block/genhd.c

   1 /*
   2  *  gendisk handling
   3  */
   4
   5 #include <linux/module.h>
   6 #include <linux/fs.h>
   7 #include <linux/genhd.h>
   8 #include <linux/kdev_t.h>
   9 #include <linux/kernel.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/init.h>
  12 #include <linux/spinlock.h>
  13 #include <linux/proc_fs.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/slab.h>
  16 #include <linux/kmod.h>
  17 #include <linux/kobj_map.h>
  18 #include <linux/buffer_head.h>
  19 #include <linux/mutex.h>
  20 #include <linux/idr.h>
  21
  22 #include "blk.h"
  23
  24 static DEFINE_MUTEX(block_class_lock);
  25 #ifndef CONFIG_SYSFS_DEPRECATED
  26 struct kobject *block_depr;
  27 #endif
  28
  29 /* for extended dynamic devt allocation, currently only one major is used */
  30 #define MAX_EXT_DEVT            (1 << MINORBITS)
  31
  32 /* For extended devt allocation.  ext_devt_mutex prevents look up
  33  * results from going away underneath its user.
  34  */
  35 static DEFINE_MUTEX(ext_devt_mutex);
  36 static DEFINE_IDR(ext_devt_idr);
  37
  38 static struct device_type disk_type;
  39
  40 /**
  41  * disk_get_part - get partition
  42  * @disk: disk to look partition from
  43  * @partno: partition number
  44  *
  45  * Look for partition @partno from @disk.  If found, increment
  46  * reference count and return it.
  47  *
  48  * CONTEXT:
  49  * Don't care.
  50  *
  51  * RETURNS:
  52  * Pointer to the found partition on success, NULL if not found.
  53  */
  54 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
  55 {
  56         struct hd_struct *part = NULL;
  57         struct disk_part_tbl *ptbl;
  58
  59         if (unlikely(partno < 0))
  60                 return NULL;
  61
  62         rcu_read_lock();
  63
  64         ptbl = rcu_dereference(disk->part_tbl);
  65         if (likely(partno < ptbl->len)) {
  66                 part = rcu_dereference(ptbl->part[partno]);
  67                 if (part)
  68                         get_device(part_to_dev(part));
  69         }
  70
  71         rcu_read_unlock();
  72
  73         return part;
  74 }
  75 EXPORT_SYMBOL_GPL(disk_get_part);
  76
  77 /**
  78  * disk_part_iter_init - initialize partition iterator
  79  * @piter: iterator to initialize
  80  * @disk: disk to iterate over
  81  * @flags: DISK_PITER_* flags
  82  *
  83  * Initialize @piter so that it iterates over partitions of @disk.
  84  *
  85  * CONTEXT:
  86  * Don't care.
  87  */
  88 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
  89                           unsigned int flags)
  90 {
  91         struct disk_part_tbl *ptbl;
  92
  93         rcu_read_lock();
  94         ptbl = rcu_dereference(disk->part_tbl);
  95
  96         piter->disk = disk;
  97         piter->part = NULL;
  98
  99         if (flags & DISK_PITER_REVERSE)
 100                 piter->idx = ptbl->len - 1;
 101         else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
 102                 piter->idx = 0;
 103         else
 104                 piter->idx = 1;
 105
 106         piter->flags = flags;
 107
 108         rcu_read_unlock();
 109 }
 110 EXPORT_SYMBOL_GPL(disk_part_iter_init);
 111
 112 /**
 113  * disk_part_iter_next - proceed iterator to the next partition and return it
 114  * @piter: iterator of interest
 115  *
 116  * Proceed @piter to the next partition and return it.
 117  *
 118  * CONTEXT:
 119  * Don't care.
 120  */
 121 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
 122 {
 123         struct disk_part_tbl *ptbl;
 124         int inc, end;
 125
 126         /* put the last partition */
 127         disk_put_part(piter->part);
 128         piter->part = NULL;
 129
 130         /* get part_tbl */
 131         rcu_read_lock();
 132         ptbl = rcu_dereference(piter->disk->part_tbl);
 133
 134         /* determine iteration parameters */
 135         if (piter->flags & DISK_PITER_REVERSE) {
 136                 inc = -1;
 137                 if (piter->flags & (DISK_PITER_INCL_PART0 |
 138                                     DISK_PITER_INCL_EMPTY_PART0))
 139                         end = -1;
 140                 else
 141                         end = 0;
 142         } else {
 143                 inc = 1;
 144                 end = ptbl->len;
 145         }
 146
 147         /* iterate to the next partition */
 148         for (; piter->idx != end; piter->idx += inc) {
 149                 struct hd_struct *part;
 150
 151                 part = rcu_dereference(ptbl->part[piter->idx]);
 152                 if (!part)
 153                         continue;
 154                 if (!part->nr_sects &&
 155                     !(piter->flags & DISK_PITER_INCL_EMPTY) &&
 156                     !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
 157                       piter->idx == 0))
 158                         continue;
 159
 160                 get_device(part_to_dev(part));
 161                 piter->part = part;
 162                 piter->idx += inc;
 163                 break;
 164         }
 165
 166         rcu_read_unlock();
 167
 168         return piter->part;
 169 }
 170 EXPORT_SYMBOL_GPL(disk_part_iter_next);
 171
 172 /**
 173  * disk_part_iter_exit - finish up partition iteration
 174  * @piter: iter of interest
 175  *
 176  * Called when iteration is over.  Cleans up @piter.
