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/genhd.h> 12 #include <linux/vmalloc.h> 13 #include <linux/blktrace_api.h> 14 #include <linux/raid/detect.h> 15 #include "check.h" 16 17 static int (*check_part[])(struct parsed_partitions *) = { 18 /* 19 * Probe partition formats with tables at disk address 0 20 * that also have an ADFS boot block at 0xdc0. 21 */ 22 #ifdef CONFIG_ACORN_PARTITION_ICS 23 adfspart_check_ICS, 24 #endif 25 #ifdef CONFIG_ACORN_PARTITION_POWERTEC 26 adfspart_check_POWERTEC, 27 #endif 28 #ifdef CONFIG_ACORN_PARTITION_EESOX 29 adfspart_check_EESOX, 30 #endif 31 32 /* 33 * Now move on to formats that only have partition info at 34 * disk address 0xdc0. Since these may also have stale 35 * PC/BIOS partition tables, they need to come before 36 * the msdos entry. 37 */ 38 #ifdef CONFIG_ACORN_PARTITION_CUMANA 39 adfspart_check_CUMANA, 40 #endif 41 #ifdef CONFIG_ACORN_PARTITION_ADFS 42 adfspart_check_ADFS, 43 #endif 44 45 #ifdef CONFIG_CMDLINE_PARTITION 46 cmdline_partition, 47 #endif 48 #ifdef CONFIG_EFI_PARTITION 49 efi_partition, /* this must come before msdos */ 50 #endif 51 #ifdef CONFIG_SGI_PARTITION 52 sgi_partition, 53 #endif 54 #ifdef CONFIG_LDM_PARTITION 55 ldm_partition, /* this must come before msdos */ 56 #endif 57 #ifdef CONFIG_MSDOS_PARTITION 58 msdos_partition, 59 #endif 60 #ifdef CONFIG_OSF_PARTITION 61 osf_partition, 62 #endif 63 #ifdef CONFIG_SUN_PARTITION 64 sun_partition, 65 #endif 66 #ifdef CONFIG_AMIGA_PARTITION 67 amiga_partition, 68 #endif 69 #ifdef CONFIG_ATARI_PARTITION 70 atari_partition, 71 #endif 72 #ifdef CONFIG_MAC_PARTITION 73 mac_partition, 74 #endif 75 #ifdef CONFIG_ULTRIX_PARTITION 76 ultrix_partition, 77 #endif 78 #ifdef CONFIG_IBM_PARTITION 79 ibm_partition, 80 #endif 81 #ifdef CONFIG_KARMA_PARTITION 82 karma_partition, 83 #endif 84 #ifdef CONFIG_SYSV68_PARTITION 85 sysv68_partition, 86 #endif 87 NULL 88 }; 89 90 static void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors) 91 { 92 spin_lock(&bdev->bd_size_lock); 93 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); 94 bdev->bd_nr_sectors = sectors; 95 spin_unlock(&bdev->bd_size_lock); 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 { 127 struct parsed_partitions *state; 128 int i, res, err; 129 130 state = allocate_partitions(hd); 131 if (!state) 132 return NULL; 133 state->pp_buf = (char *)__get_free_page(GFP_KERNEL); 134 if (!state->pp_buf) { 135 free_partitions(state); 136 return NULL; 137 } 138 state->pp_buf[0] = '\0'; 139 140 state->disk = hd; 141 snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_name); 142 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name); 143 if (isdigit(state->name[strlen(state->name)-1])) 144 sprintf(state->name, "p"); 145 146 i = res = err = 0; 147 while (!res && check_part[i]) { 148 memset(state->parts, 0, state->limit * sizeof(state->parts[0])); 149 res = check_part[i++](state); 150 if (res < 0) { 151 /* 152 * We have hit an I/O error which we don't report now. 153 * But record it, and let the others do their job. 154 */ 155 err = res; 156 res = 0; 157 } 158 159 } 160 if (res > 0) { 161 printk(KERN_INFO "%s", state->pp_buf); 162 163 free_page((unsigned long)state->pp_buf); 164 return state; 165 } 166 if (state->access_beyond_eod) 167 err = -ENOSPC; 168 /* 169 * The partition is unrecognized. So report I/O errors if there were any 170 */ 171 if (err) 172 res = err; 173 if (res) { 174 strlcat(state->pp_buf, 175 " unable to read partition table\n", PAGE_SIZE); 176 printk(KERN_INFO "%s", state->pp_buf); 177 } 178 179 free_page((unsigned long)state->pp_buf); 180 free_partitions(state); 181 return ERR_PTR(res); 182 } 183 184 static