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 spin_lock(&bdev->bd_size_lock); 92 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); 93 spin_unlock(&bdev->bd_size_lock); 94 } 95 96 static struct parsed_partitions *allocate_partitions(struct gendisk *hd) 97 { 98 struct parsed_partitions *state; 99 int nr; 100 101 state = kzalloc(sizeof(*state), GFP_KERNEL); 102 if (!state) 103 return NULL; 104 105 nr = disk_max_parts(hd); 106 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0]))); 107 if (!state->parts) { 108 kfree(state); 109 return NULL; 110 } 111 112 state->limit = nr; 113 114 return state; 115 } 116 117 static void free_partitions(struct parsed_partitions *state) 118 { 119 vfree(state->parts); 120 kfree(state); 121 } 122 123 static struct parsed_partitions *check_partition(struct gendisk *hd) 124 { 125 struct parsed_partitions *state; 126 int i, res, err; 127 128 state = allocate_partitions(hd); 129 if (!state) 130 return NULL; 131 state->pp_buf = (char *)__get_free_page(GFP_KERNEL); 132 if (!state->pp_buf) { 133 free_partitions(state); 134 return NULL; 135 } 136 state->pp_buf[0] = '\0'; 137 138 state->disk = hd; 139 snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_name); 140 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name); 141 if (isdigit(state->name[strlen(state->name)-1])) 142 sprintf(state->name, "p"); 143 144 i = res = err = 0; 145 while (!res && check_part[i]) { 146 memset(state->parts, 0, state->limit * sizeof(state->parts[0])); 147 res = check_part[i++](state); 148 if (res < 0) { 149 /* 150 * We have hit an I/O error which we don't report now. 151 * But record it, and let the others do their job. 152 */ 153 err = res; 154 res = 0; 155 } 156 157 } 158 if (res > 0) { 159 printk(KERN_INFO "%s", state->pp_buf); 160 161 free_page((unsigned long)state->pp_buf); 162 return state; 163 } 164 if (state->access_beyond_eod) 165 err = -ENOSPC; 166 /* 167 * The partition is unrecognized. So report I/O errors if there were any 168 */ 169 if (err) 170 res = err; 171 if (res) { 172 strlcat(state->pp_buf, 173 " unable to read partition table\n", PAGE_SIZE); 174 printk(KERN_INFO "%s", state->pp_buf); 175 } 176 177 free_page((unsigned long)state->pp_buf); 178 free_partitions(state); 179 return ERR_PTR(res); 180 } 181 182 static ssize_t part_partition_show(struct device *dev, 183 struct device_attribute *attr, char *buf) 184 { 185 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno); 186 } 187 188 static ssize_t part_start_show(struct device *dev, 189 struct device_attribute *attr, char *buf) 190 { 191 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect); 192 } 193 194 static ssize_t part_ro_show(struct device *dev, 195 struct device_attribute *attr, char *buf) 196 { 197 return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev))); 198 } 199 200 static ssize_t part_alignment_offset_show(struct device *dev, 201 struct device_attribute *attr, char *buf) 202 { 203 struct block_device *bdev = dev_to_bdev(dev); 204 205 return sprintf(buf, "%u\n", 206 queue_limit_alignment_offset(&bdev->bd_disk->queue->limits, 207 bdev->bd_start_sect)); 208 } 209 210 static ssize_t part_discard_alignment_show(struct device *dev, 211 struct device_attribute *attr, char *buf) 212 { 213 struct block_device *bdev = dev_to_bdev(dev); 214 215 return sprintf(buf, "%u\n", 216 queue_limit_discard_alignment(&bdev->bd_disk->queue->limits, 217 bdev->bd_start_sect)); 218 } 219 220 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL); 221 static DEVICE_ATTR(start, 0444, part_start_show, NULL); 222 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 223 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL); 224 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL); 225 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL); 226 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 227 