fs/partitions/check.c

   1 /*
   2  *  fs/partitions/check.c
   3  *
   4  *  Code extracted from drivers/block/genhd.c
   5  *  Copyright (C) 1991-1998  Linus Torvalds
   6  *  Re-organised Feb 1998 Russell King
   7  *
   8  *  We now have independent partition support from the
   9  *  block drivers, which allows all the partition code to
  10  *  be grouped in one location, and it to be mostly self
  11  *  contained.
  12  *
  13  *  Added needed MAJORS for new pairs, {hdi,hdj}, {hdk,hdl}
  14  */
  15
  16 #include <linux/init.h>
  17 #include <linux/module.h>
  18 #include <linux/fs.h>
  19 #include <linux/kmod.h>
  20 #include <linux/ctype.h>
  21 #include <linux/genhd.h>
  22
  23 #include "check.h"
  24
  25 #include "acorn.h"
  26 #include "amiga.h"
  27 #include "atari.h"
  28 #include "ldm.h"
  29 #include "mac.h"
  30 #include "msdos.h"
  31 #include "osf.h"
  32 #include "sgi.h"
  33 #include "sun.h"
  34 #include "ibm.h"
  35 #include "ultrix.h"
  36 #include "efi.h"
  37 #include "karma.h"
  38 #include "sysv68.h"
  39
  40 #ifdef CONFIG_BLK_DEV_MD
  41 extern void md_autodetect_dev(dev_t dev);
  42 #endif
  43
  44 int warn_no_part = 1; /*This is ugly: should make genhd removable media aware*/
  45
  46 static int (*check_part[])(struct parsed_partitions *, struct block_device *) = {
  47         /*
  48          * Probe partition formats with tables at disk address 0
  49          * that also have an ADFS boot block at 0xdc0.
  50          */
  51 #ifdef CONFIG_ACORN_PARTITION_ICS
  52         adfspart_check_ICS,
  53 #endif
  54 #ifdef CONFIG_ACORN_PARTITION_POWERTEC
  55         adfspart_check_POWERTEC,
  56 #endif
  57 #ifdef CONFIG_ACORN_PARTITION_EESOX
  58         adfspart_check_EESOX,
  59 #endif
  60
  61         /*
  62          * Now move on to formats that only have partition info at
  63          * disk address 0xdc0.  Since these may also have stale
  64          * PC/BIOS partition tables, they need to come before
  65          * the msdos entry.
  66          */
  67 #ifdef CONFIG_ACORN_PARTITION_CUMANA
  68         adfspart_check_CUMANA,
  69 #endif
  70 #ifdef CONFIG_ACORN_PARTITION_ADFS
  71         adfspart_check_ADFS,
  72 #endif
  73
  74 #ifdef CONFIG_EFI_PARTITION
  75         efi_partition,          /* this must come before msdos */
  76 #endif
  77 #ifdef CONFIG_SGI_PARTITION
  78         sgi_partition,
  79 #endif
  80 #ifdef CONFIG_LDM_PARTITION
  81         ldm_partition,          /* this must come before msdos */
  82 #endif
  83 #ifdef CONFIG_MSDOS_PARTITION
  84         msdos_partition,
  85 #endif
  86 #ifdef CONFIG_OSF_PARTITION
  87         osf_partition,
  88 #endif
  89 #ifdef CONFIG_SUN_PARTITION
  90         sun_partition,
  91 #endif
  92 #ifdef CONFIG_AMIGA_PARTITION
  93         amiga_partition,
  94 #endif
  95 #ifdef CONFIG_ATARI_PARTITION
  96         atari_partition,
  97 #endif
  98 #ifdef CONFIG_MAC_PARTITION
  99         mac_partition,
 100 #endif
 101 #ifdef CONFIG_ULTRIX_PARTITION
 102         ultrix_partition,
 103 #endif
 104 #ifdef CONFIG_IBM_PARTITION
 105         ibm_partition,
 106 #endif
 107 #ifdef CONFIG_KARMA_PARTITION
 108         karma_partition,
 109 #endif
 110 #ifdef CONFIG_SYSV68_PARTITION
 111         sysv68_partition,
 112 #endif
 113         NULL
 114 };
 115
 116 /*
 117  * disk_name() is used by partition check code and the genhd driver.
 118  * It formats the devicename of the indicated disk into
 119  * the supplied buffer (of size at least 32), and returns
 120  * a pointer to that same buffer (for convenience).
