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 (*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 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors) 89 { 90 spin_lock(&bdev->bd_size_lock); 91 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); 92 bdev->bd_nr_sectors = sectors; 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 = DISK_MAX_PARTS; 100 101 state = kzalloc(sizeof(*state), GFP_KERNEL); 102 if (!state) 103 return NULL; 104 105 state->parts = vzalloc(array_size(nr, sizeof(state->parts[0]))); 106 if (!state->parts) { 107 kfree(state); 108 return NULL; 109 } 110 111 state->limit = nr; 112 113 return state; 114 } 115 116 static void free_partitions(struct parsed_partitions *state) 117 { 118 vfree(state->parts); 119 kfree(state); 120 } 121 122 static struct parsed_partitions *check_partition(struct gendisk *hd) 123 { 124 struct parsed_partitions *state; 125 int i, res, err; 126 127 state = allocate_partitions(hd); 128 if (!state) 129 return NULL; 130 state->pp_buf = (char *)__get_free_page(GFP_KERNEL); 131 if (!state->pp_buf) { 132 free_partitions(state); 133 return NULL; 134 } 135 state->pp_buf[0] = '\0'; 136 137 state->disk = hd; 138 snprintf(state->name, BDEVNAME_SIZE, "%s", hd->disk_name); 139 snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name); 140 if (isdigit(state->name[strlen(state->name)-1])) 141 sprintf(state->name, "p"); 142 143 i = res = err = 0; 144 while (!res && check_part[i]) { 145 memset(state->parts, 0, state->limit * sizeof(state->parts[0])); 146 res = check_part[i++](state); 147 if (res < 0) { 148 /* 149 * We have hit an I/O error which we don't report now. 150 * But record it, and let the others do their job. 151 */ 152 err = res; 153 res = 0; 154 } 155 156 } 157 if (res > 0) { 158 printk(KERN_INFO "%s", state->pp_buf); 159 160 free_page((unsigned long)state->pp_buf); 161 return state; 162 } 163 if (state->access_beyond_eod) 164 err = -ENOSPC; 165 /* 166 * The partition is unrecognized. So report I/O errors if there were any 167 */ 168 if (err) 169 res = err; 170 if (res) { 171 strlcat(state->pp_buf, 172 " unable to read partition table\n", PAGE_SIZE); 173 printk(KERN_INFO "%s", state->pp_buf); 174 } 175 176 free_page((unsigned long)state->pp_buf); 177 free_partitions(state); 178 return ERR_PTR(res); 179 } 180 181 static ssize_t part_partition_show(struct device *dev, 182 struct device_attribute *attr, char *buf) 183 { 184 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_partno); 185 } 186 187 static ssize_t part_start_show(struct device *dev, 188 struct device_attribute *attr, char *buf) 189 { 190 return sprintf(buf, "%llu\n", dev_to_bdev(dev)->bd_start_sect); 191 } 192 193 static ssize_t part_ro_show(struct device *dev, 194 struct device_attribute *attr, char *buf) 195 { 196 return sprintf(buf, "%d\n", bdev_read_only(dev_to_bdev(dev))); 197 } 198 199 static ssize_t part_alignment_offset_show(struct device *dev, 200 struct device_attribute *attr, char *buf) 201 { 202 return sprintf(buf, "%u\n", bdev_alignment_offset(dev_to_bdev(dev))); 203 } 204 205 static ssize_t part_discard_alignment_show(struct device *dev, 206 struct device_attribute *attr, char *buf) 207 { 208 return sprintf(buf, "%u\n", bdev_discard_alignment(dev_to_bdev(dev))); 209 } 210 211 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL); 212 static DEVICE_ATTR(start, 0444, part_start_show, NULL); 213 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 214 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL); 215 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL); 216 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL); 217 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 218 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 219 #ifdef CONFIG_FAIL_MAKE_REQUEST 220 static struct