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 228 #ifdef CONFIG_FAIL_MAKE_REQUEST 229 static struct device_attribute dev_attr_fail = 230 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 231 #endif 232 233 static struct attribute *part_attrs[] = { 234 &dev_attr_partition.attr, 235 &dev_attr_start.attr, 236 &dev_attr_size.attr, 237 &dev_attr_ro.attr, 238 &dev_attr_alignment_offset.attr, 239 &dev_attr_discard_alignment.attr, 240 &dev_attr_stat.attr, 241 &dev_attr_inflight.attr, 242 #ifdef CONFIG_FAIL_MAKE_REQUEST 243 &dev_attr_fail.attr, 244 #endif 245 NULL 246 }; 247 248 static struct attribute_group part_attr_group = { 249 .attrs = part_attrs, 250 }; 251 252 static const struct attribute_group *part_attr_groups[] = { 253 &part_attr_group, 254 #ifdef CONFIG_BLK_DEV_IO_TRACE 255 &blk_trace_attr_group, 256 #endif 257 NULL 258 }; 259 260 static void part_release(struct device *dev) 261 { 262 put_disk(dev_to_bdev(dev)->bd_disk); 263 iput(dev_to_bdev(dev)->bd_inode); 264 } 265 266 static int part_uevent(struct device *dev, struct kobj_uevent_env *env) 267 { 268 struct block_device *part = dev_to_bdev(dev); 269 270 add_uevent_var(env, "PARTN=%u", part->bd_partno); 271 if (part->bd_meta_info && part->bd_meta_info->volname[0]) 272 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname); 273 return 0; 274 } 275 276 struct device_type part_type = { 277 .name = "partition", 278 .groups = part_attr_groups, 279 .release = part_release, 280 .uevent = part_uevent, 281 }; 282 283 static void delete_partition(struct block_device *part) 284 { 285 lockdep_assert_held(&part->bd_disk->open_mutex); 286 287 fsync_bdev(part); 288 __invalidate_device(part, true); 289 290 xa_erase(&part->bd_disk->part_tbl, part->bd_partno); 291 kobject_put(part->bd_holder_dir); 292 device_del(&part->bd_device); 293 294 /* 295 * Remove the block device from the inode hash, so that it cannot be 296 * looked up any more even when openers still hold references. 297 */ 298 remove_inode_hash(part->bd_inode); 299 300 put_device(&part->bd_device); 301 } 302 303 static ssize_t whole_disk_show(struct device *dev, 304 struct device_attribute *attr, char *buf) 305 { 306 return 0; 307 } 308 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 309 310 /* 311 * Must be called either with open_mutex held, before a disk can be opened or 312 * after all disk users are gone. 313 */ 314 static struct block_device *add_partition(struct gendisk *disk, int partno, 315 sector_t start, sector_t len, int flags, 316 struct partition_meta_info *info) 317 { 318 dev_t devt = MKDEV(0, 0); 319 struct device *ddev = disk_to_dev(disk); 320 struct device *pdev; 321 struct block_device *bdev; 322 const char *dname; 323 int err; 324 325 lockdep_assert_held(&disk->open_mutex); 326 327 if (partno >= disk_max_parts(disk)) 328 return ERR_PTR(-EINVAL); 329 330 /* 331 * Partitions are not supported on zoned block devices that are used as 332 * such. 333 */ 334 switch (disk->queue->limits.zoned) { 335 case BLK_ZONED_HM: 336 pr_warn("%s: partitions not supported on host managed zoned block device\n", 337 disk->disk_name); 338 return ERR_PTR(-ENXIO); 339 case BLK_ZONED_HA: 340 pr_info("%s: disabling host aware zoned block device support due to partitions\n", 341 disk->disk_name); 342 blk_queue_set_zoned(disk, BLK_ZONED_NONE); 343 break; 344 case BLK_ZONED_NONE: 345 break; 346 } 347 348 if (xa_load(&disk->part_tbl, partno)) 349 return ERR_PTR(-EBUSY); 350 351 /* ensure we always have a reference to the whole disk */ 352 get_device(disk_to_dev(disk)); 353 354 err = -ENOMEM; 355 bdev = bdev_alloc(disk, partno); 356 if (!bdev) 357 goto out_put_disk; 358 359 bdev->bd_start_sect = start; 360 bdev_set_nr_sectors(bdev, len); 361 362 pdev = &bdev->bd_device; 363 dname = dev_name(ddev); 364 if (isdigit(dname[strlen(dname) - 1])) 365 dev_set_name(pdev, "%sp%d", dname, partno); 366 else 367 