 177  *
 178  * CONTEXT:
 179  * Don't care.
 180  */
 181 void disk_part_iter_exit(struct disk_part_iter *piter)
 182 {
 183         disk_put_part(piter->part);
 184         piter->part = NULL;
 185 }
 186 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
 187
 188 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
 189 {
 190         return part->start_sect <= sector &&
 191                 sector < part->start_sect + part->nr_sects;
 192 }
 193
 194 /**
 195  * disk_map_sector_rcu - map sector to partition
 196  * @disk: gendisk of interest
 197  * @sector: sector to map
 198  *
 199  * Find out which partition @sector maps to on @disk.  This is
 200  * primarily used for stats accounting.
 201  *
 202  * CONTEXT:
 203  * RCU read locked.  The returned partition pointer is valid only
 204  * while preemption is disabled.
 205  *
 206  * RETURNS:
 207  * Found partition on success, part0 is returned if no partition matches
 208  */
 209 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
 210 {
 211         struct disk_part_tbl *ptbl;
 212         struct hd_struct *part;
 213         int i;
 214
 215         ptbl = rcu_dereference(disk->part_tbl);
 216
 217         part = rcu_dereference(ptbl->last_lookup);
 218         if (part && sector_in_part(part, sector))
 219                 return part;
 220
 221         for (i = 1; i < ptbl->len; i++) {
 222                 part = rcu_dereference(ptbl->part[i]);
 223
 224                 if (part && sector_in_part(part, sector)) {
 225                         rcu_assign_pointer(ptbl->last_lookup, part);
 226                         return part;
 227                 }
 228         }
 229         return &disk->part0;
 230 }
 231 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
 232
 233 /*
 234  * Can be deleted altogether. Later.
 235  *
 236  */
 237 static struct blk_major_name {
 238         struct blk_major_name *next;
 239         int major;
 240         char name[16];
 241 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
 242
 243 /* index in the above - for now: assume no multimajor ranges */
 244 static inline int major_to_index(int major)
 245 {
 246         return major % BLKDEV_MAJOR_HASH_SIZE;
 247 }
 248
 249 #ifdef CONFIG_PROC_FS
 250 void blkdev_show(struct seq_file *seqf, off_t offset)
 251 {
 252         struct blk_major_name *dp;
 253
 254         if (offset < BLKDEV_MAJOR_HASH_SIZE) {
 255                 mutex_lock(&block_class_lock);
 256                 for (dp = major_names[offset]; dp; dp = dp->next)
 257                         seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
 258                 mutex_unlock(&block_class_lock);
 259         }
 260 }
 261 #endif /* CONFIG_PROC_FS */
 262
 263 /**
 264  * register_blkdev - register a new block device
 265  *
 266  * @major: the requested major device number [1..255]. If @major=0, try to
 267  *         allocate any unused major number.
 268  * @name: the name of the new block device as a zero terminated string
 269  *
 270  * The @name must be unique within the system.
 271  *
 272  * The return value depends on the @major input parameter.
 273  *  - if a major device number was requested in range [1..255] then the
 274  *    function returns zero on success, or a negative error code
 275  *  - if any unused major number was requested with @major=0 parameter
 276  *    then the return value is the allocated major number in range
 277  *    [1..255] or a negative error code otherwise
 278  */
 279 int register_blkdev(unsigned int major, const char *name)
 280 {
 281         struct blk_major_name **n, *p;
 282         int index, ret = 0;
 283
 284         mutex_lock(&block_class_lock);
 285
 286         /* temporary */
 287         if (major == 0) {
 288                 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
 289                         if (major_names[index] == NULL)
 290                                 break;
 291                 }
 292
 293                 if (index == 0) {
 294                         printk("register_blkdev: failed to get major for %s\n",
 295                                name);
 296                         ret = -EBUSY;
 297                         goto out;
 298                 }
 299                 major = index;
 300                 ret = major;
 301         }
 302
 303         p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
 304         if (p == NULL) {
 305                 ret = -ENOMEM;
 306                 goto out;
 307         }
 308
 309         p->major = major;
 310         strlcpy(p->name, name, sizeof(p->name));
 311         p->next = NULL;
 312         index = major_to_index(major);
 313
 314         for (n = &major_names[index]; *n; n = &(*n)->next) {
 315                 if ((*n)->major == major)
 316                         break;
 317         }
 318         if (!*n)
 319                 *n = p;
 320         else
 321                 ret = -EBUSY;
 322
 323         if (ret < 0) {
 324                 printk("register_blkdev: cannot get major %d for %s\n",
 325                        major, name);
 326                 kfree(p);
 327         }
 328 out:
 329         mutex_unlock(&block_class_lock);
 330         return ret;
 331 }
 332
 333 EXPORT_SYMBOL(register_blkdev);
 334
 335 void unregister_blkdev(unsigned int major, const char *name)
 336 {
 337         struct blk_major_name **n;
 338         struct blk_major_name *p = NULL;
 339         int index = major_to_index(major);
 340
 341         mutex_lock(&block_class_lock);
 342         for (n = &major_names[index]; *n; n = &(*n)->next)
 343                 if ((*n)->major == major)
 344                         break;
 345         if (!*n || strcmp((*n)->name, name)) {
 346                 WARN_ON(1);
 347         } else {
 348                 p = *n;
 349                 *n = p->next;
 350         }
 351         mutex_unlock(&block_class_lock);
 352         kfree(p);
 353 }
 354
 355 EXPORT_SYMBOL(unregister_blkdev);
 356
 357 static struct kobj_map *bdev_map;
 358
 359 /**
 360  * blk_mangle_minor - scatter minor numbers apart
 361  * @minor: minor number to mangle
 362  *
 363  * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
 364  * is enabled.  Mangling twice gives the original value.
