block/partitions/core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 1991-1998  Linus Torvalds
   4  * Re-organised Feb 1998 Russell King
   5  */
   6 #include <linux/fs.h>
   7 #include <linux/slab.h>
   8 #include <linux/ctype.h>
   9 #include <linux/genhd.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/blktrace_api.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 struct parsed_partitions *allocate_partitions(struct gendisk *hd)
  89 {
  90         struct parsed_partitions *state;
  91         int nr;
  92
  93         state = kzalloc(sizeof(*state), GFP_KERNEL);
  94         if (!state)
  95                 return NULL;
  96
  97         nr = disk_max_parts(hd);
  98         state->parts = vzalloc(array_size(nr, sizeof(state->parts[0])));
  99         if (!state->parts) {
 100                 kfree(state);
 101                 return NULL;
 102         }
 103
 104         state->limit = nr;
 105
 106         return state;
 107 }
 108
 109 static void free_partitions(struct parsed_partitions *state)
 110 {
 111         vfree(state->parts);
 112         kfree(state);
 113 }
 114
 115 static struct parsed_partitions *check_partition(struct gendisk *hd,
 116                 struct block_device *bdev)
 117 {
 118         struct parsed_partitions *state;
 119         int i, res, err;
 120
 121         state = allocate_partitions(hd);
 122         if (!state)
 123                 return NULL;
 124         state->pp_buf = (char *)__get_free_page(GFP_KERNEL);
 125         if (!state->pp_buf) {
 126                 free_partitions(state);
 127                 return NULL;
 128         }
 129         state->pp_buf[0] = '\0';
 130
 131         state->bdev = bdev;
 132         disk_name(hd, 0, state->name);
 133         snprintf(state->pp_buf, PAGE_SIZE, " %s:", state->name);
 134         if (isdigit(state->name[strlen(state->name)-1]))
 135                 sprintf(state->name, "p");
 136
 137         i = res = err = 0;
 138         while (!res && check_part[i]) {
 139                 memset(state->parts, 0, state->limit * sizeof(state->parts[0]));
 140                 res = check_part[i++](state);
 141                 if (res < 0) {
 142                         /*
 143                          * We have hit an I/O error which we don't report now.
 144                          * But record it, and let the others do their job.
 145                          */
 146                         err = res;
 147                         res = 0;
 148                 }
 149
 150         }
 151         if (res > 0) {
 152                 printk(KERN_INFO "%s", state->pp_buf);
 153
 154                 free_page((unsigned long)state->pp_buf);
 155                 return state;
 156         }
 157         if (state->access_beyond_eod)
 158                 err = -ENOSPC;
 159         /*
 160          * The partition is unrecognized. So report I/O errors if there were any
 161          */
 162         if (err)
 163                 res = err;
 164         if (res) {
 165                 strlcat(state->pp_buf,
 166                         " unable to read partition table\n", PAGE_SIZE);
 167                 printk(KERN_INFO "%s", state->pp_buf);
 168         }
 169
 170         free_page((unsigned long)state->pp_buf);
 171         free_partitions(state);
 172         return ERR_PTR(res);
 173 }
 174
 175 static ssize_t part_partition_show(struct device *dev,
 176                                    struct device_attribute *attr, char *buf)
 177 {
 178         struct hd_struct *p = dev_to_part(dev);
 179
 180         return sprintf(buf, "%d\n", p->partno);
 181 }
 182
 183 static ssize_t part_start_show(struct device *dev,
 184                                struct device_attribute *attr, char *buf)
 185 {
 186         struct hd_struct *p = dev_to_part(dev);
 187
 188         return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
 189 }
 190
 191 static ssize_t part_ro_show(struct device *dev,
 192                             struct device_attribute *attr, char *buf)
 193 {
 194         struct hd_struct *p = dev_to_part(dev);
 195         return sprintf(buf, "%d\n", p->policy ? 1 : 0);
 196 }
 197
 198 static ssize_t part_alignment_offset_show(struct device *dev,
 199                                           struct device_attribute *attr, char *buf)
 200 {
 201         struct hd_struct *p = dev_to_part(dev);
