1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 1991-1998 Linus Torvalds 4 * Re-organised Feb 1998 Russell King 5 * Copyright (C) 2020 Christoph Hellwig 6 */ 7 #include <linux/fs.h> 8 #include <linux/slab.h> 9 #include <linux/ctype.h> 10 #include <linux/genhd.h> 11 #include <linux/vmalloc.h> 12 #include <linux/blktrace_api.h> 13 #include <linux/raid/detect.h> 14 #include "check.h" 15 16 static int (*check_part[])(struct parsed_partitions *) = { 17 /* 18 * Probe partition formats with tables at disk address 0 19 * that also have an ADFS boot block at 0xdc0. 20 */ 21 #ifdef CONFIG_ACORN_PARTITION_ICS 22 adfspart_check_ICS, 23 #endif 24 #ifdef CONFIG_ACORN_PARTITION_POWERTEC 25 adfspart_check_POWERTEC, 26 #endif 27 #ifdef CONFIG_ACORN_PARTITION_EESOX 28 adfspart_check_EESOX, 29 #endif 30 31 /* 32 * Now move on to formats that only have partition info at 33 * disk address 0xdc0. Since these may also have stale 34 * PC/BIOS partition tables, they need to come before 35 * the msdos entry. 36 */ 37 #ifdef CONFIG_ACORN_PARTITION_CUMANA 38 adfspart_check_CUMANA, 39 #endif 40 #ifdef CONFIG_ACORN_PARTITION_ADFS 41 adfspart_check_ADFS, 42 #endif 43 44 #ifdef CONFIG_CMDLINE_PARTITION 45 cmdline_partition, 46 #endif 47 #ifdef CONFIG_EFI_PARTITION 48 efi_partition, /* this must come before msdos */ 49 #endif 50 #ifdef CONFIG_SGI_PARTITION 51 sgi_partition, 52 #endif 53 #ifdef CONFIG_LDM_PARTITION 54 ldm_partition, /* this must come before msdos */ 55 #endif 56 #ifdef CONFIG_MSDOS_PARTITION 57 msdos_partition, 58 #endif 59 #ifdef CONFIG_OSF_PARTITION 60 osf_partition, 61 #endif 62 #ifdef CONFIG_SUN_PARTITION 63 sun_partition, 64 #endif 65 #ifdef CONFIG_AMIGA_PARTITION 66 amiga_partition, 67 #endif 68 #ifdef CONFIG_ATARI_PARTITION 69 atari_partition, 70 #endif 71 #ifdef CONFIG_MAC_PARTITION 72 mac_partition, 73 #endif 74 #ifdef CONFIG_ULTRIX_PARTITION 75 ultrix_partition, 76 #endif 77 #ifdef CONFIG_IBM_PARTITION 78 ibm_partition, 79 #endif 80 #ifdef CONFIG_KARMA_PARTITION 81 karma_partition, 82 #endif 83 #ifdef CONFIG_SYSV68_PARTITION 84 sysv68_partition, 85 #endif 86 NULL 87 }; 88 89 static void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors) 90 { 91 spin_lock(&bdev->bd_size_lock); 92 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); 93 spin_unlock(&bdev->bd_size_lock); 94 } 95 96 static struct parsed_partitions *allocate_partitions(struct gendisk *hd) 97 { 98 struct parsed_partitions *state; 99 int nr; 100 101 state = kzalloc(sizeof(*state), GFP_KERNEL); 102 if (!state) 103 return NULL; 104 105 nr = disk_max_parts(hd); 106 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0]))); 107 if (!state->parts) { 108 kfree(state); 109 return NULL; 110 } 111 112 state->limit = nr; 113 114 return state; 115 } 116 117 static void free_partitions(struct parsed_partitions *state) 118 { 119 vfree(state->parts); 120 kfree(state); 121 } 122 123 static struct parsed_partitions *check_partition(struct gendisk *hd, 124 struct block_device *bdev) 125 { 126 struct parsed_partitions *state; 127 int i, res, err; 128 129 state = allocate_partitions(hd); 130 if (!state) 131 return NULL; 132 state->pp_buf = (char *)__get_free_page(GFP_KERNEL); 133 if (!state->pp_buf) { 134 free_partitions(state); 135 return NULL; 136 } 137 state->pp_buf[0] = '\0'; 138 139 state->bdev = bdev; 140 disk_name(hd, 0, state->name); 141 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name); 142 if (isdigit(state->name[strlen(state->name)-1])) 143 sprintf(state->name, "p"); 144 145 i = res = err = 0; 146 while (!res && check_part[i]) { 147 memset(state->parts, 0, state->limit * sizeof(state->parts[0])); 148 res = check_part[i++](state); 149 if (res < 0) { 150 /* 151 * We have hit an I/O error which we don't report now. 152 * But record it, and