 365  *
 366  * RETURNS:
 367  * Mangled value.
 368  *
 369  * CONTEXT:
 370  * Don't care.
 371  */
 372 static int blk_mangle_minor(int minor)
 373 {
 374 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
 375         int i;
 376
 377         for (i = 0; i < MINORBITS / 2; i++) {
 378                 int low = minor & (1 << i);
 379                 int high = minor & (1 << (MINORBITS - 1 - i));
 380                 int distance = MINORBITS - 1 - 2 * i;
 381
 382                 minor ^= low | high;    /* clear both bits */
 383                 low <<= distance;       /* swap the positions */
 384                 high >>= distance;
 385                 minor |= low | high;    /* and set */
 386         }
 387 #endif
 388         return minor;
 389 }
 390
 391 /**
 392  * blk_alloc_devt - allocate a dev_t for a partition
 393  * @part: partition to allocate dev_t for
 394  * @devt: out parameter for resulting dev_t
 395  *
 396  * Allocate a dev_t for block device.
 397  *
 398  * RETURNS:
 399  * 0 on success, allocated dev_t is returned in *@devt.  -errno on
 400  * failure.
 401  *
 402  * CONTEXT:
 403  * Might sleep.
 404  */
 405 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
 406 {
 407         struct gendisk *disk = part_to_disk(part);
 408         int idx, rc;
 409
 410         /* in consecutive minor range? */
 411         if (part->partno < disk->minors) {
 412                 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
 413                 return 0;
 414         }
 415
 416         /* allocate ext devt */
 417         do {
 418                 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
 419                         return -ENOMEM;
 420                 rc = idr_get_new(&ext_devt_idr, part, &idx);
 421         } while (rc == -EAGAIN);
 422
 423         if (rc)
 424                 return rc;
 425
 426         if (idx > MAX_EXT_DEVT) {
 427                 idr_remove(&ext_devt_idr, idx);
 428                 return -EBUSY;
 429         }
 430
 431         *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
 432         return 0;
 433 }
 434
 435 /**
 436  * blk_free_devt - free a dev_t
 437  * @devt: dev_t to free
 438  *
 439  * Free @devt which was allocated using blk_alloc_devt().
 440  *
 441  * CONTEXT:
 442  * Might sleep.
 443  */
 444 void blk_free_devt(dev_t devt)
 445 {
 446         might_sleep();
 447
 448         if (devt == MKDEV(0, 0))
 449                 return;
 450
 451         if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
 452                 mutex_lock(&ext_devt_mutex);
 453                 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 454                 mutex_unlock(&ext_devt_mutex);
 455         }
 456 }
 457
 458 static char *bdevt_str(dev_t devt, char *buf)
 459 {
 460         if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
 461                 char tbuf[BDEVT_SIZE];
 462                 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
 463                 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
 464         } else
 465                 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
 466
 467         return buf;
 468 }
 469
 470 /*
 471  * Register device numbers dev..(dev+range-1)
 472  * range must be nonzero
 473  * The hash chain is sorted on range, so that subranges can override.
 474  */
 475 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
 476                          struct kobject *(*probe)(dev_t, int *, void *),
 477                          int (*lock)(dev_t, void *), void *data)
 478 {
 479         kobj_map(bdev_map, devt, range, module, probe, lock, data);
 480 }
 481
 482 EXPORT_SYMBOL(blk_register_region);
 483
 484 void blk_unregister_region(dev_t devt, unsigned long range)
 485 {
 486         kobj_unmap(bdev_map, devt, range);
 487 }
 488
 489 EXPORT_SYMBOL(blk_unregister_region);
 490
 491 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
 492 {
 493         struct gendisk *p = data;
 494
 495         return &disk_to_dev(p)->kobj;
 496 }
 497
 498 static int exact_lock(dev_t devt, void *data)
 499 {
 500         struct gendisk *p = data;
 501
 502         if (!get_disk(p))
 503                 return -1;
 504         return 0;
 505 }
 506
 507 /**
 508  * add_disk - add partitioning information to kernel list
 509  * @disk: per-device partitioning information
 510  *
 511  * This function registers the partitioning information in @disk
 512  * with the kernel.
