linux/block/genhd.c
<<
>>
Prefs
   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#include <linux/log2.h>
  22
  23#include "blk.h"
  24
  25static DEFINE_MUTEX(block_class_lock);
  26struct kobject *block_depr;
  27
  28/* for extended dynamic devt allocation, currently only one major is used */
  29#define MAX_EXT_DEVT            (1 << MINORBITS)
  30
  31/* For extended devt allocation.  ext_devt_mutex prevents look up
  32 * results from going away underneath its user.
  33 */
  34static DEFINE_MUTEX(ext_devt_mutex);
  35static DEFINE_IDR(ext_devt_idr);
  36
  37static struct device_type disk_type;
  38
  39static void disk_add_events(struct gendisk *disk);
  40static void disk_del_events(struct gendisk *disk);
  41static void disk_release_events(struct gendisk *disk);
  42
  43/**
  44 * disk_get_part - get partition
  45 * @disk: disk to look partition from
  46 * @partno: partition number
  47 *
  48 * Look for partition @partno from @disk.  If found, increment
  49 * reference count and return it.
  50 *
  51 * CONTEXT:
  52 * Don't care.
  53 *
  54 * RETURNS:
  55 * Pointer to the found partition on success, NULL if not found.
  56 */
  57struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
  58{
  59        struct hd_struct *part = NULL;
  60        struct disk_part_tbl *ptbl;
  61
  62        if (unlikely(partno < 0))
  63                return NULL;
  64
  65        rcu_read_lock();
  66
  67        ptbl = rcu_dereference(disk->part_tbl);
  68        if (likely(partno < ptbl->len)) {
  69                part = rcu_dereference(ptbl->part[partno]);
  70                if (part)
  71                        get_device(part_to_dev(part));
  72        }
  73
  74        rcu_read_unlock();
  75
  76        return part;
  77}
  78EXPORT_SYMBOL_GPL(disk_get_part);
  79
  80/**
  81 * disk_part_iter_init - initialize partition iterator
  82 * @piter: iterator to initialize
  83 * @disk: disk to iterate over
  84 * @flags: DISK_PITER_* flags
  85 *
  86 * Initialize @piter so that it iterates over partitions of @disk.
  87 *
  88 * CONTEXT:
  89 * Don't care.
  90 */
  91void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
  92                          unsigned int flags)
  93{
  94        struct disk_part_tbl *ptbl;
  95
  96        rcu_read_lock();
  97        ptbl = rcu_dereference(disk->part_tbl);
  98
  99        piter->disk = disk;
 100        piter->part = NULL;
 101
 102        if (flags & DISK_PITER_REVERSE)
 103                piter->idx = ptbl->len - 1;
 104        else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
 105                piter->idx = 0;
 106        else
 107                piter->idx = 1;
 108
 109        piter->flags = flags;
 110
 111        rcu_read_unlock();
 112}
 113EXPORT_SYMBOL_GPL(disk_part_iter_init);
 114
 115/**
 116 * disk_part_iter_next - proceed iterator to the next partition and return it
 117 * @piter: iterator of interest
 118 *
 119 * Proceed @piter to the next partition and return it.
 120 *
 121 * CONTEXT:
 122 * Don't care.
 123 */
 124struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
 125{
 126        struct disk_part_tbl *ptbl;
 127        int inc, end;
 128
 129        /* put the last partition */
 130        disk_put_part(piter->part);
 131        piter->part = NULL;
 132
 133        /* get part_tbl */
 134        rcu_read_lock();
 135        ptbl = rcu_dereference(piter->disk->part_tbl);
 136
 137        /* determine iteration parameters */
 138        if (piter->flags & DISK_PITER_REVERSE) {
 139                inc = -1;
 140                if (piter->flags & (DISK_PITER_INCL_PART0 |
 141                                    DISK_PITER_INCL_EMPTY_PART0))
 142                        end = -1;
 143                else
 144                        end = 0;
 145        } else {
 146                inc = 1;
 147                end = ptbl->len;
 148        }
 149
 150        /* iterate to the next partition */
 151        for (; piter->idx != end; piter->idx += inc) {
 152                struct hd_struct *part;
 153
 154                part = rcu_dereference(ptbl->part[piter->idx]);
 155                if (!part)
 156                        continue;
 157                if (!part->nr_sects &&
 158                    !(piter->flags & DISK_PITER_INCL_EMPTY) &&
 159                    !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
 160                      piter->idx == 0))
 161                        continue;
 162
 163                get_device(part_to_dev(part));
 164                piter->part = part;
 165                piter->idx += inc;
 166                break;
 167        }
 168
 169        rcu_read_unlock();
 170
 171        return piter->part;
 172}
 173EXPORT_SYMBOL_GPL(disk_part_iter_next);
 174
 175/**
 176 * disk_part_iter_exit - finish up partition iteration
 177 * @piter: iter of interest
 178 *
 179 * Called when iteration is over.  Cleans up @piter.
 180 *
 181 * CONTEXT:
 182 * Don't care.
 183 */
 184void disk_part_iter_exit(struct disk_part_iter *piter)
 185{
 186        disk_put_part(piter->part);
 187        piter->part = NULL;
 188}
 189EXPORT_SYMBOL_GPL(disk_part_iter_exit);
 190
 191static inline int sector_in_part(struct hd_struct *part, sector_t sector)
 192{
 193        return part->start_sect <= sector &&
 194                sector < part->start_sect + part->nr_sects;
 195}
 196
 197/**
 198 * disk_map_sector_rcu - map sector to partition
 199 * @disk: gendisk of interest
 200 * @sector: sector to map
 201 *
 202 * Find out which partition @sector maps to on @disk.  This is
 203 * primarily used for stats accounting.
 204 *
 205 * CONTEXT:
 206 * RCU read locked.  The returned partition pointer is valid only
 207 * while preemption is disabled.
 208 *
 209 * RETURNS:
 210 * Found partition on success, part0 is returned if no partition matches
 211 */
 212struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
 213{
 214        struct disk_part_tbl *ptbl;
 215        struct hd_struct *part;
 216        int i;
 217
 218        ptbl = rcu_dereference(disk->part_tbl);
 219
 220        part = rcu_dereference(ptbl->last_lookup);
 221        if (part && sector_in_part(part, sector))
 222                return part;
 223
 224        for (i = 1; i < ptbl->len; i++) {
 225                part = rcu_dereference(ptbl->part[i]);
 226
 227                if (part && sector_in_part(part, sector)) {
 228                        rcu_assign_pointer(ptbl->last_lookup, part);
 229                        return part;
 230                }
 231        }
 232        return &disk->part0;
 233}
 234EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
 235
 236/*
 237 * Can be deleted altogether. Later.
