block/partition-generic.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  Code extracted from drivers/block/genhd.c
   4  *  Copyright (C) 1991-1998  Linus Torvalds
   5  *  Re-organised Feb 1998 Russell King
   6  *
   7  *  We now have independent partition support from the
   8  *  block drivers, which allows all the partition code to
   9  *  be grouped in one location, and it to be mostly self
  10  *  contained.
  11  */
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/fs.h>
  16 #include <linux/slab.h>
  17 #include <linux/kmod.h>
  18 #include <linux/ctype.h>
  19 #include <linux/genhd.h>
  20 #include <linux/blktrace_api.h>
  21
  22 #include "partitions/check.h"
  23
  24 #ifdef CONFIG_BLK_DEV_MD
  25 extern void md_autodetect_dev(dev_t dev);
  26 #endif
  27
  28 /*
  29  * disk_name() is used by partition check code and the genhd driver.
  30  * It formats the devicename of the indicated disk into
  31  * the supplied buffer (of size at least 32), and returns
  32  * a pointer to that same buffer (for convenience).
  33  */
  34
  35 char *disk_name(struct gendisk *hd, int partno, char *buf)
  36 {
  37         if (!partno)
  38                 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
  39         else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
  40                 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
  41         else
  42                 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
  43
  44         return buf;
  45 }
  46
  47 const char *bdevname(struct block_device *bdev, char *buf)
  48 {
  49         return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
  50 }
  51
  52 EXPORT_SYMBOL(bdevname);
  53
  54 /*
  55  * There's very little reason to use this, you should really
  56  * have a struct block_device just about everywhere and use
  57  * bdevname() instead.
  58  */
  59 const char *__bdevname(dev_t dev, char *buffer)
  60 {
  61         scnprintf(buffer, BDEVNAME_SIZE, "unknown-block(%u,%u)",
  62                                 MAJOR(dev), MINOR(dev));
  63         return buffer;
  64 }
  65
  66 EXPORT_SYMBOL(__bdevname);
  67
  68 static ssize_t part_partition_show(struct device *dev,
  69                                    struct device_attribute *attr, char *buf)
  70 {
  71         struct hd_struct *p = dev_to_part(dev);
  72
  73         return sprintf(buf, "%d\n", p->partno);
  74 }
  75
  76 static ssize_t part_start_show(struct device *dev,
  77                                struct device_attribute *attr, char *buf)
  78 {
  79         struct hd_struct *p = dev_to_part(dev);
  80
  81         return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
  82 }
  83
  84 ssize_t part_size_show(struct device *dev,
  85                        struct device_attribute *attr, char *buf)
  86 {
  87         struct hd_struct *p = dev_to_part(dev);
  88         return sprintf(buf, "%llu\n",(unsigned long long)part_nr_sects_read(p));
  89 }
  90
  91 static ssize_t part_ro_show(struct device *dev,
  92                             struct device_attribute *attr, char *buf)
  93 {
  94         struct hd_struct *p = dev_to_part(dev);
  95         return sprintf(buf, "%d\n", p->policy ? 1 : 0);
  96 }
  97
  98 static ssize_t part_alignment_offset_show(struct device *dev,
  99                                           struct device_attribute *attr, char *buf)
 100 {
 101         struct hd_struct *p = dev_to_part(dev);
