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/dax.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         int cpu;
 117
 118         cpu = part_stat_lock();
 119         part_round_stats(cpu, p);
 120         part_stat_unlock();
 121         return sprintf(buf,
 122                 "%8lu %8lu %8llu %8u "
 123                 "%8lu %8lu %8llu %8u "
 124                 "%8u %8u %8u"
 125                 "\n",
 126                 part_stat_read(p, ios[READ]),
 127                 part_stat_read(p, merges[READ]),
 128                 (unsigned long long)part_stat_read(p, sectors[READ]),
 129                 jiffies_to_msecs(part_stat_read(p, ticks[READ])),
 130                 part_stat_read(p, ios[WRITE]),
 131                 part_stat_read(p, merges[WRITE]),
 132                 (unsigned long long)part_stat_read(p, sectors[WRITE]),
 133                 jiffies_to_msecs(part_stat_read(p, ticks[WRITE])),
 134                 part_in_flight(p),
 135                 jiffies_to_msecs(part_stat_read(p, io_ticks)),
 136                 jiffies_to_msecs(part_stat_read(p, time_in_queue)));
 137 }
 138
 139 ssize_t part_inflight_show(struct device *dev,
 140                         struct device_attribute *attr, char *buf)
 141 {
 142         struct hd_struct *p = dev_to_part(dev);
 143
 144         return sprintf(buf, "%8u %8u\n", atomic_read(&p->in_flight[0]),
 145                 atomic_read(&p->in_flight[1]));
 146 }
 147
 148 #ifdef CONFIG_FAIL_MAKE_REQUEST
 149 ssize_t part_fail_show(struct device *dev,
 150                        struct device_attribute *attr, char *buf)
 151 {
 152         struct hd_struct *p = dev_to_part(dev);
 153
 154         return sprintf(buf, "%d\n", p->make_it_fail);
 155 }
 156
 157 ssize_t part_fail_store(struct device *dev,
 158                         struct device_attribute *attr,
 159                         const char *buf, size_t count)
 160 {
 161         struct hd_struct *p = dev_to_part(dev);
 162         int i;
 163
 164         if (count > 0 && sscanf(buf, "%d", &i) > 0)
 165                 p->make_it_fail = (i == 0) ? 0 : 1;
 166
 167         return count;
 168 }
 169 #endif
 170
 171 static DEVICE_ATTR(partition, S_IRUGO, part_partition_show, NULL);
 172 static DEVICE_ATTR(start, S_IRUGO, part_start_show, NULL);
 173 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
 174 static DEVICE_ATTR(ro, S_IRUGO, part_ro_show, NULL);
 175 static DEVICE_ATTR(alignment_offset, S_IRUGO, part_alignment_offset_show, NULL);
 176 static DEVICE_ATTR(discard_alignment, S_IRUGO, part_discard_alignment_show,
 177                    NULL);
 178 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
 179 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
 180 #ifdef CONFIG_FAIL_MAKE_REQUEST
 181 static struct device_attribute dev_attr_fail =
 182         __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
 183 #endif
 184
 185 static struct attribute *part_attrs[] = {
 186         &dev_attr_partition.attr,
 187         &dev_attr_start.attr,
 188         &dev_attr_size.attr,
 189         &dev_attr_ro.attr,
 190         &dev_attr_alignment_offset.attr,
 191         &dev_attr_discard_alignment.attr,
 192         &dev_attr_stat.attr,
 193         &dev_attr_inflight.attr,
 194 #ifdef CONFIG_FAIL_MAKE_REQUEST
 195         &dev_attr_fail.attr,
 196 #endif
 197         NULL
 198 };
 199
 200 static struct attribute_group part_attr_group = {
 201         .attrs = part_attrs,
 202 };
 203
 204 static const struct attribute_group *part_attr_groups[] = {
 205         &part_attr_group,
 206 #ifdef CONFIG_BLK_DEV_IO_TRACE
 207         &blk_trace_attr_group,
 208 #endif
 209         NULL
 210 };
 211
 212 static void part_release(struct device *dev)
 213 {
 214         struct hd_struct *p = dev_to_part(dev);
 215         blk_free_devt(dev->devt);
 216         hd_free_part(p);
 217         kfree(p);
 218 }
 219
 220 static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
 221 {
 222         struct hd_struct *part = dev_to_part(dev);
 223
 224         add_uevent_var(env, "PARTN=%u", part->partno);
 225         if (part->info && part->info->volname[0])
 226                 add_uevent_var(env, "PARTNAME=%s", part->info->volname);
 227         return 0;
 228 }
 229
 230 struct device_type part_type = {
 231         .name           = "partition",
 232         .groups         = part_attr_groups,
 233         .release        = part_release,
 234         .uevent         = part_uevent,
 235 };
 236
 237 static void delete_partition_rcu_cb(struct rcu_head *head)
 238 {
 239         struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);
 240
 241         part->start_sect = 0;
 242         part->nr_sects = 0;
 243         part_stat_set_all(part, 0);
 244         put_device(part_to_dev(part));
 245 }
 246
 247 void __delete_partition(struct percpu_ref *ref)
 248 {