 202         return sprintf(buf, "%llu\n", (unsigned long long)p->alignment_offset);
 203 }
 204
 205 static ssize_t part_discard_alignment_show(struct device *dev,
 206                                            struct device_attribute *attr, char *buf)
 207 {
 208         struct hd_struct *p = dev_to_part(dev);
 209         return sprintf(buf, "%u\n", p->discard_alignment);
 210 }
 211
 212 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
 213 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
 214 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
 215 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
 216 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
 217 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
 218 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
 219 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
 220 #ifdef CONFIG_FAIL_MAKE_REQUEST
 221 static struct device_attribute dev_attr_fail =
 222         __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
 223 #endif
 224
 225 static struct attribute *part_attrs[] = {
 226         &dev_attr_partition.attr,
 227         &dev_attr_start.attr,
 228         &dev_attr_size.attr,
 229         &dev_attr_ro.attr,
 230         &dev_attr_alignment_offset.attr,
 231         &dev_attr_discard_alignment.attr,
 232         &dev_attr_stat.attr,
 233         &dev_attr_inflight.attr,
 234 #ifdef CONFIG_FAIL_MAKE_REQUEST
 235         &dev_attr_fail.attr,
 236 #endif
 237         NULL
 238 };
 239
 240 static struct attribute_group part_attr_group = {
 241         .attrs = part_attrs,
 242 };
 243
 244 static const struct attribute_group *part_attr_groups[] = {
 245         &part_attr_group,
 246 #ifdef CONFIG_BLK_DEV_IO_TRACE
 247         &blk_trace_attr_group,
 248 #endif
 249         NULL
 250 };
 251
 252 static void part_release(struct device *dev)
 253 {
 254         struct hd_struct *p = dev_to_part(dev);
 255         blk_free_devt(dev->devt);
 256         hd_free_part(p);
 257         kfree(p);
 258 }
 259
 260 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
 261 {
 262         struct hd_struct *part = dev_to_part(dev);
 263
 264         add_uevent_var(env, "PARTN=%u", part->partno);
 265         if (part->info && part->info->volname[0])
 266                 add_uevent_var(env, "PARTNAME=%s", part->info->volname);
 267         return 0;
 268 }
 269
 270 struct device_type part_type = {
 271         .name           = "partition",
 272         .groups         = part_attr_groups,
 273         .release        = part_release,
 274         .uevent         = part_uevent,
 275 };
 276
 277 static void delete_partition_work_fn(struct work_struct *work)
 278 {
 279         struct hd_struct *part = container_of(to_rcu_work(work), struct hd_struct,
 280                                         rcu_work);
 281
 282         part->start_sect = 0;
 283         part->nr_sects = 0;
 284         part_stat_set_all(part, 0);
 285         put_device(part_to_dev(part));
 286 }
 287
 288 void __delete_partition(struct percpu_ref *ref)
 289 {
 290         struct hd_struct *part = container_of(ref, struct hd_struct, ref);
 291         INIT_RCU_WORK(&part->rcu_work, delete_partition_work_fn);
 292         queue_rcu_work(system_wq, &part->rcu_work);
 293 }
 294
 295 /*
 296  * Must be called either with bd_mutex held, before a disk can be opened or
 297  * after all disk users are gone.
 298  */
 299 void delete_partition(struct gendisk *disk, int partno)
 300 {
 301         struct disk_part_tbl *ptbl =
 302                 rcu_dereference_protected(disk->part_tbl, 1);
 303         struct hd_struct *part;
 304
 305         if (partno >= ptbl->len)
 306                 return;
 307
 308         part = rcu_dereference_protected(ptbl->part[partno], 1);
 309         if (!part)
 310                 return;
 311
 312         rcu_assign_pointer(ptbl->part[partno], NULL);
 313         rcu_assign_pointer(ptbl->last_lookup, NULL);
 314         kobject_put(part->holder_dir);
 315         device_del(part_to_dev(part));
 316
 317         /*
 318          * Remove gendisk pointer from idr so that it cannot be looked up
 319          * while RCU period before freeing gendisk is running to prevent
 320          * use-after-free issues. Note that the device number stays
 321          * "in-use" until we really free the gendisk.