 121  */
 122
 123 char *disk_name(struct gendisk *hd, int partno, char *buf)
 124 {
 125         if (!partno)
 126                 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
 127         else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
 128                 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
 129         else
 130                 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
 131
 132         return buf;
 133 }
 134
 135 const char *bdevname(struct block_device *bdev, char *buf)
 136 {
 137         return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
 138 }
 139
 140 EXPORT_SYMBOL(bdevname);
 141
 142 /*
 143  * There's very little reason to use this, you should really
 144  * have a struct block_device just about everywhere and use
 145  * bdevname() instead.
 146  */
 147 const char *__bdevname(dev_t dev, char *buffer)
 148 {
 149         scnprintf(buffer, BDEVNAME_SIZE, "unknown-block(%u,%u)",
 150                                 MAJOR(dev), MINOR(dev));
 151         return buffer;
 152 }
 153
 154 EXPORT_SYMBOL(__bdevname);
 155
 156 static struct parsed_partitions *
 157 check_partition(struct gendisk *hd, struct block_device *bdev)
 158 {
 159         struct parsed_partitions *state;
 160         int i, res, err;
 161
 162         state = kmalloc(sizeof(struct parsed_partitions), GFP_KERNEL);
 163         if (!state)
 164                 return NULL;
 165
 166         disk_name(hd, 0, state->name);
 167         printk(KERN_INFO " %s:", state->name);
 168         if (isdigit(state->name[strlen(state->name)-1]))
 169                 sprintf(state->name, "p");
 170
 171         state->limit = disk_max_parts(hd);
 172         i = res = err = 0;
 173         while (!res && check_part[i]) {
 174                 memset(&state->parts, 0, sizeof(state->parts));
 175                 res = check_part[i++](state, bdev);
 176                 if (res < 0) {
 177                         /* We have hit an I/O error which we don't report now.
 178                         * But record it, and let the others do their job.
 179                         */
 180                         err = res;
 181                         res = 0;
 182                 }
 183
 184         }
 185         if (res > 0)
 186                 return state;
 187         if (err)
 188         /* The partition is unrecognized. So report I/O errors if there were any */
 189                 res = err;
 190         if (!res)
 191                 printk(" unknown partition table\n");
 192         else if (warn_no_part)
 193                 printk(" unable to read partition table\n");
 194         kfree(state);
 195         return ERR_PTR(res);
 196 }
 197
 198 static ssize_t part_partition_show(struct device *dev,
 199                                    struct device_attribute *attr, char *buf)
 200 {
 201         struct hd_struct *p = dev_to_part(dev);
 202
 203         return sprintf(buf, "%d\n", p->partno);
 204 }
 205
 206 static ssize_t part_start_show(struct device *dev,
 207                                struct device_attribute *attr, char *buf)
 208 {
 209         struct hd_struct *p = dev_to_part(dev);
 210
 211         return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
 212 }
 213
 214 ssize_t part_size_show(struct device *dev,
 215                        struct device_attribute *attr, char *buf)
 216 {
 217         struct hd_struct *p = dev_to_part(dev);
 218         return sprintf(buf, "%llu\n",(unsigned long long)p->nr_sects);
 219 }
 220
 221 ssize_t part_stat_show(struct device *dev,
 222                        struct device_attribute *attr, char *buf)
 223 {
 224         struct hd_struct *p = dev_to_part(dev);
 225         int cpu;
 226
 227         cpu = part_stat_lock();
 228         part_round_stats(cpu, p);
 229         part_stat_unlock();
 230         return sprintf(buf,
 231                 "%8lu %8lu %8llu %8u "
 232                 "%8lu %8lu %8llu %8u "
 233                 "%8u %8u %8u"
 234                 "\n",
 235                 part_stat_read(p, ios[READ]),
 236                 part_stat_read(p, merges[READ]),
 237                 (unsigned long long)part_stat_read(p, sectors[READ]),
 238                 jiffies_to_msecs(part_stat_read(p, ticks[READ])),
 239                 part_stat_read(p, ios[WRITE]),
 240                 part_stat_read(p, merges[WRITE]),
 241                 (unsigned long long)part_stat_read(p, sectors[WRITE]),
 242                 jiffies_to_msecs(part_stat_read(p, ticks[WRITE])),
 243                 p->in_flight,
 244                 jiffies_to_msecs(part_stat_read(p, io_ticks)),
 245                 jiffies_to_msecs(part_stat_read(p, time_in_queue)));
 246 }
 247
 248 #ifdef CONFIG_FAIL_MAKE_REQUEST
 249 ssize_t part_fail_show(struct device *dev,
