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 const char *bio_devname(struct bio *bio, char *buf)
  55 {
  56         return disk_name(bio->bi_disk, bio->bi_partno, buf);
  57 }
  58 EXPORT_SYMBOL(bio_devname);
  59
  60 /*
  61  * There's very little reason to use this, you should really
  62  * have a struct block_device just about everywhere and use
  63  * bdevname() instead.
  64  */
  65 const char *__bdevname(dev_t dev, char *buffer)
  66 {
  67         scnprintf(buffer, BDEVNAME_SIZE, "unknown-block(%u,%u)",
  68                                 MAJOR(dev), MINOR(dev));
  69         return buffer;
  70 }
  71
  72 EXPORT_SYMBOL(__bdevname);
  73
  74 static ssize_t part_partition_show(struct device *dev,
  75                                    struct device_attribute *attr, char *buf)
  76 {
  77         struct hd_struct *p = dev_to_part(dev);
  78
  79         return sprintf(buf, "%d\n", p->partno);
  80 }
  81
  82 static ssize_t part_start_show(struct device *dev,
  83                                struct device_attribute *attr, char *buf)
  84 {
  85         struct hd_struct *p = dev_to_part(dev);
  86
  87         return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
  88 }
  89
  90 ssize_t part_size_show(struct device *dev,
  91                        struct device_attribute *attr, char *buf)
  92 {
  93         struct hd_struct *p = dev_to_part(dev);
  94         return sprintf(buf, "%llu\n",(unsigned long long)part_nr_sects_read(p));
  95 }
  96
  97 static ssize_t part_ro_show(struct device *dev,
  98                             struct device_attribute *attr, char *buf)
  99 {
 100         struct hd_struct *p = dev_to_part(dev);
 101         return sprintf(buf, "%d\n", p->policy ? 1 : 0);
 102 }
 103
 104 static ssize_t part_alignment_offset_show(struct device *dev,
 105                                           struct device_attribute *attr, char *buf)
 106 {
 107         struct hd_struct *p = dev_to_part(dev);
 108         return sprintf(buf, "%llu\n", (unsigned long long)p->alignment_offset);
 109 }
 110
 111 static ssize_t part_discard_alignment_show(struct device *dev,
 112                                            struct device_attribute *attr, char *buf)
 113 {
 114         struct hd_struct *p = dev_to_part(dev);
 115         return sprintf(buf, "%u\n", p->discard_alignment);
 116 }
 117
 118 ssize_t part_stat_show(struct device *dev,
 119                        struct device_attribute *attr, char *buf)
 120 {
 121         struct hd_struct *p = dev_to_part(dev);
 122         struct request_queue *q = part_to_disk(p)->queue;
 123         unsigned int inflight[2];
 124         int cpu;
 125
 126         cpu = part_stat_lock();
 127         part_round_stats(q, cpu, p);
 128         part_stat_unlock();
 129         part_in_flight(q, p, inflight);
 130         return sprintf(buf,
 131                 "%8lu %8lu %8llu %8u "
 132                 "%8lu %8lu %8llu %8u "
 133                 "%8u %8u %8u"
 134                 "\n",
 135                 part_stat_read(p, ios[READ]),
 136                 part_stat_read(p, merges[READ]),
 137                 (unsigned long long)part_stat_read(p, sectors[READ]),
 138                 jiffies_to_msecs(part_stat_read(p, ticks[READ])),
 139                 part_stat_read(p, ios[WRITE]),
 140                 part_stat_read(p, merges[WRITE]),
 141                 (unsigned long long)part_stat_read(p, sectors[WRITE]),
 142                 jiffies_to_msecs(part_stat_read(p, ticks[WRITE])),
 143                 inflight[0],
 144                 jiffies_to_msecs(part_stat_read(p, io_ticks)),
 145                 jiffies_to_msecs(part_stat_read(p, time_in_queue)));
 146 }
 147
 148 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
 149                            char *buf)
 150 {
 151         struct hd_struct *p = dev_to_part(dev);