 513  *
 514  * FIXME: error handling
 515  */
 516 void add_disk(struct gendisk *disk)
 517 {
 518         struct backing_dev_info *bdi;
 519         dev_t devt;
 520         int retval;
 521
 522         /* minors == 0 indicates to use ext devt from part0 and should
 523          * be accompanied with EXT_DEVT flag.  Make sure all
 524          * parameters make sense.
 525          */
 526         WARN_ON(disk->minors && !(disk->major || disk->first_minor));
 527         WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
 528
 529         disk->flags |= GENHD_FL_UP;
 530
 531         retval = blk_alloc_devt(&disk->part0, &devt);
 532         if (retval) {
 533                 WARN_ON(1);
 534                 return;
 535         }
 536         disk_to_dev(disk)->devt = devt;
 537
 538         /* ->major and ->first_minor aren't supposed to be
 539          * dereferenced from here on, but set them just in case.
 540          */
 541         disk->major = MAJOR(devt);
 542         disk->first_minor = MINOR(devt);
 543
 544         blk_register_region(disk_devt(disk), disk->minors, NULL,
 545                             exact_match, exact_lock, disk);
 546         register_disk(disk);
 547         blk_register_queue(disk);
 548
 549         bdi = &disk->queue->backing_dev_info;
 550         bdi_register_dev(bdi, disk_devt(disk));
 551         retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
 552                                    "bdi");
 553         WARN_ON(retval);
 554 }
 555
 556 EXPORT_SYMBOL(add_disk);
 557 EXPORT_SYMBOL(del_gendisk);     /* in partitions/check.c */
 558
 559 void unlink_gendisk(struct gendisk *disk)
 560 {
 561         sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
 562         bdi_unregister(&disk->queue->backing_dev_info);
 563         blk_unregister_queue(disk);
 564         blk_unregister_region(disk_devt(disk), disk->minors);
 565 }
 566
 567 /**
 568  * get_gendisk - get partitioning information for a given device
 569  * @devt: device to get partitioning information for
 570  * @partno: returned partition index
 571  *
 572  * This function gets the structure containing partitioning
 573  * information for the given device @devt.
 574  */
 575 struct gendisk *get_gendisk(dev_t devt, int *partno)
 576 {
 577         struct gendisk *disk = NULL;
 578
 579         if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
 580                 struct kobject *kobj;
 581
 582                 kobj = kobj_lookup(bdev_map, devt, partno);
 583                 if (kobj)
 584                         disk = dev_to_disk(kobj_to_dev(kobj));
 585         } else {
 586                 struct hd_struct *part;
 587
 588                 mutex_lock(&ext_devt_mutex);
 589                 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 590                 if (part && get_disk(part_to_disk(part))) {
 591                         *partno = part->partno;
 592                         disk = part_to_disk(part);
 593                 }
 594                 mutex_unlock(&ext_devt_mutex);
 595         }
 596
 597         return disk;
 598 }
 599
 600 /**
 601  * bdget_disk - do bdget() by gendisk and partition number
 602  * @disk: gendisk of interest
 603  * @partno: partition number
 604  *
 605  * Find partition @partno from @disk, do bdget() on it.
 606  *
 607  * CONTEXT:
 608  * Don't care.
 609  *
 610  * RETURNS:
 611  * Resulting block_device on success, NULL on failure.
 612  */
 613 struct block_device *bdget_disk(struct gendisk *disk, int partno)
 614 {
 615         struct hd_struct *part;
 616         struct block_device *bdev = NULL;
 617
 618         part = disk_get_part(disk, partno);
 619         if (part)
 620                 bdev = bdget(part_devt(part));
 621         disk_put_part(part);
 622
 623         return bdev;
 624 }
 625 EXPORT_SYMBOL(bdget_disk);
 626
 627 /*
 628  * print a full list of all partitions - intended for places where the root
 629  * filesystem can't be mounted and thus to give the victim some idea of what
 630  * went wrong
 631  */
 632 void __init printk_all_partitions(void)
 633 {
 634         struct class_dev_iter iter;
 635         struct device *dev;
 636
 637         class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
 638         while ((dev = class_dev_iter_next(&iter))) {
 639                 struct gendisk *disk = dev_to_disk(dev);
 640                 struct disk_part_iter piter;
 641                 struct hd_struct *part;
 642                 char name_buf[BDEVNAME_SIZE];
 643                 char devt_buf[BDEVT_SIZE];
 644
 645                 /*
 646                  * Don't show empty devices or things that have been
 647                  * surpressed
 648                  */
 649                 if (get_capacity(disk) == 0 ||
 650                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 651                         continue;
 652
 653                 /*
 654                  * Note, unlike /proc/partitions, I am showing the
 655                  * numbers in hex - the same format as the root=
 656                  * option takes.
