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         /* Register BDI before referencing it from bdev */
 545         bdi = &disk->queue->backing_dev_info;
 546         bdi_register_dev(bdi, disk_devt(disk));
 547
 548         blk_register_region(disk_devt(disk), disk->minors, NULL,
 549                             exact_match, exact_lock, disk);
 550         register_disk(disk);
 551         blk_register_queue(disk);
 552
 553         retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
 554                                    "bdi");
 555         WARN_ON(retval);
 556 }
 557
 558 EXPORT_SYMBOL(add_disk);
 559 EXPORT_SYMBOL(del_gendisk);     /* in partitions/check.c */
 560
 561 void unlink_gendisk(struct gendisk *disk)
 562 {
 563         sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
 564         bdi_unregister(&disk->queue->backing_dev_info);
 565         blk_unregister_queue(disk);
 566         blk_unregister_region(disk_devt(disk), disk->minors);
 567 }
 568
 569 /**
 570  * get_gendisk - get partitioning information for a given device
 571  * @devt: device to get partitioning information for
 572  * @partno: returned partition index
 573  *
 574  * This function gets the structure containing partitioning
 575  * information for the given device @devt.
 576  */
 577 struct gendisk *get_gendisk(dev_t devt, int *partno)
 578 {
 579         struct gendisk *disk = NULL;
 580
 581         if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
 582                 struct kobject *kobj;
 583
 584                 kobj = kobj_lookup(bdev_map, devt, partno);
 585                 if (kobj)
 586                         disk = dev_to_disk(kobj_to_dev(kobj));
 587         } else {
 588                 struct hd_struct *part;
 589
 590                 mutex_lock(&ext_devt_mutex);
 591                 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 592                 if (part && get_disk(part_to_disk(part))) {
 593                         *partno = part->partno;
 594                         disk = part_to_disk(part);
 595                 }
 596                 mutex_unlock(&ext_devt_mutex);
 597         }
 598
 599         return disk;
 600 }
 601 EXPORT_SYMBOL(get_gendisk);
 602
 603 /**
 604  * bdget_disk - do bdget() by gendisk and partition number
 605  * @disk: gendisk of interest
 606  * @partno: partition number
 607  *
 608  * Find partition @partno from @disk, do bdget() on it.
 609  *
 610  * CONTEXT:
 611  * Don't care.
 612  *
 613  * RETURNS:
 614  * Resulting block_device on success, NULL on failure.
 615  */
 616 struct block_device *bdget_disk(struct gendisk *disk, int partno)
 617 {
 618         struct hd_struct *part;
 619         struct block_device *bdev = NULL;
 620
 621         part = disk_get_part(disk, partno);
 622         if (part)
 623                 bdev = bdget(part_devt(part));
 624         disk_put_part(part);
 625
 626         return bdev;
 627 }
 628 EXPORT_SYMBOL(bdget_disk);
 629
 630 /*
 631  * print a full list of all partitions - intended for places where the root
 632  * filesystem can't be mounted and thus to give the victim some idea of what
 633  * went wrong
 634  */
 635 void __init printk_all_partitions(void)
 636 {
 637         struct class_dev_iter iter;
 638         struct device *dev;
 639
 640         class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
 641         while ((dev = class_dev_iter_next(&iter))) {
 642                 struct gendisk *disk = dev_to_disk(dev);
 643                 struct disk_part_iter piter;
 644                 struct hd_struct *part;
 645                 char name_buf[BDEVNAME_SIZE];
 646                 char devt_buf[BDEVT_SIZE];
 647
 648                 /*
 649                  * Don't show empty devices or things that have been
 650                  * surpressed
 651                  */
 652                 if (get_capacity(disk) == 0 ||
 653                     (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 654                         continue;
 655
 656                 /*
 657                  * Note, unlike /proc/partitions, I am showing the
 658                  * numbers in hex - the same format as the root=
 659                  * option takes.