ssize_t part_partition_show(struct device *dev, 185 struct device_attribute *attr, char *buf) 186 { 187 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno); 188 } 189 190 static ssize_t part_start_show(struct device *dev, 191 struct device_attribute *attr, char *buf) 192 { 193 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect); 194 } 195 196 static ssize_t part_ro_show(struct device *dev, 197 struct device_attribute *attr, char *buf) 198 { 199 return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev))); 200 } 201 202 static ssize_t part_alignment_offset_show(struct device *dev, 203 struct device_attribute *attr, char *buf) 204 { 205 struct block_device *bdev = dev_to_bdev(dev); 206 207 return sprintf(buf, "%u\n", 208 queue_limit_alignment_offset(&bdev_get_queue(bdev)->limits, 209 bdev->bd_start_sect)); 210 } 211 212 static ssize_t part_discard_alignment_show(struct device *dev, 213 struct device_attribute *attr, char *buf) 214 { 215 struct block_device *bdev = dev_to_bdev(dev); 216 217 return sprintf(buf, "%u\n", 218 queue_limit_discard_alignment(&bdev_get_queue(bdev)->limits, 219 bdev->bd_start_sect)); 220 } 221 222 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL); 223 static DEVICE_ATTR(start, 0444, part_start_show, NULL); 224 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 225 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL); 226 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL); 227 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL); 228 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 229 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 230 #ifdef CONFIG_FAIL_MAKE_REQUEST 231 static struct device_attribute dev_attr_fail = 232 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 233 #endif 234 235 static struct attribute *part_attrs[] = { 236 &dev_attr_partition.attr, 237 &dev_attr_start.attr, 238 &dev_attr_size.attr, 239 &dev_attr_ro.attr, 240 &dev_attr_alignment_offset.attr, 241 &dev_attr_discard_alignment.attr, 242 &dev_attr_stat.attr, 243 &dev_attr_inflight.attr, 244 #ifdef CONFIG_FAIL_MAKE_REQUEST 245 &dev_attr_fail.attr, 246 #endif 247 NULL 248 }; 249 250 static struct attribute_group part_attr_group = { 251 .attrs = part_attrs, 252 }; 253 254 static const struct attribute_group *part_attr_groups[] = { 255 &part_attr_group, 256 #ifdef CONFIG_BLK_DEV_IO_TRACE 257 &blk_trace_attr_group, 258 #endif 259 NULL 260 }; 261 262 static void part_release(struct device *dev) 263 { 264 put_disk(dev_to_bdev(dev)->bd_disk); 265 iput(dev_to_bdev(dev)->bd_inode); 266 } 267 268 static int part_uevent(struct device *dev, struct kobj_uevent_env *env) 269 { 270 struct block_device *part = dev_to_bdev(dev); 271 272 add_uevent_var(env, "PARTN=%u", part->bd_partno); 273 if (part->bd_meta_info && part->bd_meta_info->volname[0]) 274 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname); 275 return 0; 276 } 277 278 struct device_type part_type = { 279 .name = "partition", 280 .groups = part_attr_groups, 281 .release = part_release, 282 .uevent = part_uevent, 283 }; 284 285 static void delete_partition(struct block_device *part) 286 { 287 lockdep_assert_held(&part->bd_disk->open_mutex); 288 289 fsync_bdev(part); 290 __invalidate_device(part, true); 291 292 xa_erase(&part->bd_disk->part_tbl, part->bd_partno); 293 kobject_put(part->bd_holder_dir); 294 device_del(&part->bd_device); 295 296 /* 297 * Remove the block device from the inode hash, so that it cannot be 298 * looked up any more even when openers still hold references. 