 657                  */
 658                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 659                 while ((part = disk_part_iter_next(&piter))) {
 660                         bool is_part0 = part == &disk->part0;
 661
 662                         printk("%s%s %10llu %s", is_part0 ? "" : "  ",
 663                                bdevt_str(part_devt(part), devt_buf),
 664                                (unsigned long long)part->nr_sects >> 1,
 665                                disk_name(disk, part->partno, name_buf));
 666                         if (is_part0) {
 667                                 if (disk->driverfs_dev != NULL &&
 668                                     disk->driverfs_dev->driver != NULL)
 669                                         printk(" driver: %s\n",
 670                                               disk->driverfs_dev->driver->name);
 671                                 else
 672                                         printk(" (driver?)\n");
 673                         } else
 674                                 printk("\n");
 675                 }
 676                 disk_part_iter_exit(&piter);
 677         }
 678         class_dev_iter_exit(&iter);
 679 }
 680
 681 #ifdef CONFIG_PROC_FS
 682 /* iterator */
 683 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
 684 {
 685         loff_t skip = *pos;
 686         struct class_dev_iter *iter;
 687         struct device *dev;
 688
 689         iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 690         if (!iter)
 691                 return ERR_PTR(-ENOMEM);
 692
 693         seqf->private = iter;
 694         class_dev_iter_init(iter, &block_class, NULL, &disk_type);
 695         do {
 696                 dev = class_dev_iter_next(iter);
 697                 if (!dev)
 698                         return NULL;
 699         } while (skip--);
 700
 701         return dev_to_disk(dev);
 702 }
 703
 704 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
 705 {
 706         struct device *dev;
 707
 708         (*pos)++;
 709         dev = class_dev_iter_next(seqf->private);
 710         if (dev)
 711                 return dev_to_disk(dev);
 712
 713         return NULL;
 714 }
 715
 716 static void disk_seqf_stop(struct seq_file *seqf, void *v)
 717 {
 718         struct class_dev_iter *iter = seqf->private;
 719
 720         /* stop is called even after start failed :-( */
 721         if (iter) {
 722                 class_dev_iter_exit(iter);
 723                 kfree(iter);
 724         }
 725 }
 726
 727 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
 728 {
 729         static void *p;
 730
 731         p = disk_seqf_start(seqf, pos);
 732         if (!IS_ERR(p) && p && !*pos)
 733                 seq_puts(seqf, "major minor  #blocks  name\n\n");
 734         return p;
 735 }
 736
 737 static int show_partition(struct seq_file *seqf, void *v)
 738 {
 739         struct gendisk *sgp = v;
 740         struct disk_part_iter piter;
 741         struct hd_struct *part;
 742         char buf[BDEVNAME_SIZE];
 743
 744         /* Don't show non-partitionable removeable devices or empty devices */
 745         if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
 746                                    (sgp->flags & GENHD_FL_REMOVABLE)))
 747                 return 0;
 748         if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
 749                 return 0;
 750
 751         /* show the full disk and all non-0 size partitions of it */
 752         disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
 753         while ((part = disk_part_iter_next(&piter)))
 754                 seq_printf(seqf, "%4d  %7d %10llu %s\n",
 755                            MAJOR(part_devt(part)), MINOR(part_devt(part)),
 756                            (unsigned long long)part->nr_sects >> 1,
 757                            disk_name(sgp, part->partno, buf));
 758         disk_part_iter_exit(&piter);
 759
 760         return 0;
 761 }
 762
 763 static const struct seq_operations partitions_op = {
 764         .start  = show_partition_start,
 765         .next   = disk_seqf_next,
 766         .stop   = disk_seqf_stop,
 767         .show   = show_partition
 768 };
 769
 770 static int partitions_open(struct inode *inode, struct file *file)
 771 {
 772         return seq_open(file, &partitions_op);
 773 }
 774
 775 static const struct file_operations proc_partitions_operations = {
 776         .open           = partitions_open,
 777         .read           = seq_read,
 778         .llseek         = seq_lseek,
 779         .release        = seq_release,
 780 };
 781 #endif
 782
 783
 784 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
 785 {
 786         if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
 787                 /* Make old-style 2.4 aliases work */
 788                 request_module("block-major-%d", MAJOR(devt));
 789         return NULL;
 790 }
 791
 792 static int __init genhd_device_init(void)
 793 {
 794         int error;
 795
 796         block_class.dev_kobj = sysfs_dev_block_kobj;
 797         error = class_register(&block_class);
 798         if (unlikely(error))
 799                 return error;
 800         bdev_map = kobj_map_init(base_probe, &block_class_lock);
 801         blk_dev_init();
 802
 803         register_blkdev(BLOCK_EXT_MAJOR, "blkext");
 804
 805 #ifndef CONFIG_SYSFS_DEPRECATED
 806         /* create top-level block dir */
 807         block_depr = kobject_create_and_add("block", NULL);
 808 #endif
 809         return 0;
 810 }
 811