 238 *
 239 */
 240static struct blk_major_name {
 241        struct blk_major_name *next;
 242        int major;
 243        char name[16];
 244} *major_names[BLKDEV_MAJOR_HASH_SIZE];
 245
 246/* index in the above - for now: assume no multimajor ranges */
 247static inline int major_to_index(unsigned major)
 248{
 249        return major % BLKDEV_MAJOR_HASH_SIZE;
 250}
 251
 252#ifdef CONFIG_PROC_FS
 253void blkdev_show(struct seq_file *seqf, off_t offset)
 254{
 255        struct blk_major_name *dp;
 256
 257        if (offset < BLKDEV_MAJOR_HASH_SIZE) {
 258                mutex_lock(&block_class_lock);
 259                for (dp = major_names[offset]; dp; dp = dp->next)
 260                        seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
 261                mutex_unlock(&block_class_lock);
 262        }
 263}
 264#endif /* CONFIG_PROC_FS */
 265
 266/**
 267 * register_blkdev - register a new block device
 268 *
 269 * @major: the requested major device number [1..255]. If @major=0, try to
 270 *         allocate any unused major number.
 271 * @name: the name of the new block device as a zero terminated string
 272 *
 273 * The @name must be unique within the system.
 274 *
 275 * The return value depends on the @major input parameter.
 276 *  - if a major device number was requested in range [1..255] then the
 277 *    function returns zero on success, or a negative error code
 278 *  - if any unused major number was requested with @major=0 parameter
 279 *    then the return value is the allocated major number in range
 280 *    [1..255] or a negative error code otherwise
 281 */
 282int register_blkdev(unsigned int major, const char *name)
 283{
 284        struct blk_major_name **n, *p;
 285        int index, ret = 0;
 286
 287        mutex_lock(&block_class_lock);
 288
 289        /* temporary */
 290        if (major == 0) {
 291                for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
 292                        if (major_names[index] == NULL)
 293                                break;
 294                }
 295
 296                if (index == 0) {
 297                        printk("register_blkdev: failed to get major for %s\n",
 298                               name);
 299                        ret = -EBUSY;
 300                        goto out;
 301                }
 302                major = index;
 303                ret = major;
 304        }
 305
 306        p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
 307        if (p == NULL) {
 308                ret = -ENOMEM;
 309                goto out;
 310        }
 311
 312        p->major = major;
 313        strlcpy(p->name, name, sizeof(p->name));
 314        p->next = NULL;
 315        index = major_to_index(major);
 316
 317        for (n = &major_names[index]; *n; n = &(*n)->next) {
 318                if ((*n)->major == major)
 319                        break;
 320        }
 321        if (!*n)
 322                *n = p;
 323        else
 324                ret = -EBUSY;
 325
 326        if (ret < 0) {
 327                printk("register_blkdev: cannot get major %d for %s\n",
 328                       major, name);
 329                kfree(p);
 330        }
 331out:
 332        mutex_unlock(&block_class_lock);
 333        return ret;
 334}
 335
 336EXPORT_SYMBOL(register_blkdev);
 337
 338void unregister_blkdev(unsigned int major, const char *name)
 339{
 340        struct blk_major_name **n;
 341        struct blk_major_name *p = NULL;
 342        int index = major_to_index(major);
 343
 344        mutex_lock(&block_class_lock);
 345        for (n = &major_names[index]; *n; n = &(*n)->next)
 346                if ((*n)->major == major)
 347                        break;
 348        if (!*n || strcmp((*n)->name, name)) {
 349                WARN_ON(1);
 350        } else {
 351                p = *n;
 352                *n = p->next;
 353        }
 354        mutex_unlock(&block_class_lock);
 355        kfree(p);
 356}
 357
 358EXPORT_SYMBOL(unregister_blkdev);
 359
 360static struct kobj_map *bdev_map;
 361
 362/**
 363 * blk_mangle_minor - scatter minor numbers apart
 364 * @minor: minor number to mangle
 365 *
 366 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
 367 * is enabled.  Mangling twice gives the original value.
 368 *
 369 * RETURNS:
 370 * Mangled value.
 371 *
 372 * CONTEXT:
 373 * Don't care.
 374 */
 375static int blk_mangle_minor(int minor)
 376{
 377#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
 378        int i;
 379
 380        for (i = 0; i < MINORBITS / 2; i++) {
 381                int low = minor & (1 << i);
 382                int high = minor & (1 << (MINORBITS - 1 - i));
 383                int distance = MINORBITS - 1 - 2 * i;
 384
 385                minor ^= low | high;    /* clear both bits */
 386                low <<= distance;       /* swap the positions */
 387                high >>= distance;
 388                minor |= low | high;    /* and set */
 389        }
 390#endif
 391        return minor;
 392}
 393
 394/**
 395 * blk_alloc_devt - allocate a dev_t for a partition
 396 * @part: partition to allocate dev_t for
 397 * @devt: out parameter for resulting dev_t
 398 *
 399 * Allocate a dev_t for block device.
 400 *
 401 * RETURNS:
 402 * 0 on success, allocated dev_t is returned in *@devt.  -errno on
 403 * failure.
 404 *
 405 * CONTEXT:
 406 * Might sleep.
 407 */
 408int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
 409{
 410        struct gendisk *disk = part_to_disk(part);
 411        int idx, rc;
 412
 413        /* in consecutive minor range? */
 414        if (part->partno < disk->minors) {
 415                *devt = MKDEV(disk->major, disk->first_minor + part->partno);
 416                return 0;
 417        }
 418
 419        /* allocate ext devt */
 420        do {
 421                if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
 422                        return -ENOMEM;
 423                rc = idr_get_new(&ext_devt_idr, part, &idx);
 424        } while (rc == -EAGAIN);
 425
 426        if (rc)
 427                return rc;
 428
 429        if (idx > MAX_EXT_DEVT) {
 430                idr_remove(&ext_devt_idr, idx);
 431                return -EBUSY;
 432        }
 433
 434        *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
 435        return 0;
 436}
 437
 438/**
 439 * blk_free_devt - free a dev_t
 440 * @devt: dev_t to free
 441 *
 442 * Free @devt which was allocated using blk_alloc_devt().
 443 *
 444 * CONTEXT:
 445 * Might sleep.
 446 */
 447void blk_free_devt(dev_t devt)
 448{
 449        might_sleep();
 450
 451        if (devt == MKDEV(0, 0))
 452                return;
 453
 454        if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
 455                mutex_lock(&ext_devt_mutex);
 456                idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 457                mutex_unlock(&ext_devt_mutex);
 458        }
 459}
 460
 461static char *bdevt_str(dev_t devt, char *buf)
 462{
 463        if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
 464                char tbuf[BDEVT_SIZE];
 465                snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
 466                snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
 467        } else
 468                snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
 469
 470        return buf;
 471}
 472
 473/*
 474 * Register device numbers dev..(dev+range-1)
 475 * range must be nonzero
 476 * The hash chain is sorted on range, so that subranges can override.