 152         struct request_queue *q = part_to_disk(p)->queue;
 153         unsigned int inflight[2];
 154
 155         part_in_flight_rw(q, p, inflight);
 156         return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
 157 }
 158
 159 #ifdef CONFIG_FAIL_MAKE_REQUEST
 160 ssize_t part_fail_show(struct device *dev,
 161                        struct device_attribute *attr, char *buf)
 162 {
 163         struct hd_struct *p = dev_to_part(dev);
 164
 165         return sprintf(buf, "%d\n", p->make_it_fail);
 166 }
 167
 168 ssize_t part_fail_store(struct device *dev,
 169                         struct device_attribute *attr,
 170                         const char *buf, size_t count)
 171 {
 172         struct hd_struct *p = dev_to_part(dev);
 173         int i;
 174
 175         if (count > 0 && sscanf(buf, "%d", &i) > 0)
 176                 p->make_it_fail = (i == 0) ? 0 : 1;
 177
 178         return count;
 179 }
 180 #endif
 181
 182 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
 183 static DEVICE_ATTR(start, 0444, part_start_show, NULL);
 184 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
 185 static DEVICE_ATTR(ro, 0444, part_ro_show, NULL);
 186 static DEVICE_ATTR(alignment_offset, 0444, part_alignment_offset_show, NULL);
 187 static DEVICE_ATTR(discard_alignment, 0444, part_discard_alignment_show, NULL);
 188 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
 189 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
 190 #ifdef CONFIG_FAIL_MAKE_REQUEST
 191 static struct device_attribute dev_attr_fail =
 192         __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
 193 #endif
 194
 195 static struct attribute *part_attrs[] = {
 196         &dev_attr_partition.attr,
 197         &dev_attr_start.attr,
 198         &dev_attr_size.attr,
 199         &dev_attr_ro.attr,
 200         &dev_attr_alignment_offset.attr,
 201         &dev_attr_discard_alignment.attr,
 202         &dev_attr_stat.attr,
 203         &dev_attr_inflight.attr,
 204 #ifdef CONFIG_FAIL_MAKE_REQUEST
 205         &dev_attr_fail.attr,
 206 #endif
 207         NULL
 208 };
 209
 210 static struct attribute_group part_attr_group = {
 211         .attrs = part_attrs,
 212 };
 213
 214 static const struct attribute_group *part_attr_groups[] = {
 215         &part_attr_group,
 216 #ifdef CONFIG_BLK_DEV_IO_TRACE
 217         &blk_trace_attr_group,
 218 #endif
 219         NULL
 220 };
 221
 222 static void part_release(struct device *dev)
 223 {
 224         struct hd_struct *p = dev_to_part(dev);
 225         blk_free_devt(dev->devt);
 226         hd_free_part(p);
 227         kfree(p);
 228 }
 229
 230 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
 231 {
 232         struct hd_struct *part = dev_to_part(dev);
 233
 234         add_uevent_var(env, "PARTN=%u", part->partno);
 235         if (part->info && part->info->volname[0])
 236                 add_uevent_var(env, "PARTNAME=%s", part->info->volname);
 237         return 0;
 238 }
 239
 240 struct device_type part_type = {
 241         .name           = "partition",
 242         .groups         = part_attr_groups,
 243         .release        = part_release,
 244         .uevent         = part_uevent,
 245 };
 246
 247 static void delete_partition_rcu_cb(struct rcu_head *head)
 248 {
 249         struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);
 250
 251         part->start_sect = 0;
 252         part->nr_sects = 0;
 253         part_stat_set_all(part, 0);
 254         put_device(part_to_dev(part));
 255 }
 256
 257 void __delete_partition(struct percpu_ref *ref)
 258 {
 259         struct hd_struct *part = container_of(ref, struct hd_struct, ref);
 260         call_rcu(&part->rcu_head, delete_partition_rcu_cb);
 261 }
 262
 263 /*
 264  * Must be called either with bd_mutex held, before a disk can be opened or
 265  * after all disk users are gone.
