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/mutex.h>
  19#include <linux/idr.h>
  20#include <linux/log2.h>
  21#include <linux/pm_runtime.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 NR_EXT_DEVT             (1 << MINORBITS)
  30
  31/* For extended devt allocation.  ext_devt_lock prevents look up
  32 * results from going away underneath its user.
  33 */
  34static DEFINE_SPINLOCK(ext_devt_lock);
  35static DEFINE_IDR(ext_devt_idr);
  36
  37static struct device_type disk_type;
  38
  39static void disk_check_events(struct disk_events *ev,
  40                              unsigned int *clearing_ptr);
  41static void disk_alloc_events(struct gendisk *disk);
  42static void disk_add_events(struct gendisk *disk);
  43static void disk_del_events(struct gendisk *disk);
  44static void disk_release_events(struct gendisk *disk);
  45
  46/**
  47 * disk_get_part - get partition
  48 * @disk: disk to look partition from
  49 * @partno: partition number
  50 *
  51 * Look for partition @partno from @disk.  If found, increment
  52 * reference count and return it.
  53 *
  54 * CONTEXT:
  55 * Don't care.
  56 *
  57 * RETURNS:
  58 * Pointer to the found partition on success, NULL if not found.
  59 */
  60struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
  61{
  62        struct hd_struct *part = NULL;
  63        struct disk_part_tbl *ptbl;
  64
  65        if (unlikely(partno < 0))
  66                return NULL;
  67
  68        rcu_read_lock();
  69
  70        ptbl = rcu_dereference(disk->part_tbl);
  71        if (likely(partno < ptbl->len)) {
  72                part = rcu_dereference(ptbl->part[partno]);
  73                if (part)
  74                        get_device(part_to_dev(part));
  75        }
  76
  77        rcu_read_unlock();
  78
  79        return part;
  80}
  81EXPORT_SYMBOL_GPL(disk_get_part);
  82
  83/**
  84 * disk_part_iter_init - initialize partition iterator
  85 * @piter: iterator to initialize
  86 * @disk: disk to iterate over
  87 * @flags: DISK_PITER_* flags
  88 *
  89 * Initialize @piter so that it iterates over partitions of @disk.
  90 *
  91 * CONTEXT:
  92 * Don't care.
  93 */
  94void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
  95                          unsigned int flags)
  96{
  97        struct disk_part_tbl *ptbl;
  98
  99        rcu_read_lock();
 100        ptbl = rcu_dereference(disk->part_tbl);
 101
 102        piter->disk = disk;
 103        piter->part = NULL;
 104
 105        if (flags & DISK_PITER_REVERSE)
 106                piter->idx = ptbl->len - 1;
 107        else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
 108                piter->idx = 0;
 109        else
 110                piter->idx = 1;
 111
 112        piter->flags = flags;
 113
 114        rcu_read_unlock();
 115}
 116EXPORT_SYMBOL_GPL(disk_part_iter_init);
 117
 118/**
 119 * disk_part_iter_next - proceed iterator to the next partition and return it
 120 * @piter: iterator of interest
 121 *
 122 * Proceed @piter to the next partition and return it.
 123 *
 124 * CONTEXT:
 125 * Don't care.
 126 */
 127struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
 128{
 129        struct disk_part_tbl *ptbl;
 130        int inc, end;
 131
 132        /* put the last partition */
 133        disk_put_part(piter->part);
 134        piter->part = NULL;
 135
 136        /* get part_tbl */
 137        rcu_read_lock();
 138        ptbl = rcu_dereference(piter->disk->part_tbl);
 139
 140        /* determine iteration parameters */
 141        if (piter->flags & DISK_PITER_REVERSE) {
 142                inc = -1;
 143                if (piter->flags & (DISK_PITER_INCL_PART0 |
 144                                    DISK_PITER_INCL_EMPTY_PART0))
 145                        end = -1;
 146                else
 147                        end = 0;
 148        } else {
 149                inc = 1;
 150                end = ptbl->len;
 151        }
 152
 153        /* iterate to the next partition */
 154        for (; piter->idx != end; piter->idx += inc) {
 155                struct hd_struct *part;
 156
 157                part = rcu_dereference(ptbl->part[piter->idx]);
 158                if (!part)
 159                        continue;
 160                if (!part_nr_sects_read(part) &&
 161                    !(piter->flags & DISK_PITER_INCL_EMPTY) &&
 162                    !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
 163                      piter->idx == 0))
 164                        continue;
 165
 166                get_device(part_to_dev(part));
 167                piter->part = part;
 168                piter->idx += inc;
 169                break;
 170        }
 171
 172        rcu_read_unlock();
 173
 174        return piter->part;
 175}
 176EXPORT_SYMBOL_GPL(disk_part_iter_next);
 177
 178/**
 179 * disk_part_iter_exit - finish up partition iteration
 180 * @piter: iter of interest
 181 *
 182 * Called when iteration is over.  Cleans up @piter.
 183 *
 184 * CONTEXT:
 185 * Don't care.
 186 */
 187void disk_part_iter_exit(struct disk_part_iter *piter)
 188{
 189        disk_put_part(piter->part);
 190        piter->part = NULL;
 191}
 192EXPORT_SYMBOL_GPL(disk_part_iter_exit);
 193
 194static inline int sector_in_part(struct hd_struct *part, sector_t sector)
 195{
 196        return part->start_sect <= sector &&
 197                sector < part->start_sect + part_nr_sects_read(part);
 198}
 199
 200/**
 201 * disk_map_sector_rcu - map sector to partition
 202 * @disk: gendisk of interest
 203 * @sector: sector to map
 204 *
 205 * Find out which partition @sector maps to on @disk.  This is
 206 * primarily used for stats accounting.
 207 *
 208 * CONTEXT:
 209 * RCU read locked.  The returned partition pointer is valid only
 210 * while preemption is disabled.
 211 *
 212 * RETURNS:
 213 * Found partition on success, part0 is returned if no partition matches
 214 */
 215struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
 216{
 217        struct disk_part_tbl *ptbl;
 218        struct hd_struct *part;
 219        int i;
 220
 221        ptbl = rcu_dereference(disk->part_tbl);
 222
 223        part = rcu_dereference(ptbl->last_lookup);
 224        if (part && sector_in_part(part, sector))
 225                return part;
 226
 227        for (i = 1; i < ptbl->len; i++) {
 228                part = rcu_dereference(ptbl->part[i]);
 229
 230                if (part && sector_in_part(part, sector)) {
 231                        rcu_assign_pointer(ptbl->last_lookup, part);
 232                        return part;
 233                }
 234        }
 235        return &disk->part0;
 236}
 237EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
 238
 239/*
 240 * Can be deleted altogether. Later.
