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/major.h> 9 #include <linux/slab.h> 10 #include <linux/string.h> 11 #include <linux/ctype.h> 12 #include <linux/vmalloc.h> 13 #include <linux/raid/detect.h> 14 #include "check.h" 15 16 static int (*const 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_OF_PARTITION 48 of_partition, /* cmdline have priority to OF */ 49 #endif 50 #ifdef CONFIG_EFI_PARTITION 51 efi_partition, /* this must come before msdos */ 52 #endif 53 #ifdef CONFIG_SGI_PARTITION 54 sgi_partition, 55 #endif 56 #ifdef CONFIG_LDM_PARTITION 57 ldm_partition, /* this must come before msdos */ 58 #endif 59 #ifdef CONFIG_MSDOS_PARTITION 60 msdos_partition, 61 #endif 62 #ifdef CONFIG_OSF_PARTITION 63 osf_partition, 64 #endif 65 #ifdef CONFIG_SUN_PARTITION 66 sun_partition, 67 #endif 68 #ifdef CONFIG_AMIGA_PARTITION 69 amiga_partition, 70 #endif 71 #ifdef CONFIG_ATARI_PARTITION 72 atari_partition, 73 #endif 74 #ifdef CONFIG_MAC_PARTITION 75 mac_partition, 76 #endif 77 #ifdef CONFIG_ULTRIX_PARTITION 78 ultrix_partition, 79 #endif 80 #ifdef CONFIG_IBM_PARTITION 81 ibm_partition, 82 #endif 83 #ifdef CONFIG_KARMA_PARTITION 84 karma_partition, 85 #endif 86 #ifdef CONFIG_SYSV68_PARTITION 87 sysv68_partition, 88 #endif 89 NULL 90 }; 91 92 static struct parsed_partitions *allocate_partitions(struct gendisk *hd) 93 { 94 struct parsed_partitions *state; 95 int nr = DISK_MAX_PARTS; 96 97 state = kzalloc_obj(*state, GFP_KERNEL); 98 if (!state) 99 return NULL; 100 101 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0]))); 102 if (!state->parts) { 103 kfree(state); 104 return NULL; 105 } 106 107 state->limit = nr; 108 109 return state; 110 } 111 112 static void free_partitions(struct parsed_partitions *state) 113 { 114 vfree(state->parts); 115 kfree(state); 116 } 117 118 static struct parsed_partitions *check_partition(struct gendisk *hd) 119 { 120 struct parsed_partitions *state; 121 int i, res, err; 122 123 state = allocate_partitions(hd); 124 if (!state) 125 return NULL; 126 state->pp_buf = (char *)__get_free_page(GFP_KERNEL); 127 if (!state->pp_buf) { 128 free_partitions(state); 129 return NULL; 130 } 131 state->pp_buf[0] = '\0'; 132 133 state->disk = hd; 134 strscpy(state->name, hd->disk_name); 135 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name); 136 if (isdigit(state->name[strlen(state->name)-1])) 137 sprintf(state->name, "p"); 138 139 i = res = err = 0; 140 while (!res && check_part[i]) { 141 memset(state->parts, 0, state->limit * sizeof(state->parts[0])); 142 res = check_part[i++](state); 143 if (res < 0) { 144 /* 145 * We have hit an I/O error which we don't report now. 146 * But record it, and let the others do their job. 147 */ 148 err = res; 149 res = 0; 150 } 151 152 } 153 if (res > 0) { 154 printk(KERN_INFO "%s", state->pp_buf); 155 156 free_page((unsigned long)state->pp_buf); 157 return state; 158 } 159 if (state->access_beyond_eod) 160 err = -ENOSPC; 161 /* 162 * The partition is unrecognized. So report I/O errors if there were any 163 */ 164 if (err) 165 res = err; 166 if (res) { 167 strlcat(state->pp_buf, 168 " unable to read partition table\n", PAGE_SIZE); 169 printk(KERN_INFO "%s", state->pp_buf); 170 } 171 172 free_page((unsigned long)state->pp_buf); 173 free_partitions(state); 174 return ERR_PTR(res); 175 } 176 177 static ssize_t part_partition_show(struct device *dev, 178 struct device_attribute *attr, char *buf) 179 { 180 return sprintf(buf, "%d\n", bdev_partno(dev_to_bdev(dev))); 181 } 182 183 static ssize_t part_start_show(struct device *dev, 184 struct device_attribute *attr, char *buf) 185 { 186 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect); 187 } 