 322          */
 323         blk_invalidate_devt(part_devt(part));
 324         hd_struct_kill(part);
 325 }
 326
 327 static ssize_t whole_disk_show(struct device *dev,
 328                                struct device_attribute *attr, char *buf)
 329 {
 330         return 0;
 331 }
 332 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
 333
 334 /*
 335  * Must be called either with bd_mutex held, before a disk can be opened or
 336  * after all disk users are gone.
 337  */
 338 struct hd_struct *add_partition(struct gendisk *disk, int partno,
 339                                 sector_t start, sector_t len, int flags,
 340                                 struct partition_meta_info *info)
 341 {
 342         struct hd_struct *p;
 343         dev_t devt = MKDEV(0, 0);
 344         struct device *ddev = disk_to_dev(disk);
 345         struct device *pdev;
 346         struct disk_part_tbl *ptbl;
 347         const char *dname;
 348         int err;
 349
 350         /*
 351          * Partitions are not supported on zoned block devices that are used as
 352          * such.
 353          */
 354         switch (disk->queue->limits.zoned) {
 355         case BLK_ZONED_HM:
 356                 pr_warn("%s: partitions not supported on host managed zoned block device\n",
 357                         disk->disk_name);
 358                 return ERR_PTR(-ENXIO);
 359         case BLK_ZONED_HA:
 360                 pr_info("%s: disabling host aware zoned block device support due to partitions\n",
 361                         disk->disk_name);
 362                 disk->queue->limits.zoned = BLK_ZONED_NONE;
 363                 break;
 364         case BLK_ZONED_NONE:
 365                 break;
 366         }
 367
 368         err = disk_expand_part_tbl(disk, partno);
 369         if (err)
 370                 return ERR_PTR(err);
 371         ptbl = rcu_dereference_protected(disk->part_tbl, 1);
 372
 373         if (ptbl->part[partno])
 374                 return ERR_PTR(-EBUSY);
 375
 376         p = kzalloc(sizeof(*p), GFP_KERNEL);
 377         if (!p)
 378                 return ERR_PTR(-EBUSY);
 379
 380         if (!init_part_stats(p)) {
 381                 err = -ENOMEM;
 382                 goto out_free;
 383         }
 384
 385         seqcount_init(&p->nr_sects_seq);
 386         pdev = part_to_dev(p);
 387
 388         p->start_sect = start;
 389         p->alignment_offset =
 390                 queue_limit_alignment_offset(&disk->queue->limits, start);
 391         p->discard_alignment =
 392                 queue_limit_discard_alignment(&disk->queue->limits, start);
 393         p->nr_sects = len;
 394         p->partno = partno;
 395         p->policy = get_disk_ro(disk);
 396
 397         if (info) {
 398                 struct partition_meta_info *pinfo;
 399
 400                 pinfo = kzalloc_node(sizeof(*pinfo), GFP_KERNEL, disk->node_id);
 401                 if (!pinfo) {
 402                         err = -ENOMEM;
 403                         goto out_free_stats;
 404                 }
 405                 memcpy(pinfo, info, sizeof(*info));
 406                 p->info = pinfo;
 407         }
 408
 409         dname = dev_name(ddev);
 410         if (isdigit(dname[strlen(dname) - 1]))
 411                 dev_set_name(pdev, "%sp%d", dname, partno);
 412         else
 413                 dev_set_name(pdev, "%s%d", dname, partno);
 414
 415         device_initialize(pdev);
 416         pdev->class = &block_class;
 417         pdev->type = &part_type;
 418         pdev->parent = ddev;
 419
 420         err = blk_alloc_devt(p, &devt);
 421         if (err)
 422                 goto out_free_info;
 423         pdev->devt = devt;
 424
 425         /* delay uevent until 'holders' subdir is created */
 426         dev_set_uevent_suppress(pdev, 1);
 427         err = device_add(pdev);
 428         if (err)
 429                 goto out_put;
 430
 431         err = -ENOMEM;
 432         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
 433         if (!p->holder_dir)