 660                  */
 661                 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 662                 while ((part = disk_part_iter_next(&piter))) {
 663                         bool is_part0 = part == &disk->part0;
 664
 665                         printk("%s%s %10llu %s", is_part0 ? "" : "  ",
 666                                bdevt_str(part_devt(part), devt_buf),
 667                                (unsigned long long)part->nr_sects >> 1,
 668                                disk_name(disk, part->partno, name_buf));
 669                         if (is_part0) {
 670                                 if (disk->driverfs_dev != NULL &&
 671                                     disk->driverfs_dev->driver != NULL)
 672                                         printk(" driver: %s\n",
 673                                               disk->driverfs_dev->driver->name);
 674                                 else
 675                                         printk(" (driver?)\n");
 676                         } else
 677                                 printk("\n");
 678                 }
 679                 disk_part_iter_exit(&piter);
 680         }
 681         class_dev_iter_exit(&iter);
 682 }
 683
 684 #ifdef CONFIG_PROC_FS
 685 /* iterator */
 686 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
 687 {
 688         loff_t skip = *pos;
 689         struct class_dev_iter *iter;
 690         struct device *dev;
 691
 692         iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 693         if (!iter)
 694                 return ERR_PTR(-ENOMEM);
 695
 696         seqf->private = iter;
 697         class_dev_iter_init(iter, &block_class, NULL, &disk_type);
 698         do {
 699                 dev = class_dev_iter_next(iter);
 700                 if (!dev)
 701                         return NULL;
 702         } while (skip--);
 703
 704         return dev_to_disk(dev);
 705 }
 706
 707 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
 708 {
 709         struct device *dev;
 710
 711         (*pos)++;
 712         dev = class_dev_iter_next(seqf->private);
 713         if (dev)
 714                 return dev_to_disk(dev);
 715
 716         return NULL;
 717 }
 718
 719 static void disk_seqf_stop(struct seq_file *seqf, void *v)
 720 {
 721         struct class_dev_iter *iter = seqf->private;
 722
 723         /* stop is called even after start failed :-( */
 724         if (iter) {
 725                 class_dev_iter_exit(iter);
 726                 kfree(iter);
 727         }
 728 }
 729
 730 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
 731 {
 732         static void *p;
 733
 734         p = disk_seqf_start(seqf, pos);
 735         if (!IS_ERR(p) && p && !*pos)
 736                 seq_puts(seqf, "major minor  #blocks  name\n\n");
 737         return p;
 738 }
 739
 740 static int show_partition(struct seq_file *seqf, void *v)
 741 {
 742         struct gendisk *sgp = v;
 743         struct disk_part_iter piter;
 744         struct hd_struct *part;
 745         char buf[BDEVNAME_SIZE];
 746
 747         /* Don't show non-partitionable removeable devices or empty devices */
 748         if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
 749                                    (sgp->flags & GENHD_FL_REMOVABLE)))
 750                 return 0;
 751         if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
 752                 return 0;
 753
 754         /* show the full disk and all non-0 size partitions of it */
 755         disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
 756         while ((part = disk_part_iter_next(&piter)))
 757                 seq_printf(seqf, "%4d  %7d %10llu %s\n",
 758                            MAJOR(part_devt(part)), MINOR(part_devt(part)),
 759                            (unsigned long long)part->nr_sects >> 1,
 760                            disk_name(sgp, part->partno, buf));
 761         disk_part_iter_exit(&piter);
 762
 763         return 0;
 764 }
 765
 766 static const struct seq_operations partitions_op = {
 767         .start  = show_partition_start,
 768         .next   = disk_seqf_next,
 769         .stop   = disk_seqf_stop,
 770         .show   = show_partition
 771 };
 772
 773 static int partitions_open(struct inode *inode, struct file *file)
 774 {
 775         return seq_open(file, &partitions_op);
 776 }
 777
 778 static const struct file_operations proc_partitions_operations = {
 779         .open           = partitions_open,
 780         .read           = seq_read,
 781         .llseek         = seq_lseek,
 782         .release        = seq_release,
 783 };
 784 #endif
 785
 786
 787 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
 788 {
 789         if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
 790                 /* Make old-style 2.4 aliases work */
 791                 request_module("block-major-%d", MAJOR(devt));
 792         return NULL;
 793 }
 794
 795 static int __init genhd_device_init(void)
 796 {
 797         int error;
 798
 799         block_class.dev_kobj = sysfs_dev_block_kobj;