299 */ 300 remove_inode_hash(part->bd_inode); 301 302 put_device(&part->bd_device); 303 } 304 305 static ssize_t whole_disk_show(struct device *dev, 306 struct device_attribute *attr, char *buf) 307 { 308 return 0; 309 } 310 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 311 312 /* 313 * Must be called either with open_mutex held, before a disk can be opened or 314 * after all disk users are gone. 315 */ 316 static struct block_device *add_partition(struct gendisk *disk, int partno, 317 sector_t start, sector_t len, int flags, 318 struct partition_meta_info *info) 319 { 320 dev_t devt = MKDEV(0, 0); 321 struct device *ddev = disk_to_dev(disk); 322 struct device *pdev; 323 struct block_device *bdev; 324 const char *dname; 325 int err; 326 327 lockdep_assert_held(&disk->open_mutex); 328 329 if (partno >= disk_max_parts(disk)) 330 return ERR_PTR(-EINVAL); 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 blk_queue_set_zoned(disk, BLK_ZONED_NONE); 345 break; 346 case BLK_ZONED_NONE: 347 break; 348 } 349 350 if (xa_load(&disk->part_tbl, partno)) 351 return ERR_PTR(-EBUSY); 352 353 /* ensure we always have a reference to the whole disk */ 354 get_device(disk_to_dev(disk)); 355 356 err = -ENOMEM; 357 bdev = bdev_alloc(disk, partno); 358 if (!bdev) 359 goto out_put_disk; 360 361 bdev->bd_start_sect = start; 362 bdev_set_nr_sectors(bdev, len); 363 364 pdev = &bdev->bd_device; 365 dname = dev_name(ddev); 366 if (isdigit(dname[strlen(dname) - 1])) 367 dev_set_name(pdev, "%sp%d", dname, partno); 368 else 369 dev_set_name(pdev, "%s%d", dname, partno); 370 371 device_initialize(pdev); 372 pdev->class = &block_class; 373 pdev->type = &part_type; 374 pdev->parent = ddev; 375 376 /* in consecutive minor range? */ 377 if (bdev->bd_partno < disk->minors) { 378 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno); 379 } else { 380 err = blk_alloc_ext_minor(); 381 if (err < 0) 382 goto out_put; 383 devt = MKDEV(BLOCK_EXT_MAJOR, err); 384 } 385 pdev->devt = devt; 386 387 if (info) { 388 err = -ENOMEM; 389 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 390 if (!bdev->bd_meta_info) 391 goto out_put; 392 } 393 394 /* delay uevent until 'holders' subdir is created */ 395 dev_set_uevent_suppress(pdev, 1); 396 err = device_add(pdev); 397 if (err) 398 goto out_put; 399 400 err = -ENOMEM; 401 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj); 402 if (!bdev->bd_holder_dir) 403 goto out_del; 404 405 dev_set_uevent_suppress(pdev, 0); 406 if (flags & ADDPART_FLAG_WHOLEDISK) { 407 err = device_create_file(pdev, &dev_attr_whole_disk); 408 if (err) 409 goto out_del; 410 } 411 412 /* everything is up and running, commence */ 413 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL); 414 if (err) 415 goto out_del; 416 bdev_add(bdev, devt); 417 418 /* suppress uevent if the disk suppresses it */ 419 if (!dev_get_uevent_suppress(ddev)) 420 kobject_uevent(&pdev->kobj, KOBJ_ADD); 421 return bdev; 422 423 out_del: 424 kobject_put(bdev->bd_holder_dir); 425 device_del(pdev); 426 out_put: 427 put_device(pdev); 428 return ERR_PTR(err); 429 out_put_disk: 430 put_disk(disk); 431 return ERR_PTR(err); 432 } 433 434 static bool partition_overlaps(struct gendisk *disk, sector_t start, 435 sector_t length, int skip_partno) 436 { 437 struct block_device *part; 438 bool overlap = false; 439 unsigned long idx; 440 441 rcu_read_lock(); 442 xa_for_each_start(&disk->part_tbl, idx, part, 1) { 443 if (part->bd_partno != skip_partno && 444 start < part->bd_start_sect + bdev_nr_sectors(part) && 445 start + length > part->bd_start_sect) { 446 overlap = true; 447 break; 448 } 449 } 450 rcu_read_unlock(); 451 452 return overlap; 453 } 454 455 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, 456 sector_t length) 457 { 458 struct block_device *part; 459 int ret; 460 