let the others do their job. 153 */ 154 err = res; 155 res = 0; 156 } 157 158 } 159 if (res > 0) { 160 printk(KERN_INFO "%s", state->pp_buf); 161 162 free_page((unsigned long)state->pp_buf); 163 return state; 164 } 165 if (state->access_beyond_eod) 166 err = -ENOSPC; 167 /* 168 * The partition is unrecognized. So report I/O errors if there were any 169 */ 170 if (err) 171 res = err; 172 if (res) { 173 strlcat(state->pp_buf, 174 " unable to read partition table\n", PAGE_SIZE); 175 printk(KERN_INFO "%s", state->pp_buf); 176 } 177 178 free_page((unsigned long)state->pp_buf); 179 free_partitions(state); 180 return ERR_PTR(res); 181 } 182 183 static ssize_t part_partition_show(struct device *dev, 184 struct device_attribute *attr, char *buf) 185 { 186 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno); 187 } 188 189 static ssize_t part_start_show(struct device *dev, 190 struct device_attribute *attr, char *buf) 191 { 192 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect); 193 } 194 195 static ssize_t part_ro_show(struct device *dev, 196 struct device_attribute *attr, char *buf) 197 { 198 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_read_only); 199 } 200 201 static ssize_t part_alignment_offset_show(struct device *dev, 202 struct device_attribute *attr, char *buf) 203 { 204 struct block_device *bdev = dev_to_bdev(dev); 205 206 return sprintf(buf, "%u\n", 207 queue_limit_alignment_offset(&bdev->bd_disk->queue->limits, 208 bdev->bd_start_sect)); 209 } 210 211 static ssize_t part_discard_alignment_show(struct device *dev, 212 struct device_attribute *attr, char *buf) 213 { 214 struct block_device *bdev = dev_to_bdev(dev); 215 216 return sprintf(buf, "%u\n", 217 queue_limit_discard_alignment(&bdev->bd_disk->queue->limits, 218 bdev->bd_start_sect)); 219 } 220 221 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL); 222 static DEVICE_ATTR(start, 0444, part_start_show, NULL); 223 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 224 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL); 225 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL); 226 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL); 227 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 228 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 229 #ifdef CONFIG_FAIL_MAKE_REQUEST 230 static struct device_attribute dev_attr_fail = 231 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 232 #endif 233 234 static struct attribute *part_attrs[] = { 235 &dev_attr_partition.attr, 236 &dev_attr_start.attr, 237 &dev_attr_size.attr, 238 &dev_attr_ro.attr, 239 &dev_attr_alignment_offset.attr, 240 &dev_attr_discard_alignment.attr, 241 &dev_attr_stat.attr, 242 &dev_attr_inflight.attr, 243 #ifdef CONFIG_FAIL_MAKE_REQUEST 244 &dev_attr_fail.attr, 245 #endif 246 NULL 247 }; 248 249 static struct attribute_group part_attr_group = { 250 .attrs = part_attrs, 251 }; 252 253 static const struct attribute_group *part_attr_groups[] = { 254 &part_attr_group, 255 #ifdef CONFIG_BLK_DEV_IO_TRACE 256 &blk_trace_attr_group, 257 #endif 258 NULL 259 }; 260 261 static void part_release(struct device *dev) 262 { 263 blk_free_devt(dev->devt); 264 bdput(dev_to_bdev(dev)); 265 } 266 267 static int part_uevent(struct device *dev, struct kobj_uevent_env *env) 268 { 269 struct block_device *part = dev_to_bdev(dev); 270 271 add_uevent_var(env, "PARTN=%u", part->bd_partno); 272 if (part->bd_meta_info && part->bd_meta_info->volname[0]) 273 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname); 274 return 0; 275 } 276 277 struct device_type part_type = { 278 .name = "partition", 279 .groups = part_attr_groups, 280 .release = part_release, 281 .uevent = part_uevent, 282 }; 283 284 /* 285 * Must be called either with bd_mutex held, before a disk can be