 241 *
 242 */
 243static struct blk_major_name {
 244        struct blk_major_name *next;
 245        int major;
 246        char name[16];
 247} *major_names[BLKDEV_MAJOR_HASH_SIZE];
 248
 249/* index in the above - for now: assume no multimajor ranges */
 250static inline int major_to_index(unsigned major)
 251{
 252        return major % BLKDEV_MAJOR_HASH_SIZE;
 253}
 254
 255#ifdef CONFIG_PROC_FS
 256void blkdev_show(struct seq_file *seqf, off_t offset)
 257{
 258        struct blk_major_name *dp;
 259
 260        if (offset < BLKDEV_MAJOR_HASH_SIZE) {
 261                mutex_lock(&block_class_lock);
 262                for (dp = major_names[offset]; dp; dp = dp->next)
 263                        seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
 264                mutex_unlock(&block_class_lock);
 265        }
 266}
 267#endif /* CONFIG_PROC_FS */
 268
 269/**
 270 * register_blkdev - register a new block device
 271 *
 272 * @major: the requested major device number [1..255]. If @major=0, try to
 273 *         allocate any unused major number.
 274 * @name: the name of the new block device as a zero terminated string
 275 *
 276 * The @name must be unique within the system.
 277 *
 278 * The return value depends on the @major input parameter.
 279 *  - if a major device number was requested in range [1..255] then the
 280 *    function returns zero on success, or a negative error code
 281 *  - if any unused major number was requested with @major=0 parameter
 282 *    then the return value is the allocated major number in range
 283 *    [1..255] or a negative error code otherwise
 284 */
 285int register_blkdev(unsigned int major, const char *name)
 286{
 287        struct blk_major_name **n, *p;
 288        int index, ret = 0;
 289
 290        mutex_lock(&block_class_lock);
 291
 292        /* temporary */
 293        if (major == 0) {
 294                for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
 295                        if (major_names[index] == NULL)
 296                                break;
 297                }
 298
 299                if (index == 0) {
 300                        printk("register_blkdev: failed to get major for %s\n",
 301                               name);
 302                        ret = -EBUSY;
 303                        goto out;
 304                }
 305                major = index;
 306                ret = major;
 307        }
 308
 309        p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
 310        if (p == NULL) {
 311                ret = -ENOMEM;
 312                goto out;
 313        }
 314
 315        p->major = major;
 316        strlcpy(p->name, name, sizeof(p->name));
 317        p->next = NULL;
 318        index = major_to_index(major);
 319
 320        for (n = &major_names[index]; *n; n = &(*n)->next) {
 321                if ((*n)->major == major)
 322                        break;
 323        }
 324        if (!*n)
 325                *n = p;
 326        else
 327                ret = -EBUSY;
 328
 329        if (ret < 0) {
 330                printk("register_blkdev: cannot get major %d for %s\n",
 331                       major, name);
 332                kfree(p);
 333        }
 334out:
 335        mutex_unlock(&block_class_lock);
 336        return ret;
 337}
 338
 339EXPORT_SYMBOL(register_blkdev);
 340
 341void unregister_blkdev(unsigned int major, const char *name)
 342{
 343        struct blk_major_name **n;
 344        struct blk_major_name *p = NULL;
 345        int index = major_to_index(major);
 346
 347        mutex_lock(&block_class_lock);
 348        for (n = &major_names[index]; *n; n = &(*n)->next)
 349                if ((*n)->major == major)
 350                        break;
 351        if (!*n || strcmp((*n)->name, name)) {
 352                WARN_ON(1);
 353        } else {
 354                p = *n;
 355                *n = p->next;
 356        }
 357        mutex_unlock(&block_class_lock);
 358        kfree(p);
 359}
 360
 361EXPORT_SYMBOL(unregister_blkdev);
 362
 363static struct kobj_map *bdev_map;
 364
 365/**
 366 * blk_mangle_minor - scatter minor numbers apart
 367 * @minor: minor number to mangle
 368 *
 369 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
 370 * is enabled.  Mangling twice gives the original value.
 371 *
 372 * RETURNS:
 373 * Mangled value.
 374 *
 375 * CONTEXT:
 376 * Don't care.
 377 */
 378static int blk_mangle_minor(int minor)
 379{
 380#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
 381        int i;
 382
 383        for (i = 0; i < MINORBITS / 2; i++) {
 384                int low = minor & (1 << i);
 385                int high = minor & (1 << (MINORBITS - 1 - i));
 386                int distance = MINORBITS - 1 - 2 * i;
 387
 388                minor ^= low | high;    /* clear both bits */
 389                low <<= distance;       /* swap the positions */
 390                high >>= distance;
 391                minor |= low | high;    /* and set */
 392        }
 393#endif
 394        return minor;
 395}
 396
 397/**
 398 * blk_alloc_devt - allocate a dev_t for a partition
 399 * @part: partition to allocate dev_t for
 400 * @devt: out parameter for resulting dev_t
 401 *
 402 * Allocate a dev_t for block device.
 403 *
 404 * RETURNS:
 405 * 0 on success, allocated dev_t is returned in *@devt.  -errno on
 406 * failure.
 407 *
 408 * CONTEXT:
 409 * Might sleep.
 410 */
 411int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
 412{
 413        struct gendisk *disk = part_to_disk(part);
 414        int idx;
 415
 416        /* in consecutive minor range? */
 417        if (part->partno < disk->minors) {
 418                *devt = MKDEV(disk->major, disk->first_minor + part->partno);
 419                return 0;
 420        }
 421
 422        /* allocate ext devt */
 423        idr_preload(GFP_KERNEL);
 424
 425        spin_lock_bh(&ext_devt_lock);
 426        idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
 427        spin_unlock_bh(&ext_devt_lock);
 428
 429        idr_preload_end();
 430        if (idx < 0)
 431                return idx == -ENOSPC ? -EBUSY : idx;
 432
 433        *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
 434        return 0;
 435}
 436
 437/**
 438 * blk_free_devt - free a dev_t
 439 * @devt: dev_t to free
 440 *
 441 * Free @devt which was allocated using blk_alloc_devt().
 442 *
 443 * CONTEXT:
 444 * Might sleep.