dev_set_name(pdev, "%s%d", dname, partno); 368 369 device_initialize(pdev); 370 pdev->class = &block_class; 371 pdev->type = &part_type; 372 pdev->parent = ddev; 373 374 /* in consecutive minor range? */ 375 if (bdev->bd_partno < disk->minors) { 376 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno); 377 } else { 378 err = blk_alloc_ext_minor(); 379 if (err < 0) 380 goto out_put; 381 devt = MKDEV(BLOCK_EXT_MAJOR, err); 382 } 383 pdev->devt = devt; 384 385 if (info) { 386 err = -ENOMEM; 387 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 388 if (!bdev->bd_meta_info) 389 goto out_put; 390 } 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 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL); 412 if (err) 413 goto out_del; 414 bdev_add(bdev, devt); 415 416 /* suppress uevent if the disk suppresses it */ 417 if (!dev_get_uevent_suppress(ddev)) 418 kobject_uevent(&pdev->kobj, KOBJ_ADD); 419 return bdev; 420 421 out_del: 422 kobject_put(bdev->bd_holder_dir); 423 device_del(pdev); 424 out_put: 425 put_device(pdev); 426 out_put_disk: 427 put_disk(disk); 428 return ERR_PTR(err); 429 } 430 431 static bool partition_overlaps(struct gendisk *disk, sector_t start, 432 sector_t length, int skip_partno) 433 { 434 struct block_device *part; 435 bool overlap = false; 436 unsigned long idx; 437 438 rcu_read_lock(); 439 xa_for_each_start(&disk->part_tbl, idx, part, 1) { 440 if (part->bd_partno != skip_partno && 441 start < part->bd_start_sect + bdev_nr_sectors(part) && 442 start + length > part->bd_start_sect) { 443 overlap = true; 444 break; 445 } 446 } 447 rcu_read_unlock(); 448 449 return overlap; 450 } 451 452 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, 453 sector_t length) 454 { 455 struct block_device *part; 456 int ret; 457 458 mutex_lock(&disk->open_mutex); 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 (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 const struct block_device_operations *bdops = disk->fops; 528 529 if (bdops->unlock_native_capacity && 530 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) { 531 printk(KERN_CONT "enabling native capacity\n"); 532 bdops->unlock_native_capacity(disk); 533 disk->flags |= GENHD_FL_NATIVE_CAPACITY; 534 return true; 535 } else { 536 printk(KERN_CONT "truncated\n"); 537 return false; 538 } 539 } 540 541 void blk_drop_partitions(struct gendisk *disk) 542 { 543 struct block_device *part; 544 unsigned long idx; 545 546 lockdep_assert_held(&disk->open_mutex); 547 548 xa_for_each_start(&disk->part_tbl, idx, part, 1) 549 delete_partition(part); 550 } 551 552 static bool blk_add_partition(struct gendisk *disk, 553 struct parsed_partitions *state, int p) 554 { 555 sector_t size = state->parts[p].size; 556 sector_t from = state->parts[p].from; 557 struct block_device *part; 558 559 if (!size) 560 return true; 561 562 if (from >= get_capacity(disk)) { 563 printk(KERN_WARNING 564 "%s: p%d start %llu is beyond EOD, ", 565 disk->disk_name, p, (unsigned long long) from); 566 if (disk_unlock_native_capacity(disk)) 567 return false; 568 return true; 569 } 570 571 if (from + size > get_capacity(disk)) { 572 printk(KERN_WARNING 573 "%s: p%d size %llu extends beyond EOD, ", 574 disk->disk_name, p, (unsigned long long) size); 575 576 if (disk_unlock_native_capacity(disk)) 577 return false; 578 579 /* 580 * We can not ignore partitions of broken tables created by for 581 * example camera firmware, but we limit them to the end of the 582 * disk to avoid creating invalid block devices. 