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 struct device_type part_type = {
 221         .name           = "partition",
 222         .groups         = part_attr_groups,
 223         .release        = part_release,
 224 };
 225
 226 static void delete_partition_rcu_cb(struct rcu_head *head)
 227 {
 228         struct hd_struct *part = container_of(head, struct hd_struct, rcu_head);
 229
 230         part->start_sect = 0;
 231         part->nr_sects = 0;
 232         part_stat_set_all(part, 0);
 233         put_device(part_to_dev(part));
 234 }
 235
 236 void __delete_partition(struct percpu_ref *ref)
 237 {
 238         struct hd_struct *part = container_of(ref, struct hd_struct, ref);
 239         call_rcu(&part->rcu_head, delete_partition_rcu_cb);
 240 }
 241
 242 void delete_partition(struct gendisk *disk, int partno)
 243 {
 244         struct disk_part_tbl *ptbl = disk->part_tbl;
 245         struct hd_struct *part;
 246
 247         if (partno >= ptbl->len)
 248                 return;
 249
 250         part = ptbl->part[partno];
 251         if (!part)
 252                 return;
 253
 254         rcu_assign_pointer(ptbl->part[partno], NULL);
 255         rcu_assign_pointer(ptbl->last_lookup, NULL);
 256         kobject_put(part->holder_dir);
 257         device_del(part_to_dev(part));
 258
 259         hd_struct_kill(part);
 260 }
 261
 262 static ssize_t whole_disk_show(struct device *dev,
 263                                struct device_attribute *attr, char *buf)
 264 {
 265         return 0;
 266 }
 267 static DEVICE_ATTR(whole_disk, S_IRUSR | S_IRGRP | S_IROTH,
 268                    whole_disk_show, NULL);
 269
 270 struct hd_struct *add_partition(struct gendisk *disk, int partno,
 271                                 sector_t start, sector_t len, int flags,
 272                                 struct partition_meta_info *info)
 273 {
 274         struct hd_struct *p;
 275         dev_t devt = MKDEV(0, 0);
 276         struct device *ddev = disk_to_dev(disk);
 277         struct device *pdev;
 278         struct disk_part_tbl *ptbl;
 279         const char *dname;
 280         int err;
 281
 282         err = disk_expand_part_tbl(disk, partno);
 283         if (err)
 284                 return ERR_PTR(err);
 285         ptbl = disk->part_tbl;
 286
 287         if (ptbl->part[partno])
 288                 return ERR_PTR(-EBUSY);
 289
 290         p = kzalloc(sizeof(*p), GFP_KERNEL);
 291         if (!p)
 292                 return ERR_PTR(-EBUSY);
 293
 294         if (!init_part_stats(p)) {
 295                 err = -ENOMEM;
 296                 goto out_free;
 297         }
 298
 299         seqcount_init(&p->nr_sects_seq);
 300         pdev = part_to_dev(p);
 301
 302         p->start_sect = start;
 303         p->alignment_offset =
 304                 queue_limit_alignment_offset(&disk->queue->limits, start);
 305         p->discard_alignment =
 306                 queue_limit_discard_alignment(&disk->queue->limits, start);
 307         p->nr_sects = len;
 308         p->partno = partno;
 309         p->policy = get_disk_ro(disk);
 310
 311         if (info) {
 312                 struct partition_meta_info *pinfo = alloc_part_info(disk);
 313                 if (!pinfo)
 314                         goto out_free_stats;
 315                 memcpy(pinfo, info, sizeof(*info));
 316                 p->info = pinfo;
 317         }
 318
 319         dname = dev_name(ddev);
 320         if (isdigit(dname[strlen(dname) - 1]))
 321                 dev_set_name(pdev, "%sp%d", dname, partno);
 322         else
 323                 dev_set_name(pdev, "%s%d", dname, partno);
 324
 325         device_initialize(pdev);
 326         pdev->class = &block_class;
 327         pdev->type = &part_type;
 328         pdev->parent = ddev;
 329
 330         err = blk_alloc_devt(p, &devt);
 331         if (err)
 332                 goto out_free_info;
 333         pdev->devt = devt;
 334
 335         /* delay uevent until 'holders' subdir is created */
 336         dev_set_uevent_suppress(pdev, 1);
 337         err = device_add(pdev);
 338         if (err)
 339                 goto out_put;
 340
 341         err = -ENOMEM;
 342         p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
 343         if (!p->holder_dir)
 344                 goto out_del;
 345
 346         dev_set_uevent_suppress(pdev, 0);
 347         if (flags & ADDPART_FLAG_WHOLEDISK) {
 348                 err = device_create_file(pdev, &dev_attr_whole_disk);