device_attribute dev_attr_fail = 221 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 222 #endif 223 224 static struct attribute *part_attrs[] = { 225 &dev_attr_partition.attr, 226 &dev_attr_start.attr, 227 &dev_attr_size.attr, 228 &dev_attr_ro.attr, 229 &dev_attr_alignment_offset.attr, 230 &dev_attr_discard_alignment.attr, 231 &dev_attr_stat.attr, 232 &dev_attr_inflight.attr, 233 #ifdef CONFIG_FAIL_MAKE_REQUEST 234 &dev_attr_fail.attr, 235 #endif 236 NULL 237 }; 238 239 static struct attribute_group part_attr_group = { 240 .attrs = part_attrs, 241 }; 242 243 static const struct attribute_group *part_attr_groups[] = { 244 &part_attr_group, 245 #ifdef CONFIG_BLK_DEV_IO_TRACE 246 &blk_trace_attr_group, 247 #endif 248 NULL 249 }; 250 251 static void part_release(struct device *dev) 252 { 253 put_disk(dev_to_bdev(dev)->bd_disk); 254 iput(dev_to_bdev(dev)->bd_inode); 255 } 256 257 static int part_uevent(struct device *dev, struct kobj_uevent_env *env) 258 { 259 struct block_device *part = dev_to_bdev(dev); 260 261 add_uevent_var(env, "PARTN=%u", part->bd_partno); 262 if (part->bd_meta_info && part->bd_meta_info->volname[0]) 263 add_uevent_var(env, "PARTNAME=%s", part->bd_meta_info->volname); 264 return 0; 265 } 266 267 struct device_type part_type = { 268 .name = "partition", 269 .groups = part_attr_groups, 270 .release = part_release, 271 .uevent = part_uevent, 272 }; 273 274 static void delete_partition(struct block_device *part) 275 { 276 lockdep_assert_held(&part->bd_disk->open_mutex); 277 278 fsync_bdev(part); 279 __invalidate_device(part, true); 280 281 xa_erase(&part->bd_disk->part_tbl, part->bd_partno); 282 kobject_put(part->bd_holder_dir); 283 device_del(&part->bd_device); 284 285 /* 286 * Remove the block device from the inode hash, so that it cannot be 287 * looked up any more even when openers still hold references. 288 */ 289 remove_inode_hash(part->bd_inode); 290 291 put_device(&part->bd_device); 292 } 293 294 static ssize_t whole_disk_show(struct device *dev, 295 struct device_attribute *attr, char *buf) 296 { 297 return 0; 298 } 299 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL); 300 301 /* 302 * Must be called either with open_mutex held, before a disk can be opened or 303 * after all disk users are gone. 304 */ 305 static struct block_device *add_partition(struct gendisk *disk, int partno, 306 sector_t start, sector_t len, int flags, 307 struct partition_meta_info *info) 308 { 309 dev_t devt = MKDEV(0, 0); 310 struct device *ddev = disk_to_dev(disk); 311 struct device *pdev; 312 struct block_device *bdev; 313 const char *dname; 314 int err; 315 316 lockdep_assert_held(&disk->open_mutex); 317 318 if (partno >= DISK_MAX_PARTS) 319 return ERR_PTR(-EINVAL); 320 321 /* 322 * Partitions are not supported on zoned block devices that are used as 323 * such. 324 */ 325 switch (disk->queue->limits.zoned) { 326 case BLK_ZONED_HM: 327 pr_warn("%s: partitions not supported on host managed zoned block device\n", 328 disk->disk_name); 329 return ERR_PTR(-ENXIO); 330 case BLK_ZONED_HA: 331 pr_info("%s: disabling host aware zoned block device support due to partitions\n", 332 disk->disk_name); 333 disk_set_zoned(disk, BLK_ZONED_NONE); 334 break; 335 case BLK_ZONED_NONE: 336 break; 337 } 338 339 if (xa_load(&disk->part_tbl, partno)) 340 return ERR_PTR(-EBUSY); 341 342 /* ensure we always have a reference to the whole disk */ 343 get_device(disk_to_dev(disk)); 344 345 err = -ENOMEM; 346 bdev = bdev_alloc(disk, partno); 347 if (!bdev) 348 goto out_put_disk; 349 350 bdev->bd_start_sect = start; 351 bdev_set_nr_sectors(bdev, len); 352 353 pdev = &bdev->bd_device; 354 dname = dev_name(ddev); 355 if (isdigit(dname[strlen(dname) - 1])) 356 dev_set_name(pdev, "%sp%d", dname, partno); 357 else 358 dev_set_name(pdev, "%s%d", dname, partno); 