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); 284 285 fsync_bdev(part); 286 __invalidate_device(part, true); 287 288 drop_partition(part); 289 } 290 291 static ssize_t whole_disk_show(struct device *dev, 292 struct device_attribute *attr, char *buf) 293 { 294 return 0; 295 } 296 static const DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 297 298 /* 299 * Must be called either with open_mutex held, before a disk can be opened or 300 * after all disk users are gone. 301 */ 302 static struct block_device *add_partition(struct gendisk *disk, int partno, 303 sector_t start, sector_t len, int flags, 304 struct partition_meta_info *info) 305 { 306 dev_t devt = MKDEV(0, 0); 307 struct device *ddev = disk_to_dev(disk); 308 struct device *pdev; 309 struct block_device *bdev; 310 const char *dname; 311 int err; 312 313 lockdep_assert_held(&disk->open_mutex); 314 315 if (partno >= DISK_MAX_PARTS) 316 return ERR_PTR(-EINVAL); 317 318 /* 319 * Partitions are not supported on zoned block devices that are used as 320 * such. 321 */ 322 switch (disk->queue->limits.zoned) { 323 case BLK_ZONED_HM: 324 pr_warn("%s: partitions not supported on host managed zoned block device\n", 325 disk->disk_name); 326 return ERR_PTR(-ENXIO); 327 case BLK_ZONED_HA: 328 pr_info("%s: disabling host aware zoned block device support due to partitions\n", 329 disk->disk_name); 330 disk_set_zoned(disk, BLK_ZONED_NONE); 331 break; 332 case BLK_ZONED_NONE: 333 break; 334 } 335 336 if (xa_load(&disk->part_tbl, partno)) 337 return ERR_PTR(-EBUSY); 338 339 /* ensure we always have a reference to the whole disk */ 340 get_device(disk_to_dev(disk)); 341 342 err = -ENOMEM; 343 bdev = bdev_alloc(disk, partno); 344 if (!bdev) 345 goto out_put_disk; 346 347 bdev->bd_start_sect = start; 348 bdev_set_nr_sectors(bdev, len); 349 350 pdev = &bdev->bd_device; 351 dname = dev_name(ddev); 352 if (isdigit(dname[strlen(dname) - 1])) 353 dev_set_name(pdev, "%sp%d", dname, partno); 354 else 355 dev_set_name(pdev, "%s%d", dname, partno); 356 357 device_initialize(pdev); 358 pdev->class = &block_class; 359 pdev->type = &part_type; 360 pdev->parent = ddev; 361 362 /* in consecutive minor range? */ 363 if (bdev->bd_partno < disk->minors) { 364 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno); 365 } else { 366 err = blk_alloc_ext_minor(); 367 if (err < 0) 368 goto out_put; 369 devt = MKDEV(BLOCK_EXT_MAJOR, err); 370 } 371 pdev->devt = devt; 372 373 if (info) { 374 err = -ENOMEM; 375 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 376 if (!bdev->bd_meta_info) 377 goto out_put; 378 } 379 380 /* delay uevent until 'holders' subdir is created */ 381 dev_set_uevent_suppress(pdev, 1); 382 err = device_add(pdev); 383 if (err) 384 goto out_put; 385 386 err = -ENOMEM; 387 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj); 388 if (!bdev->bd_holder_dir) 389 goto out_del; 390 391 dev_set_uevent_suppress(pdev, 0); 392 if (flags & ADDPART_FLAG_WHOLEDISK) { 393 err = device_create_file(pdev, &dev_attr_whole_disk); 394 if (err) 395 goto out_del; 396 } 397 398 /* everything is up and running, commence */ 399 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL); 400 if (err) 401 goto out_del; 402 bdev_add(bdev, devt); 403 404 /* suppress uevent if the disk suppresses it */ 405 if (!dev_get_uevent_suppress(ddev)) 406 kobject_uevent(&pdev->kobj, KOBJ_ADD); 407 return bdev; 408 409 out_del: 410 kobject_put(bdev->bd_holder_dir); 411 device_del(pdev); 412 out_put: 413 put_device(pdev); 414 return ERR_PTR(err); 415 out_put_disk: 416 put_disk(disk); 417 return ERR_PTR(err); 418 } 419 420 static bool partition_overlaps(struct gendisk *disk, sector_t start, 421 sector_t length, int skip_partno) 422 { 423 struct block_device *part; 424 bool overlap = false; 425 unsigned long idx; 426 427 rcu_read_lock(); 428 