 445 */
 446void blk_free_devt(dev_t devt)
 447{
 448        if (devt == MKDEV(0, 0))
 449                return;
 450
 451        if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
 452                spin_lock_bh(&ext_devt_lock);
 453                idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 454                spin_unlock_bh(&ext_devt_lock);
 455        }
 456}
 457
 458static 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 */
 475void 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
 482EXPORT_SYMBOL(blk_register_region);
 483
 484void blk_unregister_region(dev_t devt, unsigned long range)
 485{
 486        kobj_unmap(bdev_map, devt, range);
 487}
 488
 489EXPORT_SYMBOL(blk_unregister_region);
 490
 491static 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
 498static 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
 507static void register_disk(struct gendisk *disk)
 508{
 509        struct device *ddev = disk_to_dev(disk);
 510        struct block_device *bdev;
 511        struct disk_part_iter piter;
 512        struct hd_struct *part;
 513        int err;
 514
 515        ddev->parent = disk->driverfs_dev;
 516
 517        dev_set_name(ddev, "%s", disk->disk_name);
 518
 519        /* delay uevents, until we scanned partition table */
 520        dev_set_uevent_suppress(ddev, 1);
 521
 522        if (device_add(ddev))
 523                return;
 524        if (!sysfs_deprecated) {
 525                err = sysfs_create_link(block_depr, &ddev->kobj,
 526                                        kobject_name(&ddev->kobj));
 527                if (err) {
 528                        device_del(ddev);
 529                        return;
 530                }
 531        }
 532
 533        /*
 534         * avoid probable deadlock caused by allocating memory with
 535         * GFP_KERNEL in runtime_resume callback of its all ancestor
 536         * devices
 537         */
 538        pm_runtime_set_memalloc_noio(ddev, true);
 539
 540        disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
 541        disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
 542
 543        /* No minors to use for partitions */
 544        if (!disk_part_scan_enabled(disk))
 545                goto exit;
 546
 547        /* No such device (e.g., media were just removed) */
 548        if (!get_capacity(disk))
 549                goto exit;
 550
 551        bdev = bdget_disk(disk, 0);
 552        if (!bdev)
 553                goto exit;
 554
 555        bdev->bd_invalidated = 1;
 556        err = blkdev_get(bdev, FMODE_READ, NULL);
 557        if (err < 0)
 558                goto exit;
 559        blkdev_put(bdev, FMODE_READ);
 560
 561exit:
 562        /* announce disk after possible partitions are created */
 563        dev_set_uevent_suppress(ddev, 0);
 564        kobject_uevent(&ddev->kobj, KOBJ_ADD);
 565
 566        /* announce possible partitions */
 567        disk_part_iter_init(&piter, disk, 0);
 568        while ((part = disk_part_iter_next(&piter)))
 569                kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
 570        disk_part_iter_exit(&piter);
 571}
 572
 573/**
 574 * add_disk - add partitioning information to kernel list
 575 * @disk: per-device partitioning information
 576 *
 577 * This function registers the partitioning information in @disk
 578 * with the kernel.
 579 *
 580 * FIXME: error handling
 581 */
 582void add_disk(struct gendisk *disk)
 583{
 584        struct backing_dev_info *bdi;
 585        dev_t devt;
 586        int retval;
 587
 588        /* minors == 0 indicates to use ext devt from part0 and should
 589         * be accompanied with EXT_DEVT flag.  Make sure all
 590         * parameters make sense.
 591         */
 592        WARN_ON(disk->minors && !(disk->major || disk->first_minor));
 593        WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
 594
 595        disk->flags |= GENHD_FL_UP;
 596
 597        retval = blk_alloc_devt(&disk->part0, &devt);
 598        if (retval) {
 599                WARN_ON(1);
 600                return;
 601        }
 602        disk_to_dev(disk)->devt = devt;
 603
 604        /* ->major and ->first_minor aren't supposed to be
 605         * dereferenced from here on, but set them just in case.
 606         */
 607        disk->major = MAJOR(devt);
 608        disk->first_minor = MINOR(devt);
 609
 610        disk_alloc_events(disk);
 611
 612        /* Register BDI before referencing it from bdev */
 613        bdi = &disk->queue->backing_dev_info;
 614        bdi_register_dev(bdi, disk_devt(disk));
 615
 616        blk_register_region(disk_devt(disk), disk->minors, NULL,
 617                            exact_match, exact_lock, disk);
 618        register_disk(disk);
 619        blk_register_queue(disk);
 620
 621        /*
 622         * Take an extra ref on queue which will be put on disk_release()
 623         * so that it sticks around as long as @disk is there.
 624         */
 625        WARN_ON_ONCE(!blk_get_queue(disk->queue));
 626
 627        retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
 628                                   "bdi");
 629        WARN_ON(retval);
 630
 631        disk_add_events(disk);
 632}
 633EXPORT_SYMBOL(add_disk);
 634
 635void del_gendisk(struct gendisk *disk)
 636{
 637        struct disk_part_iter piter;
 638        struct hd_struct *part;
 639
 640        disk_del_events(disk);
 641
 642        /* invalidate stuff */
 643        disk_part_iter_init(&piter, disk,
 644                             DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
 645        while ((part = disk_part_iter_next(&piter))) {
 646                invalidate_partition(disk, part->partno);
 647                delete_partition(disk, part->partno);
 648        }
 649        disk_part_iter_exit(&piter);
 650
 651        invalidate_partition(disk, 0);
 652        set_capacity(disk, 0);
 653        disk->flags &= ~GENHD_FL_UP;
 654
 655        sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
 656        bdi_unregister(&disk->queue->backing_dev_info);
 657        blk_unregister_queue(disk);
 658        blk_unregister_region(disk_devt(disk), disk->minors);
 659
 660        part_stat_set_all(&disk->part0, 0);
 661        disk->part0.stamp = 0;
 662
 663        kobject_put(disk->part0.holder_dir);
 664        kobject_put(disk->slave_dir);
 665        if (!sysfs_deprecated)
 666                sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
 667        pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
 668        device_del(disk_to_dev(disk));
 669}
 670EXPORT_SYMBOL(del_gendisk);
 671
 672/**
 673 * get_gendisk - get partitioning information for a given device
 674 * @devt: device to get partitioning information for
 675 * @partno: returned partition index
 676 *
 677 * This function gets the structure containing partitioning
 678 * information for the given device @devt.
 679 */
 680struct gendisk *get_gendisk(dev_t devt, int *partno)
 681{
 682        struct gendisk *disk = NULL;
 683
 684        if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
 685                struct kobject *kobj;
 686
 687                kobj = kobj_lookup(bdev_map, devt, partno);
 688                if (kobj)
 689                        disk = dev_to_disk(kobj_to_dev(kobj));
 690        } else {
 691                struct hd_struct *part;
 692
 693                spin_lock_bh(&ext_devt_lock);
 694                part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 695                if (part && get_disk(part_to_disk(part))) {
 696                        *partno = part->partno;
 697                        disk = part_to_disk(part);
 698                }
 699                spin_unlock_bh(&ext_devt_lock);
 700        }
 701
 702        return disk;
 703}
 704EXPORT_SYMBOL(get_gendisk);
 705
 706/**
 707 * bdget_disk - do bdget() by gendisk and partition number
 708 * @disk: gendisk of interest
 709 * @partno: partition number
 710 *
 711 * Find partition @partno from @disk, do bdget() on it.
 712 *
 713 * CONTEXT:
 714 * Don't care.
 715 *
 716 * RETURNS:
 717 * Resulting block_device on success, NULL on failure.