359 360 device_initialize(pdev); 361 pdev->class = &block_class; 362 pdev->type = &part_type; 363 pdev->parent = ddev; 364 365 /* in consecutive minor range? */ 366 if (bdev->bd_partno < disk->minors) { 367 devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno); 368 } else { 369 err = blk_alloc_ext_minor(); 370 if (err < 0) 371 goto out_put; 372 devt = MKDEV(BLOCK_EXT_MAJOR, err); 373 } 374 pdev->devt = devt; 375 376 if (info) { 377 err = -ENOMEM; 378 bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL); 379 if (!bdev->bd_meta_info) 380 goto out_put; 381 } 382 383 /* delay uevent until 'holders' subdir is created */ 384 dev_set_uevent_suppress(pdev, 1); 385 err = device_add(pdev); 386 if (err) 387 goto out_put; 388 389 err = -ENOMEM; 390 bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj); 391 if (!bdev->bd_holder_dir) 392 goto out_del; 393 394 dev_set_uevent_suppress(pdev, 0); 395 if (flags & ADDPART_FLAG_WHOLEDISK) { 396 err = device_create_file(pdev, &dev_attr_whole_disk); 397 if (err) 398 goto out_del; 399 } 400 401 /* everything is up and running, commence */ 402 err = xa_insert(&disk->part_tbl, partno, bdev, GFP_KERNEL); 403 if (err) 404 goto out_del; 405 bdev_add(bdev, devt); 406 407 /* suppress uevent if the disk suppresses it */ 408 if (!dev_get_uevent_suppress(ddev)) 409 kobject_uevent(&pdev->kobj, KOBJ_ADD); 410 return bdev; 411 412 out_del: 413 kobject_put(bdev->bd_holder_dir); 414 device_del(pdev); 415 out_put: 416 put_device(pdev); 417 return ERR_PTR(err); 418 out_put_disk: 419 put_disk(disk); 420 return ERR_PTR(err); 421 } 422 423 static bool partition_overlaps(struct gendisk *disk, sector_t start, 424 sector_t length, int skip_partno) 425 { 426 struct block_device *part; 427 bool overlap = false; 428 unsigned long idx; 429 430 rcu_read_lock(); 431 xa_for_each_start(&disk->part_tbl, idx, part, 1) { 432 if (part->bd_partno != skip_partno && 433 start < part->bd_start_sect + bdev_nr_sectors(part) && 434 start + length > part->bd_start_sect) { 435 overlap = true; 436 break; 437 } 438 } 439 rcu_read_unlock(); 440 441 return overlap; 442 } 443 444 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, 445 sector_t length) 446 { 447 struct block_device *part; 448 int ret; 449 450 mutex_lock(&disk->open_mutex); 451 if (!disk_live(disk)) { 452 ret = -ENXIO; 453 goto out; 454 } 455 456 if (partition_overlaps(disk, start, length, -1)) { 457 ret = -EBUSY; 458 goto out; 459 } 460 461 part = add_partition(disk, partno, start, length, 462 ADDPART_FLAG_NONE, NULL); 463 ret = PTR_ERR_OR_ZERO(part); 464 out: 465 mutex_unlock(&disk->open_mutex); 466 return ret; 467 } 468 469 int bdev_del_partition(struct gendisk *disk, int partno) 470 { 471 struct block_device *part = NULL; 472 int ret = -ENXIO; 473 474 mutex_lock(&disk->open_mutex); 475 part = xa_load(&disk->part_tbl, partno); 476 if (!part) 477 goto out_unlock; 478 479 ret = -EBUSY; 480 if (atomic_read(&part->bd_openers)) 481 goto out_unlock; 482 483 delete_partition(part); 484 ret = 0; 485 out_unlock: 486 mutex_unlock(&disk->open_mutex); 487 return ret; 488 } 489 490 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start, 491 sector_t length) 492 { 493 struct block_device *part = NULL; 494 int ret = -ENXIO; 495 496 mutex_lock(&disk->open_mutex); 497 part = xa_load(&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(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(&disk->open_mutex); 514 return ret; 515 } 516 517 static bool disk_unlock_native_capacity(struct gendisk *disk) 518 { 519 if (!disk->fops->unlock_native_capacity || 520 test_and_set_bit(GD_NATIVE_CAPACITY, &disk->state)) { 521 printk(KERN_CONT "truncated\n"); 522 return false; 523 } 524 525 printk(KERN_CONT "enabling native capacity\n"); 526 disk->fops->unlock_native_capacity(disk); 527 return true; 528 } 529 530 void blk_drop_partitions(struct gendisk *disk) 531 { 532 struct block_device *part; 533 unsigned long idx; 534 535 lockdep_assert_held(&disk->open_mutex); 536 537 xa_for_each_start(&disk->part_tbl, idx, part, 1) 538 delete_partition(part); 539 } 540 541 static bool blk_add_partition(struct gendisk *disk, 542 struct parsed_partitions *state, int p) 543 { 544 sector_t size = state->parts[p].size; 545 sector_t from = state->parts[p].from; 546 struct block_device *part; 547 548 if (!size) 549 return true; 550 551 if (from >= get_capacity(disk)) { 552 printk(KERN_WARNING 553 "%s: p%d start %llu is beyond EOD, ", 554 disk->disk_name, p, (unsigned long long) from); 555 if (disk_unlock_native_capacity(disk)) 556 return false; 557 return true; 558 } 559 560 if (from + size > get_capacity(disk)) { 561 printk(KERN_WARNING 562 "%s: p%d size %llu extends beyond EOD, ", 563 disk->disk_name, p, (unsigned long long) size); 564 565 if (disk_unlock_native_capacity(disk)) 566 return false; 567 568 /* 569 * We can not ignore partitions of broken tables created by for 570 * example camera firmware, but we limit them to the end of the 571 * disk to avoid creating invalid block devices. 572 */ 573 size = get_capacity(disk) - from; 574 } 575 576 part = add_partition(disk, p, from, size, state->parts[p].flags, 577 &state->parts[p].info); 578 if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) { 579 printk(KERN_ERR " %s: p%d could not be added: %ld\n", 580 disk->disk_name, p, -PTR_ERR(part)); 581 return true; 582 } 583 584 if (IS_BUILTIN(CONFIG_BLK_DEV_MD) && 585 (state->parts[p].flags & ADDPART_FLAG_RAID)) 586 md_autodetect_dev(part->bd_dev); 587 588 return true; 589 } 590 591 static int blk_add_partitions(struct gendisk *disk) 592 { 593 struct parsed_partitions *state; 594 int ret = -EAGAIN, p; 595 596 if (disk->flags & GENHD_FL_NO_PART) 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 int ret = 0; 655 656 lockdep_assert_held(&disk->open_mutex); 657 658 if (!disk_live(disk)) 659 return -ENXIO; 660 661 rescan: 662 if (disk->open_partitions) 663 return -EBUSY; 664 sync_blockdev(disk->part0); 665 invalidate_bdev(disk->part0); 666 blk_drop_partitions(disk); 667 668 clear_bit(GD_NEED_PART_SCAN, &disk->state); 669 670 /* 671 * Historically we only set the capacity to zero for devices that 672 * support partitions (independ of actually having partitions created). 673 * Doing that is rather inconsistent, but changing it broke legacy 674 * udisks polling for legacy ide-cdrom devices. Use the crude check 675 * below to get the sane behavior for most device while not breaking 676 * userspace for this particular setup. 677 */ 678 if (invalidate) { 679 if (!(disk->flags & GENHD_FL_NO_PART) || 680 !(disk->flags & GENHD_FL_REMOVABLE)) 681 set_capacity(disk, 0); 682 } 683 684 if (get_capacity(disk)) { 685 ret = blk_add_partitions(disk); 686 if (ret == -EAGAIN) 687 goto rescan; 688 } else if (invalidate) { 689 /* 690 * Tell userspace that the media / partition table may have 691 * changed. 692 */ 693 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE); 694 } 695 696 return ret; 697 } 698 /* 699 * Only exported for loop and dasd for historic reasons. Don't use in new 700 * code! 701 */ 702 EXPORT_SYMBOL_GPL(bdev_disk_changed); 703 704 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p) 705 { 706 struct address_space *mapping = state->disk->part0->bd_inode->i_mapping; 707 struct folio *folio; 708 709 if (n >= get_capacity(state->disk)) { 710 state->access_beyond_eod = true; 711 goto out; 712 } 713 714 folio = read_mapping_folio(mapping, n >> PAGE_SECTORS_SHIFT, NULL); 715 if (IS_ERR(folio)) 716 goto out; 717 718 p->v = folio; 719 return folio_address(folio) + offset_in_folio(folio, n * SECTOR_SIZE); 720 out: 721 p->v = NULL; 722 return NULL; 723 } 724