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->bdev = hd->part0; 139 disk_name(hd, 0, state->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 if (MAJOR(dev->devt) == BLOCK_EXT_MAJOR) 263 blk_free_ext_minor(MINOR(dev->devt)); 264 bdput(dev_to_bdev(dev)); 265 } 266 267 static int part_uevent(struct device *dev, struct kobj_uevent_env *env) 268 { 269 struct block_device *part = dev_to_bdev(dev); 270 271 add_uevent_var(env, "PARTN=%u", part->bd_partno); 272 if (part->bd_meta_info && part->bd_meta_info->volname[0]) 273 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname); 274 return 0; 275 } 276 277 struct device_type part_type = { 278 .name = "partition", 279 .groups = part_attr_groups, 280 .release = part_release, 281 .uevent = part_uevent, 282 }; 283 284 /* 285 * Must be called either with open_mutex held, before a disk can be opened or 286 * after all disk users are gone. 287 */ 288 static void delete_partition(struct block_device *part) 289 { 290 fsync_bdev(part); 291 __invalidate_device(part, true); 292 293 xa_erase(&part->bd_disk->part_tbl, part->bd_partno); 294 kobject_put(part->bd_holder_dir); 295 device_del(&part->bd_device); 296 297 /* 298 * Remove the block device from the inode hash, so that it cannot be 299 * looked up any more even when openers still hold references. 300 */ 301 remove_inode_hash(part->bd_inode); 302 303 put_device(&part->bd_device); 304 } 305 306 static ssize_t whole_disk_show(struct device *dev, 307 struct device_attribute *attr, char *buf) 308 { 309 return 0; 310 } 311 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 312 313 /* 314 * Must be called either with open_mutex held, before a disk can be opened or 315 * after all disk users are gone. 316 */ 317 static struct block_device *add_partition(struct gendisk *disk, int partno, 318 sector_t start, sector_t len, int flags, 319 struct partition_meta_info *info) 320 { 321 dev_t devt = MKDEV(0, 0); 322 struct device *ddev = disk_to_dev(disk); 323 struct device *pdev; 324 struct block_device *bdev; 325 const char *dname; 326 int err; 327 328 lockdep_assert_held(&disk->open_mutex); 329 330 if (partno >= disk_max_parts(disk)) 331 return ERR_PTR(-EINVAL); 332 333 /* 334 * Partitions are not supported on zoned block devices that are used as 335 * such. 336 */ 337 switch (disk->queue->limits.zoned) { 338 case BLK_ZONED_HM: 339 pr_warn("%s: partitions not supported on host managed zoned block device\n", 340 disk->disk_name); 341 return ERR_PTR(-ENXIO); 342 case BLK_ZONED_HA: 343 pr_info("%s: disabling host aware zoned block device support due to partitions\n", 344 disk->disk_name); 345 blk_queue_set_zoned(disk, BLK_ZONED_NONE); 346 break; 347 case BLK_ZONED_NONE: 348 break; 349 } 350 351 if (xa_load(&disk->part_tbl, partno)) 352 return ERR_PTR(-EBUSY); 353 354 bdev = bdev_alloc(disk, partno); 355 if (!bdev) 356 return ERR_PTR(-ENOMEM); 357 358 bdev->bd_start_sect = start; 359 bdev_set_nr_sectors(bdev, len); 360 361 if (info) { 362 err = -ENOMEM; 363 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 364 if (!bdev->bd_meta_info) 365 goto out_bdput; 366 } 367 368 pdev = &bdev->bd_device; 369 dname = dev_name(ddev); 370 if (isdigit(dname[strlen(dname) - 1])) 371 dev_set_name(pdev, "%sp%d", dname, partno); 372 else 373 dev_set_name(pdev, "%s%d", dname, partno); 374 375 device_initialize(pdev); 376 pdev->class = &block_class; 377 pdev->type = &part_type; 378 pdev->parent = ddev; 379 380 /* in consecutive minor range? */ 381 if (bdev->bd_partno < disk->minors) { 382 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno); 383 } else { 384 err = blk_alloc_ext_minor(); 385 if (err < 0) 386 goto out_put; 387 devt = MKDEV(BLOCK_EXT_MAJOR, err); 388 } 389 pdev->devt = devt; 390 391 /* delay uevent until 'holders' subdir is created */ 392 dev_set_uevent_suppress(pdev, 1); 393 err = device_add(pdev); 394 if (err) 395 goto out_put; 396 397 err = -ENOMEM; 398 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj); 399 if (!bdev->bd_holder_dir) 400 goto out_del; 401 402 dev_set_uevent_suppress(pdev, 0); 403 if (flags & ADDPART_FLAG_WHOLEDISK) { 