461 mutex_lock(&disk->open_mutex); 462 if (!disk_live(disk)) { 463 ret = -ENXIO; 464 goto out; 465 } 466 467 if (partition_overlaps(disk, start, length, -1)) { 468 ret = -EBUSY; 469 goto out; 470 } 471 472 part = add_partition(disk, partno, start, length, 473 ADDPART_FLAG_NONE, NULL); 474 ret = PTR_ERR_OR_ZERO(part); 475 out: 476 mutex_unlock(&disk->open_mutex); 477 return ret; 478 } 479 480 int bdev_del_partition(struct gendisk *disk, int partno) 481 { 482 struct block_device *part = NULL; 483 int ret = -ENXIO; 484 485 mutex_lock(&disk->open_mutex); 486 part = xa_load(&disk->part_tbl, partno); 487 if (!part) 488 goto out_unlock; 489 490 ret = -EBUSY; 491 if (part->bd_openers) 492 goto out_unlock; 493 494 delete_partition(part); 495 ret = 0; 496 out_unlock: 497 mutex_unlock(&disk->open_mutex); 498 return ret; 499 } 500 501 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start, 502 sector_t length) 503 { 504 struct block_device *part = NULL; 505 int ret = -ENXIO; 506 507 mutex_lock(&disk->open_mutex); 508 part = xa_load(&disk->part_tbl, partno); 509 if (!part) 510 goto out_unlock; 511 512 ret = -EINVAL; 513 if (start != part->bd_start_sect) 514 goto out_unlock; 515 516 ret = -EBUSY; 517 if (partition_overlaps(disk, start, length, partno)) 518 goto out_unlock; 519 520 bdev_set_nr_sectors(part, length); 521 522 ret = 0; 523 out_unlock: 524 mutex_unlock(&disk->open_mutex); 525 return ret; 526 } 527 528 static bool disk_unlock_native_capacity(struct gendisk *disk) 529 { 530 const struct block_device_operations *bdops = disk->fops; 531 532 if (bdops->unlock_native_capacity && 533 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) { 534 printk(KERN_CONT "enabling native capacity\n"); 535 bdops->unlock_native_capacity(disk); 536 disk->flags |= GENHD_FL_NATIVE_CAPACITY; 537 return true; 538 } else { 539 printk(KERN_CONT "truncated\n"); 540 return false; 541 } 542 } 543 544 void blk_drop_partitions(struct gendisk *disk) 545 { 546 struct block_device *part; 547 unsigned long idx; 548 549 lockdep_assert_held(&disk->open_mutex); 550 551 xa_for_each_start(&disk->part_tbl, idx, part, 1) 552 delete_partition(part); 553 } 554 555 static bool blk_add_partition(struct gendisk *disk, 556 struct parsed_partitions *state, int p) 557 { 558 sector_t size = state->parts[p].size; 559 sector_t from = state->parts[p].from; 560 struct block_device *part; 561 562 if (!size) 563 return true; 564 565 if (from >= get_capacity(disk)) { 566 printk(KERN_WARNING 567 "%s: p%d start %llu is beyond EOD, ", 568 disk->disk_name, p, (unsigned long long) from); 569 if (disk_unlock_native_capacity(disk)) 570 return false; 571 return true; 572 } 573 574 if (from + size > get_capacity(disk)) { 575 printk(KERN_WARNING 576 "%s: p%d size %llu extends beyond EOD, ", 577 disk->disk_name, p, (unsigned long long) size); 578 579 if (disk_unlock_native_capacity(disk)) 580 return false; 581 582 /* 583 * We can not ignore partitions of broken tables created by for 584 * example camera firmware, but we limit them to the end of the 585 * disk to avoid creating invalid block devices. 586 */ 587 size = get_capacity(disk) - from; 588 } 589 590 part = add_partition(disk, p, from, size, state->parts[p].flags, 591 &state->parts[p].info); 592 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) { 593 printk(KERN_ERR " %s: p%d could not be added: %ld\n", 594 disk->disk_name, p, -PTR_ERR(part)); 595 return true; 596 } 597 598 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 599 (state->parts[p].flags & ADDPART_FLAG_RAID)) 600 md_autodetect_dev(part->bd_dev); 601 602 return true; 603 } 604 605 static int blk_add_partitions(struct gendisk *disk) 606 { 607 struct parsed_partitions *state; 608 int ret = -EAGAIN, p; 609 610 if (!disk_part_scan_enabled(disk)) 611 return 0; 612 613 state = check_partition(disk); 614 if (!state) 615 return 0; 616 if (IS_ERR(state)) { 617 /* 618 * I/O error reading the partition table. If we tried to read 619 * beyond EOD, retry after unlocking the native capacity. 