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 unsigned long flags; 92 93 spin_lock_irqsave(&bdev->bd_size_lock, flags); 94 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); 95 spin_unlock_irqrestore(&bdev->bd_size_lock, flags); 96 } 97 98 static struct parsed_partitions *allocate_partitions(struct gendisk *hd) 99 { 100 struct parsed_partitions *state; 101 int nr; 102 103 state = kzalloc(sizeof(*state), GFP_KERNEL); 104 if (!state) 105 return NULL; 106 107 nr = disk_max_parts(hd); 108 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0]))); 109 if (!state->parts) { 110 kfree(state); 111 return NULL; 112 } 113 114 state->limit = nr; 115 116 return state; 117 } 118 119 static void free_partitions(struct parsed_partitions *state) 120 { 121 vfree(state->parts); 122 kfree(state); 123 } 124 125 static struct parsed_partitions *check_partition(struct gendisk *hd, 126 struct block_device *bdev) 127 { 128 struct parsed_partitions *state; 129 int i, res, err; 130 131 state = allocate_partitions(hd); 132 if (!state) 133 return NULL; 134 state->pp_buf = (char *)__get_free_page(GFP_KERNEL); 135 if (!state->pp_buf) { 136 free_partitions(state); 137 return NULL; 138 } 139 state->pp_buf[0] = '\0'; 140 141 state->bdev = bdev; 142 disk_name(hd, 0, state->name); 143 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name); 144 if (isdigit(state->name[strlen(state->name)-1])) 145 sprintf(state->name, "p"); 146 147 i = res = err = 0; 148 while (!res && check_part[i]) { 149 memset(state->parts, 0, state->limit * sizeof(state->parts[0])); 150 res = check_part[i++](state); 151 if (res < 0) { 152 /* 153 * We have hit an I/O error which we don't report now. 154 * But record it, and let the others do their job. 155 */ 156 err = res; 157 res = 0; 158 } 159 160 } 161 if (res > 0) { 162 printk(KERN_INFO "%s", state->pp_buf); 163 164 free_page((unsigned long)state->pp_buf); 165 return state; 166 } 167 if (state->access_beyond_eod) 168 err = -ENOSPC; 169 /* 170 * The partition is unrecognized. So report I/O errors if there were any 171 */ 172 if (err) 173 res = err; 174 if (res) { 175 strlcat(state->pp_buf, 176 " unable to read partition table\n", PAGE_SIZE); 177 printk(KERN_INFO "%s", state->pp_buf); 178 } 179 180 free_page((unsigned long)state->pp_buf); 181 free_partitions(state); 182 return ERR_PTR(res); 183 } 184 185 static ssize_t part_partition_show(struct device *dev, 186 struct device_attribute *attr, char *buf) 187 { 188 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno); 189 } 190 191 static ssize_t part_start_show(struct device *dev, 192 struct device_attribute *attr, char *buf) 193 { 194 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect); 195 } 196 197 static ssize_t part_ro_show(struct device *dev, 198 struct device_attribute *attr, char *buf) 199 { 200 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_read_only); 201 } 202 203 static ssize_t part_alignment_offset_show(struct device *dev, 204 struct device_attribute *attr, char *buf) 205 { 206 struct block_device *bdev = dev_to_bdev(dev); 207 208 return sprintf(buf, "%u\n", 209 queue_limit_alignment_offset(&bdev->bd_disk->queue->limits, 210 bdev->bd_start_sect)); 211 } 212 213 static ssize_t part_discard_alignment_show(struct device *dev, 214 struct device_attribute *attr, char *buf) 215 { 216 struct block_device *bdev = dev_to_bdev(dev); 217 218 return sprintf(buf, "%u\n", 219 queue_limit_discard_alignment(&bdev->bd_disk->queue->limits, 220 bdev->bd_start_sect)); 221 } 222 223 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL); 224 static DEVICE_ATTR(start, 0444, part_start_show, NULL); 225 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 226 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL); 227 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL); 228 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL); 229 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 230 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 231 #ifdef CONFIG_FAIL_MAKE_REQUEST 232 static struct device_attribute dev_attr_fail = 233 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 234 #endif 235 236 static struct attribute *part_attrs[] = { 237 &dev_attr_partition.attr, 238 &dev_attr_start.attr, 239 &dev_attr_size.attr, 240 &dev_attr_ro.attr, 241 &dev_attr_alignment_offset.attr, 242 &dev_attr_discard_alignment.attr, 243 &dev_attr_stat.attr, 244 &dev_attr_inflight.attr, 245 #ifdef CONFIG_FAIL_MAKE_REQUEST 246 &dev_attr_fail.attr, 247 #endif 248 NULL 249 }; 250 251 static struct attribute_group part_attr_group = { 252 .attrs = part_attrs, 253 }; 254 255 static const struct attribute_group *part_attr_groups[] = { 256 &part_attr_group, 257 #ifdef CONFIG_BLK_DEV_IO_TRACE 258 &blk_trace_attr_group, 259 #endif 260 NULL 261 }; 262 263 static void part_release(struct device *dev) 264 { 265 blk_free_devt(dev->devt); 266 bdput(dev_to_bdev(dev)); 267 } 268 269 static int part_uevent(struct device *dev, struct kobj_uevent_env *env) 270 { 271 struct block_device *part = dev_to_bdev(dev); 272 273 add_uevent_var(env, "PARTN=%u", part->bd_partno); 274 if (part->bd_meta_info && part->bd_meta_info->volname[0]) 275 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname); 276 return 0; 277 } 278 279 struct device_type part_type = { 280 .name = "partition", 281 .groups = part_attr_groups, 282 .release = part_release, 283 .uevent = part_uevent, 284 }; 285 286 /* 287 * Must be called either with bd_mutex held, before a disk can be opened or 288 * after all disk users are gone. 289 */ 290 void delete_partition(struct block_device *part) 291 { 292 struct gendisk *disk = part->bd_disk; 293 struct disk_part_tbl *ptbl = 294 rcu_dereference_protected(disk->part_tbl, 1); 295 296 rcu_assign_pointer(ptbl->part[part->bd_partno], NULL); 297 rcu_assign_pointer(ptbl->last_lookup, NULL); 298 299 kobject_put(part->bd_holder_dir); 300 device_del(&part->bd_device); 301 302 /* 303 * Remove the block device from the inode hash, so that it cannot be 304 * looked up any more even when openers still hold references. 305 */ 306 remove_inode_hash(part->bd_inode); 307 308 put_device(&part->bd_device); 309 } 310 311 static ssize_t whole_disk_show(struct device *dev, 312 struct device_attribute *attr, char *buf) 313 { 314 return 0; 315 } 316 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 317 318 /* 319 * Must be called either with bd_mutex held, before a disk can be opened or 320 * after all disk users are gone. 321 */ 322 static struct block_device *add_partition(struct gendisk *disk, int partno, 323 sector_t start, sector_t len, int flags, 324 struct partition_meta_info *info) 325 { 326 dev_t devt = MKDEV(0, 0); 327 struct device *ddev = disk_to_dev(disk); 328 struct device *pdev; 329 struct block_device *bdev; 330 struct disk_part_tbl *ptbl; 331 const char *dname; 332 int err; 333 334 /* 335 * Partitions are not supported on zoned block devices that are used as 336 * such. 337 */ 338 switch (disk->queue->limits.zoned) { 339 case BLK_ZONED_HM: 340 pr_warn("%s: partitions not supported on host managed zoned block device\n", 341 disk->disk_name); 342 return ERR_PTR(-ENXIO); 343 case BLK_ZONED_HA: 344 pr_info("%s: disabling host aware zoned block device support due to partitions\n", 345 disk->disk_name); 346 disk->queue->limits.zoned = BLK_ZONED_NONE; 347 break; 348 case BLK_ZONED_NONE: 349 break; 350 } 351 352 err = disk_expand_part_tbl(disk, partno); 353 if (err) 354 return ERR_PTR(err); 355 ptbl = rcu_dereference_protected(disk->part_tbl, 1); 356 357 if (ptbl->part[partno]) 358 return ERR_PTR(-EBUSY); 359 360 bdev = bdev_alloc(disk, partno); 361 if (!bdev) 362 return ERR_PTR(-ENOMEM); 363 364 bdev->bd_start_sect = start; 365 bdev_set_nr_sectors(bdev, len); 366 bdev->bd_read_only = get_disk_ro(disk); 367 368 if (info) { 369 err = -ENOMEM; 370 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 371 if (!bdev->bd_meta_info) 372 goto out_bdput; 373 } 374 375 pdev = &bdev->bd_device; 376 dname = dev_name(ddev); 377 if (isdigit(dname[strlen(dname) - 1])) 378 dev_set_name(pdev, "%sp%d", dname, partno); 379 else 380 dev_set_name(pdev, "%s%d", dname, partno); 381 382 device_initialize(pdev); 383 pdev->class = &block_class; 384 pdev->type = &part_type; 385 pdev->parent = ddev; 386 387 err = blk_alloc_devt(bdev, &devt); 388 if (err) 389 goto out_put; 390 pdev->devt = devt; 391 392 /* delay uevent until 'holders' subdir is