 477 */
 478void blk_register_region(dev_t devt, unsigned long range, struct module *module,
 479                         struct kobject *(*probe)(dev_t, int *, void *),
 480                         int (*lock)(dev_t, void *), void *data)
 481{
 482        kobj_map(bdev_map, devt, range, module, probe, lock, data);
 483}
 484
 485EXPORT_SYMBOL(blk_register_region);
 486
 487void blk_unregister_region(dev_t devt, unsigned long range)
 488{
 489        kobj_unmap(bdev_map, devt, range);
 490}
 491
 492EXPORT_SYMBOL(blk_unregister_region);
 493
 494static struct kobject *exact_match(dev_t devt, int *partno, void *data)
 495{
 496        struct gendisk *p = data;
 497
 498        return &disk_to_dev(p)->kobj;
 499}
 500
 501static int exact_lock(dev_t devt, void *data)
 502{
 503        struct gendisk *p = data;
 504
 505        if (!get_disk(p))
 506                return -1;
 507        return 0;
 508}
 509
 510void register_disk(struct gendisk *disk)
 511{
 512        struct device *ddev = disk_to_dev(disk);
 513        struct block_device *bdev;
 514        struct disk_part_iter piter;
 515        struct hd_struct *part;
 516        int err;
 517
 518        ddev->parent = disk->driverfs_dev;
 519
 520        dev_set_name(ddev, disk->disk_name);
 521
 522        /* delay uevents, until we scanned partition table */
 523        dev_set_uevent_suppress(ddev, 1);
 524
 525        if (device_add(ddev))
 526                return;
 527        if (!sysfs_deprecated) {
 528                err = sysfs_create_link(block_depr, &ddev->kobj,
 529                                        kobject_name(&ddev->kobj));
 530                if (err) {
 531                        device_del(ddev);
 532                        return;
 533                }
 534        }
 535        disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
 536        disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
 537
 538        /* No minors to use for partitions */
 539        if (!disk_partitionable(disk))
 540                goto exit;
 541
 542        /* No such device (e.g., media were just removed) */
 543        if (!get_capacity(disk))
 544                goto exit;
 545
 546        bdev = bdget_disk(disk, 0);
 547        if (!bdev)
 548                goto exit;
 549
 550        bdev->bd_invalidated = 1;
 551        err = blkdev_get(bdev, FMODE_READ, NULL);
 552        if (err < 0)
 553                goto exit;
 554        blkdev_put(bdev, FMODE_READ);
 555
 556exit:
 557        /* announce disk after possible partitions are created */
 558        dev_set_uevent_suppress(ddev, 0);
 559        kobject_uevent(&ddev->kobj, KOBJ_ADD);
 560
 561        /* announce possible partitions */
 562        disk_part_iter_init(&piter, disk, 0);
 563        while ((part = disk_part_iter_next(&piter)))
 564                kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
 565        disk_part_iter_exit(&piter);
 566}
 567
 568/**
 569 * add_disk - add partitioning information to kernel list
 570 * @disk: per-device partitioning information
 571 *
 572 * This function registers the partitioning information in @disk
 573 * with the kernel.
 574 *
 575 * FIXME: error handling
 576 */
 577void add_disk(struct gendisk *disk)
 578{
 579        struct backing_dev_info *bdi;
 580        dev_t devt;
 581        int retval;
 582
 583        /* minors == 0 indicates to use ext devt from part0 and should
 584         * be accompanied with EXT_DEVT flag.  Make sure all
 585         * parameters make sense.
 586         */
 587        WARN_ON(disk->minors && !(disk->major || disk->first_minor));
 588        WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
 589
 590        disk->flags |= GENHD_FL_UP;
 591
 592        retval = blk_alloc_devt(&disk->part0, &devt);
 593        if (retval) {
 594                WARN_ON(1);
 595                return;
 596        }
 597        disk_to_dev(disk)->devt = devt;
 598
 599        /* ->major and ->first_minor aren't supposed to be
 600         * dereferenced from here on, but set them just in case.
 601         */
 602        disk->major = MAJOR(devt);
 603        disk->first_minor = MINOR(devt);
 604
 605        /* Register BDI before referencing it from bdev */ 
 606        bdi = &disk->queue->backing_dev_info;
 607        bdi_register_dev(bdi, disk_devt(disk));
 608
 609        blk_register_region(disk_devt(disk), disk->minors, NULL,
 610                            exact_match, exact_lock, disk);
 611        register_disk(disk);
 612        blk_register_queue(disk);
 613
 614        retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
 615                                   "bdi");
 616        WARN_ON(retval);
 617
 618        disk_add_events(disk);
 619}
 620EXPORT_SYMBOL(add_disk);
 621
 622void del_gendisk(struct gendisk *disk)
 623{
 624        struct disk_part_iter piter;
 625        struct hd_struct *part;
 626
 627        disk_del_events(disk);
 628
 629        /* invalidate stuff */
 630        disk_part_iter_init(&piter, disk,
 631                             DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
 632        while ((part = disk_part_iter_next(&piter))) {
 633                invalidate_partition(disk, part->partno);
 634                delete_partition(disk, part->partno);
 635        }
 636        disk_part_iter_exit(&piter);
 637
 638        invalidate_partition(disk, 0);
 639        blk_free_devt(disk_to_dev(disk)->devt);
 640        set_capacity(disk, 0);
 641        disk->flags &= ~GENHD_FL_UP;
 642
 643        sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
 644        bdi_unregister(&disk->queue->backing_dev_info);
 645        blk_unregister_queue(disk);
 646        blk_unregister_region(disk_devt(disk), disk->minors);
 647
 648        part_stat_set_all(&disk->part0, 0);
 649        disk->part0.stamp = 0;
 650
 651        kobject_put(disk->part0.holder_dir);
 652        kobject_put(disk->slave_dir);
 653        disk->driverfs_dev = NULL;
 654        if (!sysfs_deprecated)
 655                sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
 656        device_del(disk_to_dev(disk));
 657}
 658EXPORT_SYMBOL(del_gendisk);
 659
 660/**
 661 * get_gendisk - get partitioning information for a given device
 662 * @devt: device to get partitioning information for
 663 * @partno: returned partition index
 664 *
 665 * This function gets the structure containing partitioning
 666 * information for the given device @devt.
 667 */
 668struct gendisk *get_gendisk(dev_t devt, int *partno)
 669{
 670        struct gendisk *disk = NULL;
 671
 672        if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
 673                struct kobject *kobj;
 674
 675                kobj = kobj_lookup(bdev_map, devt, partno);
 676                if (kobj)
 677                        disk = dev_to_disk(kobj_to_dev(kobj));
 678        } else {
 679                struct hd_struct *part;
 680
 681                mutex_lock(&ext_devt_mutex);
 682                part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 683                if (part && get_disk(part_to_disk(part))) {
 684                        *partno = part->partno;
 685                        disk = part_to_disk(part);
 686                }
 687                mutex_unlock(&ext_devt_mutex);
 688        }
 689
 690        return disk;
 691}
 692EXPORT_SYMBOL(get_gendisk);
 693
 694/**
 695 * bdget_disk - do bdget() by gendisk and partition number
 696 * @disk: gendisk of interest
 697 * @partno: partition number
 698 *
 699 * Find partition @partno from @disk, do bdget() on it.