 266  */
 267 void delete_partition(struct gendisk *disk, int partno)
 268 {
 269         struct disk_part_tbl *ptbl =
 270                 rcu_dereference_protected(disk->part_tbl, 1);
 271         struct hd_struct *part;
 272
 273         if (partno >= ptbl->len)
 274                 return;
 275
 276         part = rcu_dereference_protected(ptbl->part[partno], 1);
 277         if (!part)
 278                 return;
 279
 280         rcu_assign_pointer(ptbl->part[partno], NULL);
 281         rcu_assign_pointer(ptbl->last_lookup, NULL);
 282         kobject_put(part->holder_dir);
 283         device_del(part_to_dev(part));
 284
 285         hd_struct_kill(part);
 286 }
 287
 288 static ssize_t whole_disk_show(struct device *dev,
 289                                struct device_attribute *attr, char *buf)
 290 {
 291         return 0;
 292 }
 293 static DEVICE_ATTR(whole_disk, 0444, whole_disk_show, NULL);
 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 struct hd_struct *add_partition(struct gendisk *disk, int partno,
 300                                 sector_t start, sector_t len, int flags,
 301                                 struct partition_meta_info *info)
 302 {
 303         struct hd_struct *p;
 304         dev_t devt = MKDEV(0, 0);
 305         struct device *ddev = disk_to_dev(disk);
 306         struct device *pdev;
 307         struct disk_part_tbl *ptbl;
 308         const char *dname;
 309         int err;
 310
 311         err = disk_expand_part_tbl(disk, partno);
 312         if (err)
 313                 return ERR_PTR(err);
 314         ptbl = rcu_dereference_protected(disk->part_tbl, 1);
 315
 316         if (ptbl->part[partno])
 317                 return ERR_PTR(-EBUSY);
 318
 319         p = kzalloc(sizeof(*p), GFP_KERNEL);
 320         if (!p)
 321                 return ERR_PTR(-EBUSY);
 322
 323         if (!init_part_stats(p)) {
 324                 err = -ENOMEM;
 325                 goto out_free;
 326         }
 327
 328         seqcount_init(&p->nr_sects_seq);
 329         pdev = part_to_dev(p);
 330
 331         p->start_sect = start;
 332         p->alignment_offset =
 333                 queue_limit_alignment_offset(&disk->queue->limits, start);
 334         p->discard_alignment =
 335                 queue_limit_discard_alignment(&disk->queue->limits, start);
 336         p->nr_sects = len;
 337         p->partno = partno;
 338         p->policy = get_disk_ro(disk);
 339
 340         if (info) {
 341                 struct partition_meta_info *pinfo = alloc_part_info(disk);
 342                 if (!pinfo) {
 343                         err = -ENOMEM;
 344                         goto out_free_stats;
 345                 }
 346                 memcpy(pinfo, info, sizeof(*info));
 347                 p->info = pinfo;
 348         }
 349
 350         dname = dev_name(ddev);
 351         if (isdigit(dname[strlen(dname) - 1]))
 352                 dev_set_name(pdev, "%sp%d", dname, partno);
 353         else
 354                 dev_set_name(pdev, "%s%d", dname, partno);
 355
 356         device_initialize(pdev);
 357         pdev->class = &block_class;
 358         pdev->type = &part_type;
 359         pdev->parent = ddev;
 360
 361         err = blk_alloc_devt(p, &devt);
 362         if (err)
 363                 goto out_free_info;
 364         pdev->devt = devt;
 365
 366         /* delay uevent until 'holders' subdir is created */
 367         dev_set_uevent_suppress(pdev, 1);
 368         err = device_add(pdev);
 369         if (err)
 370                 goto out_put;
 371
 372         err = -ENOMEM;
 373         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
 374         if (!p->holder_dir)
 375                 goto out_del;
 376
 377         dev_set_uevent_suppress(pdev, 0);
 378         if (flags & ADDPART_FLAG_WHOLEDISK) {
 379                 err = device_create_file(pdev, &dev_attr_whole_disk);
 380                 if (err)
 381                         goto out_del;
 382         }
 383
 384         err = hd_ref_init(p);
 385         if (err) {
 386                 if (flags & ADDPART_FLAG_WHOLEDISK)
 387                         goto out_remove_file;
 388                 goto out_del;
 389         }
 390
 391         /* everything is up and running, commence */
 392         rcu_assign_pointer(ptbl->part[partno], p);
 393
 394         /* suppress uevent if the disk suppresses it */
 395         if (!dev_get_uevent_suppress(ddev))
 396                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 397         return p;
 398
 399 out_free_info:
 400         free_part_info(p);
 401 out_free_stats:
 402         free_part_stats(p);
 403 out_free:
 404         kfree(p);
 405         return ERR_PTR(err);
 406 out_remove_file:
 407         device_remove_file(pdev, &dev_attr_whole_disk);
 408 out_del:
 409         kobject_put(p->holder_dir);
 410         device_del(pdev);
 411 out_put:
 412         put_device(pdev);
 413         return ERR_PTR(err);
 414 }
 415
