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/ctype.h> 11 #include <linux/vmalloc.h> 12 #include <linux/raid/detect.h> 13 #include "check.h" 14 15 static int (*const check_part[])(struct parsed_partitions *) = { 16 /* 17 * Probe partition formats with tables at disk address 0 18 * that also have an ADFS boot block at 0xdc0. 19 */ 20 #ifdef CONFIG_ACORN_PARTITION_ICS 21 adfspart_check_ICS, 22 #endif 23 #ifdef CONFIG_ACORN_PARTITION_POWERTEC 24 adfspart_check_POWERTEC, 25 #endif 26 #ifdef CONFIG_ACORN_PARTITION_EESOX 27 adfspart_check_EESOX, 28 #endif 29 30 /* 31 * Now move on to formats that only have partition info at 32 * disk address 0xdc0. Since these may also have stale 33 * PC/BIOS partition tables, they need to come before 34 * the msdos entry. 35 */ 36 #ifdef CONFIG_ACORN_PARTITION_CUMANA 37 adfspart_check_CUMANA, 38 #endif 39 #ifdef CONFIG_ACORN_PARTITION_ADFS 40 adfspart_check_ADFS, 41 #endif 42 43 #ifdef CONFIG_CMDLINE_PARTITION 44 cmdline_partition, 45 #endif 46 #ifdef CONFIG_EFI_PARTITION 47 efi_partition, /* this must come before msdos */ 48 #endif 49 #ifdef CONFIG_SGI_PARTITION 50 sgi_partition, 51 #endif 52 #ifdef CONFIG_LDM_PARTITION 53 ldm_partition, /* this must come before msdos */ 54 #endif 55 #ifdef CONFIG_MSDOS_PARTITION 56 msdos_partition, 57 #endif 58 #ifdef CONFIG_OSF_PARTITION 59 osf_partition, 60 #endif 61 #ifdef CONFIG_SUN_PARTITION 62 sun_partition, 63 #endif 64 #ifdef CONFIG_AMIGA_PARTITION 65 amiga_partition, 66 #endif 67 #ifdef CONFIG_ATARI_PARTITION 68 atari_partition, 69 #endif 70 #ifdef CONFIG_MAC_PARTITION 71 mac_partition, 72 #endif 73 #ifdef CONFIG_ULTRIX_PARTITION 74 ultrix_partition, 75 #endif 76 #ifdef CONFIG_IBM_PARTITION 77 ibm_partition, 78 #endif 79 #ifdef CONFIG_KARMA_PARTITION 80 karma_partition, 81 #endif 82 #ifdef CONFIG_SYSV68_PARTITION 83 sysv68_partition, 84 #endif 85 NULL 86 }; 87 88 static struct parsed_partitions *allocate_partitions(struct gendisk *hd) 89 { 90 struct parsed_partitions *state; 91 int nr = DISK_MAX_PARTS; 92 93 state = kzalloc(sizeof(*state), GFP_KERNEL); 94 if (!state) 95 return NULL; 96 97 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0]))); 98 if (!state->parts) { 99 kfree(state); 100 return NULL; 101 } 102 103 state->limit = nr; 104 105 return state; 106 } 107 108 static void free_partitions(struct parsed_partitions *state) 109 { 110 vfree(state->parts); 111 kfree(state); 112 } 113 114 static struct parsed_partitions *check_partition(struct gendisk *hd) 115 { 116 struct parsed_partitions *state; 117 int i, res, err; 118 119 state = allocate_partitions(hd); 120 if (!state) 121 return NULL; 122 state->pp_buf = (char *)__get_free_page(GFP_KERNEL); 123 if (!state->pp_buf) { 124 free_partitions(state); 125 return NULL; 126 } 127 state->pp_buf[0] = '\0'; 128 129 state->disk = hd; 130 snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_name); 131 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name); 132 if (isdigit(state->name[strlen(state->name)-1])) 133 sprintf(state->name, "p"); 134 135 i = res = err = 0; 136 while (!res && check_part[i]) { 137 memset(state->parts, 0, state->limit * sizeof(state->parts[0])); 138 res = check_part[i++](state); 139 if (res < 0) { 140 /* 141 * We have hit an I/O error which we don't report now. 142 * But record it, and let the others do their job. 143 */ 144 err = res; 145 res = 0; 146 } 147 148 } 149 if (res > 0) { 150 printk(KERN_INFO "%s", state->pp_buf); 151 152 free_page((unsigned long)state->pp_buf); 153 return state; 154 } 155 if (state->access_beyond_eod) 156 err = -ENOSPC; 157 /* 158 * The partition is unrecognized. So report I/O errors if there were any 159 */ 160 if (err) 161 res = err; 162 if (res) { 163 strlcat(state->pp_buf, 164 " unable to read partition