xa_for_each_start(&disk->part_tbl, idx, part, 1) { 429 if (part->bd_partno != skip_partno && 430 start < part->bd_start_sect + bdev_nr_sectors(part) && 431 start + length > part->bd_start_sect) { 432 overlap = true; 433 break; 434 } 435 } 436 rcu_read_unlock(); 437 438 return overlap; 439 } 440 441 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, 442 sector_t length) 443 { 444 sector_t capacity = get_capacity(disk), end; 445 struct block_device *part; 446 int ret; 447 448 mutex_lock(&disk->open_mutex); 449 if (check_add_overflow(start, length, &end)) { 450 ret = -EINVAL; 451 goto out; 452 } 453 454 if (start >= capacity || end > capacity) { 455 ret = -EINVAL; 456 goto out; 457 } 458 459 if (!disk_live(disk)) { 460 ret = -ENXIO; 461 goto out; 462 } 463 464 if (partition_overlaps(disk, start, length, -1)) { 465 ret = -EBUSY; 466 goto out; 467 } 468 469 part = add_partition(disk, partno, start, length, 470 ADDPART_FLAG_NONE, NULL); 471 ret = PTR_ERR_OR_ZERO(part); 472 out: 473 mutex_unlock(&disk->open_mutex); 474 return ret; 475 } 476 477 int bdev_del_partition(struct gendisk *disk, int partno) 478 { 479 struct block_device *part = NULL; 480 int ret = -ENXIO; 481 482 mutex_lock(&disk->open_mutex); 483 part = xa_load(&disk->part_tbl, partno); 484 if (!part) 485 goto out_unlock; 486 487 ret = -EBUSY; 488 if (atomic_read(&part->bd_openers)) 489 goto out_unlock; 490 491 delete_partition(part); 492 ret = 0; 493 out_unlock: 494 mutex_unlock(&disk->open_mutex); 495 return ret; 496 } 497 498 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start, 499 sector_t length) 500 { 501 struct block_device *part = NULL; 502 int ret = -ENXIO; 503 504 mutex_lock(&disk->open_mutex); 505 part = xa_load(&disk->part_tbl, partno); 506 if (!part) 507 goto out_unlock; 508 509 ret = -EINVAL; 510 if (start != part->bd_start_sect) 511 goto out_unlock; 512 513 ret = -EBUSY; 514 if (partition_overlaps(disk, start, length, partno)) 515 goto out_unlock; 516 517 bdev_set_nr_sectors(part, length); 518 519 ret = 0; 520 out_unlock: 521 mutex_unlock(&disk->open_mutex); 522 return ret; 523 } 524 525 static bool disk_unlock_native_capacity(struct gendisk *disk) 526 { 527 if (!disk->fops->unlock_native_capacity || 528 test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) { 529 printk(KERN_CONT "truncated\n"); 530 return false; 531 } 532 533 printk(KERN_CONT "enabling native capacity\n"); 534 disk->fops->unlock_native_capacity(disk); 535 return true; 536 } 537 538 static bool blk_add_partition(struct gendisk *disk, 539 struct parsed_partitions *state, int p) 540 { 541 sector_t size = state->parts[p].size; 542 sector_t from = state->parts[p].from; 543 struct block_device *part; 544 545 if (!size) 546 return true; 547 548 if (from >= get_capacity(disk)) { 549 printk(KERN_WARNING 550 "%s: p%d start %llu is beyond EOD, ", 551 disk->disk_name, p, (unsigned long long) from); 552 if (disk_unlock_native_capacity(disk)) 553 return false; 554 return true; 555 } 556 557 if (from + size > get_capacity(disk)) { 558 printk(KERN_WARNING 559 "%s: p%d size %llu extends beyond EOD, ", 560 disk->disk_name, p, (unsigned long long) size); 561 562 if (disk_unlock_native_capacity(disk)) 563 return false; 564 565 /* 566 * We can not ignore partitions of broken tables created by for 567 * example camera firmware, but we limit them to the end of the 568 * disk to avoid creating invalid block devices. 