 249         struct hd_struct *part = container_of(ref, struct hd_struct, ref);
 250         call_rcu(&part->rcu_head, delete_partition_rcu_cb);
 251 }
 252
 253 void delete_partition(struct gendisk *disk, int partno)
 254 {
 255         struct disk_part_tbl *ptbl = disk->part_tbl;
 256         struct hd_struct *part;
 257
 258         if (partno >= ptbl->len)
 259                 return;
 260
 261         part = ptbl->part[partno];
 262         if (!part)
 263                 return;
 264
 265         rcu_assign_pointer(ptbl->part[partno], NULL);
 266         rcu_assign_pointer(ptbl->last_lookup, NULL);
 267         kobject_put(part->holder_dir);
 268         device_del(part_to_dev(part));
 269
 270         hd_struct_kill(part);
 271 }
 272
 273 static ssize_t whole_disk_show(struct device *dev,
 274                                struct device_attribute *attr, char *buf)
 275 {
 276         return 0;
 277 }
 278 static DEVICE_ATTR(whole_disk, S_IRUSR | S_IRGRP | S_IROTH,
 279                    whole_disk_show, NULL);
 280
 281 struct hd_struct *add_partition(struct gendisk *disk, int partno,
 282                                 sector_t start, sector_t len, int flags,
 283                                 struct partition_meta_info *info)
 284 {
 285         struct hd_struct *p;
 286         dev_t devt = MKDEV(0, 0);
 287         struct device *ddev = disk_to_dev(disk);
 288         struct device *pdev;
 289         struct disk_part_tbl *ptbl;
 290         const char *dname;
 291         int err;
 292
 293         err = disk_expand_part_tbl(disk, partno);
 294         if (err)
 295                 return ERR_PTR(err);
 296         ptbl = disk->part_tbl;
 297
 298         if (ptbl->part[partno])
 299                 return ERR_PTR(-EBUSY);
 300
 301         p = kzalloc(sizeof(*p), GFP_KERNEL);
 302         if (!p)
 303                 return ERR_PTR(-EBUSY);
 304
 305         if (!init_part_stats(p)) {
 306                 err = -ENOMEM;
 307                 goto out_free;
 308         }
 309
 310         seqcount_init(&p->nr_sects_seq);
 311         pdev = part_to_dev(p);
 312
 313         p->start_sect = start;
 314         p->alignment_offset =
 315                 queue_limit_alignment_offset(&disk->queue->limits, start);
 316         p->discard_alignment =
 317                 queue_limit_discard_alignment(&disk->queue->limits, start);
 318         p->nr_sects = len;
 319         p->partno = partno;
 320         p->policy = get_disk_ro(disk);
 321
 322         if (info) {
 323                 struct partition_meta_info *pinfo = alloc_part_info(disk);
 324                 if (!pinfo)
 325                         goto out_free_stats;
 326                 memcpy(pinfo, info, sizeof(*info));
 327                 p->info = pinfo;
 328         }
 329
 330         dname = dev_name(ddev);
 331         if (isdigit(dname[strlen(dname) - 1]))
 332                 dev_set_name(pdev, "%sp%d", dname, partno);
 333         else
 334                 dev_set_name(pdev, "%s%d", dname, partno);
 335
 336         device_initialize(pdev);
 337         pdev->class = &block_class;
 338         pdev->type = &part_type;
 339         pdev->parent = ddev;
 340
 341         err = blk_alloc_devt(p, &devt);
 342         if (err)
 343                 goto out_free_info;
 344         pdev->devt = devt;
 345
 346         /* delay uevent until 'holders' subdir is created */
 347         dev_set_uevent_suppress(pdev, 1);
 348         err = device_add(pdev);
 349         if (err)
 350                 goto out_put;
 351
 352         err = -ENOMEM;
 353         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
 354         if (!p->holder_dir)
 355                 goto out_del;
 356
 357         dev_set_uevent_suppress(pdev, 0);
 358         if (flags & ADDPART_FLAG_WHOLEDISK) {
 359                 err = device_create_file(pdev, &dev_attr_whole_disk);
 360                 if (err)
 361                         goto out_del;
 362         }
 363
 364         err = hd_ref_init(p);
 365         if (err) {
 366                 if (flags & ADDPART_FLAG_WHOLEDISK)
 367                         goto out_remove_file;
 368                 goto out_del;
 369         }
 370
 371         /* everything is up and running, commence */
 372         rcu_assign_pointer(ptbl->part[partno], p);
 373
 374         /* suppress uevent if the disk suppresses it */
 375         if (!dev_get_uevent_suppress(ddev))
 376                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 377         return p;
 378
 379 out_free_info:
 380         free_part_info(p);
 381 out_free_stats:
 382         free_part_stats(p);
 383 out_free:
 384         kfree(p);
 385         return ERR_PTR(err);
 386 out_remove_file:
 387         device_remove_file(pdev, &dev_attr_whole_disk);
 388 out_del:
 389         kobject_put(p->holder_dir);
 390         device_del(pdev);