 718 */
 719struct block_device *bdget_disk(struct gendisk *disk, int partno)
 720{
 721        struct hd_struct *part;
 722        struct block_device *bdev = NULL;
 723
 724        part = disk_get_part(disk, partno);
 725        if (part)
 726                bdev = bdget(part_devt(part));
 727        disk_put_part(part);
 728
 729        return bdev;
 730}
 731EXPORT_SYMBOL(bdget_disk);
 732
 733/*
 734 * print a full list of all partitions - intended for places where the root
 735 * filesystem can't be mounted and thus to give the victim some idea of what
 736 * went wrong
 737 */
 738void __init printk_all_partitions(void)
 739{
 740        struct class_dev_iter iter;
 741        struct device *dev;
 742
 743        class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
 744        while ((dev = class_dev_iter_next(&iter))) {
 745                struct gendisk *disk = dev_to_disk(dev);
 746                struct disk_part_iter piter;
 747                struct hd_struct *part;
 748                char name_buf[BDEVNAME_SIZE];
 749                char devt_buf[BDEVT_SIZE];
 750
 751                /*
 752                 * Don't show empty devices or things that have been
 753                 * suppressed
 754                 */
 755                if (get_capacity(disk) == 0 ||
 756                    (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 757                        continue;
 758
 759                /*
 760                 * Note, unlike /proc/partitions, I am showing the
 761                 * numbers in hex - the same format as the root=
 762                 * option takes.
 763                 */
 764                disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 765                while ((part = disk_part_iter_next(&piter))) {
 766                        bool is_part0 = part == &disk->part0;
 767
 768                        printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
 769                               bdevt_str(part_devt(part), devt_buf),
 770                               (unsigned long long)part_nr_sects_read(part) >> 1
 771                               , disk_name(disk, part->partno, name_buf),
 772                               part->info ? part->info->uuid : "");
 773                        if (is_part0) {
 774                                if (disk->driverfs_dev != NULL &&
 775                                    disk->driverfs_dev->driver != NULL)
 776                                        printk(" driver: %s\n",
 777                                              disk->driverfs_dev->driver->name);
 778                                else
 779                                        printk(" (driver?)\n");
 780                        } else
 781                                printk("\n");
 782                }
 783                disk_part_iter_exit(&piter);
 784        }
 785        class_dev_iter_exit(&iter);
 786}
 787
 788#ifdef CONFIG_PROC_FS
 789/* iterator */
 790static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
 791{
 792        loff_t skip = *pos;
 793        struct class_dev_iter *iter;
 794        struct device *dev;
 795
 796        iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 797        if (!iter)
 798                return ERR_PTR(-ENOMEM);
 799
 800        seqf->private = iter;
 801        class_dev_iter_init(iter, &block_class, NULL, &disk_type);
 802        do {
 803                dev = class_dev_iter_next(iter);
 804                if (!dev)
 805                        return NULL;
 806        } while (skip--);
 807
 808        return dev_to_disk(dev);
 809}
 810
 811static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
 812{
 813        struct device *dev;
 814
 815        (*pos)++;
 816        dev = class_dev_iter_next(seqf->private);
 817        if (dev)
 818                return dev_to_disk(dev);
 819
 820        return NULL;
 821}
 822
 823static void disk_seqf_stop(struct seq_file *seqf, void *v)
 824{
 825        struct class_dev_iter *iter = seqf->private;
 826
 827        /* stop is called even after start failed :-( */
 828        if (iter) {
 829                class_dev_iter_exit(iter);
 830                kfree(iter);
 831                seqf->private = NULL;
 832        }
 833}
 834
 835static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
 836{
 837        void *p;
 838
 839        p = disk_seqf_start(seqf, pos);
 840        if (!IS_ERR_OR_NULL(p) && !*pos)
 841                seq_puts(seqf, "major minor  #blocks  name\n\n");
 842        return p;
 843}
 844
 845static int show_partition(struct seq_file *seqf, void *v)
 846{
 847        struct gendisk *sgp = v;
 848        struct disk_part_iter piter;
 849        struct hd_struct *part;
 850        char buf[BDEVNAME_SIZE];
 851
 852        /* Don't show non-partitionable removeable devices or empty devices */
 853        if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
 854                                   (sgp->flags & GENHD_FL_REMOVABLE)))
 855                return 0;
 856        if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
 857                return 0;
 858
 859        /* show the full disk and all non-0 size partitions of it */
 860        disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
 861        while ((part = disk_part_iter_next(&piter)))
 862                seq_printf(seqf, "%4d  %7d %10llu %s\n",
 863                           MAJOR(part_devt(part)), MINOR(part_devt(part)),
 864                           (unsigned long long)part_nr_sects_read(part) >> 1,
 865                           disk_name(sgp, part->partno, buf));
 866        disk_part_iter_exit(&piter);
 867
 868        return 0;
 869}
 870
 871static const struct seq_operations partitions_op = {
 872        .start  = show_partition_start,
 873        .next   = disk_seqf_next,
 874        .stop   = disk_seqf_stop,
 875        .show   = show_partition
 876};
 877
 878static int partitions_open(struct inode *inode, struct file *file)
 879{
 880        return seq_open(file, &partitions_op);
 881}
 882
 883static const struct file_operations proc_partitions_operations = {
 884        .open           = partitions_open,
 885        .read           = seq_read,
 886        .llseek         = seq_lseek,
 887        .release        = seq_release,
 888};
 889#endif
 890
 891
 892static struct kobject *base_probe(dev_t devt, int *partno, void *data)
 893{
 894        if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
 895                /* Make old-style 2.4 aliases work */
 896                request_module("block-major-%d", MAJOR(devt));
 897        return NULL;
 898}
 899
 900static int __init genhd_device_init(void)
 901{
 902        int error;
 903
 904        block_class.dev_kobj = sysfs_dev_block_kobj;
 905        error = class_register(&block_class);
 906        if (unlikely(error))
 907                return error;
 908        bdev_map = kobj_map_init(base_probe, &block_class_lock);
 909        blk_dev_init();
 910
 911        register_blkdev(BLOCK_EXT_MAJOR, "blkext");
 912
 913        /* create top-level block dir */
 914        if (!sysfs_deprecated)
 915                block_depr = kobject_create_and_add("block", NULL);
 916        return 0;
 917}
 918
 919subsys_initcall(genhd_device_init);
 920
 921static ssize_t disk_range_show(struct device *dev,
 922                               struct device_attribute *attr, char *buf)
 923{
 924        struct gendisk *disk = dev_to_disk(dev);
 925
 926        return sprintf(buf, "%d\n", disk->minors);
 927}
 928
 929static ssize_t disk_ext_range_show(struct device *dev,
 930                                   struct device_attribute *attr, char *buf)
 931{
 932        struct gendisk *disk = dev_to_disk(dev);
 933
 934        return sprintf(buf, "%d\n", disk_max_parts(disk));
 935}
 936
 937static ssize_t disk_removable_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",
 943                       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
 944}
 945
 946static ssize_t disk_ro_show(struct device *dev,
 947                                   struct device_attribute *attr, char *buf)
 948{
 949        struct gendisk *disk = dev_to_disk(dev);
 950
 951        return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
 952}
 953
 954static ssize_t disk_capability_show(struct device *dev,
 955                                    struct device_attribute *attr, char *buf)
 956{
 957        struct gendisk *disk = dev_to_disk(dev);
 958
 959        return sprintf(buf, "%x\n", disk->flags);
 960}
 961
 962static ssize_t disk_alignment_offset_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_alignment_offset(disk->queue));
 969}
 970
 971static ssize_t disk_discard_alignment_show(struct device *dev,
 972                                           struct device_attribute *attr,
 973                                           char *buf)
 974{
 975        struct gendisk *disk = dev_to_disk(dev);
 976
 977        return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
 978}
 979
 980static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
 981static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
 982static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
 983static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
 984static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
 985static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
 986static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
 987                   NULL);
 988static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
 989static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
 990static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
 991#ifdef CONFIG_FAIL_MAKE_REQUEST
 992static struct device_attribute dev_attr_fail =
 993        __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
 994#endif
 995#ifdef CONFIG_FAIL_IO_TIMEOUT
 996static struct device_attribute dev_attr_fail_timeout =
 997        __ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
 998                part_timeout_store);
 999#endif
1000
1001static struct attribute *disk_attrs[] = {
1002        &dev_attr_range.attr,
1003        &dev_attr_ext_range.attr,
1004        &dev_attr_removable.attr,
1005        &dev_attr_ro.attr,
1006        &dev_attr_size.attr,
1007        &dev_attr_alignment_offset.attr,
1008        &dev_attr_discard_alignment.attr,
1009        &dev_attr_capability.attr,
1010        &dev_attr_stat.attr,
1011        &dev_attr_inflight.attr,
1012#ifdef CONFIG_FAIL_MAKE_REQUEST
1013        &dev_attr_fail.attr,
1014#endif
1015#ifdef CONFIG_FAIL_IO_TIMEOUT
1016        &dev_attr_fail_timeout.attr,
1017#endif
1018        NULL
1019};
1020
1021static struct attribute_group disk_attr_group = {
1022        .attrs = disk_attrs,
1023};
1024
1025static const struct attribute_group *disk_attr_groups[] = {
1026        &disk_attr_group,
1027        NULL
1028};
1029
1030/**
1031 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1032 * @disk: disk to replace part_tbl for
1033 * @new_ptbl: new part_tbl to install
1034 *
1035 * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
1036 * original ptbl is freed using RCU callback.