404 err = device_create_file(pdev, &dev_attr_whole_disk); 405 if (err) 406 goto out_del; 407 } 408 409 /* everything is up and running, commence */ 410 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL); 411 if (err) 412 goto out_del; 413 bdev_add(bdev, devt); 414 415 /* suppress uevent if the disk suppresses it */ 416 if (!dev_get_uevent_suppress(ddev)) 417 kobject_uevent(&pdev->kobj, KOBJ_ADD); 418 return bdev; 419 420 out_bdput: 421 bdput(bdev); 422 return ERR_PTR(err); 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 } 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 block_device *bdev, int partno, 453 sector_t start, sector_t length) 454 { 455 struct block_device *part; 456 457 mutex_lock(&bdev->bd_disk->open_mutex); 458 if (partition_overlaps(bdev->bd_disk, start, length, -1)) { 459 mutex_unlock(&bdev->bd_disk->open_mutex); 460 return -EBUSY; 461 } 462 463 part = add_partition(bdev->bd_disk, partno, start, length, 464 ADDPART_FLAG_NONE, NULL); 465 mutex_unlock(&bdev->bd_disk->open_mutex); 466 return PTR_ERR_OR_ZERO(part); 467 } 468 469 int bdev_del_partition(struct block_device *bdev, int partno) 470 { 471 struct block_device *part = NULL; 472 int ret = -ENXIO; 473 474 mutex_lock(&bdev->bd_disk->open_mutex); 475 part = xa_load(&bdev->bd_disk->part_tbl, partno); 476 if (!part) 477 goto out_unlock; 478 479 ret = -EBUSY; 480 if (part->bd_openers) 481 goto out_unlock; 482 483 delete_partition(part); 484 ret = 0; 485 out_unlock: 486 mutex_unlock(&bdev->bd_disk->open_mutex); 487 return ret; 488 } 489 490 int bdev_resize_partition(struct block_device *bdev, int partno, 491 sector_t start, sector_t length) 492 { 493 struct block_device *part = NULL; 494 int ret = -ENXIO; 495 496 mutex_lock(&bdev->bd_disk->open_mutex); 497 part = xa_load(&bdev->bd_disk->part_tbl, partno); 498 if (!part) 499 goto out_unlock; 500 501 ret = -EINVAL; 502 if (start != part->bd_start_sect) 503 goto out_unlock; 504 505 ret = -EBUSY; 506 if (partition_overlaps(bdev->bd_disk, start, length, partno)) 507 goto out_unlock; 508 509 bdev_set_nr_sectors(part, length); 510 511 ret = 0; 512 out_unlock: 513 mutex_unlock(&bdev->bd_disk->open_mutex); 514 return ret; 515 } 516 517 static bool disk_unlock_native_capacity(struct gendisk *disk) 518 { 519 const struct block_device_operations *bdops = disk->fops; 520 521 if (bdops->unlock_native_capacity && 522 !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) { 523 printk(KERN_CONT "enabling native capacity\n"); 524 bdops->unlock_native_capacity(disk); 525 disk->flags |= GENHD_FL_NATIVE_CAPACITY; 526 return true; 527 } else { 528 printk(KERN_CONT "truncated\n"); 529 return false; 530 } 531 } 532 533 void blk_drop_partitions(struct gendisk *disk) 534 { 535 struct block_device *part; 536 unsigned long idx; 537 538 lockdep_assert_held(&disk->open_mutex); 539 540 xa_for_each_start(&disk->part_tbl, idx, part, 1) { 541 if (!bdgrab(part)) 542 continue; 543 delete_partition(part); 544 bdput(part); 545 } 546 } 547 548 static bool blk_add_partition(struct gendisk *disk, 549 struct parsed_partitions *state, int p) 550 { 551 sector_t size = state->parts[p].size; 552 sector_t from = state->parts[p].from; 553 struct block_device *part; 554 555 if (!size) 556 return true; 557 558 if (from >= get_capacity(disk)) { 559 printk(KERN_WARNING 560 "%s: p%d start %llu is beyond EOD, ", 561 disk->disk_name, p, (unsigned long long) from); 562 if (disk_unlock_native_capacity(disk)) 563 return false; 564 return true; 565 } 566 567 if (from + size > get_capacity(disk)) { 568 printk(KERN_WARNING 569 "%s: p%d size %llu extends beyond EOD, ", 570 disk->disk_name, p, (unsigned long long) size); 571 572 if (disk_unlock_native_capacity(disk)) 573 return false; 574 575 /* 576 * We can not ignore partitions of broken tables created by for 577 * example camera firmware, but we limit them to the end of the 578 * disk to avoid creating invalid block devices. 