 800         error = class_register(&block_class);
 801         if (unlikely(error))
 802                 return error;
 803         bdev_map = kobj_map_init(base_probe, &block_class_lock);
 804         blk_dev_init();
 805
 806         register_blkdev(BLOCK_EXT_MAJOR, "blkext");
 807
 808 #ifndef CONFIG_SYSFS_DEPRECATED
 809         /* create top-level block dir */
 810         block_depr = kobject_create_and_add("block", NULL);
 811 #endif
 812         return 0;
 813 }
 814
 815 subsys_initcall(genhd_device_init);
 816
 817 static ssize_t disk_range_show(struct device *dev,
 818                                struct device_attribute *attr, char *buf)
 819 {
 820         struct gendisk *disk = dev_to_disk(dev);
 821
 822         return sprintf(buf, "%d\n", disk->minors);
 823 }
 824
 825 static ssize_t disk_ext_range_show(struct device *dev,
 826                                    struct device_attribute *attr, char *buf)
 827 {
 828         struct gendisk *disk = dev_to_disk(dev);
 829
 830         return sprintf(buf, "%d\n", disk_max_parts(disk));
 831 }
 832
 833 static ssize_t disk_removable_show(struct device *dev,
 834                                    struct device_attribute *attr, char *buf)
 835 {
 836         struct gendisk *disk = dev_to_disk(dev);
 837
 838         return sprintf(buf, "%d\n",
 839                        (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
 840 }
 841
 842 static ssize_t disk_ro_show(struct device *dev,
 843                                    struct device_attribute *attr, char *buf)
 844 {
 845         struct gendisk *disk = dev_to_disk(dev);
 846
 847         return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
 848 }
 849
 850 static ssize_t disk_capability_show(struct device *dev,
 851                                     struct device_attribute *attr, char *buf)
 852 {
 853         struct gendisk *disk = dev_to_disk(dev);
 854
 855         return sprintf(buf, "%x\n", disk->flags);
 856 }
 857
 858 static ssize_t disk_alignment_offset_show(struct device *dev,
 859                                           struct device_attribute *attr,
 860                                           char *buf)
 861 {
 862         struct gendisk *disk = dev_to_disk(dev);
 863
 864         return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
 865 }
 866
 867 static ssize_t disk_discard_alignment_show(struct device *dev,
 868                                            struct device_attribute *attr,
 869                                            char *buf)
 870 {
 871         struct gendisk *disk = dev_to_disk(dev);
 872
 873         return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
 874 }
 875
 876 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
 877 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
 878 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
 879 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
 880 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
 881 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
 882 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
 883                    NULL);
 884 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
 885 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
 886 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
 887 #ifdef CONFIG_FAIL_MAKE_REQUEST
 888 static struct device_attribute dev_attr_fail =
 889         __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
 890 #endif
 891 #ifdef CONFIG_FAIL_IO_TIMEOUT
 892 static struct device_attribute dev_attr_fail_timeout =
 893         __ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
 894                 part_timeout_store);
 895 #endif
 896
 897 static struct attribute *disk_attrs[] = {
 898         &dev_attr_range.attr,
 899         &dev_attr_ext_range.attr,
 900         &dev_attr_removable.attr,
 901         &dev_attr_ro.attr,
 902         &dev_attr_size.attr,
 903         &dev_attr_alignment_offset.attr,
 904         &dev_attr_discard_alignment.attr,
 905         &dev_attr_capability.attr,
 906         &dev_attr_stat.attr,
 907         &dev_attr_inflight.attr,
 908 #ifdef CONFIG_FAIL_MAKE_REQUEST
 909         &dev_attr_fail.attr,
 910 #endif
 911 #ifdef CONFIG_FAIL_IO_TIMEOUT
 912         &dev_attr_fail_timeout.attr,
 913 #endif
 914         NULL
 915 };
 916
 917 static struct attribute_group disk_attr_group = {
 918         .attrs = disk_attrs,
 919 };
 920
 921 static const struct attribute_group *disk_attr_groups[] = {
 922         &disk_attr_group,
 923         NULL
 924 };
 925
 926 static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
 927 {
 928         struct disk_part_tbl *ptbl =
 929                 container_of(head, struct disk_part_tbl, rcu_head);
 930
 931         kfree(ptbl);
 932 }
 933
 934 /**
 935  * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
 936  * @disk: disk to replace part_tbl for
 937  * @new_ptbl: new part_tbl to install
 938  *
 939  * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
 940  * original ptbl is freed using RCU callback.