620 */ 621 if (PTR_ERR(state) == -ENOSPC) { 622 printk(KERN_WARNING "%s: partition table beyond EOD, ", 623 disk->disk_name); 624 if (disk_unlock_native_capacity(disk)) 625 return -EAGAIN; 626 } 627 return -EIO; 628 } 629 630 /* 631 * Partitions are not supported on host managed zoned block devices. 632 */ 633 if (disk->queue->limits.zoned == BLK_ZONED_HM) { 634 pr_warn("%s: ignoring partition table on host managed zoned block device\n", 635 disk->disk_name); 636 ret = 0; 637 goto out_free_state; 638 } 639 640 /* 641 * If we read beyond EOD, try unlocking native capacity even if the 642 * partition table was successfully read as we could be missing some 643 * partitions. 644 */ 645 if (state->access_beyond_eod) { 646 printk(KERN_WARNING 647 "%s: partition table partially beyond EOD, ", 648 disk->disk_name); 649 if (disk_unlock_native_capacity(disk)) 650 goto out_free_state; 651 } 652 653 /* tell userspace that the media / partition table may have changed */ 654 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 655 656 for (p = 1; p < state->limit; p++) 657 if (!blk_add_partition(disk, state, p)) 658 goto out_free_state; 659 660 ret = 0; 661 out_free_state: 662 free_partitions(state); 663 return ret; 664 } 665 666 int bdev_disk_changed(struct gendisk *disk, bool invalidate) 667 { 668 int ret = 0; 669 670 lockdep_assert_held(&disk->open_mutex); 671 672 if (!disk_live(disk)) 673 return -ENXIO; 674 675 rescan: 676 if (disk->open_partitions) 677 return -EBUSY; 678 sync_blockdev(disk->part0); 679 invalidate_bdev(disk->part0); 680 blk_drop_partitions(disk); 681 682 clear_bit(GD_NEED_PART_SCAN, &disk->state); 683 684 /* 685 * Historically we only set the capacity to zero for devices that 686 * support partitions (independ of actually having partitions created). 687 * Doing that is rather inconsistent, but changing it broke legacy 688 * udisks polling for legacy ide-cdrom devices. Use the crude check 689 * below to get the sane behavior for most device while not breaking 690 * userspace for this particular setup. 691 */ 692 if (invalidate) { 693 if (disk_part_scan_enabled(disk) || 694 !(disk->flags & GENHD_FL_REMOVABLE)) 695 set_capacity(disk, 0); 696 } 697 698 if (get_capacity(disk)) { 699 ret = blk_add_partitions(disk); 700 if (ret == -EAGAIN) 701 goto rescan; 702 } else if (invalidate) { 703 /* 704 * Tell userspace that the media / partition table may have 705 * changed. 706 */ 707 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 708 } 709 710 return ret; 711 } 712 /* 713 * Only exported for loop and dasd for historic reasons. Don't use in new 714 * code! 715 */ 716 EXPORT_SYMBOL_GPL(bdev_disk_changed); 717 718 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 719 { 720 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping; 721 struct page *page; 722 723 if (n >= get_capacity(state->disk)) { 724 state->access_beyond_eod = true; 725 return NULL; 726 } 727 728 page = read_mapping_page(mapping, 729 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL); 730 if (IS_ERR(page)) 731 goto out; 732 if (PageError(page)) 733 goto out_put_page; 734 735 p->v = page; 736 return (unsigned char *)page_address(page) + 737 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT); 738 out_put_page: 739 put_page(page); 740 out: 741 p->v = NULL; 742 return NULL; 743 } 744