569 */ 570 size = get_capacity(disk) - from; 571 } 572 573 part = add_partition(disk, p, from, size, state->parts[p].flags, 574 &state->parts[p].info); 575 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) { 576 printk(KERN_ERR " %s: p%d could not be added: %ld\n", 577 disk->disk_name, p, -PTR_ERR(part)); 578 return true; 579 } 580 581 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 582 (state->parts[p].flags & ADDPART_FLAG_RAID)) 583 md_autodetect_dev(part->bd_dev); 584 585 return true; 586 } 587 588 static int blk_add_partitions(struct gendisk *disk) 589 { 590 struct parsed_partitions *state; 591 int ret = -EAGAIN, p; 592 593 if (disk->flags & GENHD_FL_NO_PART) 594 return 0; 595 596 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state)) 597 return 0; 598 599 state = check_partition(disk); 600 if (!state) 601 return 0; 602 if (IS_ERR(state)) { 603 /* 604 * I/O error reading the partition table. If we tried to read 605 * beyond EOD, retry after unlocking the native capacity. 606 */ 607 if (PTR_ERR(state) == -ENOSPC) { 608 printk(KERN_WARNING "%s: partition table beyond EOD, ", 609 disk->disk_name); 610 if (disk_unlock_native_capacity(disk)) 611 return -EAGAIN; 612 } 613 return -EIO; 614 } 615 616 /* 617 * Partitions are not supported on host managed zoned block devices. 618 */ 619 if (disk->queue->limits.zoned == BLK_ZONED_HM) { 620 pr_warn("%s: ignoring partition table on host managed zoned block device\n", 621 disk->disk_name); 622 ret = 0; 623 goto out_free_state; 624 } 625 626 /* 627 * If we read beyond EOD, try unlocking native capacity even if the 628 * partition table was successfully read as we could be missing some 629 * partitions. 630 */ 631 if (state->access_beyond_eod) { 632 printk(KERN_WARNING 633 "%s: partition table partially beyond EOD, ", 634 disk->disk_name); 635 if (disk_unlock_native_capacity(disk)) 636 goto out_free_state; 637 } 638 639 /* tell userspace that the media / partition table may have changed */ 640 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 641 642 for (p = 1; p < state->limit; p++) 643 if (!blk_add_partition(disk, state, p)) 644 goto out_free_state; 645 646 ret = 0; 647 out_free_state: 648 free_partitions(state); 649 return ret; 650 } 651 652 int bdev_disk_changed(struct gendisk *disk, bool invalidate) 653 { 654 struct block_device *part; 655 unsigned long idx; 656 int ret = 0; 657 658 lockdep_assert_held(&disk->open_mutex); 659 660 if (!disk_live(disk)) 661 return -ENXIO; 662 663 rescan: 664 if (disk->open_partitions) 665 return -EBUSY; 666 sync_blockdev(disk->part0); 667 invalidate_bdev(disk->part0); 668 669 xa_for_each_start(&disk->part_tbl, idx, part, 1) 670 delete_partition(part); 671 clear_bit(GD_NEED_PART_SCAN, &disk->state); 672 673 /* 674 * Historically we only set the capacity to zero for devices that 675 * support partitions (independ of actually having partitions created). 676 * Doing that is rather inconsistent, but changing it broke legacy 677 * udisks polling for legacy ide-cdrom devices. Use the crude check 678 * below to get the sane behavior for most device while not breaking 679 * userspace for this particular setup. 680 */ 681 if (invalidate) { 682 if (!(disk->flags & GENHD_FL_NO_PART) || 683 !(disk->flags & GENHD_FL_REMOVABLE)) 684 set_capacity(disk, 0); 685 } 686 687 if (get_capacity(disk)) { 688 ret = blk_add_partitions(disk); 689 if (ret == -EAGAIN) 690 goto rescan; 691 } else if (invalidate) { 692 /* 693 * Tell userspace that the media / partition table may have 694 * changed. 695 */ 696 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 697 } 698 699 return ret; 700 } 701 /* 702 * Only exported for loop and dasd for historic reasons. Don't use in new 703 * code! 704 */ 705 EXPORT_SYMBOL_GPL(bdev_disk_changed); 706 707 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 708 { 709 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping; 710 struct folio *folio; 711 712 if (n >= get_capacity(state->disk)) { 713 state->access_beyond_eod = true; 714 goto out; 715 } 716 717 folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL); 718 if (IS_ERR(folio)) 719 goto out; 720 721 p->v = folio; 722 return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE); 723 out: 724 p->v = NULL; 725 return NULL; 726 } 727