table\n", PAGE_SIZE); 165 printk(KERN_INFO "%s", state->pp_buf); 166 } 167 168 free_page((unsigned long)state->pp_buf); 169 free_partitions(state); 170 return ERR_PTR(res); 171 } 172 173 static ssize_t part_partition_show(struct device *dev, 174 struct device_attribute *attr, char *buf) 175 { 176 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno); 177 } 178 179 static ssize_t part_start_show(struct device *dev, 180 struct device_attribute *attr, char *buf) 181 { 182 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect); 183 } 184 185 static ssize_t part_ro_show(struct device *dev, 186 struct device_attribute *attr, char *buf) 187 { 188 return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev))); 189 } 190 191 static ssize_t part_alignment_offset_show(struct device *dev, 192 struct device_attribute *attr, char *buf) 193 { 194 return sprintf(buf, "%u\n", bdev_alignment_offset(dev_to_bdev(dev))); 195 } 196 197 static ssize_t part_discard_alignment_show(struct device *dev, 198 struct device_attribute *attr, char *buf) 199 { 200 return sprintf(buf, "%u\n", bdev_discard_alignment(dev_to_bdev(dev))); 201 } 202 203 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL); 204 static DEVICE_ATTR(start, 0444, part_start_show, NULL); 205 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 206 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL); 207 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL); 208 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL); 209 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 210 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 211 #ifdef CONFIG_FAIL_MAKE_REQUEST 212 static struct device_attribute dev_attr_fail = 213 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 214 #endif 215 216 static struct attribute *part_attrs[] = { 217 &dev_attr_partition.attr, 218 &dev_attr_start.attr, 219 &dev_attr_size.attr, 220 &dev_attr_ro.attr, 221 &dev_attr_alignment_offset.attr, 222 &dev_attr_discard_alignment.attr, 223 &dev_attr_stat.attr, 224 &dev_attr_inflight.attr, 225 #ifdef CONFIG_FAIL_MAKE_REQUEST 226 &dev_attr_fail.attr, 227 #endif 228 NULL 229 }; 230 231 static const struct attribute_group part_attr_group = { 232 .attrs = part_attrs, 233 }; 234 235 static const struct attribute_group *part_attr_groups[] = { 236 &part_attr_group, 237 #ifdef CONFIG_BLK_DEV_IO_TRACE 238 &blk_trace_attr_group, 239 #endif 240 NULL 241 }; 242 243 static void part_release(struct device *dev) 244 { 245 put_disk(dev_to_bdev(dev)->bd_disk); 246 iput(dev_to_bdev(dev)->bd_inode); 247 } 248 249 static int part_uevent(const struct device *dev, struct kobj_uevent_env *env) 250 { 251 const struct block_device *part = dev_to_bdev(dev); 252 253 add_uevent_var(env, "PARTN=%u", part->bd_partno); 254 if (part->bd_meta_info && part->bd_meta_info->volname[0]) 255 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname); 256 return 0; 257 } 258 259 const struct device_type part_type = { 260 .name = "partition", 261 .groups = part_attr_groups, 262 .release = part_release, 263 .uevent = part_uevent, 264 }; 265 266 void drop_partition(struct block_device *part) 267 { 268 lockdep_assert_held(&part->bd_disk->open_mutex); 269 270 xa_erase(&part->bd_disk->part_tbl, part->bd_partno); 271 kobject_put(part->bd_holder_dir); 272 273 device_del(&part->bd_device); 274 put_device(&part->bd_device); 275 } 276 277 static void delete_partition(struct block_device *part) 278 { 279 /* 280 * Remove the block device from the inode hash, so that it cannot be 281 * looked up any more even when openers still hold references. 