1037 *
1038 * LOCKING:
1039 * Matching bd_mutx locked.
1040 */
1041static void disk_replace_part_tbl(struct gendisk *disk,
1042                                  struct disk_part_tbl *new_ptbl)
1043{
1044        struct disk_part_tbl *old_ptbl = disk->part_tbl;
1045
1046        rcu_assign_pointer(disk->part_tbl, new_ptbl);
1047
1048        if (old_ptbl) {
1049                rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1050                kfree_rcu(old_ptbl, rcu_head);
1051        }
1052}
1053
1054/**
1055 * disk_expand_part_tbl - expand disk->part_tbl
1056 * @disk: disk to expand part_tbl for
1057 * @partno: expand such that this partno can fit in
1058 *
1059 * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
1060 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1061 *
1062 * LOCKING:
1063 * Matching bd_mutex locked, might sleep.
1064 *
1065 * RETURNS:
1066 * 0 on success, -errno on failure.
1067 */
1068int disk_expand_part_tbl(struct gendisk *disk, int partno)
1069{
1070        struct disk_part_tbl *old_ptbl = disk->part_tbl;
1071        struct disk_part_tbl *new_ptbl;
1072        int len = old_ptbl ? old_ptbl->len : 0;
1073        int i, target;
1074        size_t size;
1075
1076        /*
1077         * check for int overflow, since we can get here from blkpg_ioctl()
1078         * with a user passed 'partno'.
1079         */
1080        target = partno + 1;
1081        if (target < 0)
1082                return -EINVAL;
1083
1084        /* disk_max_parts() is zero during initialization, ignore if so */
1085        if (disk_max_parts(disk) && target > disk_max_parts(disk))
1086                return -EINVAL;
1087
1088        if (target <= len)
1089                return 0;
1090
1091        size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1092        new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1093        if (!new_ptbl)
1094                return -ENOMEM;
1095
1096        new_ptbl->len = target;
1097
1098        for (i = 0; i < len; i++)
1099                rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1100
1101        disk_replace_part_tbl(disk, new_ptbl);
1102        return 0;
1103}
1104
1105static void disk_release(struct device *dev)
1106{
1107        struct gendisk *disk = dev_to_disk(dev);
1108
1109        blk_free_devt(dev->devt);
1110        disk_release_events(disk);
1111        kfree(disk->random);
1112        disk_replace_part_tbl(disk, NULL);
1113        free_part_stats(&disk->part0);
1114        free_part_info(&disk->part0);
1115        if (disk->queue)
1116                blk_put_queue(disk->queue);
1117        kfree(disk);
1118}
1119struct class block_class = {
1120        .name           = "block",
1121};
1122
1123static char *block_devnode(struct device *dev, umode_t *mode,
1124                           kuid_t *uid, kgid_t *gid)
1125{
1126        struct gendisk *disk = dev_to_disk(dev);
1127
1128        if (disk->devnode)
1129                return disk->devnode(disk, mode);
1130        return NULL;
1131}
1132
1133static struct device_type disk_type = {
1134        .name           = "disk",
1135        .groups         = disk_attr_groups,
1136        .release        = disk_release,
1137        .devnode        = block_devnode,
1138};
1139
1140#ifdef CONFIG_PROC_FS
1141/*
1142 * aggregate disk stat collector.  Uses the same stats that the sysfs
1143 * entries do, above, but makes them available through one seq_file.
1144 *
1145 * The output looks suspiciously like /proc/partitions with a bunch of
1146 * extra fields.
1147 */
1148static int diskstats_show(struct seq_file *seqf, void *v)
1149{
1150        struct gendisk *gp = v;
1151        struct disk_part_iter piter;
1152        struct hd_struct *hd;
1153        char buf[BDEVNAME_SIZE];
1154        int cpu;
1155
1156        /*
1157        if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1158                seq_puts(seqf,  "major minor name"
1159                                "     rio rmerge rsect ruse wio wmerge "
1160                                "wsect wuse running use aveq"
1161                                "\n\n");
1162        */
1163
1164        disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1165        while ((hd = disk_part_iter_next(&piter))) {
1166                cpu = part_stat_lock();
1167                part_round_stats(cpu, hd);
1168                part_stat_unlock();
1169                seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1170                           "%u %lu %lu %lu %u %u %u %u\n",
1171                           MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1172                           disk_name(gp, hd->partno, buf),
1173                           part_stat_read(hd, ios[READ]),
1174                           part_stat_read(hd, merges[READ]),
1175                           part_stat_read(hd, sectors[READ]),
1176                           jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1177                           part_stat_read(hd, ios[WRITE]),
1178                           part_stat_read(hd, merges[WRITE]),
1179                           part_stat_read(hd, sectors[WRITE]),
1180                           jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1181                           part_in_flight(hd),
1182                           jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1183                           jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1184                        );
1185        }
1186        disk_part_iter_exit(&piter);
1187
1188        return 0;
1189}
1190
1191static const struct seq_operations diskstats_op = {
1192        .start  = disk_seqf_start,
1193        .next   = disk_seqf_next,
1194        .stop   = disk_seqf_stop,
1195        .show   = diskstats_show
1196};
1197
1198static int diskstats_open(struct inode *inode, struct file *file)
1199{
1200        return seq_open(file, &diskstats_op);
1201}
1202
1203static const struct file_operations proc_diskstats_operations = {
1204        .open           = diskstats_open,
1205        .read           = seq_read,
1206        .llseek         = seq_lseek,
1207        .release        = seq_release,
1208};
1209
1210static int __init proc_genhd_init(void)
1211{
1212        proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1213        proc_create("partitions", 0, NULL, &proc_partitions_operations);
1214        return 0;
1215}
1216module_init(proc_genhd_init);
1217#endif /* CONFIG_PROC_FS */
1218
1219dev_t blk_lookup_devt(const char *name, int partno)
1220{
1221        dev_t devt = MKDEV(0, 0);
1222        struct class_dev_iter iter;
1223        struct device *dev;
1224
1225        class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1226        while ((dev = class_dev_iter_next(&iter))) {
1227                struct gendisk *disk = dev_to_disk(dev);
1228                struct hd_struct *part;
1229
1230                if (strcmp(dev_name(dev), name))
1231                        continue;
1232
1233                if (partno < disk->minors) {
1234                        /* We need to return the right devno, even
1235                         * if the partition doesn't exist yet.