188 189 static ssize_t part_ro_show(struct device *dev, 190 struct device_attribute *attr, char *buf) 191 { 192 return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev))); 193 } 194 195 static ssize_t part_alignment_offset_show(struct device *dev, 196 struct device_attribute *attr, char *buf) 197 { 198 return sprintf(buf, "%u\n", bdev_alignment_offset(dev_to_bdev(dev))); 199 } 200 201 static ssize_t part_discard_alignment_show(struct device *dev, 202 struct device_attribute *attr, char *buf) 203 { 204 return sprintf(buf, "%u\n", bdev_discard_alignment(dev_to_bdev(dev))); 205 } 206 207 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL); 208 static DEVICE_ATTR(start, 0444, part_start_show, NULL); 209 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 210 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL); 211 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL); 212 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL); 213 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 214 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 215 #ifdef CONFIG_FAIL_MAKE_REQUEST 216 static struct device_attribute dev_attr_fail = 217 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 218 #endif 219 220 static struct attribute *part_attrs[] = { 221 &dev_attr_partition.attr, 222 &dev_attr_start.attr, 223 &dev_attr_size.attr, 224 &dev_attr_ro.attr, 225 &dev_attr_alignment_offset.attr, 226 &dev_attr_discard_alignment.attr, 227 &dev_attr_stat.attr, 228 &dev_attr_inflight.attr, 229 #ifdef CONFIG_FAIL_MAKE_REQUEST 230 &dev_attr_fail.attr, 231 #endif 232 NULL 233 }; 234 235 static const struct attribute_group part_attr_group = { 236 .attrs = part_attrs, 237 }; 238 239 static const struct attribute_group *part_attr_groups[] = { 240 &part_attr_group, 241 #ifdef CONFIG_BLK_DEV_IO_TRACE 242 &blk_trace_attr_group, 243 #endif 244 NULL 245 }; 246 247 static void part_release(struct device *dev) 248 { 249 put_disk(dev_to_bdev(dev)->bd_disk); 250 bdev_drop(dev_to_bdev(dev)); 251 } 252 253 static int part_uevent(const struct device *dev, struct kobj_uevent_env *env) 254 { 255 const struct block_device *part = dev_to_bdev(dev); 256 257 add_uevent_var(env, "PARTN=%u", bdev_partno(part)); 258 if (part->bd_meta_info && part->bd_meta_info->volname[0]) 259 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname); 260 if (part->bd_meta_info && part->bd_meta_info->uuid[0]) 261 add_uevent_var(env, "PARTUUID=%s", part->bd_meta_info->uuid); 262 return 0; 263 } 264 265 const struct device_type part_type = { 266 .name = "partition", 267 .groups = part_attr_groups, 268 .release = part_release, 269 .uevent = part_uevent, 270 }; 271 272 void drop_partition(struct block_device *part) 273 { 274 lockdep_assert_held(&part->bd_disk->open_mutex); 275 276 xa_erase(&part->bd_disk->part_tbl, bdev_partno(part)); 277 kobject_put(part->bd_holder_dir); 278 279 device_del(&part->bd_device); 280 put_device(&part->bd_device); 281 } 282 283 static ssize_t whole_disk_show(struct device *dev, 284 struct device_attribute *attr, char *buf) 285 { 286 return 0; 287 } 288 static const DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 289 290 /* 291 * Must be called either with open_mutex held, before a disk can be opened or 292 * after all disk users are gone. 293 */ 294 static struct block_device *add_partition(struct gendisk *disk, int partno, 295 sector_t start, sector_t len, int flags, 296 struct partition_meta_info *info) 297 { 298 dev_t devt = MKDEV(0, 0); 299 struct device *ddev = disk_to_dev(disk); 300 struct device *pdev; 301 struct block_device *bdev; 302 const char *dname; 303 int err; 304 305 lockdep_assert_held(&disk->open_mutex); 