 812 subsys_initcall(genhd_device_init);
 813
 814 static ssize_t disk_range_show(struct device *dev,
 815                                struct device_attribute *attr, char *buf)
 816 {
 817         struct gendisk *disk = dev_to_disk(dev);
 818
 819         return sprintf(buf, "%d\n", disk->minors);
 820 }
 821
 822 static ssize_t disk_ext_range_show(struct device *dev,
 823                                    struct device_attribute *attr, char *buf)
 824 {
 825         struct gendisk *disk = dev_to_disk(dev);
 826
 827         return sprintf(buf, "%d\n", disk_max_parts(disk));
 828 }
 829
 830 static ssize_t disk_removable_show(struct device *dev,
 831                                    struct device_attribute *attr, char *buf)
 832 {
 833         struct gendisk *disk = dev_to_disk(dev);
 834
 835         return sprintf(buf, "%d\n",
 836                        (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
 837 }
 838
 839 static ssize_t disk_ro_show(struct device *dev,
 840                                    struct device_attribute *attr, char *buf)
 841 {
 842         struct gendisk *disk = dev_to_disk(dev);
 843
 844         return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
 845 }
 846
 847 static ssize_t disk_capability_show(struct device *dev,
 848                                     struct device_attribute *attr, char *buf)
 849 {
 850         struct gendisk *disk = dev_to_disk(dev);
 851
 852         return sprintf(buf, "%x\n", disk->flags);
 853 }
 854
 855 static ssize_t disk_alignment_offset_show(struct device *dev,
 856                                           struct device_attribute *attr,
 857                                           char *buf)
 858 {
 859         struct gendisk *disk = dev_to_disk(dev);
 860
 861         return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
 862 }
 863
 864 static ssize_t disk_discard_alignment_show(struct device *dev,
 865                                            struct device_attribute *attr,
 866                                            char *buf)
 867 {
 868         struct gendisk *disk = dev_to_disk(dev);
 869
 870         return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
 871 }
 872
 873 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
 874 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
 875 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
 876 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
 877 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
 878 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
 879 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
 880                    NULL);
 881 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
 882 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
 883 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
 884 #ifdef CONFIG_FAIL_MAKE_REQUEST
 885 static struct device_attribute dev_attr_fail =
 886         __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
 887 #endif
 888 #ifdef CONFIG_FAIL_IO_TIMEOUT
 889 static struct device_attribute dev_attr_fail_timeout =
 890         __ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
 891                 part_timeout_store);
 892 #endif
 893
 894 static struct attribute *disk_attrs[] = {
 895         &dev_attr_range.attr,
 896         &dev_attr_ext_range.attr,
 897         &dev_attr_removable.attr,
 898         &dev_attr_ro.attr,
 899         &dev_attr_size.attr,
 900         &dev_attr_alignment_offset.attr,
 901         &dev_attr_discard_alignment.attr,
 902         &dev_attr_capability.attr,
 903         &dev_attr_stat.attr,
 904         &dev_attr_inflight.attr,
 905 #ifdef CONFIG_FAIL_MAKE_REQUEST
 906         &dev_attr_fail.attr,
 907 #endif
 908 #ifdef CONFIG_FAIL_IO_TIMEOUT
 909         &dev_attr_fail_timeout.attr,
 910 #endif
 911         NULL
 912 };
 913
 914 static struct attribute_group disk_attr_group = {
 915         .attrs = disk_attrs,
 916 };
 917
 918 static const struct attribute_group *disk_attr_groups[] = {
 919         &disk_attr_group,
 920         NULL
 921 };
 922
 923 static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
 924 {
 925         struct disk_part_tbl *ptbl =
 926                 container_of(head, struct disk_part_tbl, rcu_head);
 927
 928         kfree(ptbl);
 929 }
 930
 931 /**
 932  * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
 933  * @disk: disk to replace part_tbl for
 934  * @new_ptbl: new part_tbl to install
 935  *
 936  * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
 937  * original ptbl is freed using RCU callback.
 938  *
 939  * LOCKING:
 940  * Matching bd_mutx locked.
 941  */
 942 static void disk_replace_part_tbl(struct gendisk *disk,
 943                                   struct disk_part_tbl *new_ptbl)
 944 {
 945         struct disk_part_tbl *old_ptbl = disk->part_tbl;
 946
 947         rcu_assign_pointer(disk->part_tbl, new_ptbl);
 948
 949         if (old_ptbl) {
 950                 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
 951                 call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
 952         }
 953 }
 954
 955 /**
 956  * disk_expand_part_tbl - expand disk->part_tbl
 957  * @disk: disk to expand part_tbl for
 958  * @partno: expand such that this partno can fit in
 959  *
 960  * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
 961  * uses RCU to allow unlocked dereferencing for stats and other stuff.
 962  *
 963  * LOCKING:
 964  * Matching bd_mutex locked, might sleep.
 965  *
 966  * RETURNS:
 967  * 0 on success, -errno on failure.