583 */ 584 size = get_capacity(disk) - from; 585 } 586 587 part = add_partition(disk, p, from, size, state->parts[p].flags, 588 &state->parts[p].info); 589 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) { 590 printk(KERN_ERR " %s: p%d could not be added: %ld\n", 591 disk->disk_name, p, -PTR_ERR(part)); 592 return true; 593 } 594 595 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 596 (state->parts[p].flags & ADDPART_FLAG_RAID)) 597 md_autodetect_dev(part->bd_dev); 598 599 return true; 600 } 601 602 static int blk_add_partitions(struct gendisk *disk) 603 { 604 struct parsed_partitions *state; 605 int ret = -EAGAIN, p; 606 607 if (!disk_part_scan_enabled(disk)) 608 return 0; 609 610 state = check_partition(disk); 611 if (!state) 612 return 0; 613 if (IS_ERR(state)) { 614 /* 615 * I/O error reading the partition table. If we tried to read 616 * beyond EOD, retry after unlocking the native capacity. 617 */ 618 if (PTR_ERR(state) == -ENOSPC) { 619 printk(KERN_WARNING "%s: partition table beyond EOD, ", 620 disk->disk_name); 621 if (disk_unlock_native_capacity(disk)) 622 return -EAGAIN; 623 } 624 return -EIO; 625 } 626 627 /* 628 * Partitions are not supported on host managed zoned block devices. 629 */ 630 if (disk->queue->limits.zoned == BLK_ZONED_HM) { 631 pr_warn("%s: ignoring partition table on host managed zoned block device\n", 632 disk->disk_name); 633 ret = 0; 634 goto out_free_state; 635 } 636 637 /* 638 * If we read beyond EOD, try unlocking native capacity even if the 639 * partition table was successfully read as we could be missing some 640 * partitions. 641 */ 642 if (state->access_beyond_eod) { 643 printk(KERN_WARNING 644 "%s: partition table partially beyond EOD, ", 645 disk->disk_name); 646 if (disk_unlock_native_capacity(disk)) 647 goto out_free_state; 648 } 649 650 /* tell userspace that the media / partition table may have changed */ 651 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 652 653 for (p = 1; p < state->limit; p++) 654 if (!blk_add_partition(disk, state, p)) 655 goto out_free_state; 656 657 ret = 0; 658 out_free_state: 659 free_partitions(state); 660 return ret; 661 } 662 663 int bdev_disk_changed(struct gendisk *disk, bool invalidate) 664 { 665 int ret = 0; 666 667 lockdep_assert_held(&disk->open_mutex); 668 669 if (!disk_live(disk)) 670 return -ENXIO; 671 672 rescan: 673 if (disk->open_partitions) 674 return -EBUSY; 675 sync_blockdev(disk->part0); 676 invalidate_bdev(disk->part0); 677 blk_drop_partitions(disk); 678 679 clear_bit(GD_NEED_PART_SCAN, &disk->state); 680 681 /* 682 * Historically we only set the capacity to zero for devices that 683 * support partitions (independ of actually having partitions created). 684 * Doing that is rather inconsistent, but changing it broke legacy 685 * udisks polling for legacy ide-cdrom devices. Use the crude check 686 * below to get the sane behavior for most device while not breaking 687 * userspace for this particular setup. 688 */ 689 if (invalidate) { 690 if (disk_part_scan_enabled(disk) || 691 !(disk->flags & GENHD_FL_REMOVABLE)) 692 set_capacity(disk, 0); 693 } 694 695 if (get_capacity(disk)) { 696 ret = blk_add_partitions(disk); 697 if (ret == -EAGAIN) 698 goto rescan; 699 } else if (invalidate) { 700 /* 701 * Tell userspace that the media / partition table may have 702 * changed. 703 */ 704 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 705 } 706 707 return ret; 708 } 709 /* 710 * Only exported for loop and dasd for historic reasons. Don't use in new 711 * code! 712 */ 713 EXPORT_SYMBOL_GPL(bdev_disk_changed); 714 715 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 716 { 717 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping; 718 struct page *page; 719 720 if (n >= get_capacity(state->disk)) { 721 state->access_beyond_eod = true; 722 return NULL; 723 } 724 725 page = read_mapping_page(mapping, 726 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL); 727 if (IS_ERR(page)) 728 goto out; 729 if (PageError(page)) 730 goto out_put_page; 731 732 p->v = page; 733 return (unsigned char *)page_address(page) + 734 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT); 735 out_put_page: 736 put_page(page); 737 out: 738 p->v = NULL; 739 return NULL; 740 } 741