1236                         */
1237                        devt = MKDEV(MAJOR(dev->devt),
1238                                     MINOR(dev->devt) + partno);
1239                        break;
1240                }
1241                part = disk_get_part(disk, partno);
1242                if (part) {
1243                        devt = part_devt(part);
1244                        disk_put_part(part);
1245                        break;
1246                }
1247                disk_put_part(part);
1248        }
1249        class_dev_iter_exit(&iter);
1250        return devt;
1251}
1252EXPORT_SYMBOL(blk_lookup_devt);
1253
1254struct gendisk *alloc_disk(int minors)
1255{
1256        return alloc_disk_node(minors, NUMA_NO_NODE);
1257}
1258EXPORT_SYMBOL(alloc_disk);
1259
1260struct gendisk *alloc_disk_node(int minors, int node_id)
1261{
1262        struct gendisk *disk;
1263
1264        disk = kmalloc_node(sizeof(struct gendisk),
1265                                GFP_KERNEL | __GFP_ZERO, node_id);
1266        if (disk) {
1267                if (!init_part_stats(&disk->part0)) {
1268                        kfree(disk);
1269                        return NULL;
1270                }
1271                disk->node_id = node_id;
1272                if (disk_expand_part_tbl(disk, 0)) {
1273                        free_part_stats(&disk->part0);
1274                        kfree(disk);
1275                        return NULL;
1276                }
1277                disk->part_tbl->part[0] = &disk->part0;
1278
1279                /*
1280                 * set_capacity() and get_capacity() currently don't use
1281                 * seqcounter to read/update the part0->nr_sects. Still init
1282                 * the counter as we can read the sectors in IO submission
1283                 * patch using seqence counters.
1284                 *
1285                 * TODO: Ideally set_capacity() and get_capacity() should be
1286                 * converted to make use of bd_mutex and sequence counters.
1287                 */
1288                seqcount_init(&disk->part0.nr_sects_seq);
1289                hd_ref_init(&disk->part0);
1290
1291                disk->minors = minors;
1292                rand_initialize_disk(disk);
1293                disk_to_dev(disk)->class = &block_class;
1294                disk_to_dev(disk)->type = &disk_type;
1295                device_initialize(disk_to_dev(disk));
1296        }
1297        return disk;
1298}
1299EXPORT_SYMBOL(alloc_disk_node);
1300
1301struct kobject *get_disk(struct gendisk *disk)
1302{
1303        struct module *owner;
1304        struct kobject *kobj;
1305
1306        if (!disk->fops)
1307                return NULL;
1308        owner = disk->fops->owner;
1309        if (owner && !try_module_get(owner))
1310                return NULL;
1311        kobj = kobject_get(&disk_to_dev(disk)->kobj);
1312        if (kobj == NULL) {
1313                module_put(owner);
1314                return NULL;
1315        }
1316        return kobj;
1317
1318}
1319
1320EXPORT_SYMBOL(get_disk);
1321
1322void put_disk(struct gendisk *disk)
1323{
1324        if (disk)
1325                kobject_put(&disk_to_dev(disk)->kobj);
1326}
1327
1328EXPORT_SYMBOL(put_disk);
1329
1330static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1331{
1332        char event[] = "DISK_RO=1";
1333        char *envp[] = { event, NULL };
1334
1335        if (!ro)
1336                event[8] = '0';
1337        kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1338}
1339
1340void set_device_ro(struct block_device *bdev, int flag)
1341{
1342        bdev->bd_part->policy = flag;
1343}
1344
1345EXPORT_SYMBOL(set_device_ro);
1346
1347void set_disk_ro(struct gendisk *disk, int flag)
1348{
1349        struct disk_part_iter piter;
1350        struct hd_struct *part;
1351
1352        if (disk->part0.policy != flag) {
1353                set_disk_ro_uevent(disk, flag);
1354                disk->part0.policy = flag;
1355        }
1356
1357        disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1358        while ((part = disk_part_iter_next(&piter)))
1359                part->policy = flag;
1360        disk_part_iter_exit(&piter);
1361}
1362
1363EXPORT_SYMBOL(set_disk_ro);
1364
1365int bdev_read_only(struct block_device *bdev)
1366{
1367        if (!bdev)
1368                return 0;
1369        return bdev->bd_part->policy;
1370}
1371
1372EXPORT_SYMBOL(bdev_read_only);
1373
1374int invalidate_partition(struct gendisk *disk, int partno)
1375{
1376        int res = 0;
1377        struct block_device *bdev = bdget_disk(disk, partno);
1378        if (bdev) {
1379                fsync_bdev(bdev);
1380                res = __invalidate_device(bdev, true);
1381                bdput(bdev);
1382        }
1383        return res;
1384}
1385
1386EXPORT_SYMBOL(invalidate_partition);
1387
1388/*
1389 * Disk events - monitor disk events like media change and eject request.
1390 */
1391struct disk_events {
1392        struct list_head        node;           /* all disk_event's */
1393        struct gendisk          *disk;          /* the associated disk */
1394        spinlock_t              lock;
1395
1396        struct mutex            block_mutex;    /* protects blocking */
1397        int                     block;          /* event blocking depth */
1398        unsigned int            pending;        /* events already sent out */
1399        unsigned int            clearing;       /* events being cleared */
1400
1401        long                    poll_msecs;     /* interval, -1 for default */
1402        struct delayed_work     dwork;
1403};
1404
1405static const char *disk_events_strs[] = {
1406        [ilog2(DISK_EVENT_MEDIA_CHANGE)]        = "media_change",
1407        [ilog2(DISK_EVENT_EJECT_REQUEST)]       = "eject_request",
1408};
1409
1410static char *disk_uevents[] = {
1411        [ilog2(DISK_EVENT_MEDIA_CHANGE)]        = "DISK_MEDIA_CHANGE=1",
1412        [ilog2(DISK_EVENT_EJECT_REQUEST)]       = "DISK_EJECT_REQUEST=1",
1413};
1414
1415/* list of all disk_events */
1416static DEFINE_MUTEX(disk_events_mutex);
1417static LIST_HEAD(disk_events);
1418
1419/* disable in-kernel polling by default */
1420static unsigned long disk_events_dfl_poll_msecs = 0;
1421
1422static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1423{
1424        struct disk_events *ev = disk->ev;
1425        long intv_msecs = 0;
1426
1427        /*
1428         * If device-specific poll interval is set, always use it.  If
1429         * the default is being used, poll iff there are events which
1430         * can't be monitored asynchronously.