 250                        struct device_attribute *attr, char *buf)
 251 {
 252         struct hd_struct *p = dev_to_part(dev);
 253
 254         return sprintf(buf, "%d\n", p->make_it_fail);
 255 }
 256
 257 ssize_t part_fail_store(struct device *dev,
 258                         struct device_attribute *attr,
 259                         const char *buf, size_t count)
 260 {
 261         struct hd_struct *p = dev_to_part(dev);
 262         int i;
 263
 264         if (count > 0 && sscanf(buf, "%d", &i) > 0)
 265                 p->make_it_fail = (i == 0) ? 0 : 1;
 266
 267         return count;
 268 }
 269 #endif
 270
 271 static DEVICE_ATTR(partition, S_IRUGO, part_partition_show, NULL);
 272 static DEVICE_ATTR(start, S_IRUGO, part_start_show, NULL);
 273 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
 274 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
 275 #ifdef CONFIG_FAIL_MAKE_REQUEST
 276 static struct device_attribute dev_attr_fail =
 277         __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
 278 #endif
 279
 280 static struct attribute *part_attrs[] = {
 281         &dev_attr_partition.attr,
 282         &dev_attr_start.attr,
 283         &dev_attr_size.attr,
 284         &dev_attr_stat.attr,
 285 #ifdef CONFIG_FAIL_MAKE_REQUEST
 286         &dev_attr_fail.attr,
 287 #endif
 288         NULL
 289 };
 290
 291 static struct attribute_group part_attr_group = {
 292         .attrs = part_attrs,
 293 };
 294
 295 static struct attribute_group *part_attr_groups[] = {
 296         &part_attr_group,
 297         NULL
 298 };
 299
 300 static void part_release(struct device *dev)
 301 {
 302         struct hd_struct *p = dev_to_part(dev);
 303         free_part_stats(p);
 304         kfree(p);
 305 }
 306
 307 struct device_type part_type = {
 308         .name           = "partition",
 309         .groups         = part_attr_groups,
 310         .release        = part_release,
 311 };
 312
 313 static void delete_partition_rcu_cb(struct rcu_head *head)
 314 {
 315         struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);
 316
 317         part->start_sect = 0;
 318         part->nr_sects = 0;
 319         part_stat_set_all(part, 0);
 320         put_device(part_to_dev(part));
 321 }
 322
 323 void delete_partition(struct gendisk *disk, int partno)
 324 {
 325         struct disk_part_tbl *ptbl = disk->part_tbl;
 326         struct hd_struct *part;
 327
 328         if (partno >= ptbl->len)
 329                 return;
 330
 331         part = ptbl->part[partno];
 332         if (!part)
 333                 return;
 334
 335         blk_free_devt(part_devt(part));
 336         rcu_assign_pointer(ptbl->part[partno], NULL);
 337         rcu_assign_pointer(ptbl->last_lookup, NULL);
 338         kobject_put(part->holder_dir);
 339         device_del(part_to_dev(part));
 340
 341         call_rcu(&part->rcu_head, delete_partition_rcu_cb);
 342 }
 343
 344 static ssize_t whole_disk_show(struct device *dev,
 345                                struct device_attribute *attr, char *buf)
 346 {
 347         return 0;
 348 }
 349 static DEVICE_ATTR(whole_disk, S_IRUSR | S_IRGRP | S_IROTH,
 350                    whole_disk_show, NULL);
 351
 352 struct hd_struct *add_partition(struct gendisk *disk, int partno,
 353                                 sector_t start, sector_t len, int flags)
 354 {
 355         struct hd_struct *p;
 356         dev_t devt = MKDEV(0, 0);
 357         struct device *ddev = disk_to_dev(disk);
 358         struct device *pdev;
 359         struct disk_part_tbl *ptbl;
 360         const char *dname;
 361         int err;
 362
 363         err = disk_expand_part_tbl(disk, partno);
 364         if (err)
 365                 return ERR_PTR(err);
 366         ptbl = disk->part_tbl;
 367
 368         if (ptbl->part[partno])
 369                 return ERR_PTR(-EBUSY);
 370
 371         p = kzalloc(sizeof(*p), GFP_KERNEL);
 372         if (!p)
 373                 return ERR_PTR(-EBUSY);
 374
 375         if (!init_part_stats(p)) {
 376                 err = -ENOMEM;
 377                 goto out_free;
 378         }
 379         pdev = part_to_dev(p);
 380
 381         p->start_sect = start;
 382         p->nr_sects = len;
 383         p->partno = partno;
 384         p->policy = get_disk_ro(disk);
 385
 386         dname = dev_name(ddev);
 387         if (isdigit(dname[strlen(dname) - 1]))
 388                 dev_set_name(pdev, "%sp%d", dname, partno);
 389         else
 390                 dev_set_name(pdev, "%s%d", dname, partno);
 391
 392         device_initialize(pdev);
 393         pdev->class = &block_class;
 394         pdev->type = &part_type;
 395         pdev->parent = ddev;
 396
 397         err = blk_alloc_devt(p, &devt);
 398         if (err)
 399                 goto out_free_stats;