579 */ 580 size = get_capacity(disk) - from; 581 } 582 583 part = add_partition(disk, p, from, size, state->parts[p].flags, 584 &state->parts[p].info); 585 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) { 586 printk(KERN_ERR " %s: p%d could not be added: %ld\n", 587 disk->disk_name, p, -PTR_ERR(part)); 588 return true; 589 } 590 591 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 592 (state->parts[p].flags & ADDPART_FLAG_RAID)) 593 md_autodetect_dev(part->bd_dev); 594 595 return true; 596 } 597 598 static int blk_add_partitions(struct gendisk *disk) 599 { 600 struct parsed_partitions *state; 601 int ret = -EAGAIN, p; 602 603 if (!disk_part_scan_enabled(disk)) 604 return 0; 605 606 state = check_partition(disk); 607 if (!state) 608 return 0; 609 if (IS_ERR(state)) { 610 /* 611 * I/O error reading the partition table. If we tried to read 612 * beyond EOD, retry after unlocking the native capacity. 613 */ 614 if (PTR_ERR(state) == -ENOSPC) { 615 printk(KERN_WARNING "%s: partition table beyond EOD, ", 616 disk->disk_name); 617 if (disk_unlock_native_capacity(disk)) 618 return -EAGAIN; 619 } 620 return -EIO; 621 } 622 623 /* 624 * Partitions are not supported on host managed zoned block devices. 625 */ 626 if (disk->queue->limits.zoned == BLK_ZONED_HM) { 627 pr_warn("%s: ignoring partition table on host managed zoned block device\n", 628 disk->disk_name); 629 ret = 0; 630 goto out_free_state; 631 } 632 633 /* 634 * If we read beyond EOD, try unlocking native capacity even if the 635 * partition table was successfully read as we could be missing some 636 * partitions. 637 */ 638 if (state->access_beyond_eod) { 639 printk(KERN_WARNING 640 "%s: partition table partially beyond EOD, ", 641 disk->disk_name); 642 if (disk_unlock_native_capacity(disk)) 643 goto out_free_state; 644 } 645 646 /* tell userspace that the media / partition table may have changed */ 647 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 648 649 for (p = 1; p < state->limit; p++) 650 if (!blk_add_partition(disk, state, p)) 651 goto out_free_state; 652 653 ret = 0; 654 out_free_state: 655 free_partitions(state); 656 return ret; 657 } 658 659 int bdev_disk_changed(struct gendisk *disk, bool invalidate) 660 { 661 int ret = 0; 662 663 lockdep_assert_held(&disk->open_mutex); 664 665 if (!(disk->flags & GENHD_FL_UP)) 666 return -ENXIO; 667 668 rescan: 669 if (disk->open_partitions) 670 return -EBUSY; 671 sync_blockdev(disk->part0); 672 invalidate_bdev(disk->part0); 673 blk_drop_partitions(disk); 674 675 clear_bit(GD_NEED_PART_SCAN, &disk->state); 676 677 /* 678 * Historically we only set the capacity to zero for devices that 679 * support partitions (independ of actually having partitions created). 680 * Doing that is rather inconsistent, but changing it broke legacy 681 * udisks polling for legacy ide-cdrom devices. Use the crude check 682 * below to get the sane behavior for most device while not breaking 683 * userspace for this particular setup. 684 */ 685 if (invalidate) { 686 if (disk_part_scan_enabled(disk) || 687 !(disk->flags & GENHD_FL_REMOVABLE)) 688 set_capacity(disk, 0); 689 } 690 691 if (get_capacity(disk)) { 692 ret = blk_add_partitions(disk); 693 if (ret == -EAGAIN) 694 goto rescan; 695 } else if (invalidate) { 696 /* 697 * Tell userspace that the media / partition table may have 698 * changed. 699 */ 700 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 701 } 702 703 return ret; 704 } 705 /* 706 * Only exported for loop and dasd for historic reasons. Don't use in new 707 * code! 708 */ 709 EXPORT_SYMBOL_GPL(bdev_disk_changed); 710 711 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 712 { 713 struct address_space *mapping = state->bdev->bd_inode->i_mapping; 714 struct page *page; 715 716 if (n >= get_capacity(state->bdev->bd_disk)) { 717 state->access_beyond_eod = true; 718 return NULL; 719 } 720 721 page = read_mapping_page(mapping, 722 (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL); 723 if (IS_ERR(page)) 724 goto out; 725 if (PageError(page)) 726 goto out_put_page; 727 728 p->v = page; 729 return (unsigned char *)page_address(page) + 730 ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT); 731 out_put_page: 732 put_page(page); 733 out: 734 p->v = NULL; 735 return NULL; 736 } 737