282 */ 283 remove_inode_hash(part->bd_inode); 284 bdev_mark_dead(part, false); 285 drop_partition(part); 286 } 287 288 static ssize_t whole_disk_show(struct device *dev, 289 struct device_attribute *attr, char *buf) 290 { 291 return 0; 292 } 293 static const DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 294 295 /* 296 * Must be called either with open_mutex held, before a disk can be opened or 297 * after all disk users are gone. 298 */ 299 static struct block_device *add_partition(struct gendisk *disk, int partno, 300 sector_t start, sector_t len, int flags, 301 struct partition_meta_info *info) 302 { 303 dev_t devt = MKDEV(0, 0); 304 struct device *ddev = disk_to_dev(disk); 305 struct device *pdev; 306 struct block_device *bdev; 307 const char *dname; 308 int err; 309 310 lockdep_assert_held(&disk->open_mutex); 311 312 if (partno >= DISK_MAX_PARTS) 313 return ERR_PTR(-EINVAL); 314 315 /* 316 * Partitions are not supported on zoned block devices that are used as 317 * such. 318 */ 319 switch (disk->queue->limits.zoned) { 320 case BLK_ZONED_HM: 321 pr_warn("%s: partitions not supported on host managed zoned block device\n", 322 disk->disk_name); 323 return ERR_PTR(-ENXIO); 324 case BLK_ZONED_HA: 325 pr_info("%s: disabling host aware zoned block device support due to partitions\n", 326 disk->disk_name); 327 disk_set_zoned(disk, BLK_ZONED_NONE); 328 break; 329 case BLK_ZONED_NONE: 330 break; 331 } 332 333 if (xa_load(&disk->part_tbl, partno)) 334 return ERR_PTR(-EBUSY); 335 336 /* ensure we always have a reference to the whole disk */ 337 get_device(disk_to_dev(disk)); 338 339 err = -ENOMEM; 340 bdev = bdev_alloc(disk, partno); 341 if (!bdev) 342 goto out_put_disk; 343 344 bdev->bd_start_sect = start; 345 bdev_set_nr_sectors(bdev, len); 346 347 pdev = &bdev->bd_device; 348 dname = dev_name(ddev); 349 if (isdigit(dname[strlen(dname) - 1])) 350 dev_set_name(pdev, "%sp%d", dname, partno); 351 else 352 dev_set_name(pdev, "%s%d", dname, partno); 353 354 device_initialize(pdev); 355 pdev->class = &block_class; 356 pdev->type = &part_type; 357 pdev->parent = ddev; 358 359 /* in consecutive minor range? */ 360 if (bdev->bd_partno < disk->minors) { 361 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno); 362 } else { 363 err = blk_alloc_ext_minor(); 364 if (err < 0) 365 goto out_put; 366 devt = MKDEV(BLOCK_EXT_MAJOR, err); 367 } 368 pdev->devt = devt; 369 370 if (info) { 371 err = -ENOMEM; 372 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 373 if (!bdev->bd_meta_info) 374 goto out_put; 375 } 376 377 /* delay uevent until 'holders' subdir is created */ 378 dev_set_uevent_suppress(pdev, 1); 379 err = device_add(pdev); 380 if (err) 381 goto out_put; 382 383 err = -ENOMEM; 384 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj); 385 if (!bdev->bd_holder_dir) 386 goto out_del; 387 388 dev_set_uevent_suppress(pdev, 0); 389 if (flags & ADDPART_FLAG_WHOLEDISK) { 390 err = device_create_file(pdev, &dev_attr_whole_disk); 391 if (err) 392 goto out_del; 393 } 394 395 /* everything is up and running, commence */ 396 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL); 397 if (err) 398 goto out_del; 399 bdev_add(bdev, devt); 400 401 /* suppress uevent if the disk suppresses it */ 402 if (!dev_get_uevent_suppress(ddev)) 403 kobject_uevent(&pdev->kobj, KOBJ_ADD); 404 return bdev; 405 406 out_del: 407 kobject_put(bdev->bd_holder_dir); 408 device_del(pdev); 409 out_put: 410 put_device(pdev); 411 return ERR_PTR(err); 412 out_put_disk: 413 put_disk(disk); 414 return ERR_PTR(err); 415 } 416 417 static bool partition_overlaps(struct gendisk *disk, sector_t start, 418 sector_t length, int skip_partno) 419 { 420 struct block_device *part; 421 bool overlap = false; 422 unsigned long idx; 423 424 rcu_read_lock(); 425 xa_for_each_start(&disk->part_tbl, idx, part, 1) { 426 if (part->bd_partno != skip_partno && 427 start < part->bd_start_sect + bdev_nr_sectors(part) && 