 700 *
 701 * CONTEXT:
 702 * Don't care.
 703 *
 704 * RETURNS:
 705 * Resulting block_device on success, NULL on failure.
 706 */
 707struct block_device *bdget_disk(struct gendisk *disk, int partno)
 708{
 709        struct hd_struct *part;
 710        struct block_device *bdev = NULL;
 711
 712        part = disk_get_part(disk, partno);
 713        if (part)
 714                bdev = bdget(part_devt(part));
 715        disk_put_part(part);
 716
 717        return bdev;
 718}
 719EXPORT_SYMBOL(bdget_disk);
 720
 721/*
 722 * print a full list of all partitions - intended for places where the root
 723 * filesystem can't be mounted and thus to give the victim some idea of what
 724 * went wrong
 725 */
 726void __init printk_all_partitions(void)
 727{
 728        struct class_dev_iter iter;
 729        struct device *dev;
 730
 731        class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
 732        while ((dev = class_dev_iter_next(&iter))) {
 733                struct gendisk *disk = dev_to_disk(dev);
 734                struct disk_part_iter piter;
 735                struct hd_struct *part;
 736                char name_buf[BDEVNAME_SIZE];
 737                char devt_buf[BDEVT_SIZE];
 738                u8 uuid[PARTITION_META_INFO_UUIDLTH * 2 + 1];
 739
 740                /*
 741                 * Don't show empty devices or things that have been
 742                 * suppressed
 743                 */
 744                if (get_capacity(disk) == 0 ||
 745                    (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 746                        continue;
 747
 748                /*
 749                 * Note, unlike /proc/partitions, I am showing the
 750                 * numbers in hex - the same format as the root=
 751                 * option takes.
 752                 */
 753                disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 754                while ((part = disk_part_iter_next(&piter))) {
 755                        bool is_part0 = part == &disk->part0;
 756
 757                        uuid[0] = 0;
 758                        if (part->info)
 759                                part_unpack_uuid(part->info->uuid, uuid);
 760
 761                        printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
 762                               bdevt_str(part_devt(part), devt_buf),
 763                               (unsigned long long)part->nr_sects >> 1,
 764                               disk_name(disk, part->partno, name_buf), uuid);
 765                        if (is_part0) {
 766                                if (disk->driverfs_dev != NULL &&
 767                                    disk->driverfs_dev->driver != NULL)
 768                                        printk(" driver: %s\n",
 769                                              disk->driverfs_dev->driver->name);
 770                                else
 771                                        printk(" (driver?)\n");
 772                        } else
 773                                printk("\n");
 774                }
 775                disk_part_iter_exit(&piter);
 776        }
 777        class_dev_iter_exit(&iter);
 778}
 779
 780#ifdef CONFIG_PROC_FS
 781/* iterator */
 782static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
 783{
 784        loff_t skip = *pos;
 785        struct class_dev_iter *iter;
 786        struct device *dev;
 787
 788        iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 789        if (!iter)
 790                return ERR_PTR(-ENOMEM);
 791
 792        seqf->private = iter;
 793        class_dev_iter_init(iter, &block_class, NULL, &disk_type);
 794        do {
 795                dev = class_dev_iter_next(iter);
 796                if (!dev)
 797                        return NULL;
 798        } while (skip--);
 799
 800        return dev_to_disk(dev);
 801}
 802
 803static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
 804{
 805        struct device *dev;
 806
 807        (*pos)++;
 808        dev = class_dev_iter_next(seqf->private);
 809        if (dev)
 810                return dev_to_disk(dev);
 811
 812        return NULL;
 813}
 814
 815static void disk_seqf_stop(struct seq_file *seqf, void *v)
 816{
 817        struct class_dev_iter *iter = seqf->private;
 818
 819        /* stop is called even after start failed :-( */
 820        if (iter) {
 821                class_dev_iter_exit(iter);
 822                kfree(iter);
 823        }
 824}
 825
 826static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
 827{
 828        static void *p;
 829
 830        p = disk_seqf_start(seqf, pos);
 831        if (!IS_ERR_OR_NULL(p) && !*pos)
 832                seq_puts(seqf, "major minor  #blocks  name\n\n");
 833        return p;
 834}
 835
 836static int show_partition(struct seq_file *seqf, void *v)
 837{
 838        struct gendisk *sgp = v;
 839        struct disk_part_iter piter;
 840        struct hd_struct *part;
 841        char buf[BDEVNAME_SIZE];
 842
 843        /* Don't show non-partitionable removeable devices or empty devices */
 844        if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
 845                                   (sgp->flags & GENHD_FL_REMOVABLE)))
 846                return 0;
 847        if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
 848                return 0;
 849
 850        /* show the full disk and all non-0 size partitions of it */
 851        disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
 852        while ((part = disk_part_iter_next(&piter)))
 853                seq_printf(seqf, "%4d  %7d %10llu %s\n",
 854                           MAJOR(part_devt(part)), MINOR(part_devt(part)),
 855                           (unsigned long long)part->nr_sects >> 1,
 856                           disk_name(sgp, part->partno, buf));
 857        disk_part_iter_exit(&piter);
 858
 859        return 0;
 860}
 861
 862static const struct seq_operations partitions_op = {
 863        .start  = show_partition_start,
 864        .next   = disk_seqf_next,
 865        .stop   = disk_seqf_stop,
 866        .show   = show_partition
 867};
 868
 869static int partitions_open(struct inode *inode, struct file *file)
 870{
 871        return seq_open(file, &partitions_op);
 872}
 873
 874static const struct file_operations proc_partitions_operations = {
 875        .open           = partitions_open,
 876        .read           = seq_read,
 877        .llseek         = seq_lseek,
 878        .release        = seq_release,
 879};
 880#endif
 881
 882
 883static struct kobject *base_probe(dev_t devt, int *partno, void *data)
 884{
 885        if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
 886                /* Make old-style 2.4 aliases work */
 887                request_module("block-major-%d", MAJOR(devt));
 888        return NULL;
 889}
 890
 891static int __init genhd_device_init(void)
 892{
 893        int error;
 894
 895        block_class.dev_kobj = sysfs_dev_block_kobj;
 896        error = class_register(&block_class);
 897        if (unlikely(error))
 898                return error;
 899        bdev_map = kobj_map_init(base_probe, &block_class_lock);
 900        blk_dev_init();
 901
 902        register_blkdev(BLOCK_EXT_MAJOR, "blkext");
 903
 904        /* create top-level block dir */
 905        if (!sysfs_deprecated)
 906                block_depr = kobject_create_and_add("block", NULL);
 907        return 0;
 908}
 909
 910subsys_initcall(genhd_device_init);
 911
 912static ssize_t disk_range_show(struct device *dev,