 434                 goto out_del;
 435
 436         dev_set_uevent_suppress(pdev, 0);
 437         if (flags & ADDPART_FLAG_WHOLEDISK) {
 438                 err = device_create_file(pdev, &dev_attr_whole_disk);
 439                 if (err)
 440                         goto out_del;
 441         }
 442
 443         err = hd_ref_init(p);
 444         if (err) {
 445                 if (flags & ADDPART_FLAG_WHOLEDISK)
 446                         goto out_remove_file;
 447                 goto out_del;
 448         }
 449
 450         /* everything is up and running, commence */
 451         rcu_assign_pointer(ptbl->part[partno], p);
 452
 453         /* suppress uevent if the disk suppresses it */
 454         if (!dev_get_uevent_suppress(ddev))
 455                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 456         return p;
 457
 458 out_free_info:
 459         kfree(p->info);
 460 out_free_stats:
 461         free_part_stats(p);
 462 out_free:
 463         kfree(p);
 464         return ERR_PTR(err);
 465 out_remove_file:
 466         device_remove_file(pdev, &dev_attr_whole_disk);
 467 out_del:
 468         kobject_put(p->holder_dir);
 469         device_del(pdev);
 470 out_put:
 471         put_device(pdev);
 472         return ERR_PTR(err);
 473 }
 474
 475 static bool disk_unlock_native_capacity(struct gendisk *disk)
 476 {
 477         const struct block_device_operations *bdops = disk->fops;
 478
 479         if (bdops->unlock_native_capacity &&
 480             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
 481                 printk(KERN_CONT "enabling native capacity\n");
 482                 bdops->unlock_native_capacity(disk);
 483                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
 484                 return true;
 485         } else {
 486                 printk(KERN_CONT "truncated\n");
 487                 return false;
 488         }
 489 }
 490
 491 int blk_drop_partitions(struct gendisk *disk, struct block_device *bdev)
 492 {
 493         struct disk_part_iter piter;
 494         struct hd_struct *part;
 495         int res;
 496
 497         if (!disk_part_scan_enabled(disk))
 498                 return 0;
 499         if (bdev->bd_part_count || bdev->bd_openers > 1)
 500                 return -EBUSY;
 501         res = invalidate_partition(disk, 0);
 502         if (res)
 503                 return res;
 504
 505         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 506         while ((part = disk_part_iter_next(&piter)))
 507                 delete_partition(disk, part->partno);
 508         disk_part_iter_exit(&piter);
 509
 510         return 0;
 511 }
 512
 513 static bool blk_add_partition(struct gendisk *disk, struct block_device *bdev,
 514                 struct parsed_partitions *state, int p)
 515 {
 516         sector_t size = state->parts[p].size;
 517         sector_t from = state->parts[p].from;
 518         struct hd_struct *part;
 519
 520         if (!size)
 521                 return true;
 522
 523         if (from >= get_capacity(disk)) {
 524                 printk(KERN_WARNING
 525                        "%s: p%d start %llu is beyond EOD, ",
 526                        disk->disk_name, p, (unsigned long long) from);
 527                 if (disk_unlock_native_capacity(disk))
 528                         return false;
 529                 return true;
 530         }
 531
 532         if (from + size > get_capacity(disk)) {
 533                 printk(KERN_WARNING
 534                        "%s: p%d size %llu extends beyond EOD, ",
 535                        disk->disk_name, p, (unsigned long long) size);
 536
 537                 if (disk_unlock_native_capacity(disk))
 538                         return false;
 539
 540                 /*
 541                  * We can not ignore partitions of broken tables created by for
 542                  * example camera firmware, but we limit them to the end of the
 543                  * disk to avoid creating invalid block devices.