 400         pdev->devt = devt;
 401
 402         /* delay uevent until 'holders' subdir is created */
 403         dev_set_uevent_suppress(pdev, 1);
 404         err = device_add(pdev);
 405         if (err)
 406                 goto out_put;
 407
 408         err = -ENOMEM;
 409         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
 410         if (!p->holder_dir)
 411                 goto out_del;
 412
 413         dev_set_uevent_suppress(pdev, 0);
 414         if (flags & ADDPART_FLAG_WHOLEDISK) {
 415                 err = device_create_file(pdev, &dev_attr_whole_disk);
 416                 if (err)
 417                         goto out_del;
 418         }
 419
 420         /* everything is up and running, commence */
 421         INIT_RCU_HEAD(&p->rcu_head);
 422         rcu_assign_pointer(ptbl->part[partno], p);
 423
 424         /* suppress uevent if the disk supresses it */
 425         if (!dev_get_uevent_suppress(pdev))
 426                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 427
 428         return p;
 429
 430 out_free_stats:
 431         free_part_stats(p);
 432 out_free:
 433         kfree(p);
 434         return ERR_PTR(err);
 435 out_del:
 436         kobject_put(p->holder_dir);
 437         device_del(pdev);
 438 out_put:
 439         put_device(pdev);
 440         blk_free_devt(devt);
 441         return ERR_PTR(err);
 442 }
 443
 444 /* Not exported, helper to add_disk(). */
 445 void register_disk(struct gendisk *disk)
 446 {
 447         struct device *ddev = disk_to_dev(disk);
 448         struct block_device *bdev;
 449         struct disk_part_iter piter;
 450         struct hd_struct *part;
 451         int err;
 452
 453         ddev->parent = disk->driverfs_dev;
 454
 455         dev_set_name(ddev, disk->disk_name);
 456
 457         /* delay uevents, until we scanned partition table */
 458         dev_set_uevent_suppress(ddev, 1);
 459
 460         if (device_add(ddev))
 461                 return;
 462 #ifndef CONFIG_SYSFS_DEPRECATED
 463         err = sysfs_create_link(block_depr, &ddev->kobj,
 464                                 kobject_name(&ddev->kobj));
 465         if (err) {
 466                 device_del(ddev);
 467                 return;
 468         }
 469 #endif
 470         disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
 471         disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
 472
 473         /* No minors to use for partitions */
 474         if (!disk_partitionable(disk))
 475                 goto exit;
 476
 477         /* No such device (e.g., media were just removed) */
 478         if (!get_capacity(disk))
 479                 goto exit;
 480
 481         bdev = bdget_disk(disk, 0);
 482         if (!bdev)
 483                 goto exit;
 484
 485         bdev->bd_invalidated = 1;
 486         err = blkdev_get(bdev, FMODE_READ);
 487         if (err < 0)
 488                 goto exit;
 489         blkdev_put(bdev, FMODE_READ);
 490
 491 exit:
 492         /* announce disk after possible partitions are created */
 493         dev_set_uevent_suppress(ddev, 0);
 494         kobject_uevent(&ddev->kobj, KOBJ_ADD);
 495
 496         /* announce possible partitions */
 497         disk_part_iter_init(&piter, disk, 0);
 498         while ((part = disk_part_iter_next(&piter)))
 499                 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
 500         disk_part_iter_exit(&piter);
 501 }
 502
 503 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 504 {
 505         struct disk_part_iter piter;
 506         struct hd_struct *part;
 507         struct parsed_partitions *state;
 508         int p, highest, res;
 509
 510         if (bdev->bd_part_count)
 511                 return -EBUSY;
 512         res = invalidate_partition(disk, 0);
 513         if (res)
 514                 return res;
 515
 516         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 517         while ((part = disk_part_iter_next(&piter)))
 518                 delete_partition(disk, part->partno);
 519         disk_part_iter_exit(&piter);
 520
 521         if (disk->fops->revalidate_disk)
 522                 disk->fops->revalidate_disk(disk);
 523         check_disk_size_change(disk, bdev);
 524         bdev->bd_invalidated = 0;
 525         if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
 526                 return 0;
 527         if (IS_ERR(state))      /* I/O error reading the partition table */
 528                 return -EIO;
 529
 530         /* tell userspace that the media / partition table may have changed */
 531         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 532
 533         /* Detect the highest partition number and preallocate
 534          * disk->part_tbl.  This is an optimization and not strictly
 535          * necessary.