306 307 if (partno >= DISK_MAX_PARTS) 308 return ERR_PTR(-EINVAL); 309 310 /* 311 * Partitions are not supported on zoned block devices that are used as 312 * such. 313 */ 314 if (bdev_is_zoned(disk->part0)) { 315 pr_warn("%s: partitions not supported on host managed zoned block device\n", 316 disk->disk_name); 317 return ERR_PTR(-ENXIO); 318 } 319 320 if (xa_load(&disk->part_tbl, partno)) 321 return ERR_PTR(-EBUSY); 322 323 /* ensure we always have a reference to the whole disk */ 324 get_device(disk_to_dev(disk)); 325 326 err = -ENOMEM; 327 bdev = bdev_alloc(disk, partno); 328 if (!bdev) 329 goto out_put_disk; 330 331 bdev->bd_start_sect = start; 332 bdev_set_nr_sectors(bdev, len); 333 334 pdev = &bdev->bd_device; 335 dname = dev_name(ddev); 336 if (isdigit(dname[strlen(dname) - 1])) 337 dev_set_name(pdev, "%sp%d", dname, partno); 338 else 339 dev_set_name(pdev, "%s%d", dname, partno); 340 341 device_initialize(pdev); 342 pdev->class = &block_class; 343 pdev->type = &part_type; 344 pdev->parent = ddev; 345 346 /* in consecutive minor range? */ 347 if (bdev_partno(bdev) < disk->minors) { 348 devt = MKDEV(disk->major, disk->first_minor + bdev_partno(bdev)); 349 } else { 350 err = blk_alloc_ext_minor(); 351 if (err < 0) 352 goto out_put; 353 devt = MKDEV(BLOCK_EXT_MAJOR, err); 354 } 355 pdev->devt = devt; 356 357 if (info) { 358 err = -ENOMEM; 359 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 360 if (!bdev->bd_meta_info) 361 goto out_put; 362 } 363 364 /* delay uevent until 'holders' subdir is created */ 365 dev_set_uevent_suppress(pdev, 1); 366 err = device_add(pdev); 367 if (err) 368 goto out_put; 369 370 err = -ENOMEM; 371 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj); 372 if (!bdev->bd_holder_dir) 373 goto out_del; 374 375 dev_set_uevent_suppress(pdev, 0); 376 if (flags & ADDPART_FLAG_WHOLEDISK) { 377 err = device_create_file(pdev, &dev_attr_whole_disk); 378 if (err) 379 goto out_del; 380 } 381 382 if (flags & ADDPART_FLAG_READONLY) 383 bdev_set_flag(bdev, BD_READ_ONLY); 384 385 /* everything is up and running, commence */ 386 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL); 387 if (err) 388 goto out_del; 389 bdev_add(bdev, devt); 390 391 /* suppress uevent if the disk suppresses it */ 392 if (!dev_get_uevent_suppress(ddev)) 393 kobject_uevent(&pdev->kobj, KOBJ_ADD); 394 return bdev; 395 396 out_del: 397 kobject_put(bdev->bd_holder_dir); 398 device_del(pdev); 399 out_put: 400 put_device(pdev); 401 return ERR_PTR(err); 402 out_put_disk: 403 put_disk(disk); 404 return ERR_PTR(err); 405 } 406 407 static bool partition_overlaps(struct gendisk *disk, sector_t start, 408 sector_t length, int skip_partno) 409 { 410 struct block_device *part; 411 bool overlap = false; 412 unsigned long idx; 413 414 rcu_read_lock(); 415 xa_for_each_start(&disk->part_tbl, idx, part, 1) { 416 if (bdev_partno(part) != skip_partno && 417 start < part->bd_start_sect + bdev_nr_sectors(part) && 418 start + length > part->bd_start_sect) { 419 overlap = true; 420 break; 421 } 422 } 423 rcu_read_unlock(); 424 425 return overlap; 426 } 427 428 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, 429 sector_t length) 430 { 431 struct block_device *part; 432 int ret; 433 434 mutex_lock(&disk->open_mutex); 435 if (!disk_live(disk)) { 436 ret = -ENXIO; 437 goto out; 438 } 439 440 if (disk->flags & GENHD_FL_NO_PART) { 441 ret = -EINVAL; 442 goto out; 443 } 444 445 if (partition_overlaps(disk, start, length, -1)) { 446 ret = -EBUSY; 447 goto out; 448 } 449 450 part = add_partition(disk, partno, start, length, 451 ADDPART_FLAG_NONE, NULL); 452 ret = PTR_ERR_OR_ZERO(part); 453 out: 454 mutex_unlock(&disk->open_mutex); 