 102         return sprintf(buf, "%llu\n", (unsigned long long)p->alignment_offset);
 103 }
 104
 105 static ssize_t part_discard_alignment_show(struct device *dev,
 106                                            struct device_attribute *attr, char *buf)
 107 {
 108         struct hd_struct *p = dev_to_part(dev);
 109         return sprintf(buf, "%u\n", p->discard_alignment);
 110 }
 111
 112 ssize_t part_stat_show(struct device *dev,
 113                        struct device_attribute *attr, char *buf)
 114 {
 115         struct hd_struct *p = dev_to_part(dev);
 116         struct request_queue *q = part_to_disk(p)->queue;
 117         unsigned int inflight[2];
 118         int cpu;
 119
 120         cpu = part_stat_lock();
 121         part_round_stats(q, cpu, p);
 122         part_stat_unlock();
 123         part_in_flight(q, p, inflight);
 124         return sprintf(buf,
 125                 "%8lu %8lu %8llu %8u "
 126                 "%8lu %8lu %8llu %8u "
 127                 "%8u %8u %8u"
 128                 "\n",
 129                 part_stat_read(p, ios[READ]),
 130                 part_stat_read(p, merges[READ]),
 131                 (unsigned long long)part_stat_read(p, sectors[READ]),
 132                 jiffies_to_msecs(part_stat_read(p, ticks[READ])),
 133                 part_stat_read(p, ios[WRITE]),
 134                 part_stat_read(p, merges[WRITE]),
 135                 (unsigned long long)part_stat_read(p, sectors[WRITE]),
 136                 jiffies_to_msecs(part_stat_read(p, ticks[WRITE])),
 137                 inflight[0],
 138                 jiffies_to_msecs(part_stat_read(p, io_ticks)),
 139                 jiffies_to_msecs(part_stat_read(p, time_in_queue)));
 140 }
 141
 142 ssize_t part_inflight_show(struct device *dev,
 143                         struct device_attribute *attr, char *buf)
 144 {
 145         struct hd_struct *p = dev_to_part(dev);
 146
 147         return sprintf(buf, "%8u %8u\n", atomic_read(&p->in_flight[0]),
 148                 atomic_read(&p->in_flight[1]));
 149 }
 150
 151 #ifdef CONFIG_FAIL_MAKE_REQUEST
 152 ssize_t part_fail_show(struct device *dev,
 153                        struct device_attribute *attr, char *buf)
 154 {
 155         struct hd_struct *p = dev_to_part(dev);
 156
 157         return sprintf(buf, "%d\n", p->make_it_fail);
 158 }
 159
 160 ssize_t part_fail_store(struct device *dev,
 161                         struct device_attribute *attr,
 162                         const char *buf, size_t count)
 163 {
 164         struct hd_struct *p = dev_to_part(dev);
 165         int i;
 166
 167         if (count > 0 && sscanf(buf, "%d", &i) > 0)
 168                 p->make_it_fail = (i == 0) ? 0 : 1;
 169
 170         return count;
 171 }
 172 #endif
 173
 174 static DEVICE_ATTR(partition, S_IRUGO, part_partition_show, NULL);
 175 static DEVICE_ATTR(start, S_IRUGO, part_start_show, NULL);
 176 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
 177 static DEVICE_ATTR(ro, S_IRUGO, part_ro_show, NULL);
 178 static DEVICE_ATTR(alignment_offset, S_IRUGO, part_alignment_offset_show, NULL);
 179 static DEVICE_ATTR(discard_alignment, S_IRUGO, part_discard_alignment_show,
 180                    NULL);
 181 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
 182 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
 183 #ifdef CONFIG_FAIL_MAKE_REQUEST
 184 static struct device_attribute dev_attr_fail =