1431         */
1432        if (ev->poll_msecs >= 0)
1433                intv_msecs = ev->poll_msecs;
1434        else if (disk->events & ~disk->async_events)
1435                intv_msecs = disk_events_dfl_poll_msecs;
1436
1437        return msecs_to_jiffies(intv_msecs);
1438}
1439
1440/**
1441 * disk_block_events - block and flush disk event checking
1442 * @disk: disk to block events for
1443 *
1444 * On return from this function, it is guaranteed that event checking
1445 * isn't in progress and won't happen until unblocked by
1446 * disk_unblock_events().  Events blocking is counted and the actual
1447 * unblocking happens after the matching number of unblocks are done.
1448 *
1449 * Note that this intentionally does not block event checking from
1450 * disk_clear_events().
1451 *
1452 * CONTEXT:
1453 * Might sleep.
1454 */
1455void disk_block_events(struct gendisk *disk)
1456{
1457        struct disk_events *ev = disk->ev;
1458        unsigned long flags;
1459        bool cancel;
1460
1461        if (!ev)
1462                return;
1463
1464        /*
1465         * Outer mutex ensures that the first blocker completes canceling
1466         * the event work before further blockers are allowed to finish.
1467         */
1468        mutex_lock(&ev->block_mutex);
1469
1470        spin_lock_irqsave(&ev->lock, flags);
1471        cancel = !ev->block++;
1472        spin_unlock_irqrestore(&ev->lock, flags);
1473
1474        if (cancel)
1475                cancel_delayed_work_sync(&disk->ev->dwork);
1476
1477        mutex_unlock(&ev->block_mutex);
1478}
1479
1480static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1481{
1482        struct disk_events *ev = disk->ev;
1483        unsigned long intv;
1484        unsigned long flags;
1485
1486        spin_lock_irqsave(&ev->lock, flags);
1487
1488        if (WARN_ON_ONCE(ev->block <= 0))
1489                goto out_unlock;
1490
1491        if (--ev->block)
1492                goto out_unlock;
1493
1494        /*
1495         * Not exactly a latency critical operation, set poll timer
1496         * slack to 25% and kick event check.
1497         */
1498        intv = disk_events_poll_jiffies(disk);
1499        set_timer_slack(&ev->dwork.timer, intv / 4);
1500        if (check_now)
1501                queue_delayed_work(system_freezable_wq, &ev->dwork, 0);
1502        else if (intv)
1503                queue_delayed_work(system_freezable_wq, &ev->dwork, intv);
1504out_unlock:
1505        spin_unlock_irqrestore(&ev->lock, flags);
1506}
1507
1508/**
1509 * disk_unblock_events - unblock disk event checking
1510 * @disk: disk to unblock events for
1511 *
1512 * Undo disk_block_events().  When the block count reaches zero, it
1513 * starts events polling if configured.
1514 *
1515 * CONTEXT:
1516 * Don't care.  Safe to call from irq context.
1517 */
1518void disk_unblock_events(struct gendisk *disk)
1519{
1520        if (disk->ev)
1521                __disk_unblock_events(disk, false);
1522}
1523
1524/**
1525 * disk_flush_events - schedule immediate event checking and flushing
1526 * @disk: disk to check and flush events for
1527 * @mask: events to flush
1528 *
1529 * Schedule immediate event checking on @disk if not blocked.  Events in
1530 * @mask are scheduled to be cleared from the driver.  Note that this
1531 * doesn't clear the events from @disk->ev.
1532 *
1533 * CONTEXT:
1534 * If @mask is non-zero must be called with bdev->bd_mutex held.
1535 */
1536void disk_flush_events(struct gendisk *disk, unsigned int mask)
1537{
1538        struct disk_events *ev = disk->ev;
1539
1540        if (!ev)
1541                return;
1542
1543        spin_lock_irq(&ev->lock);
1544        ev->clearing |= mask;
1545        if (!ev->block)
1546                mod_delayed_work(system_freezable_wq, &ev->dwork, 0);
1547        spin_unlock_irq(&ev->lock);
1548}
1549
1550/**
1551 * disk_clear_events - synchronously check, clear and return pending events
1552 * @disk: disk to fetch and clear events from
1553 * @mask: mask of events to be fetched and clearted
1554 *
1555 * Disk events are synchronously checked and pending events in @mask
1556 * are cleared and returned.  This ignores the block count.
1557 *
1558 * CONTEXT:
1559 * Might sleep.
1560 */
1561unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1562{
1563        const struct block_device_operations *bdops = disk->fops;
1564        struct disk_events *ev = disk->ev;
1565        unsigned int pending;
1566        unsigned int clearing = mask;
1567
1568        if (!ev) {
1569                /* for drivers still using the old ->media_changed method */
1570                if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1571                    bdops->media_changed && bdops->media_changed(disk))
1572                        return DISK_EVENT_MEDIA_CHANGE;
1573                return 0;
1574        }
1575
1576        disk_block_events(disk);
1577
1578        /*
1579         * store the union of mask and ev->clearing on the stack so that the
1580         * race with disk_flush_events does not cause ambiguity (ev->clearing
1581         * can still be modified even if events are blocked).
1582         */
1583        spin_lock_irq(&ev->lock);
1584        clearing |= ev->clearing;
1585        ev->clearing = 0;
1586        spin_unlock_irq(&ev->lock);
1587
1588        disk_check_events(ev, &clearing);
1589        /*
1590         * if ev->clearing is not 0, the disk_flush_events got called in the
1591         * middle of this function, so we want to run the workfn without delay.
1592         */
1593        __disk_unblock_events(disk, ev->clearing ? true : false);
1594
1595        /* then, fetch and clear pending events */
1596        spin_lock_irq(&ev->lock);
1597        pending = ev->pending & mask;
1598        ev->pending &= ~mask;
1599        spin_unlock_irq(&ev->lock);
1600        WARN_ON_ONCE(clearing & mask);
1601
1602        return pending;
1603}
1604
1605/*
1606 * Separate this part out so that a different pointer for clearing_ptr can be
1607 * passed in for disk_clear_events.
