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