(rnd, PROBABILITY_BASE); 531 if (rem < fc->random_write_corrupt) 532 corrupt_random = true; 533 } 534 if (corrupt_fixed || corrupt_random) { 535 struct bio *clone = clone_bio(ti, fc, bio); 536 if (clone) { 537 if (corrupt_fixed) 538 corrupt_bio_data(clone, fc); 539 if (corrupt_random) 540 corrupt_bio_random(clone); 541 submit_bio(clone); 542 return DM_MAPIO_SUBMITTED; 543 } 544 } 545 } 546 547map_bio: 548 flakey_map_bio(ti, bio); 549 550 return DM_MAPIO_REMAPPED; 551} 552 553static int flakey_end_io(struct dm_target *ti, struct bio *bio, 554 blk_status_t *error) 555{ 556 struct flakey_c *fc = ti->private; 557 struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); 558 559 if (op_is_zone_mgmt(bio_op(bio))) 560 return DM_ENDIO_DONE; 561 562 if (!*error && pb->bio_submitted && (bio_data_dir(bio) == READ)) { 563 if (fc->corrupt_bio_byte) { 564 if ((fc->corrupt_bio_rw == READ) && 565 all_corrupt_bio_flags_match(bio, fc)) { 566 /* 567 * Corrupt successful matching READs while in down state. 568 */ 569 corrupt_bio_data(bio, fc); 570 } 571 } 572 if (fc->random_read_corrupt) { 573 u64 rnd = get_random_u64(); 574 u32 rem = do_div(rnd, PROBABILITY_BASE); 575 if (rem < fc->random_read_corrupt) 576 corrupt_bio_random(bio); 577 } 578 if (test_bit(ERROR_READS, &fc->flags)) { 579 /* 580 * Error read during the down_interval if drop_writes 581 * and error_writes were not configured. 582 */ 583 *error = BLK_STS_IOERR; 584 } 585 } 586 587 return DM_ENDIO_DONE; 588} 589 590static void flakey_status(struct dm_target *ti, status_type_t type, 591 unsigned int status_flags, char *result, unsigned int maxlen) 592{ 593 unsigned int sz = 0; 594 struct flakey_c *fc = ti->private; 595 unsigned int error_reads, drop_writes, error_writes; 596 597 switch (type) { 598 case STATUSTYPE_INFO: 599 result[0] = '\0'; 600 break; 601 602 case STATUSTYPE_TABLE: 603 DMEMIT("%s %llu %u %u", fc->dev->name, 604 (unsigned long long)fc->start, fc->up_interval, 605 fc->down_interval); 606 607 error_reads = test_bit(ERROR_READS, &fc->flags); 608 drop_writes = test_bit(DROP_WRITES, &fc->flags); 609 error_writes = test_bit(ERROR_WRITES, &fc->flags); 610 DMEMIT(" %u", error_reads + drop_writes + error_writes + 611 (fc->corrupt_bio_byte > 0) * 5 + 612 (fc->random_read_corrupt > 0) * 2 + 613 (fc->random_write_corrupt > 0) * 2); 614 615 if (error_reads) 616 DMEMIT(" error_reads"); 617 if (drop_writes) 618 DMEMIT(" drop_writes"); 619 else if (error_writes) 620 DMEMIT(" error_writes"); 621 622 if (fc->corrupt_bio_byte) 623 DMEMIT(" corrupt_bio_byte %u %c %u %u", 624 fc->corrupt_bio_byte, 625 (fc->corrupt_bio_rw == WRITE) ? 'w' : 'r', 626 fc->corrupt_bio_value, fc->corrupt_bio_flags); 627 628 if (fc->random_read_corrupt > 0) 629 DMEMIT(" random_read_corrupt %u", fc->random_read_corrupt); 630 if (fc->random_write_corrupt > 0) 631 DMEMIT(" random_write_corrupt %u", fc->random_write_corrupt); 632 633 break; 634 635 case STATUSTYPE_IMA: 636 result[0] = '\0'; 637 break; 638 } 639} 640 641static int flakey_prepare_ioctl(struct dm_target *ti, struct block_device **bdev) 642{ 643 struct flakey_c *fc = ti->private; 644 645 *bdev = fc->dev->bdev; 646 647 /* 648 * Only pass ioctls through if the device sizes match exactly. 649 */ 650 if (fc->start || ti->len != bdev_nr_sectors((*bdev))) 651 return 1; 652 return 0; 653} 654 655#ifdef CONFIG_BLK_DEV_ZONED 656static int flakey_report_zones(struct dm_target *ti, 657 struct dm_report_zones_args *args, unsigned int nr_zones) 658{ 659 struct flakey_c *fc = ti->private; 660 661 return dm_report_zones(fc->dev->bdev, fc->start, 662 flakey_map_sector(ti, args->next_sector), 663 args, nr_zones); 664} 665#else 666#define flakey_report_zones NULL 667#endif 668 669static int flakey_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn, void *data) 670{ 671 struct flakey_c *fc = ti->private; 672 673 return fn(ti, fc->dev, fc->start, ti->len, data); 674} 675 676static struct target_type flakey_target = { 677 .name = "flakey", 678 .version = {1, 5, 0}, 679 .features = DM_TARGET_ZONED_HM | DM_TARGET_PASSES_CRYPTO, 680 .report_zones = flakey_report_zones, 681 .module = THIS_MODULE, 682 .ctr = flakey_ctr, 683 .dtr = flakey_dtr, 684 .map = flakey_map, 685 .end_io = flakey_end_io, 686 .status = flakey_status, 687 .prepare_ioctl = flakey_prepare_ioctl, 688 .iterate_devices = flakey_iterate_devices, 689}; 690module_dm(flakey); 691 692MODULE_DESCRIPTION(DM_NAME " flakey target"); 693MODULE_AUTHOR("Joe Thornber <dm-devel@lists.linux.dev>"); 694MODULE_LICENSE("GPL"); 695