created */ 393 dev_set_uevent_suppress(pdev, 1); 394 err = device_add(pdev); 395 if (err) 396 goto out_put; 397 398 err = -ENOMEM; 399 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj); 400 if (!bdev->bd_holder_dir) 401 goto out_del; 402 403 dev_set_uevent_suppress(pdev, 0); 404 if (flags & ADDPART_FLAG_WHOLEDISK) { 405 err = device_create_file(pdev, &dev_attr_whole_disk); 406 if (err) 407 goto out_del; 408 } 409 410 /* everything is up and running, commence */ 411 bdev_add(bdev, devt); 412 rcu_assign_pointer(ptbl->part[partno], bdev); 413 414 /* suppress uevent if the disk suppresses it */ 415 if (!dev_get_uevent_suppress(ddev)) 416 kobject_uevent(&pdev->kobj, KOBJ_ADD); 417 return bdev; 418 419 out_bdput: 420 bdput(bdev); 421 return ERR_PTR(err); 422 out_del: 423 kobject_put(bdev->bd_holder_dir); 424 device_del(pdev); 425 out_put: 426 put_device(pdev); 427 return ERR_PTR(err); 428 } 429 430 static bool partition_overlaps(struct gendisk *disk, sector_t start, 431 sector_t length, int skip_partno) 432 { 433 struct disk_part_iter piter; 434 struct block_device *part; 435 bool overlap = false; 436 437 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY); 438 while ((part = disk_part_iter_next(&piter))) { 439 if (part->bd_partno == skip_partno || 440 start >= part->bd_start_sect + bdev_nr_sectors(part) || 441 start + length <= part->bd_start_sect) 442 continue; 443 overlap = true; 444 break; 445 } 446 447 disk_part_iter_exit(&piter); 448 return overlap; 449 } 450 451 int bdev_add_partition(struct block_device *bdev, int partno, 452 sector_t start, sector_t length) 453 { 454 struct block_device *part; 455 456 mutex_lock(&bdev->bd_mutex); 457 if (partition_overlaps(bdev->bd_disk, start, length, -1)) { 458 mutex_unlock(&bdev->bd_mutex); 459 return -EBUSY; 460 } 461 462 part = add_partition(bdev->bd_disk, partno, start, length, 463 ADDPART_FLAG_NONE, NULL); 464 mutex_unlock(&bdev->bd_mutex); 465 return PTR_ERR_OR_ZERO(part); 466 } 467 468 int bdev_del_partition(struct block_device *bdev, int partno) 469 { 470 struct block_device *part; 471 int ret; 472 473 part = bdget_disk(bdev->bd_disk, partno); 474 if (!part) 475 return -ENXIO; 476 477 mutex_lock(&part->bd_mutex); 478 mutex_lock_nested(&bdev->bd_mutex, 1); 479 480 ret = -EBUSY; 481 if (part->bd_openers) 482 goto out_unlock; 483 484 sync_blockdev(part); 485 invalidate_bdev(part); 486 487 delete_partition(part); 488 ret = 0; 489 out_unlock: 490 mutex_unlock(&bdev->bd_mutex); 491 mutex_unlock(&part->bd_mutex); 492 bdput(part); 493 return ret; 494 } 495 496 int bdev_resize_partition(struct block_device *bdev, int partno, 497 sector_t start, sector_t length) 498 { 499 struct block_device *part; 500 int ret = 0; 501 502 part = bdget_disk(bdev->bd_disk, partno); 503 if (!part) 504 return -ENXIO; 505 506 mutex_lock(&part->bd_mutex); 507 mutex_lock_nested(&bdev->bd_mutex, 1); 508 ret = -EINVAL; 509 if (start != part->bd_start_sect) 510 goto out_unlock; 511 512 ret = -EBUSY; 513 if (partition_overlaps(bdev->bd_disk, start, length, partno)) 514 goto out_unlock; 515 516 bdev_set_nr_sectors(part, length); 517 518 ret = 0; 519 out_unlock: 520 mutex_unlock(&part->bd_mutex); 521 mutex_unlock(&bdev->bd_mutex); 522 bdput(part); 523 return ret; 524 } 525 526 static bool disk_unlock_native_capacity(struct gendisk *disk) 527 { 528 const struct block_device_operations *bdops = disk->fops; 529 530 if (bdops->unlock_native_capacity && 531 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) { 532 printk(KERN_CONT "enabling native capacity\n"); 533 bdops->unlock_native_capacity(disk); 534 disk->flags |= GENHD_FL_NATIVE_CAPACITY; 535 return true; 536 } else { 537 printk(KERN_CONT "truncated\n"); 538 return false; 539 } 540 } 541 542 int blk_drop_partitions(struct block_device *bdev) 543 { 544 struct disk_part_iter piter; 545 struct block_device *part; 546 547 if (bdev->bd_part_count) 548 return -EBUSY; 549 550 sync_blockdev(bdev); 551 invalidate_bdev(bdev); 552 553 disk_part_iter_init(&piter, bdev->bd_disk, DISK_PITER_INCL_EMPTY); 554 while ((part = disk_part_iter_next(&piter))) 555 delete_partition(part); 556 disk_part_iter_exit(&piter); 557 558 return 0; 559 } 560 #ifdef CONFIG_S390 561 /* for