 941  *
 942  * LOCKING:
 943  * Matching bd_mutx locked.
 944  */
 945 static void disk_replace_part_tbl(struct gendisk *disk,
 946                                   struct disk_part_tbl *new_ptbl)
 947 {
 948         struct disk_part_tbl *old_ptbl = disk->part_tbl;
 949
 950         rcu_assign_pointer(disk->part_tbl, new_ptbl);
 951
 952         if (old_ptbl) {
 953                 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
 954                 call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
 955         }
 956 }
 957
 958 /**
 959  * disk_expand_part_tbl - expand disk->part_tbl
 960  * @disk: disk to expand part_tbl for
 961  * @partno: expand such that this partno can fit in
 962  *
 963  * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
 964  * uses RCU to allow unlocked dereferencing for stats and other stuff.
 965  *
 966  * LOCKING:
 967  * Matching bd_mutex locked, might sleep.
 968  *
 969  * RETURNS:
 970  * 0 on success, -errno on failure.
 971  */
 972 int disk_expand_part_tbl(struct gendisk *disk, int partno)
 973 {
 974         struct disk_part_tbl *old_ptbl = disk->part_tbl;
 975         struct disk_part_tbl *new_ptbl;
 976         int len = old_ptbl ? old_ptbl->len : 0;
 977         int target = partno + 1;
 978         size_t size;
 979         int i;
 980
 981         /* disk_max_parts() is zero during initialization, ignore if so */
 982         if (disk_max_parts(disk) && target > disk_max_parts(disk))
 983                 return -EINVAL;
 984
 985         if (target <= len)
 986                 return 0;
 987
 988         size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
 989         new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
 990         if (!new_ptbl)
 991                 return -ENOMEM;
 992
 993         new_ptbl->len = target;
 994
 995         for (i = 0; i < len; i++)
 996                 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
 997
 998         disk_replace_part_tbl(disk, new_ptbl);
 999         return 0;
1000 }
1001
1002 static void disk_release(struct device *dev)
1003 {
1004         struct gendisk *disk = dev_to_disk(dev);
1005
1006         kfree(disk->random);
1007         disk_replace_part_tbl(disk, NULL);
1008         free_part_stats(&disk->part0);
1009         kfree(disk);
1010 }
1011 struct class block_class = {
1012         .name           = "block",
1013 };
1014
1015 static char *block_devnode(struct device *dev, mode_t *mode)
1016 {
1017         struct gendisk *disk = dev_to_disk(dev);
1018
1019         if (disk->devnode)
1020                 return disk->devnode(disk, mode);
1021         return NULL;
1022 }
1023
1024 static struct device_type disk_type = {
1025         .name           = "disk",
1026         .groups         = disk_attr_groups,
1027         .release        = disk_release,
1028         .devnode        = block_devnode,
1029 };
1030
1031 #ifdef CONFIG_PROC_FS
1032 /*
1033  * aggregate disk stat collector.  Uses the same stats that the sysfs
1034  * entries do, above, but makes them available through one seq_file.
1035  *
1036  * The output looks suspiciously like /proc/partitions with a bunch of
1037  * extra fields.