 913                               struct device_attribute *attr, char *buf)
 914{
 915        struct gendisk *disk = dev_to_disk(dev);
 916
 917        return sprintf(buf, "%d\n", disk->minors);
 918}
 919
 920static ssize_t disk_ext_range_show(struct device *dev,
 921                                   struct device_attribute *attr, char *buf)
 922{
 923        struct gendisk *disk = dev_to_disk(dev);
 924
 925        return sprintf(buf, "%d\n", disk_max_parts(disk));
 926}
 927
 928static ssize_t disk_removable_show(struct device *dev,
 929                                   struct device_attribute *attr, char *buf)
 930{
 931        struct gendisk *disk = dev_to_disk(dev);
 932
 933        return sprintf(buf, "%d\n",
 934                       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
 935}
 936
 937static ssize_t disk_ro_show(struct device *dev,
 938                                   struct device_attribute *attr, char *buf)
 939{
 940        struct gendisk *disk = dev_to_disk(dev);
 941
 942        return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
 943}
 944
 945static ssize_t disk_capability_show(struct device *dev,
 946                                    struct device_attribute *attr, char *buf)
 947{
 948        struct gendisk *disk = dev_to_disk(dev);
 949
 950        return sprintf(buf, "%x\n", disk->flags);
 951}
 952
 953static ssize_t disk_alignment_offset_show(struct device *dev,
 954                                          struct device_attribute *attr,
 955                                          char *buf)
 956{
 957        struct gendisk *disk = dev_to_disk(dev);
 958
 959        return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
 960}
 961
 962static ssize_t disk_discard_alignment_show(struct device *dev,
 963                                           struct device_attribute *attr,
 964                                           char *buf)
 965{
 966        struct gendisk *disk = dev_to_disk(dev);
 967
 968        return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
 969}
 970
 971static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
 972static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
 973static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
 974static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
 975static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
 976static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
 977static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
 978                   NULL);
 979static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
 980static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
 981static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
 982#ifdef CONFIG_FAIL_MAKE_REQUEST
 983static struct device_attribute dev_attr_fail =
 984        __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
 985#endif
 986#ifdef CONFIG_FAIL_IO_TIMEOUT
 987static struct device_attribute dev_attr_fail_timeout =
 988        __ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
 989                part_timeout_store);
 990#endif
 991
 992static struct attribute *disk_attrs[] = {
 993        &dev_attr_range.attr,
 994        &dev_attr_ext_range.attr,
 995        &dev_attr_removable.attr,
 996        &dev_attr_ro.attr,
 997        &dev_attr_size.attr,
 998        &dev_attr_alignment_offset.attr,
 999        &dev_attr_discard_alignment.attr,
1000        &dev_attr_capability.attr,
1001        &dev_attr_stat.attr,
1002        &dev_attr_inflight.attr,
1003#ifdef CONFIG_FAIL_MAKE_REQUEST
1004        &dev_attr_fail.attr,
1005#endif
1006#ifdef CONFIG_FAIL_IO_TIMEOUT
1007        &dev_attr_fail_timeout.attr,
1008#endif
1009        NULL
1010};
1011
1012static struct attribute_group disk_attr_group = {
1013        .attrs = disk_attrs,
1014};
1015
1016static const struct attribute_group *disk_attr_groups[] = {
1017        &disk_attr_group,
1018        NULL
1019};
1020
1021static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
1022{
1023        struct disk_part_tbl *ptbl =
1024                container_of(head, struct disk_part_tbl, rcu_head);
1025
1026        kfree(ptbl);
1027}
1028
1029/**
1030 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1031 * @disk: disk to replace part_tbl for
1032 * @new_ptbl: new part_tbl to install
1033 *
1034 * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
1035 * original ptbl is freed using RCU callback.
1036 *
1037 * LOCKING:
1038 * Matching bd_mutx locked.
1039 */
1040static void disk_replace_part_tbl(struct gendisk *disk,
1041                                  struct disk_part_tbl *new_ptbl)
1042{
1043        struct disk_part_tbl *old_ptbl = disk->part_tbl;
1044
1045        rcu_assign_pointer(disk->part_tbl, new_ptbl);
1046
1047        if (old_ptbl) {
1048                rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1049                call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
1050        }
1051}
1052
1053/**
1054 * disk_expand_part_tbl - expand disk->part_tbl
1055 * @disk: disk to expand part_tbl for
1056 * @partno: expand such that this partno can fit in
1057 *
1058 * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
1059 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1060 *
1061 * LOCKING:
1062 * Matching bd_mutex locked, might sleep.
1063 *
1064 * RETURNS:
1065 * 0 on success, -errno on failure.
1066 */
1067int disk_expand_part_tbl(struct gendisk *disk, int partno)
1068{
1069        struct disk_part_tbl *old_ptbl = disk->part_tbl;
1070        struct disk_part_tbl *new_ptbl;
1071        int len = old_ptbl ? old_ptbl->len : 0;
1072        int target = partno + 1;
1073        size_t size;
1074        int i;
1075
1076        /* disk_max_parts() is zero during initialization, ignore if so */
1077        if (disk_max_parts(disk) && target > disk_max_parts(disk))
1078                return -EINVAL;
1079
1080        if (target <= len)
1081                return 0;
1082
1083        size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1084        new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1085        if (!new_ptbl)
1086                return -ENOMEM;
1087
1088        new_ptbl->len = target;
1089
1090        for (i = 0; i < len; i++)
1091                rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1092
1093        disk_replace_part_tbl(disk, new_ptbl);
1094        return 0;
1095}
1096
1097static void disk_release(struct device *dev)
1098{
1099        struct gendisk *disk = dev_to_disk(dev);
1100
1101        disk_release_events(disk);
1102        kfree(disk->random);
1103        disk_replace_part_tbl(disk, NULL);
1104        free_part_stats(&disk->part0);
1105        free_part_info(&disk->part0);
1106        kfree(disk);
1107}
1108struct class block_class = {
1109        .name           = "block",
1110};
1111
1112static char *block_devnode(struct device *dev, mode_t *mode)
1113{
1114        struct gendisk *disk = dev_to_disk(dev);
1115
1116        if (disk->devnode)
1117                return disk->devnode(disk, mode);
1118        return NULL;
1119}
1120
1121static struct device_type disk_type = {
1122        .name           = "disk",
1123        .groups         = disk_attr_groups,
1124        .release        = disk_release,
1125        .devnode        = block_devnode,
1126};
1127
1128#ifdef CONFIG_PROC_FS
1129/*
1130 * aggregate disk stat collector.  Uses the same stats that the sysfs
1131 * entries do, above, but makes them available through one seq_file.
1132 *
1133 * The output looks suspiciously like /proc/partitions with a bunch of
1134 * extra fields.