opened or 286 * after all disk users are gone. 287 */ 288 void delete_partition(struct block_device *part) 289 { 290 struct gendisk *disk = part->bd_disk; 291 struct disk_part_tbl *ptbl = 292 rcu_dereference_protected(disk->part_tbl, 1); 293 294 rcu_assign_pointer(ptbl->part[part->bd_partno], NULL); 295 rcu_assign_pointer(ptbl->last_lookup, NULL); 296 297 kobject_put(part->bd_holder_dir); 298 device_del(&part->bd_device); 299 300 /* 301 * Remove the block device from the inode hash, so that it cannot be 302 * looked up any more even when openers still hold references. 303 */ 304 remove_inode_hash(part->bd_inode); 305 306 put_device(&part->bd_device); 307 } 308 309 static ssize_t whole_disk_show(struct device *dev, 310 struct device_attribute *attr, char *buf) 311 { 312 return 0; 313 } 314 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 315 316 /* 317 * Must be called either with bd_mutex held, before a disk can be opened or 318 * after all disk users are gone. 319 */ 320 static struct block_device *add_partition(struct gendisk *disk, int partno, 321 sector_t start, sector_t len, int flags, 322 struct partition_meta_info *info) 323 { 324 dev_t devt = MKDEV(0, 0); 325 struct device *ddev = disk_to_dev(disk); 326 struct device *pdev; 327 struct block_device *bdev; 328 struct disk_part_tbl *ptbl; 329 const char *dname; 330 int err; 331 332 /* 333 * Partitions are not supported on zoned block devices that are used as 334 * such. 335 */ 336 switch (disk->queue->limits.zoned) { 337 case BLK_ZONED_HM: 338 pr_warn("%s: partitions not supported on host managed zoned block device\n", 339 disk->disk_name); 340 return ERR_PTR(-ENXIO); 341 case BLK_ZONED_HA: 342 pr_info("%s: disabling host aware zoned block device support due to partitions\n", 343 disk->disk_name); 344 disk->queue->limits.zoned = BLK_ZONED_NONE; 345 break; 346 case BLK_ZONED_NONE: 347 break; 348 } 349 350 err = disk_expand_part_tbl(disk, partno); 351 if (err) 352 return ERR_PTR(err); 353 ptbl = rcu_dereference_protected(disk->part_tbl, 1); 354 355 if (ptbl->part[partno]) 356 return ERR_PTR(-EBUSY); 357 358 bdev = bdev_alloc(disk, partno); 359 if (!bdev) 360 return ERR_PTR(-ENOMEM); 361 362 bdev->bd_start_sect = start; 363 bdev_set_nr_sectors(bdev, len); 364 bdev->bd_read_only = get_disk_ro(disk); 365 366 if (info) { 367 err = -ENOMEM; 368 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 369 if (!bdev->bd_meta_info) 370 goto out_bdput; 371 } 372 373 pdev = &bdev->bd_device; 374 dname = dev_name(ddev); 375 if (isdigit(dname[strlen(dname) - 1])) 376 dev_set_name(pdev, "%sp%d", dname, partno); 377 else 378 dev_set_name(pdev, "%s%d", dname, partno); 379 380 device_initialize(pdev); 381 pdev->class = &block_class; 382 pdev->type = &part_type; 383 pdev->parent = ddev; 384 385 err = blk_alloc_devt(bdev, &devt); 386 if (err) 387 goto out_bdput; 388 pdev->devt = devt; 389 390 /* delay uevent until 'holders' subdir is created */ 391 dev_set_uevent_suppress(pdev, 1); 392 err = device_add(pdev); 393 if (err) 394 goto out_put; 395 396 err = -ENOMEM; 397 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj); 398 if (!bdev->bd_holder_dir) 399 goto out_del; 400 401 dev_set_uevent_suppress(pdev, 0); 402 if (flags & ADDPART_FLAG_WHOLEDISK) { 403 err = device_create_file(pdev, &dev_attr_whole_disk); 404 if (err) 405 goto out_del; 406 } 407 408 /* everything is up and running, commence */ 409 bdev_add(bdev, devt); 410 rcu_assign_pointer(ptbl->part[partno], bdev); 411 412 /* suppress uevent if the disk suppresses it */ 413 if (!dev_get_uevent_suppress(ddev)) 414 kobject_uevent(&pdev->kobj, KOBJ_ADD); 415 return bdev; 416 417 out_bdput: 418 bdput(bdev); 419 return ERR_PTR(err); 420 out_del: 421 kobject_put(bdev->bd_holder_dir); 422 device_del(pdev); 423 out_put: 