 416 static bool disk_unlock_native_capacity(struct gendisk *disk)
 417 {
 418         const struct block_device_operations *bdops = disk->fops;
 419
 420         if (bdops->unlock_native_capacity &&
 421             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
 422                 printk(KERN_CONT "enabling native capacity\n");
 423                 bdops->unlock_native_capacity(disk);
 424                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
 425                 return true;
 426         } else {
 427                 printk(KERN_CONT "truncated\n");
 428                 return false;
 429         }
 430 }
 431
 432 static int drop_partitions(struct gendisk *disk, struct block_device *bdev)
 433 {
 434         struct disk_part_iter piter;
 435         struct hd_struct *part;
 436         int res;
 437
 438         if (bdev->bd_part_count || bdev->bd_super)
 439                 return -EBUSY;
 440         res = invalidate_partition(disk, 0);
 441         if (res)
 442                 return res;
 443
 444         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 445         while ((part = disk_part_iter_next(&piter)))
 446                 delete_partition(disk, part->partno);
 447         disk_part_iter_exit(&piter);
 448
 449         return 0;
 450 }
 451
 452 static bool part_zone_aligned(struct gendisk *disk,
 453                               struct block_device *bdev,
 454                               sector_t from, sector_t size)
 455 {
 456         unsigned int zone_sectors = bdev_zone_sectors(bdev);
 457
 458         /*
 459          * If this function is called, then the disk is a zoned block device
 460          * (host-aware or host-managed). This can be detected even if the
 461          * zoned block device support is disabled (CONFIG_BLK_DEV_ZONED not
 462          * set). In this case, however, only host-aware devices will be seen
 463          * as a block device is not created for host-managed devices. Without
 464          * zoned block device support, host-aware drives can still be used as
 465          * regular block devices (no zone operation) and their zone size will
 466          * be reported as 0. Allow this case.
 467          */
 468         if (!zone_sectors)
 469                 return true;
 470
 471         /*
 472          * Check partition start and size alignement. If the drive has a
 473          * smaller last runt zone, ignore it and allow the partition to
 474          * use it. Check the zone size too: it should be a power of 2 number
 475          * of sectors.
 476          */
 477         if (WARN_ON_ONCE(!is_power_of_2(zone_sectors))) {
 478                 u32 rem;
 479
 480                 div_u64_rem(from, zone_sectors, &rem);
 481                 if (rem)
 482                         return false;
 483                 if ((from + size) < get_capacity(disk)) {
 484                         div_u64_rem(size, zone_sectors, &rem);
 485                         if (rem)
 486                                 return false;
 487                 }
 488
 489         } else {
 490
 491                 if (from & (zone_sectors - 1))
 492                         return false;
 493                 if ((from + size) < get_capacity(disk) &&
 494                     (size & (zone_sectors - 1)))
 495                         return false;
 496
 497         }
 498
 499         return true;
 500 }
 501
 502 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 503 {
 504         struct parsed_partitions *state = NULL;
 505         struct hd_struct *part;
 506         int p, highest, res;
 507 rescan:
 508         if (state && !IS_ERR(state)) {
 509                 free_partitions(state);
 510                 state = NULL;
 511         }
 512
 513         res = drop_partitions(disk, bdev);
 514         if (res)
 515                 return res;
 516
 517         if (disk->fops->revalidate_disk)
 518                 disk->fops->revalidate_disk(disk);
 519         check_disk_size_change(disk, bdev, true);
 520         bdev->bd_invalidated = 0;
 521         if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
 522                 return 0;
 523         if (IS_ERR(state)) {
 524                 /*
 525                  * I/O error reading the partition table.  If any
 526                  * partition code tried to read beyond EOD, retry
 527                  * after unlocking native capacity.
 528                  */
 529                 if (PTR_ERR(state) == -ENOSPC) {
 530                         printk(KERN_WARNING "%s: partition table beyond EOD, ",
 531                                disk->disk_name);
 532                         if (disk_unlock_native_capacity(disk))
 533                                 goto rescan;
 534                 }
 535                 return -EIO;
 536         }
 537         /*
 538          * If any partition code tried to read beyond EOD, try
 539          * unlocking native capacity even if partition table is
 540          * successfully read as we could be missing some partitions.