 185         __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
 186 #endif
 187
 188 static struct attribute *part_attrs[] = {
 189         &dev_attr_partition.attr,
 190         &dev_attr_start.attr,
 191         &dev_attr_size.attr,
 192         &dev_attr_ro.attr,
 193         &dev_attr_alignment_offset.attr,
 194         &dev_attr_discard_alignment.attr,
 195         &dev_attr_stat.attr,
 196         &dev_attr_inflight.attr,
 197 #ifdef CONFIG_FAIL_MAKE_REQUEST
 198         &dev_attr_fail.attr,
 199 #endif
 200         NULL
 201 };
 202
 203 static struct attribute_group part_attr_group = {
 204         .attrs = part_attrs,
 205 };
 206
 207 static const struct attribute_group *part_attr_groups[] = {
 208         &part_attr_group,
 209 #ifdef CONFIG_BLK_DEV_IO_TRACE
 210         &blk_trace_attr_group,
 211 #endif
 212         NULL
 213 };
 214
 215 static void part_release(struct device *dev)
 216 {
 217         struct hd_struct *p = dev_to_part(dev);
 218         blk_free_devt(dev->devt);
 219         hd_free_part(p);
 220         kfree(p);
 221 }
 222
 223 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
 224 {
 225         struct hd_struct *part = dev_to_part(dev);
 226
 227         add_uevent_var(env, "PARTN=%u", part->partno);
 228         if (part->info && part->info->volname[0])
 229                 add_uevent_var(env, "PARTNAME=%s", part->info->volname);
 230         return 0;
 231 }
 232
 233 struct device_type part_type = {
 234         .name           = "partition",
 235         .groups         = part_attr_groups,
 236         .release        = part_release,
 237         .uevent         = part_uevent,
 238 };
 239
 240 static void delete_partition_rcu_cb(struct rcu_head *head)
 241 {
 242         struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);
 243
 244         part->start_sect = 0;
 245         part->nr_sects = 0;
 246         part_stat_set_all(part, 0);
 247         put_device(part_to_dev(part));
 248 }
 249
 250 void __delete_partition(struct percpu_ref *ref)
 251 {
 252         struct hd_struct *part = container_of(ref, struct hd_struct, ref);
 253         call_rcu(&part->rcu_head, delete_partition_rcu_cb);
 254 }
 255
 256 /*
 257  * Must be called either with bd_mutex held, before a disk can be opened or
 258  * after all disk users are gone.
 259  */
 260 void delete_partition(struct gendisk *disk, int partno)
 261 {
 262         struct disk_part_tbl *ptbl =
 263                 rcu_dereference_protected(disk->part_tbl, 1);
 264         struct hd_struct *part;
 265
 266         if (partno >= ptbl->len)
 267                 return;
 268
 269         part = rcu_dereference_protected(ptbl->part[partno], 1);
 270         if (!part)
 271                 return;
 272
 273         rcu_assign_pointer(ptbl->part[partno], NULL);
 274         rcu_assign_pointer(ptbl->last_lookup, NULL);
 275         kobject_put(part->holder_dir);
 276         device_del(part_to_dev(part));
 277
 278         hd_struct_kill(part);
 279 }
 280
 281 static ssize_t whole_disk_show(struct device *dev,
 282                                struct device_attribute *attr, char *buf)
 283 {
 284         return 0;
 285 }
 286 static DEVICE_ATTR(whole_disk, S_IRUSR | S_IRGRP | S_IROTH,
 287                    whole_disk_show, NULL);
 288
 289 /*
 290  * Must be called either with bd_mutex held, before a disk can be opened or
 291  * after all disk users are gone.