1135 */
1136static int diskstats_show(struct seq_file *seqf, void *v)
1137{
1138        struct gendisk *gp = v;
1139        struct disk_part_iter piter;
1140        struct hd_struct *hd;
1141        char buf[BDEVNAME_SIZE];
1142        int cpu;
1143
1144        /*
1145        if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1146                seq_puts(seqf,  "major minor name"
1147                                "     rio rmerge rsect ruse wio wmerge "
1148                                "wsect wuse running use aveq"
1149                                "\n\n");
1150        */
1151 
1152        disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1153        while ((hd = disk_part_iter_next(&piter))) {
1154                cpu = part_stat_lock();
1155                part_round_stats(cpu, hd);
1156                part_stat_unlock();
1157                seq_printf(seqf, "%4d %7d %s %lu %lu %llu "
1158                           "%u %lu %lu %llu %u %u %u %u\n",
1159                           MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1160                           disk_name(gp, hd->partno, buf),
1161                           part_stat_read(hd, ios[READ]),
1162                           part_stat_read(hd, merges[READ]),
1163                           (unsigned long long)part_stat_read(hd, sectors[READ]),
1164                           jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1165                           part_stat_read(hd, ios[WRITE]),
1166                           part_stat_read(hd, merges[WRITE]),
1167                           (unsigned long long)part_stat_read(hd, sectors[WRITE]),
1168                           jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1169                           part_in_flight(hd),
1170                           jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1171                           jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1172                        );
1173        }
1174        disk_part_iter_exit(&piter);
1175 
1176        return 0;
1177}
1178
1179static const struct seq_operations diskstats_op = {
1180        .start  = disk_seqf_start,
1181        .next   = disk_seqf_next,
1182        .stop   = disk_seqf_stop,
1183        .show   = diskstats_show
1184};
1185
1186static int diskstats_open(struct inode *inode, struct file *file)
1187{
1188        return seq_open(file, &diskstats_op);
1189}
1190
1191static const struct file_operations proc_diskstats_operations = {
1192        .open           = diskstats_open,
1193        .read           = seq_read,
1194        .llseek         = seq_lseek,
1195        .release        = seq_release,
1196};
1197
1198static int __init proc_genhd_init(void)
1199{
1200        proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1201        proc_create("partitions", 0, NULL, &proc_partitions_operations);
1202        return 0;
1203}
1204module_init(proc_genhd_init);
1205#endif /* CONFIG_PROC_FS */
1206
1207dev_t blk_lookup_devt(const char *name, int partno)
1208{
1209        dev_t devt = MKDEV(0, 0);
1210        struct class_dev_iter iter;
1211        struct device *dev;
1212
1213        class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1214        while ((dev = class_dev_iter_next(&iter))) {
1215                struct gendisk *disk = dev_to_disk(dev);
1216                struct hd_struct *part;
1217
1218                if (strcmp(dev_name(dev), name))
1219                        continue;
1220
1221                if (partno < disk->minors) {
1222                        /* We need to return the right devno, even
1223                         * if the partition doesn't exist yet.
1224                         */
1225                        devt = MKDEV(MAJOR(dev->devt),
1226                                     MINOR(dev->devt) + partno);
1227                        break;
1228                }
1229                part = disk_get_part(disk, partno);
1230                if (part) {
1231                        devt = part_devt(part);
1232                        disk_put_part(part);
1233                        break;
1234                }
1235                disk_put_part(part);
1236        }
1237        class_dev_iter_exit(&iter);
1238        return devt;
1239}
1240EXPORT_SYMBOL(blk_lookup_devt);
1241
1242struct gendisk *alloc_disk(int minors)
1243{
1244        return alloc_disk_node(minors, -1);
1245}
1246EXPORT_SYMBOL(alloc_disk);
1247
1248struct gendisk *alloc_disk_node(int minors, int node_id)
1249{
1250        struct gendisk *disk;
1251
1252        disk = kmalloc_node(sizeof(struct gendisk),
1253                                GFP_KERNEL | __GFP_ZERO, node_id);
1254        if (disk) {
1255                if (!init_part_stats(&disk->part0)) {
1256                        kfree(disk);
1257                        return NULL;
1258                }
1259                disk->node_id = node_id;
1260                if (disk_expand_part_tbl(disk, 0)) {
1261                        free_part_stats(&disk->part0);
1262                        kfree(disk);
1263                        return NULL;
1264                }
1265                disk->part_tbl->part[0] = &disk->part0;
1266
1267                hd_ref_init(&disk->part0);
1268
1269                disk->minors = minors;
1270                rand_initialize_disk(disk);
1271                disk_to_dev(disk)->class = &block_class;
1272                disk_to_dev(disk)->type = &disk_type;
1273                device_initialize(disk_to_dev(disk));
1274        }
1275        return disk;
1276}
1277EXPORT_SYMBOL(alloc_disk_node);
1278
1279struct kobject *get_disk(struct gendisk *disk)
1280{
1281        struct module *owner;
1282        struct kobject *kobj;
1283
1284        if (!disk->fops)
1285                return NULL;
1286        owner = disk->fops->owner;
1287        if (owner && !try_module_get(owner))
1288                return NULL;
1289        kobj = kobject_get(&disk_to_dev(disk)->kobj);
1290        if (kobj == NULL) {
1291                module_put(owner);
1292                return NULL;
1293        }
1294        return kobj;
1295
1296}
1297
1298EXPORT_SYMBOL(get_disk);
1299
1300void put_disk(struct gendisk *disk)
1301{
1302        if (disk)
1303                kobject_put(&disk_to_dev(disk)->kobj);
1304}
1305
1306EXPORT_SYMBOL(put_disk);
1307
1308static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1309{
1310        char event[] = "DISK_RO=1";
1311        char *envp[] = { event, NULL };
1312
1313        if (!ro)
1314                event[8] = '0';
1315        kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1316}
1317
1318void set_device_ro(struct block_device *bdev, int flag)
1319{
1320        bdev->bd_part->policy = flag;
1321}
1322
1323EXPORT_SYMBOL(set_device_ro);
1324
1325void set_disk_ro(struct gendisk *disk, int flag)
1326{
1327        struct disk_part_iter piter;
1328        struct hd_struct *part;
1329
1330        if (disk->part0.policy != flag) {
1331                set_disk_ro_uevent(disk, flag);
1332                disk->part0.policy = flag;
1333        }
1334
1335        disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1336        while ((part = disk_part_iter_next(&piter)))
1337                part->policy = flag;
1338        disk_part_iter_exit(&piter);
1339}
1340
1341EXPORT_SYMBOL(set_disk_ro);
1342
1343int bdev_read_only(struct block_device *bdev)
1344{
1345        if (!bdev)
1346                return 0;
1347        return bdev->bd_part->policy;
1348}
1349
1350EXPORT_SYMBOL(bdev_read_only);
1351
1352int invalidate_partition(struct gendisk *disk, int partno)
1353{
1354        int res = 0;
1355        struct block_device *bdev = bdget_disk(disk, partno);
1356        if (bdev) {
1357                fsync_bdev(bdev);
1358                res = __invalidate_device(bdev, true);
1359                bdput(bdev);
1360        }
1361        return res;
1362}
1363
1364EXPORT_SYMBOL(invalidate_partition);
1365
1366/*
1367 * Disk events - monitor disk events like media change and eject request.