 968  */
 969 int disk_expand_part_tbl(struct gendisk *disk, int partno)
 970 {
 971         struct disk_part_tbl *old_ptbl = disk->part_tbl;
 972         struct disk_part_tbl *new_ptbl;
 973         int len = old_ptbl ? old_ptbl->len : 0;
 974         int target = partno + 1;
 975         size_t size;
 976         int i;
 977
 978         /* disk_max_parts() is zero during initialization, ignore if so */
 979         if (disk_max_parts(disk) && target > disk_max_parts(disk))
 980                 return -EINVAL;
 981
 982         if (target <= len)
 983                 return 0;
 984
 985         size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
 986         new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
 987         if (!new_ptbl)
 988                 return -ENOMEM;
 989
 990         INIT_RCU_HEAD(&new_ptbl->rcu_head);
 991         new_ptbl->len = target;
 992
 993         for (i = 0; i < len; i++)
 994                 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
 995
 996         disk_replace_part_tbl(disk, new_ptbl);
 997         return 0;
 998 }
 999
1000 static void disk_release(struct device *dev)
1001 {
1002         struct gendisk *disk = dev_to_disk(dev);
1003
1004         kfree(disk->random);
1005         disk_replace_part_tbl(disk, NULL);
1006         free_part_stats(&disk->part0);
1007         kfree(disk);
1008 }
1009 struct class block_class = {
1010         .name           = "block",
1011 };
1012
1013 static char *block_devnode(struct device *dev, mode_t *mode)
1014 {
1015         struct gendisk *disk = dev_to_disk(dev);
1016
1017         if (disk->devnode)
1018                 return disk->devnode(disk, mode);
1019         return NULL;
1020 }
1021
1022 static struct device_type disk_type = {
1023         .name           = "disk",
1024         .groups         = disk_attr_groups,
1025         .release        = disk_release,
1026         .devnode        = block_devnode,
1027 };
1028
1029 #ifdef CONFIG_PROC_FS
1030 /*
1031  * aggregate disk stat collector.  Uses the same stats that the sysfs
1032  * entries do, above, but makes them available through one seq_file.
1033  *
1034  * The output looks suspiciously like /proc/partitions with a bunch of
1035  * extra fields.
1036  */
1037 static int diskstats_show(struct seq_file *seqf, void *v)
1038 {
1039         struct gendisk *gp = v;
1040         struct disk_part_iter piter;
1041         struct hd_struct *hd;
1042         char buf[BDEVNAME_SIZE];
1043         int cpu;
1044
1045         /*
1046         if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1047                 seq_puts(seqf,  "major minor name"
1048                                 "     rio rmerge rsect ruse wio wmerge "
1049                                 "wsect wuse running use aveq"
1050                                 "\n\n");
1051         */
1052
1053         disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1054         while ((hd = disk_part_iter_next(&piter))) {
1055                 cpu = part_stat_lock();
1056                 part_round_stats(cpu, hd);
1057                 part_stat_unlock();
1058                 seq_printf(seqf, "%4d %7d %s %lu %lu %llu "
1059                            "%u %lu %lu %llu %u %u %u %u\n",
1060                            MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1061                            disk_name(gp, hd->partno, buf),
1062                            part_stat_read(hd, ios[0]),
1063                            part_stat_read(hd, merges[0]),
1064                            (unsigned long long)part_stat_read(hd, sectors[0]),
1065                            jiffies_to_msecs(part_stat_read(hd, ticks[0])),
1066                            part_stat_read(hd, ios[1]),
1067                            part_stat_read(hd, merges[1]),
1068                            (unsigned long long)part_stat_read(hd, sectors[1]),
1069                            jiffies_to_msecs(part_stat_read(hd, ticks[1])),
1070                            part_in_flight(hd),
1071                            jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1072                            jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1073                         );
1074         }
1075         disk_part_iter_exit(&piter);
1076
1077         return 0;
1078 }
1079
1080 static const struct seq_operations diskstats_op = {
1081         .start  = disk_seqf_start,
1082         .next   = disk_seqf_next,
1083         .stop   = disk_seqf_stop,
1084         .show   = diskstats_show
1085 };
1086
1087 static int diskstats_open(struct inode *inode, struct file *file)
1088 {
1089         return seq_open(file, &diskstats_op);
1090 }
1091
1092 static const struct file_operations proc_diskstats_operations = {
1093         .open           = diskstats_open,
1094         .read           = seq_read,
1095         .llseek         = seq_lseek,
1096         .release        = seq_release,
1097 };
1098
1099 static int __init proc_genhd_init(void)
1100 {
1101         proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1102         proc_create("partitions", 0, NULL, &proc_partitions_operations);
1103         return 0;
1104 }
1105 module_init(proc_genhd_init);
1106 #endif /* CONFIG_PROC_FS */
1107
1108 static void media_change_notify_thread(struct work_struct *work)
1109 {
1110         struct gendisk *gd = container_of(work, struct gendisk, async_notify);
1111         char event[] = "MEDIA_CHANGE=1";
1112         char *envp[] = { event, NULL };
1113
1114         /*
1115          * set enviroment vars to indicate which event this is for
1116          * so that user space will know to go check the media status.
