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         /* everything is up and running, commence */
 365         rcu_assign_pointer(ptbl->part[partno], p);
 366
 367         /* suppress uevent if the disk suppresses it */
 368         if (!dev_get_uevent_suppress(ddev))
 369                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 370
 371         if (!hd_ref_init(p))
 372                 return p;
 373
 374 out_free_info:
 375         free_part_info(p);
 376 out_free_stats:
 377         free_part_stats(p);
 378 out_free:
 379         kfree(p);
 380         return ERR_PTR(err);
 381 out_del:
 382         kobject_put(p->holder_dir);
 383         device_del(pdev);
 384 out_put:
 385         put_device(pdev);
 386         blk_free_devt(devt);
 387         return ERR_PTR(err);
 388 }
 389
 390 static bool disk_unlock_native_capacity(struct gendisk *disk)
 391 {
 392         const struct block_device_operations *bdops = disk->fops;
 393
 394         if (bdops->unlock_native_capacity &&
 395             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
 396                 printk(KERN_CONT "enabling native capacity\n");
 397                 bdops->unlock_native_capacity(disk);
 398                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
 399                 return true;
 400         } else {
 401                 printk(KERN_CONT "truncated\n");
 402                 return false;
 403         }
 404 }
 405
 406 static int drop_partitions(struct gendisk *disk, struct block_device *bdev)
 407 {
 408         struct disk_part_iter piter;
 409         struct hd_struct *part;
 410         int res;
 411
 412         if (bdev->bd_part_count || bdev->bd_super)
 413                 return -EBUSY;
 414         res = invalidate_partition(disk, 0);
 415         if (res)
 416                 return res;
 417
 418         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 419         while ((part = disk_part_iter_next(&piter)))
 420                 delete_partition(disk, part->partno);
 421         disk_part_iter_exit(&piter);
 422
 423         return 0;
 424 }
 425
 426 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 427 {
 428         struct parsed_partitions *state = NULL;
 429         struct hd_struct *part;
 430         int p, highest, res;
 431 rescan:
 432         if (state && !IS_ERR(state)) {
 433                 free_partitions(state);
 434                 state = NULL;
 435         }
 436
 437         res = drop_partitions(disk, bdev);
 438         if (res)
 439                 return res;
 440
 441         if (disk->fops->revalidate_disk)
 442                 disk->fops->revalidate_disk(disk);
 443         blk_integrity_revalidate(disk);
 444         check_disk_size_change(disk, bdev);
 445         bdev->bd_invalidated = 0;
 446         if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
 447                 return 0;
 448         if (IS_ERR(state)) {
 449                 /*
 450                  * I/O error reading the partition table.  If any
 451                  * partition code tried to read beyond EOD, retry
 452                  * after unlocking native capacity.
 453                  */
 454                 if (PTR_ERR(state) == -ENOSPC) {
 455                         printk(KERN_WARNING "%s: partition table beyond EOD, ",
 456                                disk->disk_name);
 457                         if (disk_unlock_native_capacity(disk))
 458                                 goto rescan;
 459                 }
 460                 return -EIO;
 461         }
 462         /*
 463          * If any partition code tried to read beyond EOD, try
 464          * unlocking native capacity even if partition table is
 465          * successfully read as we could be missing some partitions.
 466          */
 467         if (state->access_beyond_eod) {
 468                 printk(KERN_WARNING
 469                        "%s: partition table partially beyond EOD, ",
 470                        disk->disk_name);
 471                 if (disk_unlock_native_capacity(disk))
 472                         goto rescan;
 473         }
 474
 475         /* tell userspace that the media / partition table may have changed */
 476         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 477
 478         /* Detect the highest partition number and preallocate
 479          * disk->part_tbl.  This is an optimization and not strictly
 480          * necessary.
 481          */
 482         for (p = 1, highest = 0; p < state->limit; p++)
 483                 if (state->parts[p].size)
 484                         highest = p;
 485
 486         disk_expand_part_tbl(disk, highest);
 487
 488         /* add partitions */
 489         for (p = 1; p < state->limit; p++) {
 490                 sector_t size, from;
 491                 struct partition_meta_info *info = NULL;
 492
 493                 size = state->parts[p].size;
 494                 if (!size)
 495                         continue;
 496
 497                 from = state->parts[p].from;
 498                 if (from >= get_capacity(disk)) {
 499                         printk(KERN_WARNING
 500                                "%s: p%d start %llu is beyond EOD, ",
 501                                disk->disk_name, p, (unsigned long long) from);
 502                         if (disk_unlock_native_capacity(disk))
 503                                 goto rescan;
 504                         continue;
 505                 }
 506
 507                 if (from + size > get_capacity(disk)) {
 508                         printk(KERN_WARNING
 509                                "%s: p%d size %llu extends beyond EOD, ",
 510                                disk->disk_name, p, (unsigned long long) size);
 511
 512                         if (disk_unlock_native_capacity(disk)) {
 513                                 /* free state and restart */
 514                                 goto rescan;
 515                         } else {
 516                                 /*
 517                                  * we can not ignore partitions of broken tables
 518                                  * created by for example camera firmware, but
 519                                  * we limit them to the end of the disk to avoid
 520                                  * creating invalid block devices
 521                                  */
 522                                 size = get_capacity(disk) - from;
 523                         }
 524                 }
 525
 526                 if (state->parts[p].has_info)
 527                         info = &state->parts[p].info;
 528                 part = add_partition(disk, p, from, size,
 529                                      state->parts[p].flags,
 530                                      &state->parts[p].info);
 531                 if (IS_ERR(part)) {
 532                         printk(KERN_ERR " %s: p%d could not be added: %ld\n",
 533                                disk->disk_name, p, -PTR_ERR(part));
 534                         continue;
 535                 }
 536 #ifdef CONFIG_BLK_DEV_MD
 537                 if (state->parts[p].flags & ADDPART_FLAG_RAID)
 538                         md_autodetect_dev(part_to_dev(part)->devt);
 539 #endif
 540         }
 541         free_partitions(state);
 542         return 0;
 543 }
 544
 545 int invalidate_partitions(struct gendisk *disk, struct block_device *bdev)
 546 {
 547         int res;
 548
 549         if (!bdev->bd_invalidated)
 550                 return 0;
 551
 552         res = drop_partitions(disk, bdev);
 553         if (res)
 554                 return res;
 555
 556         set_capacity(disk, 0);
 557         check_disk_size_change(disk, bdev);
 558         bdev->bd_invalidated = 0;
 559         /* tell userspace that the media / partition table may have changed */
 560         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 561
 562         return 0;
 563 }
 564
 565 static struct page *read_pagecache_sector(struct block_device *bdev, sector_t n)
 566 {
 567         struct address_space *mapping = bdev->bd_inode->i_mapping;
 568
 569         return read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_CACHE_SHIFT-9)),
 570                         NULL);
 571 }
 572
 573 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
 574 {
 575         struct page *page;
 576
 577         /* don't populate page cache for dax capable devices */
 578         if (IS_DAX(bdev->bd_inode))
 579                 page = read_dax_sector(bdev, n);
 580         else
 581                 page = read_pagecache_sector(bdev, n);
 582
 583         if (!IS_ERR(page)) {
 584                 if (PageError(page))
 585                         goto fail;
 586                 p->v = page;
 587                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_CACHE_SHIFT - 9)) - 1)) << 9);
 588 fail:
 589                 page_cache_release(page);
 590         }
 591         p->v = NULL;
 592         return NULL;
 593 }
 594
 595 EXPORT_SYMBOL(read_dev_sector);