block/partition-generic.c

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