1368 */
1369struct disk_events {
1370        struct list_head        node;           /* all disk_event's */
1371        struct gendisk          *disk;          /* the associated disk */
1372        spinlock_t              lock;
1373
1374        int                     block;          /* event blocking depth */
1375        unsigned int            pending;        /* events already sent out */
1376        unsigned int            clearing;       /* events being cleared */
1377
1378        long                    poll_msecs;     /* interval, -1 for default */
1379        struct delayed_work     dwork;
1380};
1381
1382static const char *disk_events_strs[] = {
1383        [ilog2(DISK_EVENT_MEDIA_CHANGE)]        = "media_change",
1384        [ilog2(DISK_EVENT_EJECT_REQUEST)]       = "eject_request",
1385};
1386
1387static char *disk_uevents[] = {
1388        [ilog2(DISK_EVENT_MEDIA_CHANGE)]        = "DISK_MEDIA_CHANGE=1",
1389        [ilog2(DISK_EVENT_EJECT_REQUEST)]       = "DISK_EJECT_REQUEST=1",
1390};
1391
1392/* list of all disk_events */
1393static DEFINE_MUTEX(disk_events_mutex);
1394static LIST_HEAD(disk_events);
1395
1396/* disable in-kernel polling by default */
1397static unsigned long disk_events_dfl_poll_msecs = 0;
1398
1399static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1400{
1401        struct disk_events *ev = disk->ev;
1402        long intv_msecs = 0;
1403
1404        /*
1405         * If device-specific poll interval is set, always use it.  If
1406         * the default is being used, poll iff there are events which
1407         * can't be monitored asynchronously.
1408         */
1409        if (ev->poll_msecs >= 0)
1410                intv_msecs = ev->poll_msecs;
1411        else if (disk->events & ~disk->async_events)
1412                intv_msecs = disk_events_dfl_poll_msecs;
1413
1414        return msecs_to_jiffies(intv_msecs);
1415}
1416
1417static void __disk_block_events(struct gendisk *disk, bool sync)
1418{
1419        struct disk_events *ev = disk->ev;
1420        unsigned long flags;
1421        bool cancel;
1422
1423        spin_lock_irqsave(&ev->lock, flags);
1424        cancel = !ev->block++;
1425        spin_unlock_irqrestore(&ev->lock, flags);
1426
1427        if (cancel) {
1428                if (sync)
1429                        cancel_delayed_work_sync(&disk->ev->dwork);
1430                else
1431                        cancel_delayed_work(&disk->ev->dwork);
1432        }
1433}
1434
1435static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1436{
1437        struct disk_events *ev = disk->ev;
1438        unsigned long intv;
1439        unsigned long flags;
1440
1441        spin_lock_irqsave(&ev->lock, flags);
1442
1443        if (WARN_ON_ONCE(ev->block <= 0))
1444                goto out_unlock;
1445
1446        if (--ev->block)
1447                goto out_unlock;
1448
1449        /*
1450         * Not exactly a latency critical operation, set poll timer
1451         * slack to 25% and kick event check.
1452         */
1453        intv = disk_events_poll_jiffies(disk);
1454        set_timer_slack(&ev->dwork.timer, intv / 4);
1455        if (check_now)
1456                queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1457        else if (intv)
1458                queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1459out_unlock:
1460        spin_unlock_irqrestore(&ev->lock, flags);
1461}
1462
1463/**
1464 * disk_block_events - block and flush disk event checking
1465 * @disk: disk to block events for
1466 *
1467 * On return from this function, it is guaranteed that event checking
1468 * isn't in progress and won't happen until unblocked by
1469 * disk_unblock_events().  Events blocking is counted and the actual
1470 * unblocking happens after the matching number of unblocks are done.
1471 *
1472 * Note that this intentionally does not block event checking from
1473 * disk_clear_events().
1474 *
1475 * CONTEXT:
1476 * Might sleep.
1477 */
1478void disk_block_events(struct gendisk *disk)
1479{
1480        if (disk->ev)
1481                __disk_block_events(disk, true);
1482}
1483
1484/**
1485 * disk_unblock_events - unblock disk event checking
1486 * @disk: disk to unblock events for
1487 *
1488 * Undo disk_block_events().  When the block count reaches zero, it
1489 * starts events polling if configured.
1490 *
1491 * CONTEXT:
1492 * Don't care.  Safe to call from irq context.
1493 */
1494void disk_unblock_events(struct gendisk *disk)
1495{
1496        if (disk->ev)
1497                __disk_unblock_events(disk, false);
1498}
1499
1500/**
1501 * disk_check_events - schedule immediate event checking
1502 * @disk: disk to check events for
1503 *
1504 * Schedule immediate event checking on @disk if not blocked.
1505 *
1506 * CONTEXT:
1507 * Don't care.  Safe to call from irq context.
1508 */
1509void disk_check_events(struct gendisk *disk)
1510{
1511        if (disk->ev) {
1512                __disk_block_events(disk, false);
1513                __disk_unblock_events(disk, true);
1514        }
1515}
1516EXPORT_SYMBOL_GPL(disk_check_events);
1517
1518/**
1519 * disk_clear_events - synchronously check, clear and return pending events
1520 * @disk: disk to fetch and clear events from
1521 * @mask: mask of events to be fetched and clearted
1522 *
1523 * Disk events are synchronously checked and pending events in @mask
1524 * are cleared and returned.  This ignores the block count.
1525 *
1526 * CONTEXT:
1527 * Might sleep.
1528 */
1529unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1530{
1531        const struct block_device_operations *bdops = disk->fops;
1532        struct disk_events *ev = disk->ev;
1533        unsigned int pending;
1534
1535        if (!ev) {
1536                /* for drivers still using the old ->media_changed method */
1537                if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1538                    bdops->media_changed && bdops->media_changed(disk))
1539                        return DISK_EVENT_MEDIA_CHANGE;
1540                return 0;
1541        }
1542
1543        /* tell the workfn about the events being cleared */
1544        spin_lock_irq(&ev->lock);
1545        ev->clearing |= mask;
1546        spin_unlock_irq(&ev->lock);
1547
1548        /* uncondtionally schedule event check and wait for it to finish */
1549        __disk_block_events(disk, true);
1550        queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1551        flush_delayed_work(&ev->dwork);
1552        __disk_unblock_events(disk, false);
1553
1554        /* then, fetch and clear pending events */
1555        spin_lock_irq(&ev->lock);
1556        WARN_ON_ONCE(ev->clearing & mask);      /* cleared by workfn */
1557        pending = ev->pending & mask;
1558        ev->pending &= ~mask;
1559        spin_unlock_irq(&ev->lock);
1560
1561        return pending;
1562}
1563
1564static void disk_events_workfn(struct work_struct *work)
1565{
1566        struct delayed_work *dwork = to_delayed_work(work);
1567        struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1568        struct gendisk *disk = ev->disk;
1569        char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1570        unsigned int clearing = ev->clearing;
1571        unsigned int events;
1572        unsigned long intv;
1573        int nr_events = 0, i;
1574
1575        /* check events */
1576        events = disk->fops->check_events(disk, clearing);
1577
1578        /* accumulate pending events and schedule next poll if necessary */
1579        spin_lock_irq(&ev->lock);
1580
1581        events &= ~ev->pending;
1582        ev->pending |= events;
1583        ev->clearing &= ~clearing;
1584
1585        intv = disk_events_poll_jiffies(disk);
1586        if (!ev->block && intv)
1587                queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1588
1589        spin_unlock_irq(&ev->lock);
1590
1591        /*
1592         * Tell userland about new events.  Only the events listed in
1593         * @disk->events are reported.  Unlisted events are processed the
1594         * same internally but never get reported to userland.