424 put_device(pdev); 425 return ERR_PTR(err); 426 } 427 428 static bool partition_overlaps(struct gendisk *disk, sector_t start, 429 sector_t length, int skip_partno) 430 { 431 struct disk_part_iter piter; 432 struct block_device *part; 433 bool overlap = false; 434 435 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY); 436 while ((part = disk_part_iter_next(&piter))) { 437 if (part->bd_partno == skip_partno || 438 start >= part->bd_start_sect + bdev_nr_sectors(part) || 439 start + length <= part->bd_start_sect) 440 continue; 441 overlap = true; 442 break; 443 } 444 445 disk_part_iter_exit(&piter); 446 return overlap; 447 } 448 449 int bdev_add_partition(struct block_device *bdev, int partno, 450 sector_t start, sector_t length) 451 { 452 struct block_device *part; 453 454 mutex_lock(&bdev->bd_mutex); 455 if (partition_overlaps(bdev->bd_disk, start, length, -1)) { 456 mutex_unlock(&bdev->bd_mutex); 457 return -EBUSY; 458 } 459 460 part = add_partition(bdev->bd_disk, partno, start, length, 461 ADDPART_FLAG_NONE, NULL); 462 mutex_unlock(&bdev->bd_mutex); 463 return PTR_ERR_OR_ZERO(part); 464 } 465 466 int bdev_del_partition(struct block_device *bdev, int partno) 467 { 468 struct block_device *part; 469 int ret; 470 471 part = bdget_disk(bdev->bd_disk, partno); 472 if (!part) 473 return -ENXIO; 474 475 mutex_lock(&part->bd_mutex); 476 mutex_lock_nested(&bdev->bd_mutex, 1); 477 478 ret = -EBUSY; 479 if (part->bd_openers) 480 goto out_unlock; 481 482 sync_blockdev(part); 483 invalidate_bdev(part); 484 485 delete_partition(part); 486 ret = 0; 487 out_unlock: 488 mutex_unlock(&bdev->bd_mutex); 489 mutex_unlock(&part->bd_mutex); 490 bdput(part); 491 return ret; 492 } 493 494 int bdev_resize_partition(struct block_device *bdev, int partno, 495 sector_t start, sector_t length) 496 { 497 struct block_device *part; 498 int ret = 0; 499 500 part = bdget_disk(bdev->bd_disk, partno); 501 if (!part) 502 return -ENXIO; 503 504 mutex_lock(&part->bd_mutex); 505 mutex_lock_nested(&bdev->bd_mutex, 1); 506 ret = -EINVAL; 507 if (start != part->bd_start_sect) 508 goto out_unlock; 509 510 ret = -EBUSY; 511 if (partition_overlaps(bdev->bd_disk, start, length, partno)) 512 goto out_unlock; 513 514 bdev_set_nr_sectors(part, length); 515 516 ret = 0; 517 out_unlock: 518 mutex_unlock(&part->bd_mutex); 519 mutex_unlock(&bdev->bd_mutex); 520 bdput(part); 521 return ret; 522 } 523 524 static bool disk_unlock_native_capacity(struct gendisk *disk) 525 { 526 const struct block_device_operations *bdops = disk->fops; 527 528 if (bdops->unlock_native_capacity && 529 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) { 530 printk(KERN_CONT "enabling native capacity\n"); 531 bdops->unlock_native_capacity(disk); 532 disk->flags |= GENHD_FL_NATIVE_CAPACITY; 533 return true; 534 } else { 535 printk(KERN_CONT "truncated\n"); 536 return false; 537 } 538 } 539 540 int blk_drop_partitions(struct block_device *bdev) 541 { 542 struct disk_part_iter piter; 543 struct block_device *part; 544 545 if (bdev->bd_part_count) 546 return -EBUSY; 547 548 sync_blockdev(bdev); 549 invalidate_bdev(bdev); 550 551 disk_part_iter_init(&piter, bdev->bd_disk, DISK_PITER_INCL_EMPTY); 552 while ((part = disk_part_iter_next(&piter))) 553 delete_partition(part); 554 disk_part_iter_exit(&piter); 555 556 return 0; 557 } 558 #ifdef CONFIG_S390 559 /* for historic reasons in the DASD driver */ 560 EXPORT_SYMBOL_GPL(blk_drop_partitions); 561 #endif 562 563 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev, 564 struct parsed_partitions *state, int p) 565 { 566 sector_t size = state->parts[p].size; 567 sector_t from = state->parts[p].from; 568 struct block_device *part; 569 570 if (!size) 571 return true; 572 573 if (from >= get_capacity(disk)) { 574 printk(KERN_WARNING 