1038  */
1039 static int diskstats_show(struct seq_file *seqf, void *v)
1040 {
1041         struct gendisk *gp = v;
1042         struct disk_part_iter piter;
1043         struct hd_struct *hd;
1044         char buf[BDEVNAME_SIZE];
1045         int cpu;
1046
1047         /*
1048         if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1049                 seq_puts(seqf,  "major minor name"
1050                                 "     rio rmerge rsect ruse wio wmerge "
1051                                 "wsect wuse running use aveq"
1052                                 "\n\n");
1053         */
1054
1055         disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1056         while ((hd = disk_part_iter_next(&piter))) {
1057                 cpu = part_stat_lock();
1058                 part_round_stats(cpu, hd);
1059                 part_stat_unlock();
1060                 seq_printf(seqf, "%4d %7d %s %lu %lu %llu "
1061                            "%u %lu %lu %llu %u %u %u %u\n",
1062                            MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1063                            disk_name(gp, hd->partno, buf),
1064                            part_stat_read(hd, ios[0]),
1065                            part_stat_read(hd, merges[0]),
1066                            (unsigned long long)part_stat_read(hd, sectors[0]),
1067                            jiffies_to_msecs(part_stat_read(hd, ticks[0])),
1068                            part_stat_read(hd, ios[1]),
1069                            part_stat_read(hd, merges[1]),
1070                            (unsigned long long)part_stat_read(hd, sectors[1]),
1071                            jiffies_to_msecs(part_stat_read(hd, ticks[1])),
1072                            part_in_flight(hd),
1073                            jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1074                            jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1075                         );
1076         }
1077         disk_part_iter_exit(&piter);
1078
1079         return 0;
1080 }
1081
1082 static const struct seq_operations diskstats_op = {
1083         .start  = disk_seqf_start,
1084         .next   = disk_seqf_next,
1085         .stop   = disk_seqf_stop,
1086         .show   = diskstats_show
1087 };
1088
1089 static int diskstats_open(struct inode *inode, struct file *file)
1090 {
1091         return seq_open(file, &diskstats_op);
1092 }
1093
1094 static const struct file_operations proc_diskstats_operations = {
1095         .open           = diskstats_open,
1096         .read           = seq_read,
1097         .llseek         = seq_lseek,
1098         .release        = seq_release,
1099 };
1100
1101 static int __init proc_genhd_init(void)
1102 {
1103         proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1104         proc_create("partitions", 0, NULL, &proc_partitions_operations);
1105         return 0;
1106 }
1107 module_init(proc_genhd_init);
1108 #endif /* CONFIG_PROC_FS */
1109
1110 static void media_change_notify_thread(struct work_struct *work)
1111 {
1112         struct gendisk *gd = container_of(work, struct gendisk, async_notify);
1113         char event[] = "MEDIA_CHANGE=1";
1114         char *envp[] = { event, NULL };
1115
1116         /*
1117          * set enviroment vars to indicate which event this is for
1118          * so that user space will know to go check the media status.
1119          */
1120         kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1121         put_device(gd->driverfs_dev);
1122 }
1123
1124 #if 0
1125 void genhd_media_change_notify(struct gendisk *disk)
1126 {
1127         get_device(disk->driverfs_dev);
1128         schedule_work(&disk->async_notify);
1129 }
1130 EXPORT_SYMBOL_GPL(genhd_media_change_notify);
1131 #endif  /*  0  */
1132
1133 dev_t blk_lookup_devt(const char *name, int partno)
1134 {
1135         dev_t devt = MKDEV(0, 0);
1136         struct class_dev_iter iter;
1137         struct device *dev;
1138
1139         class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1140         while ((dev = class_dev_iter_next(&iter))) {
1141                 struct gendisk *disk = dev_to_disk(dev);
1142                 struct hd_struct *part;
1143
1144                 if (strcmp(dev_name(dev), name))
1145                         continue;
1146
1147                 if (partno < disk->minors) {
1148                         /* We need to return the right devno, even
1149                          * if the partition doesn't exist yet.