 544                  */
 545                 size = get_capacity(disk) - from;
 546         }
 547
 548         part = add_partition(disk, p, from, size, state->parts[p].flags,
 549                              &state->parts[p].info);
 550         if (IS_ERR(part) && PTR_ERR(part) != -ENXIO) {
 551                 printk(KERN_ERR " %s: p%d could not be added: %ld\n",
 552                        disk->disk_name, p, -PTR_ERR(part));
 553                 return true;
 554         }
 555
 556         if (IS_BUILTIN(CONFIG_BLK_DEV_MD) &&
 557             (state->parts[p].flags & ADDPART_FLAG_RAID))
 558                 md_autodetect_dev(part_to_dev(part)->devt);
 559
 560         return true;
 561 }
 562
 563 int blk_add_partitions(struct gendisk *disk, struct block_device *bdev)
 564 {
 565         struct parsed_partitions *state;
 566         int ret = -EAGAIN, p, highest;
 567
 568         if (!disk_part_scan_enabled(disk))
 569                 return 0;
 570
 571         state = check_partition(disk, bdev);
 572         if (!state)
 573                 return 0;
 574         if (IS_ERR(state)) {
 575                 /*
 576                  * I/O error reading the partition table.  If we tried to read
 577                  * beyond EOD, retry after unlocking the native capacity.
 578                  */
 579                 if (PTR_ERR(state) == -ENOSPC) {
 580                         printk(KERN_WARNING "%s: partition table beyond EOD, ",
 581                                disk->disk_name);
 582                         if (disk_unlock_native_capacity(disk))
 583                                 return -EAGAIN;
 584                 }
 585                 return -EIO;
 586         }
 587
 588         /*
 589          * Partitions are not supported on host managed zoned block devices.
 590          */
 591         if (disk->queue->limits.zoned == BLK_ZONED_HM) {
 592                 pr_warn("%s: ignoring partition table on host managed zoned block device\n",
 593                         disk->disk_name);
 594                 ret = 0;
 595                 goto out_free_state;
 596         }
 597
 598         /*
 599          * If we read beyond EOD, try unlocking native capacity even if the
 600          * partition table was successfully read as we could be missing some
 601          * partitions.
 602          */
 603         if (state->access_beyond_eod) {
 604                 printk(KERN_WARNING
 605                        "%s: partition table partially beyond EOD, ",
 606                        disk->disk_name);
 607                 if (disk_unlock_native_capacity(disk))
 608                         goto out_free_state;
 609         }
 610
 611         /* tell userspace that the media / partition table may have changed */
 612         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 613
 614         /*
 615          * Detect the highest partition number and preallocate disk->part_tbl.
 616          * This is an optimization and not strictly necessary.
 617          */
 618         for (p = 1, highest = 0; p < state->limit; p++)
 619                 if (state->parts[p].size)
 620                         highest = p;
 621         disk_expand_part_tbl(disk, highest);
 622
 623         for (p = 1; p < state->limit; p++)
 624                 if (!blk_add_partition(disk, bdev, state, p))
 625                         goto out_free_state;
 626
 627         ret = 0;
 628 out_free_state:
 629         free_partitions(state);
 630         return ret;
 631 }
 632
 633 void *read_part_sector(struct parsed_partitions *state, sector_t n, Sector *p)
 634 {
 635         struct address_space *mapping = state->bdev->bd_inode->i_mapping;
 636         struct page *page;
 637
 638         if (n >= get_capacity(state->bdev->bd_disk)) {
 639                 state->access_beyond_eod = true;
 640                 return NULL;
 641         }
 642
 643         page = read_mapping_page(mapping,
 644                         (pgoff_t)(n >> (PAGE_SHIFT - 9)), NULL);
 645         if (IS_ERR(page))
 646                 goto out;
 647         if (PageError(page))
 648                 goto out_put_page;
 649
 650         p->v = page;
 651         return (unsigned char *)page_address(page) +
 652                         ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << SECTOR_SHIFT);
 653 out_put_page:
 654         put_page(page);
 655 out:
 656         p->v = NULL;
 657         return NULL;
 658 }