1595         */
1596        for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1597                if (events & disk->events & (1 << i))
1598                        envp[nr_events++] = disk_uevents[i];
1599
1600        if (nr_events)
1601                kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1602}
1603
1604/*
1605 * A disk events enabled device has the following sysfs nodes under
1606 * its /sys/block/X/ directory.
1607 *
1608 * events               : list of all supported events
1609 * events_async         : list of events which can be detected w/o polling
1610 * events_poll_msecs    : polling interval, 0: disable, -1: system default
1611 */
1612static ssize_t __disk_events_show(unsigned int events, char *buf)
1613{
1614        const char *delim = "";
1615        ssize_t pos = 0;
1616        int i;
1617
1618        for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1619                if (events & (1 << i)) {
1620                        pos += sprintf(buf + pos, "%s%s",
1621                                       delim, disk_events_strs[i]);
1622                        delim = " ";
1623                }
1624        if (pos)
1625                pos += sprintf(buf + pos, "\n");
1626        return pos;
1627}
1628
1629static ssize_t disk_events_show(struct device *dev,
1630                                struct device_attribute *attr, char *buf)
1631{
1632        struct gendisk *disk = dev_to_disk(dev);
1633
1634        return __disk_events_show(disk->events, buf);
1635}
1636
1637static ssize_t disk_events_async_show(struct device *dev,
1638                                      struct device_attribute *attr, char *buf)
1639{
1640        struct gendisk *disk = dev_to_disk(dev);
1641
1642        return __disk_events_show(disk->async_events, buf);
1643}
1644
1645static ssize_t disk_events_poll_msecs_show(struct device *dev,
1646                                           struct device_attribute *attr,
1647                                           char *buf)
1648{
1649        struct gendisk *disk = dev_to_disk(dev);
1650
1651        return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1652}
1653
1654static ssize_t disk_events_poll_msecs_store(struct device *dev,
1655                                            struct device_attribute *attr,
1656                                            const char *buf, size_t count)
1657{
1658        struct gendisk *disk = dev_to_disk(dev);
1659        long intv;
1660
1661        if (!count || !sscanf(buf, "%ld", &intv))
1662                return -EINVAL;
1663
1664        if (intv < 0 && intv != -1)
1665                return -EINVAL;
1666
1667        __disk_block_events(disk, true);
1668        disk->ev->poll_msecs = intv;
1669        __disk_unblock_events(disk, true);
1670
1671        return count;
1672}
1673
1674static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1675static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1676static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1677                         disk_events_poll_msecs_show,
1678                         disk_events_poll_msecs_store);
1679
1680static const struct attribute *disk_events_attrs[] = {
1681        &dev_attr_events.attr,
1682        &dev_attr_events_async.attr,
1683        &dev_attr_events_poll_msecs.attr,
1684        NULL,
1685};
1686
1687/*
1688 * The default polling interval can be specified by the kernel
1689 * parameter block.events_dfl_poll_msecs which defaults to 0
1690 * (disable).  This can also be modified runtime by writing to
1691 * /sys/module/block/events_dfl_poll_msecs.
1692 */
1693static int disk_events_set_dfl_poll_msecs(const char *val,
1694                                          const struct kernel_param *kp)
1695{
1696        struct disk_events *ev;
1697        int ret;
1698
1699        ret = param_set_ulong(val, kp);
1700        if (ret < 0)
1701                return ret;
1702
1703        mutex_lock(&disk_events_mutex);
1704
1705        list_for_each_entry(ev, &disk_events, node)
1706                disk_check_events(ev->disk);
1707
1708        mutex_unlock(&disk_events_mutex);
1709
1710        return 0;
1711}
1712
1713static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1714        .set    = disk_events_set_dfl_poll_msecs,
1715        .get    = param_get_ulong,
1716};
1717
1718#undef MODULE_PARAM_PREFIX
1719#define MODULE_PARAM_PREFIX     "block."
1720
1721module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1722                &disk_events_dfl_poll_msecs, 0644);
1723
1724/*
1725 * disk_{add|del|release}_events - initialize and destroy disk_events.
1726 */
1727static void disk_add_events(struct gendisk *disk)
1728{
1729        struct disk_events *ev;
1730
1731        if (!disk->fops->check_events)
1732                return;
1733
1734        ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1735        if (!ev) {
1736                pr_warn("%s: failed to initialize events\n", disk->disk_name);
1737                return;
1738        }
1739
1740        if (sysfs_create_files(&disk_to_dev(disk)->kobj,
1741                               disk_events_attrs) < 0) {
1742                pr_warn("%s: failed to create sysfs files for events\n",
1743                        disk->disk_name);
1744                kfree(ev);
1745                return;
1746        }
1747
1748        disk->ev = ev;
1749
1750        INIT_LIST_HEAD(&ev->node);
1751        ev->disk = disk;
1752        spin_lock_init(&ev->lock);
1753        ev->block = 1;
1754        ev->poll_msecs = -1;
1755        INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1756
1757        mutex_lock(&disk_events_mutex);
1758        list_add_tail(&ev->node, &disk_events);
1759        mutex_unlock(&disk_events_mutex);
1760
1761        /*
1762         * Block count is initialized to 1 and the following initial
1763         * unblock kicks it into action.
1764         */
1765        __disk_unblock_events(disk, true);
1766}
1767
1768static void disk_del_events(struct gendisk *disk)
1769{
1770        if (!disk->ev)
1771                return;
1772
1773        __disk_block_events(disk, true);
1774
1775        mutex_lock(&disk_events_mutex);
1776        list_del_init(&disk->ev->node);
1777        mutex_unlock(&disk_events_mutex);
1778
1779        sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1780}
1781
1782static void disk_release_events(struct gendisk *disk)
1783{
1784        /* the block count should be 1 from disk_del_events() */
1785        WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1786        kfree(disk->ev);
1787}
1788