1117          */
1118         kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1119         put_device(gd->driverfs_dev);
1120 }
1121
1122 #if 0
1123 void genhd_media_change_notify(struct gendisk *disk)
1124 {
1125         get_device(disk->driverfs_dev);
1126         schedule_work(&disk->async_notify);
1127 }
1128 EXPORT_SYMBOL_GPL(genhd_media_change_notify);
1129 #endif  /*  0  */
1130
1131 dev_t blk_lookup_devt(const char *name, int partno)
1132 {
1133         dev_t devt = MKDEV(0, 0);
1134         struct class_dev_iter iter;
1135         struct device *dev;
1136
1137         class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1138         while ((dev = class_dev_iter_next(&iter))) {
1139                 struct gendisk *disk = dev_to_disk(dev);
1140                 struct hd_struct *part;
1141
1142                 if (strcmp(dev_name(dev), name))
1143                         continue;
1144
1145                 if (partno < disk->minors) {
1146                         /* We need to return the right devno, even
1147                          * if the partition doesn't exist yet.
1148                          */
1149                         devt = MKDEV(MAJOR(dev->devt),
1150                                      MINOR(dev->devt) + partno);
1151                         break;
1152                 }
1153                 part = disk_get_part(disk, partno);
1154                 if (part) {
1155                         devt = part_devt(part);
1156                         disk_put_part(part);
1157                         break;
1158                 }
1159                 disk_put_part(part);
1160         }
1161         class_dev_iter_exit(&iter);
1162         return devt;
1163 }
1164 EXPORT_SYMBOL(blk_lookup_devt);
1165
1166 struct gendisk *alloc_disk(int minors)
1167 {
1168         return alloc_disk_node(minors, -1);
1169 }
1170 EXPORT_SYMBOL(alloc_disk);
1171
1172 struct gendisk *alloc_disk_node(int minors, int node_id)
1173 {
1174         struct gendisk *disk;
1175
1176         disk = kmalloc_node(sizeof(struct gendisk),
1177                                 GFP_KERNEL | __GFP_ZERO, node_id);
1178         if (disk) {
1179                 if (!init_part_stats(&disk->part0)) {
1180                         kfree(disk);
1181                         return NULL;
1182                 }
1183                 disk->node_id = node_id;
1184                 if (disk_expand_part_tbl(disk, 0)) {
1185                         free_part_stats(&disk->part0);
1186                         kfree(disk);
1187                         return NULL;
1188                 }
1189                 disk->part_tbl->part[0] = &disk->part0;
1190
1191                 disk->minors = minors;
1192                 rand_initialize_disk(disk);
1193                 disk_to_dev(disk)->class = &block_class;
1194                 disk_to_dev(disk)->type = &disk_type;
1195                 device_initialize(disk_to_dev(disk));
1196                 INIT_WORK(&disk->async_notify,
1197                         media_change_notify_thread);
1198         }
1199         return disk;
1200 }
1201 EXPORT_SYMBOL(alloc_disk_node);
1202
1203 struct kobject *get_disk(struct gendisk *disk)
1204 {
1205         struct module *owner;
1206         struct kobject *kobj;
1207
1208         if (!disk->fops)
1209                 return NULL;
1210         owner = disk->fops->owner;
1211         if (owner && !try_module_get(owner))
1212                 return NULL;
1213         kobj = kobject_get(&disk_to_dev(disk)->kobj);
1214         if (kobj == NULL) {
1215                 module_put(owner);
1216                 return NULL;
1217         }
1218         return kobj;
1219
1220 }
1221
1222 EXPORT_SYMBOL(get_disk);
1223
1224 void put_disk(struct gendisk *disk)
1225 {
1226         if (disk)
1227                 kobject_put(&disk_to_dev(disk)->kobj);
1228 }
1229
1230 EXPORT_SYMBOL(put_disk);
1231
1232 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1233 {
1234         char event[] = "DISK_RO=1";
1235         char *envp[] = { event, NULL };
1236
1237         if (!ro)
1238                 event[8] = '0';
1239         kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1240 }
1241
1242 void set_device_ro(struct block_device *bdev, int flag)
1243 {
1244         bdev->bd_part->policy = flag;
1245 }
1246
1247 EXPORT_SYMBOL(set_device_ro);
1248
1249 void set_disk_ro(struct gendisk *disk, int flag)
1250 {
1251         struct disk_part_iter piter;
1252         struct hd_struct *part;
1253
1254         if (disk->part0.policy != flag) {
1255                 set_disk_ro_uevent(disk, flag);
1256                 disk->part0.policy = flag;
1257         }
1258
1259         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1260         while ((part = disk_part_iter_next(&piter)))
1261                 part->policy = flag;
1262         disk_part_iter_exit(&piter);
1263 }
1264
1265 EXPORT_SYMBOL(set_disk_ro);
1266
1267 int bdev_read_only(struct block_device *bdev)
1268 {
1269         if (!bdev)
1270                 return 0;
1271         return bdev->bd_part->policy;
1272 }
1273
1274 EXPORT_SYMBOL(bdev_read_only);
1275
1276 int invalidate_partition(struct gendisk *disk, int partno)
1277 {
1278         int res = 0;
1279         struct block_device *bdev = bdget_disk(disk, partno);
1280         if (bdev) {
1281                 fsync_bdev(bdev);
1282                 res = __invalidate_device(bdev);
1283                 bdput(bdev);
1284         }
1285         return res;
1286 }
1287
1288 EXPORT_SYMBOL(invalidate_partition);