428 start + length > part->bd_start_sect) { 429 overlap = true; 430 break; 431 } 432 } 433 rcu_read_unlock(); 434 435 return overlap; 436 } 437 438 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, 439 sector_t length) 440 { 441 sector_t capacity = get_capacity(disk), end; 442 struct block_device *part; 443 int ret; 444 445 mutex_lock(&disk->open_mutex); 446 if (check_add_overflow(start, length, &end)) { 447 ret = -EINVAL; 448 goto out; 449 } 450 451 if (start >= capacity || end > capacity) { 452 ret = -EINVAL; 453 goto out; 454 } 455 456 if (!disk_live(disk)) { 457 ret = -ENXIO; 458 goto out; 459 } 460 461 if (disk->flags & GENHD_FL_NO_PART) { 462 ret = -EINVAL; 463 goto out; 464 } 465 466 if (partition_overlaps(disk, start, length, -1)) { 467 ret = -EBUSY; 468 goto out; 469 } 470 471 part = add_partition(disk, partno, start, length, 472 ADDPART_FLAG_NONE, NULL); 473 ret = PTR_ERR_OR_ZERO(part); 474 out: 475 mutex_unlock(&disk->open_mutex); 476 return ret; 477 } 478 479 int bdev_del_partition(struct gendisk *disk, int partno) 480 { 481 struct block_device *part = NULL; 482 int ret = -ENXIO; 483 484 mutex_lock(&disk->open_mutex); 485 part = xa_load(&disk->part_tbl, partno); 486 if (!part) 487 goto out_unlock; 488 489 ret = -EBUSY; 490 if (atomic_read(&part->bd_openers)) 491 goto out_unlock; 492 493 delete_partition(part); 494 ret = 0; 495 out_unlock: 496 mutex_unlock(&disk->open_mutex); 497 return ret; 498 } 499 500 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start, 501 sector_t length) 502 { 503 struct block_device *part = NULL; 504 int ret = -ENXIO; 505 506 mutex_lock(&disk->open_mutex); 507 part = xa_load(&disk->part_tbl, partno); 508 if (!part) 509 goto out_unlock; 510 511 ret = -EINVAL; 512 if (start != part->bd_start_sect) 513 goto out_unlock; 514 515 ret = -EBUSY; 516 if (partition_overlaps(disk, start, length, partno)) 517 goto out_unlock; 518 519 bdev_set_nr_sectors(part, length); 520 521 ret = 0; 522 out_unlock: 523 mutex_unlock(&disk->open_mutex); 524 return ret; 525 } 526 527 static bool disk_unlock_native_capacity(struct gendisk *disk) 528 { 529 if (!disk->fops->unlock_native_capacity || 530 test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) { 531 printk(KERN_CONT "truncated\n"); 532 return false; 533 } 534 535 printk(KERN_CONT "enabling native capacity\n"); 536 disk->fops->unlock_native_capacity(disk); 537 return true; 538 } 539 540 static bool blk_add_partition(struct gendisk *disk, 541 struct parsed_partitions *state, int p) 542 { 543 sector_t size = state->parts[p].size; 544 sector_t from = state->parts[p].from; 545 struct block_device *part; 546 547 if (!size) 548 return true; 549 550 if (from >= get_capacity(disk)) { 551 printk(KERN_WARNING 552 "%s: p%d start %llu is beyond EOD, ", 553 disk->disk_name, p, (unsigned long long) from); 554 if (disk_unlock_native_capacity(disk)) 555 return false; 556 return true; 557 } 558 559 if (from + size > get_capacity(disk)) { 560 printk(KERN_WARNING 561 "%s: p%d size %llu extends beyond EOD, ", 562 disk->disk_name, p, (unsigned long long) size); 563 564 if (disk_unlock_native_capacity(disk)) 565 return false; 566 567 /* 568 * We can not ignore partitions of broken tables created by for 569 * example camera firmware, but we limit them to the end of the 570 * disk to avoid creating invalid block devices. 