1150                          */
1151                         devt = MKDEV(MAJOR(dev->devt),
1152                                      MINOR(dev->devt) + partno);
1153                         break;
1154                 }
1155                 part = disk_get_part(disk, partno);
1156                 if (part) {
1157                         devt = part_devt(part);
1158                         disk_put_part(part);
1159                         break;
1160                 }
1161                 disk_put_part(part);
1162         }
1163         class_dev_iter_exit(&iter);
1164         return devt;
1165 }
1166 EXPORT_SYMBOL(blk_lookup_devt);
1167
1168 struct gendisk *alloc_disk(int minors)
1169 {
1170         return alloc_disk_node(minors, -1);
1171 }
1172 EXPORT_SYMBOL(alloc_disk);
1173
1174 struct gendisk *alloc_disk_node(int minors, int node_id)
1175 {
1176         struct gendisk *disk;
1177
1178         disk = kmalloc_node(sizeof(struct gendisk),
1179                                 GFP_KERNEL | __GFP_ZERO, node_id);
1180         if (disk) {
1181                 if (!init_part_stats(&disk->part0)) {
1182                         kfree(disk);
1183                         return NULL;
1184                 }
1185                 disk->node_id = node_id;
1186                 if (disk_expand_part_tbl(disk, 0)) {
1187                         free_part_stats(&disk->part0);
1188                         kfree(disk);
1189                         return NULL;
1190                 }
1191                 disk->part_tbl->part[0] = &disk->part0;
1192
1193                 disk->minors = minors;
1194                 rand_initialize_disk(disk);
1195                 disk_to_dev(disk)->class = &block_class;
1196                 disk_to_dev(disk)->type = &disk_type;
1197                 device_initialize(disk_to_dev(disk));
1198                 INIT_WORK(&disk->async_notify,
1199                         media_change_notify_thread);
1200         }
1201         return disk;
1202 }
1203 EXPORT_SYMBOL(alloc_disk_node);
1204
1205 struct kobject *get_disk(struct gendisk *disk)
1206 {
1207         struct module *owner;
1208         struct kobject *kobj;
1209
1210         if (!disk->fops)
1211                 return NULL;
1212         owner = disk->fops->owner;
1213         if (owner && !try_module_get(owner))
1214                 return NULL;
1215         kobj = kobject_get(&disk_to_dev(disk)->kobj);
1216         if (kobj == NULL) {
1217                 module_put(owner);
1218                 return NULL;
1219         }
1220         return kobj;
1221
1222 }
1223
1224 EXPORT_SYMBOL(get_disk);
1225
1226 void put_disk(struct gendisk *disk)
1227 {
1228         if (disk)
1229                 kobject_put(&disk_to_dev(disk)->kobj);
1230 }
1231
1232 EXPORT_SYMBOL(put_disk);
1233
1234 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1235 {
1236         char event[] = "DISK_RO=1";
1237         char *envp[] = { event, NULL };
1238
1239         if (!ro)
1240                 event[8] = '0';
1241         kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1242 }
1243
1244 void set_device_ro(struct block_device *bdev, int flag)
1245 {
1246         bdev->bd_part->policy = flag;
1247 }
1248
1249 EXPORT_SYMBOL(set_device_ro);
1250
1251 void set_disk_ro(struct gendisk *disk, int flag)
1252 {
1253         struct disk_part_iter piter;
1254         struct hd_struct *part;
1255
1256         if (disk->part0.policy != flag) {
1257                 set_disk_ro_uevent(disk, flag);
1258                 disk->part0.policy = flag;
1259         }
1260
1261         disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1262         while ((part = disk_part_iter_next(&piter)))
1263                 part->policy = flag;
1264         disk_part_iter_exit(&piter);
1265 }
1266
1267 EXPORT_SYMBOL(set_disk_ro);
1268
1269 int bdev_read_only(struct block_device *bdev)
1270 {
1271         if (!bdev)
1272                 return 0;
1273         return bdev->bd_part->policy;
1274 }
1275
1276 EXPORT_SYMBOL(bdev_read_only);
1277
1278 int invalidate_partition(struct gendisk *disk, int partno)
1279 {
1280         int res = 0;
1281         struct block_device *bdev = bdget_disk(disk, partno);
1282         if (bdev) {
1283                 fsync_bdev(bdev);
1284                 res = __invalidate_device(bdev);
1285                 bdput(bdev);
1286         }
1287         return res;
1288 }
1289
1290 EXPORT_SYMBOL(invalidate_partition);