 391 out_put:
 392         put_device(pdev);
 393         blk_free_devt(devt);
 394         return ERR_PTR(err);
 395 }
 396
 397 static bool disk_unlock_native_capacity(struct gendisk *disk)
 398 {
 399         const struct block_device_operations *bdops = disk->fops;
 400
 401         if (bdops->unlock_native_capacity &&
 402             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
 403                 printk(KERN_CONT "enabling native capacity\n");
 404                 bdops->unlock_native_capacity(disk);
 405                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
 406                 return true;
 407         } else {
 408                 printk(KERN_CONT "truncated\n");
 409                 return false;
 410         }
 411 }
 412
 413 static int drop_partitions(struct gendisk *disk, struct block_device *bdev)
 414 {
 415         struct disk_part_iter piter;
 416         struct hd_struct *part;
 417         int res;
 418
 419         if (bdev->bd_part_count || bdev->bd_super)
 420                 return -EBUSY;
 421         res = invalidate_partition(disk, 0);
 422         if (res)
 423                 return res;
 424
 425         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 426         while ((part = disk_part_iter_next(&piter)))
 427                 delete_partition(disk, part->partno);
 428         disk_part_iter_exit(&piter);
 429
 430         return 0;
 431 }
 432
 433 static bool part_zone_aligned(struct gendisk *disk,
 434                               struct block_device *bdev,
 435                               sector_t from, sector_t size)
 436 {
 437         unsigned int zone_size = bdev_zone_size(bdev);
 438
 439         /*
 440          * If this function is called, then the disk is a zoned block device
 441          * (host-aware or host-managed). This can be detected even if the
 442          * zoned block device support is disabled (CONFIG_BLK_DEV_ZONED not
 443          * set). In this case, however, only host-aware devices will be seen
 444          * as a block device is not created for host-managed devices. Without
 445          * zoned block device support, host-aware drives can still be used as
 446          * regular block devices (no zone operation) and their zone size will
 447          * be reported as 0. Allow this case.
 448          */
 449         if (!zone_size)
 450                 return true;
 451
 452         /*
 453          * Check partition start and size alignement. If the drive has a
 454          * smaller last runt zone, ignore it and allow the partition to
 455          * use it. Check the zone size too: it should be a power of 2 number
 456          * of sectors.
 457          */
 458         if (WARN_ON_ONCE(!is_power_of_2(zone_size))) {
 459                 u32 rem;
 460
 461                 div_u64_rem(from, zone_size, &rem);
 462                 if (rem)
 463                         return false;
 464                 if ((from + size) < get_capacity(disk)) {
 465                         div_u64_rem(size, zone_size, &rem);
 466                         if (rem)
 467                                 return false;
 468                 }
 469
 470         } else {
 471
 472                 if (from & (zone_size - 1))
 473                         return false;
 474                 if ((from + size) < get_capacity(disk) &&
 475                     (size & (zone_size - 1)))
 476                         return false;
 477
 478         }
 479
 480         return true;
 481 }
 482
 483 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 484 {
 485         struct parsed_partitions *state = NULL;
 486         struct hd_struct *part;
 487         int p, highest, res;
 488 rescan:
 489         if (state && !IS_ERR(state)) {
 490                 free_partitions(state);
 491                 state = NULL;
 492         }
 493
 494         res = drop_partitions(disk, bdev);
 495         if (res)
 496                 return res;
 497
 498         if (disk->fops->revalidate_disk)
 499                 disk->fops->revalidate_disk(disk);
 500         blk_integrity_revalidate(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 static struct page *read_pagecache_sector(struct block_device *bdev, sector_t n)
 635 {
 636         struct address_space *mapping = bdev->bd_inode->i_mapping;
 637
 638         return read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_SHIFT-9)),
 639                                  NULL);
 640 }
 641
 642 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
 643 {
 644         struct page *page;
 645
 646         /* don't populate page cache for dax capable devices */
 647         if (IS_DAX(bdev->bd_inode))
 648                 page = read_dax_sector(bdev, n);
 649         else
 650                 page = read_pagecache_sector(bdev, n);
 651
 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);