455 return ret; 456 } 457 458 int bdev_del_partition(struct gendisk *disk, int partno) 459 { 460 struct block_device *part = NULL; 461 int ret = -ENXIO; 462 463 mutex_lock(&disk->open_mutex); 464 part = xa_load(&disk->part_tbl, partno); 465 if (!part) 466 goto out_unlock; 467 468 ret = -EBUSY; 469 if (atomic_read(&part->bd_openers)) 470 goto out_unlock; 471 472 /* 473 * We verified that @part->bd_openers is zero above and so 474 * @part->bd_holder{_ops} can't be set. And since we hold 475 * @disk->open_mutex the device can't be claimed by anyone. 476 * 477 * So no need to call @part->bd_holder_ops->mark_dead() here. 478 * Just delete the partition and invalidate it. 479 */ 480 481 bdev_unhash(part); 482 invalidate_bdev(part); 483 drop_partition(part); 484 ret = 0; 485 out_unlock: 486 mutex_unlock(&disk->open_mutex); 487 return ret; 488 } 489 490 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start, 491 sector_t length) 492 { 493 struct block_device *part = NULL; 494 int ret = -ENXIO; 495 496 mutex_lock(&disk->open_mutex); 497 part = xa_load(&disk->part_tbl, partno); 498 if (!part) 499 goto out_unlock; 500 501 ret = -EINVAL; 502 if (start != part->bd_start_sect) 503 goto out_unlock; 504 505 ret = -EBUSY; 506 if (partition_overlaps(disk, start, length, partno)) 507 goto out_unlock; 508 509 bdev_set_nr_sectors(part, length); 510 511 ret = 0; 512 out_unlock: 513 mutex_unlock(&disk->open_mutex); 514 return ret; 515 } 516 517 static bool disk_unlock_native_capacity(struct gendisk *disk) 518 { 519 if (!disk->fops->unlock_native_capacity || 520 test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) { 521 printk(KERN_CONT "truncated\n"); 522 return false; 523 } 524 525 printk(KERN_CONT "enabling native capacity\n"); 526 disk->fops->unlock_native_capacity(disk); 527 return true; 528 } 529 530 static bool blk_add_partition(struct gendisk *disk, 531 struct parsed_partitions *state, int p) 532 { 533 sector_t size = state->parts[p].size; 534 sector_t from = state->parts[p].from; 535 struct block_device *part; 536 537 if (!size) 538 return true; 539 540 if (from >= get_capacity(disk)) { 541 printk(KERN_WARNING 542 "%s: p%d start %llu is beyond EOD, ", 543 disk->disk_name, p, (unsigned long long) from); 544 if (disk_unlock_native_capacity(disk)) 545 return false; 546 return true; 547 } 548 549 if (from + size > get_capacity(disk)) { 550 printk(KERN_WARNING 551 "%s: p%d size %llu extends beyond EOD, ", 552 disk->disk_name, p, (unsigned long long) size); 553 554 if (disk_unlock_native_capacity(disk)) 555 return false; 556 557 /* 558 * We can not ignore partitions of broken tables created by for 559 * example camera firmware, but we limit them to the end of the 560 * disk to avoid creating invalid block devices. 561 */ 562 size = get_capacity(disk) - from; 563 } 564 565 part = add_partition(disk, p, from, size, state->parts[p].flags, 566 &state->parts[p].info); 567 if (IS_ERR(part)) { 568 if (PTR_ERR(part) != -ENXIO) { 569 printk(KERN_ERR " %s: p%d could not be added: %pe\n", 570 disk->disk_name, p, part); 571 } 572 return true; 573 } 574 575 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 576 (state->parts[p].flags & ADDPART_FLAG_RAID)) 577 md_autodetect_dev(part->bd_dev); 578 579 return true; 580 } 581 582 static int blk_add_partitions(struct gendisk *disk) 583 { 584 struct parsed_partitions *state; 585 int ret = -EAGAIN, p; 586 587 if (!disk_has_partscan(disk)) 588 return 0; 589 590 state = check_partition(disk); 591 if (!state) 592 return 0; 593 if (IS_ERR(state)) { 594 /* 595 * I/O error reading the partition table. If we tried to read 596 * beyond EOD, retry after unlocking the native capacity. 