575 "%s: p%d start %llu is beyond EOD, ", 576 disk->disk_name, p, (unsigned long long) from); 577 if (disk_unlock_native_capacity(disk)) 578 return false; 579 return true; 580 } 581 582 if (from + size > get_capacity(disk)) { 583 printk(KERN_WARNING 584 "%s: p%d size %llu extends beyond EOD, ", 585 disk->disk_name, p, (unsigned long long) size); 586 587 if (disk_unlock_native_capacity(disk)) 588 return false; 589 590 /* 591 * We can not ignore partitions of broken tables created by for 592 * example camera firmware, but we limit them to the end of the 593 * disk to avoid creating invalid block devices. 594 */ 595 size = get_capacity(disk) - from; 596 } 597 598 part = add_partition(disk, p, from, size, state->parts[p].flags, 599 &state->parts[p].info); 600 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) { 601 printk(KERN_ERR " %s: p%d could not be added: %ld\n", 602 disk->disk_name, p, -PTR_ERR(part)); 603 return true; 604 } 605 606 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 607 (state->parts[p].flags & ADDPART_FLAG_RAID)) 608 md_autodetect_dev(part->bd_dev); 609 610 return true; 611 } 612 613 int blk_add_partitions(struct gendisk *disk, struct block_device *bdev) 614 { 615 struct parsed_partitions *state; 616 int ret = -EAGAIN, p, highest; 617 618 if (!disk_part_scan_enabled(disk)) 619 return 0; 620 621 state = check_partition(disk, bdev); 622 if (!state) 623 return 0; 624 if (IS_ERR(state)) { 625 /* 626 * I/O error reading the partition table. If we tried to read 627 * beyond EOD, retry after unlocking the native capacity. 628 */ 629 if (PTR_ERR(state) == -ENOSPC) { 630 printk(KERN_WARNING "%s: partition table beyond EOD, ", 631 disk->disk_name); 632 if (disk_unlock_native_capacity(disk)) 633 return -EAGAIN; 634 } 635 return -EIO; 636 } 637 638 /* 639 * Partitions are not supported on host managed zoned block devices. 640 */ 641 if (disk->queue->limits.zoned == BLK_ZONED_HM) { 642 pr_warn("%s: ignoring partition table on host managed zoned block device\n", 643 disk->disk_name); 644 ret = 0; 645 goto out_free_state; 646 } 647 648 /* 649 * If we read beyond EOD, try unlocking native capacity even if the 650 * partition table was successfully read as we could be missing some 651 * partitions. 652 */ 653 if (state->access_beyond_eod) { 654 printk(KERN_WARNING 655 "%s: partition table partially beyond EOD, ", 656 disk->disk_name); 657 if (disk_unlock_native_capacity(disk)) 658 goto out_free_state; 659 } 660 661 /* tell userspace that the media / partition table may have changed */ 662 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 663 664 /* 665 * Detect the highest partition number and preallocate disk->part_tbl. 666 * This is an optimization and not strictly necessary. 667 */ 668 for (p = 1, highest = 0; p < state->limit; p++) 669 if (state->parts[p].size) 670 highest = p; 671 disk_expand_part_tbl(disk, highest); 672 673 for (p = 1; p < state->limit; p++) 674 if (!blk_add_partition(disk, bdev, state, p)) 675 goto out_free_state; 676 677 ret = 0; 678 out_free_state: 679 free_partitions(state); 680 return ret; 681 } 682 683 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 684 { 685 struct address_space *mapping = state->bdev->bd_inode->i_mapping; 686 struct page *page; 687 688 if (n >= get_capacity(state->bdev->bd_disk)) { 689 state->access_beyond_eod = true; 690 return NULL; 691 } 692 693 page = read_mapping_page(mapping, 694 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL); 695 if (IS_ERR(page)) 696 goto out; 697 if (PageError(page)) 698 goto out_put_page; 699 700 p->v = page; 701 return (unsigned char *)page_address(page) + 702 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT); 703 out_put_page: 704 put_page(page); 705 out: 706 p->v = NULL; 707 return NULL; 708 } 709