 349                 if (err)
 350                         goto out_del;
 351         }
 352
 353         /* everything is up and running, commence */
 354         rcu_assign_pointer(ptbl->part[partno], p);
 355
 356         /* suppress uevent if the disk suppresses it */
 357         if (!dev_get_uevent_suppress(ddev))
 358                 kobject_uevent(&pdev->kobj, KOBJ_ADD);
 359
 360         if (!hd_ref_init(p))
 361                 return p;
 362
 363 out_free_info:
 364         free_part_info(p);
 365 out_free_stats:
 366         free_part_stats(p);
 367 out_free:
 368         kfree(p);
 369         return ERR_PTR(err);
 370 out_del:
 371         kobject_put(p->holder_dir);
 372         device_del(pdev);
 373 out_put:
 374         put_device(pdev);
 375         blk_free_devt(devt);
 376         return ERR_PTR(err);
 377 }
 378
 379 static bool disk_unlock_native_capacity(struct gendisk *disk)
 380 {
 381         const struct block_device_operations *bdops = disk->fops;
 382
 383         if (bdops->unlock_native_capacity &&
 384             !(disk->flags & GENHD_FL_NATIVE_CAPACITY)) {
 385                 printk(KERN_CONT "enabling native capacity\n");
 386                 bdops->unlock_native_capacity(disk);
 387                 disk->flags |= GENHD_FL_NATIVE_CAPACITY;
 388                 return true;
 389         } else {
 390                 printk(KERN_CONT "truncated\n");
 391                 return false;
 392         }
 393 }
 394
 395 static int drop_partitions(struct gendisk *disk, struct block_device *bdev)
 396 {
 397         struct disk_part_iter piter;
 398         struct hd_struct *part;
 399         int res;
 400
 401         if (bdev->bd_part_count || bdev->bd_super)
 402                 return -EBUSY;
 403         res = invalidate_partition(disk, 0);
 404         if (res)
 405                 return res;
 406
 407         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 408         while ((part = disk_part_iter_next(&piter)))
 409                 delete_partition(disk, part->partno);
 410         disk_part_iter_exit(&piter);
 411
 412         return 0;
 413 }
 414
 415 int rescan_partitions(struct gendisk *disk, struct block_device *bdev)
 416 {
 417         struct parsed_partitions *state = NULL;
 418         struct hd_struct *part;
 419         int p, highest, res;
 420 rescan:
 421         if (state && !IS_ERR(state)) {
 422                 free_partitions(state);
 423                 state = NULL;
 424         }
 425
 426         res = drop_partitions(disk, bdev);
 427         if (res)
 428                 return res;
 429
 430         if (disk->fops->revalidate_disk)
 431                 disk->fops->revalidate_disk(disk);
 432         blk_integrity_revalidate(disk);
 433         check_disk_size_change(disk, bdev);
 434         bdev->bd_invalidated = 0;
 435         if (!get_capacity(disk) || !(state = check_partition(disk, bdev)))
 436                 return 0;
 437         if (IS_ERR(state)) {
 438                 /*
 439                  * I/O error reading the partition table.  If any
 440                  * partition code tried to read beyond EOD, retry
 441                  * after unlocking native capacity.
 442                  */
 443                 if (PTR_ERR(state) == -ENOSPC) {
 444                         printk(KERN_WARNING "%s: partition table beyond EOD, ",
 445                                disk->disk_name);
 446                         if (disk_unlock_native_capacity(disk))
 447                                 goto rescan;
 448                 }
 449                 return -EIO;
 450         }
 451         /*
 452          * If any partition code tried to read beyond EOD, try
 453          * unlocking native capacity even if partition table is
 454          * successfully read as we could be missing some partitions.
 455          */
 456         if (state->access_beyond_eod) {
 457                 printk(KERN_WARNING
 458                        "%s: partition table partially beyond EOD, ",
 459                        disk->disk_name);
 460                 if (disk_unlock_native_capacity(disk))
 461                         goto rescan;
 462         }
 463
 464         /* tell userspace that the media / partition table may have changed */
 465         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 466
 467         /* Detect the highest partition number and preallocate
 468          * disk->part_tbl.  This is an optimization and not strictly
 469          * necessary.