571 */ 572 size = get_capacity(disk) - from; 573 } 574 575 part = add_partition(disk, p, from, size, state->parts[p].flags, 576 &state->parts[p].info); 577 if (IS_ERR(part)) { 578 if (PTR_ERR(part) != -ENXIO) { 579 printk(KERN_ERR " %s: p%d could not be added: %pe\n", 580 disk->disk_name, p, part); 581 } 582 return true; 583 } 584 585 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 586 (state->parts[p].flags & ADDPART_FLAG_RAID)) 587 md_autodetect_dev(part->bd_dev); 588 589 return true; 590 } 591 592 static int blk_add_partitions(struct gendisk *disk) 593 { 594 struct parsed_partitions *state; 595 int ret = -EAGAIN, p; 596 597 if (!disk_has_partscan(disk)) 598 return 0; 599 600 state = check_partition(disk); 601 if (!state) 602 return 0; 603 if (IS_ERR(state)) { 604 /* 605 * I/O error reading the partition table. If we tried to read 606 * beyond EOD, retry after unlocking the native capacity. 607 */ 608 if (PTR_ERR(state) == -ENOSPC) { 609 printk(KERN_WARNING "%s: partition table beyond EOD, ", 610 disk->disk_name); 611 if (disk_unlock_native_capacity(disk)) 612 return -EAGAIN; 613 } 614 return -EIO; 615 } 616 617 /* 618 * Partitions are not supported on host managed zoned block devices. 619 */ 620 if (disk->queue->limits.zoned == BLK_ZONED_HM) { 621 pr_warn("%s: ignoring partition table on host managed zoned block device\n", 622 disk->disk_name); 623 ret = 0; 624 goto out_free_state; 625 } 626 627 /* 628 * If we read beyond EOD, try unlocking native capacity even if the 629 * partition table was successfully read as we could be missing some 630 * partitions. 631 */ 632 if (state->access_beyond_eod) { 633 printk(KERN_WARNING 634 "%s: partition table partially beyond EOD, ", 635 disk->disk_name); 636 if (disk_unlock_native_capacity(disk)) 637 goto out_free_state; 638 } 639 640 /* tell userspace that the media / partition table may have changed */ 641 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 642 643 for (p = 1; p < state->limit; p++) 644 if (!blk_add_partition(disk, state, p)) 645 goto out_free_state; 646 647 ret = 0; 648 out_free_state: 649 free_partitions(state); 650 return ret; 651 } 652 653 int bdev_disk_changed(struct gendisk *disk, bool invalidate) 654 { 655 struct block_device *part; 656 unsigned long idx; 657 int ret = 0; 658 659 lockdep_assert_held(&disk->open_mutex); 660 661 if (!disk_live(disk)) 662 return -ENXIO; 663 664 rescan: 665 if (disk->open_partitions) 666 return -EBUSY; 667 sync_blockdev(disk->part0); 668 invalidate_bdev(disk->part0); 669 670 xa_for_each_start(&disk->part_tbl, idx, part, 1) 671 delete_partition(part); 672 clear_bit(GD_NEED_PART_SCAN, &disk->state); 673 674 /* 675 * Historically we only set the capacity to zero for devices that 676 * support partitions (independ of actually having partitions created). 677 * Doing that is rather inconsistent, but changing it broke legacy 678 * udisks polling for legacy ide-cdrom devices. Use the crude check 679 * below to get the sane behavior for most device while not breaking 680 * userspace for this particular setup. 681 */ 682 if (invalidate) { 683 if (!(disk->flags & GENHD_FL_NO_PART) || 684 !(disk->flags & GENHD_FL_REMOVABLE)) 685 set_capacity(disk, 0); 686 } 687 688 if (get_capacity(disk)) { 689 ret = blk_add_partitions(disk); 690 if (ret == -EAGAIN) 691 goto rescan; 692 } else if (invalidate) { 693 /* 694 * Tell userspace that the media / partition table may have 695 * changed. 696 */ 697 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 698 } 699 700 return ret; 701 } 702 /* 703 * Only exported for loop and dasd for historic reasons. Don't use in new 704 * code! 705 */ 706 EXPORT_SYMBOL_GPL(bdev_disk_changed); 707 708 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 709 { 710 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping; 711 struct folio *folio; 712 713 if (n >= get_capacity(state->disk)) { 714 state->access_beyond_eod = true; 715 goto out; 716 } 717 718 folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL); 719 if (IS_ERR(folio)) 720 goto out; 721 722 p->v = folio; 723 return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE); 724 out: 725 p->v = NULL; 726 return NULL; 727 } 728