597 */ 598 if (PTR_ERR(state) == -ENOSPC) { 599 printk(KERN_WARNING "%s: partition table beyond EOD, ", 600 disk->disk_name); 601 if (disk_unlock_native_capacity(disk)) 602 return -EAGAIN; 603 } 604 return -EIO; 605 } 606 607 /* 608 * Partitions are not supported on host managed zoned block devices. 609 */ 610 if (bdev_is_zoned(disk->part0)) { 611 pr_warn("%s: ignoring partition table on host managed zoned block device\n", 612 disk->disk_name); 613 ret = 0; 614 goto out_free_state; 615 } 616 617 /* 618 * If we read beyond EOD, try unlocking native capacity even if the 619 * partition table was successfully read as we could be missing some 620 * partitions. 621 */ 622 if (state->access_beyond_eod) { 623 printk(KERN_WARNING 624 "%s: partition table partially beyond EOD, ", 625 disk->disk_name); 626 if (disk_unlock_native_capacity(disk)) 627 goto out_free_state; 628 } 629 630 /* tell userspace that the media / partition table may have changed */ 631 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 632 633 for (p = 1; p < state->limit; p++) 634 if (!blk_add_partition(disk, state, p)) 635 goto out_free_state; 636 637 ret = 0; 638 out_free_state: 639 free_partitions(state); 640 return ret; 641 } 642 643 int bdev_disk_changed(struct gendisk *disk, bool invalidate) 644 { 645 struct block_device *part; 646 unsigned long idx; 647 int ret = 0; 648 649 lockdep_assert_held(&disk->open_mutex); 650 651 if (!disk_live(disk)) 652 return -ENXIO; 653 654 rescan: 655 if (disk->open_partitions) 656 return -EBUSY; 657 sync_blockdev(disk->part0); 658 invalidate_bdev(disk->part0); 659 660 xa_for_each_start(&disk->part_tbl, idx, part, 1) { 661 /* 662 * Remove the block device from the inode hash, so that 663 * it cannot be looked up any more even when openers 664 * still hold references. 665 */ 666 bdev_unhash(part); 667 668 /* 669 * If @disk->open_partitions isn't elevated but there's 670 * still an active holder of that block device things 671 * are broken. 672 */ 673 WARN_ON_ONCE(atomic_read(&part->bd_openers)); 674 invalidate_bdev(part); 675 drop_partition(part); 676 } 677 clear_bit(GD_NEED_PART_SCAN, &disk->state); 678 679 /* 680 * Historically we only set the capacity to zero for devices that 681 * support partitions (independ of actually having partitions created). 682 * Doing that is rather inconsistent, but changing it broke legacy 683 * udisks polling for legacy ide-cdrom devices. Use the crude check 684 * below to get the sane behavior for most device while not breaking 685 * userspace for this particular setup. 686 */ 687 if (invalidate) { 688 if (!(disk->flags & GENHD_FL_NO_PART) || 689 !(disk->flags & GENHD_FL_REMOVABLE)) 690 set_capacity(disk, 0); 691 } 692 693 if (get_capacity(disk)) { 694 ret = blk_add_partitions(disk); 695 if (ret == -EAGAIN) 696 goto rescan; 697 } else if (invalidate) { 698 /* 699 * Tell userspace that the media / partition table may have 700 * changed. 701 */ 702 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 703 } 704 705 return ret; 706 } 707 /* 708 * Only exported for loop and dasd for historic reasons. Don't use in new 709 * code! 710 */ 711 EXPORT_SYMBOL_GPL(bdev_disk_changed); 712 713 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 714 { 715 struct address_space *mapping = state->disk->part0->bd_mapping; 716 struct folio *folio; 717 718 if (n >= get_capacity(state->disk)) { 719 state->access_beyond_eod = true; 720 goto out; 721 } 722 723 folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL); 724 if (IS_ERR(folio)) 725 goto out; 726 727 p->v = folio; 728 return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE); 729 out: 730 p->v = NULL; 731 return NULL; 732 } 733