 292  */
 293 struct hd_struct *add_partition(struct gendisk *disk, int partno,
 294                                 sector_t start, sector_t len, int flags,
 295                                 struct partition_meta_info *info)
 296 {
 297         struct hd_struct *p;
 298         dev_t devt = MKDEV(0, 0);
 299         struct device *ddev = disk_to_dev(disk);
 300         struct device *pdev;
 301         struct disk_part_tbl *ptbl;
 302         const char *dname;
 303         int err;
 304
 305         err = disk_expand_part_tbl(disk, partno);
 306         if (err)
 307                 return ERR_PTR(err);
 308         ptbl = rcu_dereference_protected(disk->part_tbl, 1);
 309
 310         if (ptbl->part[partno])
 311                 return ERR_PTR(-EBUSY);
 312
 313         p = kzalloc(sizeof(*p), GFP_KERNEL);
 314         if (!p)
 315                 return ERR_PTR(-EBUSY);
 316
 317         if (!init_part_stats(p)) {
 318                 err = -ENOMEM;
 319                 goto out_free;
 320         }
 321
 322         seqcount_init(&p->nr_sects_seq);
 323         pdev = part_to_dev(p);
 324
 325         p->start_sect = start;
 326         p->alignment_offset =
 327                 queue_limit_alignment_offset(&disk->queue->limits, start);
 328         p->discard_alignment =
 329                 queue_limit_discard_alignment(&disk->queue->limits, start);
 330         p->nr_sects = len;
 331         p->partno = partno;
 332         p->policy = get_disk_ro(disk);
 333
 334         if (info) {
 335                 struct partition_meta_info *pinfo = alloc_part_info(disk);
 336                 if (!pinfo) {
 337                         err = -ENOMEM;
 338                         goto out_free_stats;
 339                 }
 340                 memcpy(pinfo, info, sizeof(*info));
 341                 p->info = pinfo;
 342         }
 343
 344         dname = dev_name(ddev);
 345         if (isdigit(dname[strlen(dname) - 1]))
 346                 dev_set_name(pdev, "%sp%d", dname, partno);
 347         else
 348                 dev_set_name(pdev, "%s%d", dname, partno);
 349
 350         device_initialize(pdev);
 351         pdev->class = &block_class;
 352         pdev->type = &part_type;
 353         pdev->parent = ddev;
 354
 355         err = blk_alloc_devt(p, &devt);
 356         if (err)
 357                 goto out_free_info;
 358         pdev->devt = devt;
 359
 360         /* delay uevent until 'holders' subdir is created */
 361         dev_set_uevent_suppress(pdev, 1);
 362         err = device_add(pdev);
 363         if (err)
 364                 goto out_put;
 365
 366         err = -ENOMEM;
 367         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
 368         if (!p->holder_dir)
 369                 goto out_del;
 370
 371         dev_set_uevent_suppress(pdev, 0);
 372         if (flags & ADDPART_FLAG_WHOLEDISK) {
 373                 err = device_create_file(pdev, &dev_attr_whole_disk);
 374                 if (err)
 375                         goto out_del;
 376         }
 377
 378         err = hd_ref_init(p);
 379         if (err) {
 380                 if (flags & ADDPART_FLAG_WHOLEDISK)
 381                         goto out_remove_file;
 382                 goto out_del;
 383         }
 384
 385         /* everything is up and running, commence */
 386         rcu_assign_pointer(ptbl->part[partno], p);
 387
 388         /* suppress uevent if the disk suppresses it */
 389         if (!dev_get_uevent_suppress(ddev))
 390                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 391         return p;
 392
 393 out_free_info:
 394         free_part_info(p);
 395 out_free_stats:
 396         free_part_stats(p);
 397 out_free:
 398         kfree(p);
 399         return ERR_PTR(err);
 400 out_remove_file:
 401         device_remove_file(pdev, &dev_attr_whole_disk);
 402 out_del:
 403         kobject_put(p->holder_dir);
 404         device_del(pdev);
 405 out_put:
 406         put_device(pdev);
 407         return ERR_PTR(err);
 408 }
 409
 410 static bool disk_unlock_native_capacity(struct gendisk *disk)
 411 {
 412         const struct block_device_operations *bdops = disk->fops;
 413
 414         if (bdops->unlock_native_capacity &&
 415             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
 416                 printk(KERN_CONT "enabling native capacity\n");
 417                 bdops->unlock_native_capacity(disk);
 418                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
 419                 return true;
 420         } else {
 421                 printk(KERN_CONT "truncated\n");
 422                 return false;
 423         }
 424 }
 425
 426 static int drop_partitions(struct gendisk *disk, struct block_device *bdev)
 427 {
 428         struct disk_part_iter piter;
 429         struct hd_struct *part;
 430         int res;
 431
 432         if (bdev->bd_part_count || bdev->bd_super)
 433                 return -EBUSY;
 434         res = invalidate_partition(disk, 0);
 435         if (res)
 436                 return res;
 437
 438         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 439         while ((part = disk_part_iter_next(&piter)))
 440                 delete_partition(disk, part->partno);
 441         disk_part_iter_exit(&piter);
 442
 443         return 0;
 444 }
 445
 446 static bool part_zone_aligned(struct gendisk *disk,
 447                               struct block_device *bdev,
 448                               sector_t from, sector_t size)
 449 {
 450         unsigned int zone_sectors = bdev_zone_sectors(bdev);
 451
 452         /*
 453          * If this function is called, then the disk is a zoned block device
 454          * (host-aware or host-managed). This can be detected even if the
 455          * zoned block device support is disabled (CONFIG_BLK_DEV_ZONED not
 456          * set). In this case, however, only host-aware devices will be seen
 457          * as a block device is not created for host-managed devices. Without
 458          * zoned block device support, host-aware drives can still be used as
 459          * regular block devices (no zone operation) and their zone size will
 460          * be reported as 0. Allow this case.