historic reasons in the DASD driver */ 562 EXPORT_SYMBOL_GPL(blk_drop_partitions); 563 #endif 564 565 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev, 566 struct parsed_partitions *state, int p) 567 { 568 sector_t size = state->parts[p].size; 569 sector_t from = state->parts[p].from; 570 struct block_device *part; 571 572 if (!size) 573 return true; 574 575 if (from >= get_capacity(disk)) { 576 printk(KERN_WARNING 577 "%s: p%d start %llu is beyond EOD, ", 578 disk->disk_name, p, (unsigned long long) from); 579 if (disk_unlock_native_capacity(disk)) 580 return false; 581 return true; 582 } 583 584 if (from + size > get_capacity(disk)) { 585 printk(KERN_WARNING 586 "%s: p%d size %llu extends beyond EOD, ", 587 disk->disk_name, p, (unsigned long long) size); 588 589 if (disk_unlock_native_capacity(disk)) 590 return false; 591 592 /* 593 * We can not ignore partitions of broken tables created by for 594 * example camera firmware, but we limit them to the end of the 595 * disk to avoid creating invalid block devices. 596 */ 597 size = get_capacity(disk) - from; 598 } 599 600 part = add_partition(disk, p, from, size, state->parts[p].flags, 601 &state->parts[p].info); 602 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) { 603 printk(KERN_ERR " %s: p%d could not be added: %ld\n", 604 disk->disk_name, p, -PTR_ERR(part)); 605 return true; 606 } 607 608 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 609 (state->parts[p].flags & ADDPART_FLAG_RAID)) 610 md_autodetect_dev(part->bd_dev); 611 612 return true; 613 } 614 615 int blk_add_partitions(struct gendisk *disk, struct block_device *bdev) 616 { 617 struct parsed_partitions *state; 618 int ret = -EAGAIN, p, highest; 619 620 if (!disk_part_scan_enabled(disk)) 621 return 0; 622 623 state = check_partition(disk, bdev); 624 if (!state) 625 return 0; 626 if (IS_ERR(state)) { 627 /* 628 * I/O error reading the partition table. If we tried to read 629 * beyond EOD, retry after unlocking the native capacity. 630 */ 631 if (PTR_ERR(state) == -ENOSPC) { 632 printk(KERN_WARNING "%s: partition table beyond EOD, ", 633 disk->disk_name); 634 if (disk_unlock_native_capacity(disk)) 635 return -EAGAIN; 636 } 637 return -EIO; 638 } 639 640 /* 641 * Partitions are not supported on host managed zoned block devices. 642 */ 643 if (disk->queue->limits.zoned == BLK_ZONED_HM) { 644 pr_warn("%s: ignoring partition table on host managed zoned block device\n", 645 disk->disk_name); 646 ret = 0; 647 goto out_free_state; 648 } 649 650 /* 651 * If we read beyond EOD, try unlocking native capacity even if the 652 * partition table was successfully read as we could be missing some 653 * partitions. 654 */ 655 if (state->access_beyond_eod) { 656 printk(KERN_WARNING 657 "%s: partition table partially beyond EOD, ", 658 disk->disk_name); 659 if (disk_unlock_native_capacity(disk)) 660 goto out_free_state; 661 } 662 663 /* tell userspace that the media / partition table may have changed */ 664 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 665 666 /* 667 * Detect the highest partition number and preallocate disk->part_tbl. 668 * This is an optimization and not strictly necessary. 669 */ 670 for (p = 1, highest = 0; p < state->limit; p++) 671 if (state->parts[p].size) 672 highest = p; 673 disk_expand_part_tbl(disk, highest); 674 675 for (p = 1; p < state->limit; p++) 676 if (!blk_add_partition(disk, bdev, state, p)) 677 goto out_free_state; 678 679 ret = 0; 680 out_free_state: 681 free_partitions(state); 682 return ret; 683 } 684 685 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 686 { 687 struct address_space *mapping = state->bdev->bd_inode->i_mapping; 688 struct page *page; 689 690 if (n >= get_capacity(state->bdev->bd_disk)) { 691 state->access_beyond_eod = true; 692 return NULL; 693 } 694 695 page = read_mapping_page(mapping, 696 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL); 697 if (IS_ERR(page)) 698 goto out; 699 if (PageError(page)) 700 goto out_put_page; 701 702 p->v = page; 703 return (unsigned char *)page_address(page) + 704 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT); 705 out_put_page: 706 put_page(page); 707 out: 708 p->v = NULL; 709 return NULL; 710 } 711