1608 */
1609static void disk_events_workfn(struct work_struct *work)
1610{
1611        struct delayed_work *dwork = to_delayed_work(work);
1612        struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1613
1614        disk_check_events(ev, &ev->clearing);
1615}
1616
1617static void disk_check_events(struct disk_events *ev,
1618                              unsigned int *clearing_ptr)
1619{
1620        struct gendisk *disk = ev->disk;
1621        char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1622        unsigned int clearing = *clearing_ptr;
1623        unsigned int events;
1624        unsigned long intv;
1625        int nr_events = 0, i;
1626
1627        /* check events */
1628        events = disk->fops->check_events(disk, clearing);
1629
1630        /* accumulate pending events and schedule next poll if necessary */
1631        spin_lock_irq(&ev->lock);
1632
1633        events &= ~ev->pending;
1634        ev->pending |= events;
1635        *clearing_ptr &= ~clearing;
1636
1637        intv = disk_events_poll_jiffies(disk);
1638        if (!ev->block && intv)
1639                queue_delayed_work(system_freezable_wq, &ev->dwork, intv);
1640
1641        spin_unlock_irq(&ev->lock);
1642
1643        /*
1644         * Tell userland about new events.  Only the events listed in
1645         * @disk->events are reported.  Unlisted events are processed the
1646         * same internally but never get reported to userland.
1647         */
1648        for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1649                if (events & disk->events & (1 << i))
1650                        envp[nr_events++] = disk_uevents[i];
1651
1652        if (nr_events)
1653                kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1654}
1655
1656/*
1657 * A disk events enabled device has the following sysfs nodes under
1658 * its /sys/block/X/ directory.
1659 *
1660 * events               : list of all supported events
1661 * events_async         : list of events which can be detected w/o polling
1662 * events_poll_msecs    : polling interval, 0: disable, -1: system default
1663 */
1664static ssize_t __disk_events_show(unsigned int events, char *buf)
1665{
1666        const char *delim = "";
1667        ssize_t pos = 0;
1668        int i;
1669
1670        for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1671                if (events & (1 << i)) {
1672                        pos += sprintf(buf + pos, "%s%s",
1673                                       delim, disk_events_strs[i]);
1674                        delim = " ";
1675                }
1676        if (pos)
1677                pos += sprintf(buf + pos, "\n");
1678        return pos;
1679}
1680
1681static ssize_t disk_events_show(struct device *dev,
1682                                struct device_attribute *attr, char *buf)
1683{
1684        struct gendisk *disk = dev_to_disk(dev);
1685
1686        return __disk_events_show(disk->events, buf);
1687}
1688
1689static ssize_t disk_events_async_show(struct device *dev,
1690                                      struct device_attribute *attr, char *buf)
1691{
1692        struct gendisk *disk = dev_to_disk(dev);
1693
1694        return __disk_events_show(disk->async_events, buf);
1695}
1696
1697static ssize_t disk_events_poll_msecs_show(struct device *dev,
1698                                           struct device_attribute *attr,
1699                                           char *buf)
1700{
1701        struct gendisk *disk = dev_to_disk(dev);
1702
1703        return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1704}
1705
1706static ssize_t disk_events_poll_msecs_store(struct device *dev,
1707                                            struct device_attribute *attr,
1708                                            const char *buf, size_t count)
1709{
1710        struct gendisk *disk = dev_to_disk(dev);
1711        long intv;
1712
1713        if (!count || !sscanf(buf, "%ld", &intv))
1714                return -EINVAL;
1715
1716        if (intv < 0 && intv != -1)
1717                return -EINVAL;
1718
1719        disk_block_events(disk);
1720        disk->ev->poll_msecs = intv;
1721        __disk_unblock_events(disk, true);
1722
1723        return count;
1724}
1725
1726static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1727static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1728static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1729                         disk_events_poll_msecs_show,
1730                         disk_events_poll_msecs_store);
1731
1732static const struct attribute *disk_events_attrs[] = {
1733        &dev_attr_events.attr,
1734        &dev_attr_events_async.attr,
1735        &dev_attr_events_poll_msecs.attr,
1736        NULL,
1737};
1738
1739/*
1740 * The default polling interval can be specified by the kernel
1741 * parameter block.events_dfl_poll_msecs which defaults to 0
1742 * (disable).  This can also be modified runtime by writing to
1743 * /sys/module/block/events_dfl_poll_msecs.
1744 */
1745static int disk_events_set_dfl_poll_msecs(const char *val,
1746                                          const struct kernel_param *kp)
1747{
1748        struct disk_events *ev;
1749        int ret;
1750
1751        ret = param_set_ulong(val, kp);
1752        if (ret < 0)
1753                return ret;
1754
1755        mutex_lock(&disk_events_mutex);
1756
1757        list_for_each_entry(ev, &disk_events, node)
1758                disk_flush_events(ev->disk, 0);
1759
1760        mutex_unlock(&disk_events_mutex);
1761
1762        return 0;
1763}
1764
1765static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1766        .set    = disk_events_set_dfl_poll_msecs,
1767        .get    = param_get_ulong,
1768};
1769
1770#undef MODULE_PARAM_PREFIX
1771#define MODULE_PARAM_PREFIX     "block."
1772
1773module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1774                &disk_events_dfl_poll_msecs, 0644);
1775
1776/*
1777 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1778 */
1779static void disk_alloc_events(struct gendisk *disk)
1780{
1781        struct disk_events *ev;
1782
1783        if (!disk->fops->check_events)
1784                return;
1785
1786        ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1787        if (!ev) {
1788                pr_warn("%s: failed to initialize events\n", disk->disk_name);
1789                return;
1790        }
1791
1792        INIT_LIST_HEAD(&ev->node);
1793        ev->disk = disk;
1794        spin_lock_init(&ev->lock);
1795        mutex_init(&ev->block_mutex);
1796        ev->block = 1;
1797        ev->poll_msecs = -1;
1798        INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1799
1800        disk->ev = ev;
1801}
1802
1803static void disk_add_events(struct gendisk *disk)
1804{
1805        if (!disk->ev)
1806                return;
1807
1808        /* FIXME: error handling */
1809        if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1810                pr_warn("%s: failed to create sysfs files for events\n",
1811                        disk->disk_name);
1812
1813        mutex_lock(&disk_events_mutex);
1814        list_add_tail(&disk->ev->node, &disk_events);
1815        mutex_unlock(&disk_events_mutex);
1816
1817        /*
1818         * Block count is initialized to 1 and the following initial
1819         * unblock kicks it into action.
1820         */
1821        __disk_unblock_events(disk, true);
1822}
1823
1824static void disk_del_events(struct gendisk *disk)
1825{
1826        if (!disk->ev)
1827                return;
1828
1829        disk_block_events(disk);
1830
1831        mutex_lock(&disk_events_mutex);
1832        list_del_init(&disk->ev->node);
1833        mutex_unlock(&disk_events_mutex);
1834
1835        sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1836}
1837
1838static void disk_release_events(struct gendisk *disk)
1839{
1840        /* the block count should be 1 from disk_del_events() */
1841        WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1842        kfree(disk->ev);
1843}
1844