 461          */
 462         if (!zone_sectors)
 463                 return true;
 464
 465         /*
 466          * Check partition start and size alignement. If the drive has a
 467          * smaller last runt zone, ignore it and allow the partition to
 468          * use it. Check the zone size too: it should be a power of 2 number
 469          * of sectors.
 470          */
 471         if (WARN_ON_ONCE(!is_power_of_2(zone_sectors))) {
 472                 u32 rem;
 473
 474                 div_u64_rem(from, zone_sectors, &rem);
 475                 if (rem)
 476                         return false;
 477                 if ((from + size) < get_capacity(disk)) {
 478                         div_u64_rem(size, zone_sectors, &rem);
 479                         if (rem)
 480                                 return false;
 481                 }
 482
 483         } else {
 484
 485                 if (from & (zone_sectors - 1))
 486                         return false;
 487                 if ((from + size) < get_capacity(disk) &&
 488                     (size & (zone_sectors - 1)))
 489                         return false;
 490
 491         }
 492
 493         return true;
 494 }
 495
 496 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 497 {
 498         struct parsed_partitions *state = NULL;
 499         struct hd_struct *part;
 500         int p, highest, res;
 501 rescan:
 502         if (state && !IS_ERR(state)) {
 503                 free_partitions(state);
 504                 state = NULL;
 505         }
 506
 507         res = drop_partitions(disk, bdev);
 508         if (res)
 509                 return res;
 510
 511         if (disk->fops->revalidate_disk)
 512                 disk->fops->revalidate_disk(disk);
 513         check_disk_size_change(disk, bdev);
 514         bdev->bd_invalidated = 0;
 515         if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
 516                 return 0;
 517         if (IS_ERR(state)) {
 518                 /*
 519                  * I/O error reading the partition table.  If any
 520                  * partition code tried to read beyond EOD, retry
 521                  * after unlocking native capacity.
 522                  */
 523                 if (PTR_ERR(state) == -ENOSPC) {
 524                         printk(KERN_WARNING "%s: partition table beyond EOD, ",
 525                                disk->disk_name);
 526                         if (disk_unlock_native_capacity(disk))
 527                                 goto rescan;
 528                 }
 529                 return -EIO;
 530         }
 531         /*
 532          * If any partition code tried to read beyond EOD, try
 533          * unlocking native capacity even if partition table is
 534          * successfully read as we could be missing some partitions.
 535          */
 536         if (state->access_beyond_eod) {
 537                 printk(KERN_WARNING
 538                        "%s: partition table partially beyond EOD, ",
 539                        disk->disk_name);
 540                 if (disk_unlock_native_capacity(disk))
 541                         goto rescan;
 542         }
 543
 544         /* tell userspace that the media / partition table may have changed */
 545         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 546
 547         /* Detect the highest partition number and preallocate
 548          * disk->part_tbl.  This is an optimization and not strictly
 549          * necessary.