 470          */
 471         for (p = 1, highest = 0; p < state->limit; p++)
 472                 if (state->parts[p].size)
 473                         highest = p;
 474
 475         disk_expand_part_tbl(disk, highest);
 476
 477         /* add partitions */
 478         for (p = 1; p < state->limit; p++) {
 479                 sector_t size, from;
 480                 struct partition_meta_info *info = NULL;
 481
 482                 size = state->parts[p].size;
 483                 if (!size)
 484                         continue;
 485
 486                 from = state->parts[p].from;
 487                 if (from >= get_capacity(disk)) {
 488                         printk(KERN_WARNING
 489                                "%s: p%d start %llu is beyond EOD, ",
 490                                disk->disk_name, p, (unsigned long long) from);
 491                         if (disk_unlock_native_capacity(disk))
 492                                 goto rescan;
 493                         continue;
 494                 }
 495
 496                 if (from + size > get_capacity(disk)) {
 497                         printk(KERN_WARNING
 498                                "%s: p%d size %llu extends beyond EOD, ",
 499                                disk->disk_name, p, (unsigned long long) size);
 500
 501                         if (disk_unlock_native_capacity(disk)) {
 502                                 /* free state and restart */
 503                                 goto rescan;
 504                         } else {
 505                                 /*
 506                                  * we can not ignore partitions of broken tables
 507                                  * created by for example camera firmware, but
 508                                  * we limit them to the end of the disk to avoid
 509                                  * creating invalid block devices
 510                                  */
 511                                 size = get_capacity(disk) - from;
 512                         }
 513                 }
 514
 515                 if (state->parts[p].has_info)
 516                         info = &state->parts[p].info;
 517                 part = add_partition(disk, p, from, size,
 518                                      state->parts[p].flags,
 519                                      &state->parts[p].info);
 520                 if (IS_ERR(part)) {
 521                         printk(KERN_ERR " %s: p%d could not be added: %ld\n",
 522                                disk->disk_name, p, -PTR_ERR(part));
 523                         continue;
 524                 }
 525 #ifdef CONFIG_BLK_DEV_MD
 526                 if (state->parts[p].flags & ADDPART_FLAG_RAID)
 527                         md_autodetect_dev(part_to_dev(part)->devt);
 528 #endif
 529         }
 530         free_partitions(state);
 531         return 0;
 532 }
 533
 534 int invalidate_partitions(struct gendisk *disk, struct block_device *bdev)
 535 {
 536         int res;
 537
 538         if (!bdev->bd_invalidated)
 539                 return 0;
 540
 541         res = drop_partitions(disk, bdev);
 542         if (res)
 543                 return res;
 544
 545         set_capacity(disk, 0);
 546         check_disk_size_change(disk, bdev);
 547         bdev->bd_invalidated = 0;
 548         /* tell userspace that the media / partition table may have changed */
 549         kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
 550
 551         return 0;
 552 }
 553
 554 static struct page *read_pagecache_sector(struct block_device *bdev, sector_t n)
 555 {
 556         struct address_space *mapping = bdev->bd_inode->i_mapping;
 557
 558         return read_mapping_page(mapping, (pgoff_t)(n >> (PAGE_CACHE_SHIFT-9)),
 559                         NULL);
 560 }
 561
 562 unsigned char *read_dev_sector(struct block_device *bdev, sector_t n, Sector *p)
 563 {
 564         struct page *page;
 565
 566         /* don't populate page cache for dax capable devices */
 567         if (IS_DAX(bdev->bd_inode))
 568                 page = read_dax_sector(bdev, n);
 569         else
 570                 page = read_pagecache_sector(bdev, n);
 571
 572         if (!IS_ERR(page)) {
 573                 if (PageError(page))
 574                         goto fail;
 575                 p->v = page;
 576                 return (unsigned char *)page_address(page) +  ((n & ((1 << (PAGE_CACHE_SHIFT - 9)) - 1)) << 9);
 577 fail:
 578                 page_cache_release(page);
 579         }
 580         p->v = NULL;
 581         return NULL;
 582 }
 583
 584 EXPORT_SYMBOL(read_dev_sector);