 536          */
 537         for (p = 1, highest = 0; p < state->limit; p++)
 538                 if (state->parts[p].size)
 539                         highest = p;
 540
 541         disk_expand_part_tbl(disk, highest);
 542
 543         /* add partitions */
 544         for (p = 1; p < state->limit; p++) {
 545                 sector_t size = state->parts[p].size;
 546                 sector_t from = state->parts[p].from;
 547                 if (!size)
 548                         continue;
 549                 if (from >= get_capacity(disk)) {
 550                         printk(KERN_WARNING
 551                                "%s: p%d ignored, start %llu is behind the end of the disk\n",
 552                                disk->disk_name, p, (unsigned long long) from);
 553                         continue;
 554                 }
 555                 if (from + size > get_capacity(disk)) {
 556                         /*
 557                          * we can not ignore partitions of broken tables
 558                          * created by for example camera firmware, but we
 559                          * limit them to the end of the disk to avoid
 560                          * creating invalid block devices
 561                          */
 562                         printk(KERN_WARNING
 563                                "%s: p%d size %llu limited to end of disk\n",
 564                                disk->disk_name, p, (unsigned long long) size);
 565                         size = get_capacity(disk) - from;
 566                 }
 567                 part = add_partition(disk, p, from, size,
 568                                      state->parts[p].flags);
 569                 if (IS_ERR(part)) {
 570                         printk(KERN_ERR " %s: p%d could not be added: %ld\n",
 571                                disk->disk_name, p, -PTR_ERR(part));
 572                         continue;
 573                 }
 574 #ifdef CONFIG_BLK_DEV_MD
 575                 if (state->parts[p].flags & ADDPART_FLAG_RAID)
 576                         md_autodetect_dev(part_to_dev(part)->devt);
 577 #endif
 578         }
 579         kfree(state);
 580         return 0;
 581 }
 582
 583 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
 584 {
 585         struct address_space *mapping = bdev->bd_inode->i_mapping;
 586         struct page *page;
 587
 588         page = read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_CACHE_SHIFT-9)),
 589                                  NULL);
 590         if (!IS_ERR(page)) {
 591                 if (PageError(page))
 592                         goto fail;
 593                 p->v = page;
 594                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_CACHE_SHIFT - 9)) - 1)) << 9);
 595 fail:
 596                 page_cache_release(page);
 597         }
 598         p->v = NULL;
 599         return NULL;
 600 }
 601
 602 EXPORT_SYMBOL(read_dev_sector);
 603
 604 void del_gendisk(struct gendisk *disk)
 605 {
 606         struct disk_part_iter piter;
 607         struct hd_struct *part;
 608
 609         /* invalidate stuff */
 610         disk_part_iter_init(&piter, disk,
 611                              DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
 612         while ((part = disk_part_iter_next(&piter))) {
 613                 invalidate_partition(disk, part->partno);
 614                 delete_partition(disk, part->partno);
 615         }
 616         disk_part_iter_exit(&piter);
 617
 618         invalidate_partition(disk, 0);
 619         blk_free_devt(disk_to_dev(disk)->devt);
 620         set_capacity(disk, 0);
 621         disk->flags &= ~GENHD_FL_UP;
 622         unlink_gendisk(disk);
 623         part_stat_set_all(&disk->part0, 0);
 624         disk->part0.stamp = 0;
 625
 626         kobject_put(disk->part0.holder_dir);
 627         kobject_put(disk->slave_dir);
 628         disk->driverfs_dev = NULL;
 629 #ifndef CONFIG_SYSFS_DEPRECATED
 630         sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
 631 #endif
 632         device_del(disk_to_dev(disk));
 633 }