 550          */
 551         for (p = 1, highest = 0; p < state->limit; p++)
 552                 if (state->parts[p].size)
 553                         highest = p;
 554
 555         disk_expand_part_tbl(disk, highest);
 556
 557         /* add partitions */
 558         for (p = 1; p < state->limit; p++) {
 559                 sector_t size, from;
 560
 561                 size = state->parts[p].size;
 562                 if (!size)
 563                         continue;
 564
 565                 from = state->parts[p].from;
 566                 if (from >= get_capacity(disk)) {
 567                         printk(KERN_WARNING
 568                                "%s: p%d start %llu is beyond EOD, ",
 569                                disk->disk_name, p, (unsigned long long) from);
 570                         if (disk_unlock_native_capacity(disk))
 571                                 goto rescan;
 572                         continue;
 573                 }
 574
 575                 if (from + size > get_capacity(disk)) {
 576                         printk(KERN_WARNING
 577                                "%s: p%d size %llu extends beyond EOD, ",
 578                                disk->disk_name, p, (unsigned long long) size);
 579
 580                         if (disk_unlock_native_capacity(disk)) {
 581                                 /* free state and restart */
 582                                 goto rescan;
 583                         } else {
 584                                 /*
 585                                  * we can not ignore partitions of broken tables
 586                                  * created by for example camera firmware, but
 587                                  * we limit them to the end of the disk to avoid
 588                                  * creating invalid block devices
 589                                  */
 590                                 size = get_capacity(disk) - from;
 591                         }
 592                 }
 593
 594                 /*
 595                  * On a zoned block device, partitions should be aligned on the
 596                  * device zone size (i.e. zone boundary crossing not allowed).
 597                  * Otherwise, resetting the write pointer of the last zone of
 598                  * one partition may impact the following partition.
 599                  */
 600                 if (bdev_is_zoned(bdev) &&
 601                     !part_zone_aligned(disk, bdev, from, size)) {
 602                         printk(KERN_WARNING
 603                                "%s: p%d start %llu+%llu is not zone aligned\n",
 604                                disk->disk_name, p, (unsigned long long) from,
 605                                (unsigned long long) size);
 606                         continue;
 607                 }
 608
 609                 part = add_partition(disk, p, from, size,
 610                                      state->parts[p].flags,
 611                                      &state->parts[p].info);
 612                 if (IS_ERR(part)) {
 613                         printk(KERN_ERR " %s: p%d could not be added: %ld\n",
 614                                disk->disk_name, p, -PTR_ERR(part));
 615                         continue;
 616                 }
 617 #ifdef CONFIG_BLK_DEV_MD
 618                 if (state->parts[p].flags & ADDPART_FLAG_RAID)
 619                         md_autodetect_dev(part_to_dev(part)->devt);
 620 #endif
 621         }
 622         free_partitions(state);
 623         return 0;
 624 }
 625
 626 int invalidate_partitions(struct gendisk *disk, struct block_device *bdev)
 627 {
 628         int res;
 629
 630         if (!bdev->bd_invalidated)
 631                 return 0;
 632
 633         res = drop_partitions(disk, bdev);
 634         if (res)
 635                 return res;
 636
 637         set_capacity(disk, 0);
 638         check_disk_size_change(disk, bdev);
 639         bdev->bd_invalidated = 0;
 640         /* tell userspace that the media / partition table may have changed */
 641         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 642
 643         return 0;
 644 }
 645
 646 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
 647 {
 648         struct address_space *mapping = bdev->bd_inode->i_mapping;
 649         struct page *page;
 650
 651         page = read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_SHIFT-9)), NULL);
 652         if (!IS_ERR(page)) {
 653                 if (PageError(page))
 654                         goto fail;
 655                 p->v = page;
 656                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << 9);
 657 fail:
 658                 put_page(page);
 659         }
 660         p->v = NULL;
 661         return NULL;
 662 }
 663
 664 EXPORT_SYMBOL(read_dev_sector);