 541          */
 542         if (state->access_beyond_eod) {
 543                 printk(KERN_WARNING
 544                        "%s: partition table partially beyond EOD, ",
 545                        disk->disk_name);
 546                 if (disk_unlock_native_capacity(disk))
 547                         goto rescan;
 548         }
 549
 550         /* tell userspace that the media / partition table may have changed */
 551         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 552
 553         /* Detect the highest partition number and preallocate
 554          * disk->part_tbl.  This is an optimization and not strictly
 555          * necessary.
 556          */
 557         for (p = 1, highest = 0; p < state->limit; p++)
 558                 if (state->parts[p].size)
 559                         highest = p;
 560
 561         disk_expand_part_tbl(disk, highest);
 562
 563         /* add partitions */
 564         for (p = 1; p < state->limit; p++) {
 565                 sector_t size, from;
 566
 567                 size = state->parts[p].size;
 568                 if (!size)
 569                         continue;
 570
 571                 from = state->parts[p].from;
 572                 if (from >= get_capacity(disk)) {
 573                         printk(KERN_WARNING
 574                                "%s: p%d start %llu is beyond EOD, ",
 575                                disk->disk_name, p, (unsigned long long) from);
 576                         if (disk_unlock_native_capacity(disk))
 577                                 goto rescan;
 578                         continue;
 579                 }
 580
 581                 if (from + size > get_capacity(disk)) {
 582                         printk(KERN_WARNING
 583                                "%s: p%d size %llu extends beyond EOD, ",
 584                                disk->disk_name, p, (unsigned long long) size);
 585
 586                         if (disk_unlock_native_capacity(disk)) {
 587                                 /* free state and restart */
 588                                 goto rescan;
 589                         } else {
 590                                 /*
 591                                  * we can not ignore partitions of broken tables
 592                                  * created by for example camera firmware, but
 593                                  * we limit them to the end of the disk to avoid
 594                                  * creating invalid block devices
 595                                  */
 596                                 size = get_capacity(disk) - from;
 597                         }
 598                 }
 599
 600                 /*
 601                  * On a zoned block device, partitions should be aligned on the
 602                  * device zone size (i.e. zone boundary crossing not allowed).
 603                  * Otherwise, resetting the write pointer of the last zone of
 604                  * one partition may impact the following partition.
 605                  */
 606                 if (bdev_is_zoned(bdev) &&
 607                     !part_zone_aligned(disk, bdev, from, size)) {
 608                         printk(KERN_WARNING
 609                                "%s: p%d start %llu+%llu is not zone aligned\n",
 610                                disk->disk_name, p, (unsigned long long) from,
 611                                (unsigned long long) size);
 612                         continue;
 613                 }
 614
 615                 part = add_partition(disk, p, from, size,
 616                                      state->parts[p].flags,
 617                                      &state->parts[p].info);
 618                 if (IS_ERR(part)) {
 619                         printk(KERN_ERR " %s: p%d could not be added: %ld\n",
 620                                disk->disk_name, p, -PTR_ERR(part));
 621                         continue;
 622                 }
 623 #ifdef CONFIG_BLK_DEV_MD
 624                 if (state->parts[p].flags & ADDPART_FLAG_RAID)
 625                         md_autodetect_dev(part_to_dev(part)->devt);
 626 #endif
 627         }
 628         free_partitions(state);
 629         return 0;
 630 }
 631
 632 int invalidate_partitions(struct gendisk *disk, struct block_device *bdev)
 633 {
 634         int res;
 635
 636         if (!bdev->bd_invalidated)
 637                 return 0;
 638
 639         res = drop_partitions(disk, bdev);
 640         if (res)
 641                 return res;
 642
 643         set_capacity(disk, 0);
 644         check_disk_size_change(disk, bdev, false);
 645         bdev->bd_invalidated = 0;
 646         /* tell userspace that the media / partition table may have changed */
 647         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 648
 649         return 0;
 650 }
 651
 652 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
 653 {
 654         struct address_space *mapping = bdev->bd_inode->i_mapping;
 655         struct page *page;
 656
 657         page = read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_SHIFT-9)), NULL);
 658         if (!IS_ERR(page)) {
 659                 if (PageError(page))
 660                         goto fail;
 661                 p->v = page;
 662                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_SHIFT - 9)) - 1)) << 9);
 663 fail:
 664                 put_page(page);
 665         }
 666         p->v = NULL;
 667         return NULL;
 668 }
 669
 670 EXPORT_SYMBOL(read_dev_sector);