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/backing-dev.h>
  12#include <linux/init.h>
  13#include <linux/spinlock.h>
  14#include <linux/proc_fs.h>
  15#include <linux/seq_file.h>
  16#include <linux/slab.h>
  17#include <linux/kmod.h>
  18#include <linux/kobj_map.h>
  19#include <linux/mutex.h>
  20#include <linux/idr.h>
  21#include <linux/log2.h>
  22#include <linux/pm_runtime.h>
  23#include <linux/badblocks.h>
  24
  25#include "blk.h"
  26
  27static DEFINE_MUTEX(block_class_lock);
  28struct kobject *block_depr;
  29
  30/* for extended dynamic devt allocation, currently only one major is used */
  31#define NR_EXT_DEVT		(1 << MINORBITS)
  32
  33/* For extended devt allocation.  ext_devt_lock prevents look up
  34 * results from going away underneath its user.
  35 */
  36static DEFINE_SPINLOCK(ext_devt_lock);
  37static DEFINE_IDR(ext_devt_idr);
  38
  39static struct device_type disk_type;
  40
  41static void disk_check_events(struct disk_events *ev,
  42			      unsigned int *clearing_ptr);
  43static void disk_alloc_events(struct gendisk *disk);
  44static void disk_add_events(struct gendisk *disk);
  45static void disk_del_events(struct gendisk *disk);
  46static void disk_release_events(struct gendisk *disk);
  47
  48/**
  49 * disk_get_part - get partition
  50 * @disk: disk to look partition from
  51 * @partno: partition number
  52 *
  53 * Look for partition @partno from @disk.  If found, increment
  54 * reference count and return it.
  55 *
  56 * CONTEXT:
  57 * Don't care.
  58 *
  59 * RETURNS:
  60 * Pointer to the found partition on success, NULL if not found.
  61 */
  62struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
  63{
  64	struct hd_struct *part = NULL;
  65	struct disk_part_tbl *ptbl;
  66
  67	if (unlikely(partno < 0))
  68		return NULL;
  69
  70	rcu_read_lock();
  71
  72	ptbl = rcu_dereference(disk->part_tbl);
  73	if (likely(partno < ptbl->len)) {
  74		part = rcu_dereference(ptbl->part[partno]);
  75		if (part)
  76			get_device(part_to_dev(part));
  77	}
  78
  79	rcu_read_unlock();
  80
  81	return part;
  82}
  83EXPORT_SYMBOL_GPL(disk_get_part);
  84
  85/**
  86 * disk_part_iter_init - initialize partition iterator
  87 * @piter: iterator to initialize
  88 * @disk: disk to iterate over
  89 * @flags: DISK_PITER_* flags
  90 *
  91 * Initialize @piter so that it iterates over partitions of @disk.
  92 *
  93 * CONTEXT:
  94 * Don't care.
  95 */
  96void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
  97			  unsigned int flags)
  98{
  99	struct disk_part_tbl *ptbl;
 100
 101	rcu_read_lock();
 102	ptbl = rcu_dereference(disk->part_tbl);
 103
 104	piter->disk = disk;
 105	piter->part = NULL;
 106
 107	if (flags & DISK_PITER_REVERSE)
 108		piter->idx = ptbl->len - 1;
 109	else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
 110		piter->idx = 0;
 111	else
 112		piter->idx = 1;
 113
 114	piter->flags = flags;
 115
 116	rcu_read_unlock();
 117}
 118EXPORT_SYMBOL_GPL(disk_part_iter_init);
 119
 120/**
 121 * disk_part_iter_next - proceed iterator to the next partition and return it
 122 * @piter: iterator of interest
 123 *
 124 * Proceed @piter to the next partition and return it.
 125 *
 126 * CONTEXT:
 127 * Don't care.
 128 */
 129struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
 130{
 131	struct disk_part_tbl *ptbl;
 132	int inc, end;
 133
 134	/* put the last partition */
 135	disk_put_part(piter->part);
 136	piter->part = NULL;
 137
 138	/* get part_tbl */
 139	rcu_read_lock();
 140	ptbl = rcu_dereference(piter->disk->part_tbl);
 141
 142	/* determine iteration parameters */
 143	if (piter->flags & DISK_PITER_REVERSE) {
 144		inc = -1;
 145		if (piter->flags & (DISK_PITER_INCL_PART0 |
 146				    DISK_PITER_INCL_EMPTY_PART0))
 147			end = -1;
 148		else
 149			end = 0;
 150	} else {
 151		inc = 1;
 152		end = ptbl->len;
 153	}
 154
 155	/* iterate to the next partition */
 156	for (; piter->idx != end; piter->idx += inc) {
 157		struct hd_struct *part;
 158
 159		part = rcu_dereference(ptbl->part[piter->idx]);
 160		if (!part)
 161			continue;
 162		if (!part_nr_sects_read(part) &&
 163		    !(piter->flags & DISK_PITER_INCL_EMPTY) &&
 164		    !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
 165		      piter->idx == 0))
 166			continue;
 167
 168		get_device(part_to_dev(part));
 169		piter->part = part;
 170		piter->idx += inc;
 171		break;
 172	}
 173
 174	rcu_read_unlock();
 175
 176	return piter->part;
 177}
 178EXPORT_SYMBOL_GPL(disk_part_iter_next);
 179
 180/**
 181 * disk_part_iter_exit - finish up partition iteration
 182 * @piter: iter of interest
 183 *
 184 * Called when iteration is over.  Cleans up @piter.
 185 *
 186 * CONTEXT:
 187 * Don't care.
 188 */
 189void disk_part_iter_exit(struct disk_part_iter *piter)
 190{
 191	disk_put_part(piter->part);
 192	piter->part = NULL;
 193}
 194EXPORT_SYMBOL_GPL(disk_part_iter_exit);
 195
 196static inline int sector_in_part(struct hd_struct *part, sector_t sector)
 197{
 198	return part->start_sect <= sector &&
 199		sector < part->start_sect + part_nr_sects_read(part);
 200}
 201
 202/**
 203 * disk_map_sector_rcu - map sector to partition
 204 * @disk: gendisk of interest
 205 * @sector: sector to map
 206 *
 207 * Find out which partition @sector maps to on @disk.  This is
 208 * primarily used for stats accounting.
 209 *
 210 * CONTEXT:
 211 * RCU read locked.  The returned partition pointer is valid only
 212 * while preemption is disabled.
 213 *
 214 * RETURNS:
 215 * Found partition on success, part0 is returned if no partition matches
 216 */
 217struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
 218{
 219	struct disk_part_tbl *ptbl;
 220	struct hd_struct *part;
 221	int i;
 222
 223	ptbl = rcu_dereference(disk->part_tbl);
 224
 225	part = rcu_dereference(ptbl->last_lookup);
 226	if (part && sector_in_part(part, sector))
 227		return part;
 228
 229	for (i = 1; i < ptbl->len; i++) {
 230		part = rcu_dereference(ptbl->part[i]);
 231
 232		if (part && sector_in_part(part, sector)) {
 233			rcu_assign_pointer(ptbl->last_lookup, part);
 234			return part;
 235		}
 236	}
 237	return &disk->part0;
 238}
 239EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
 240
 241/*
 242 * Can be deleted altogether. Later.
 243 *
 244 */
 245static struct blk_major_name {
 246	struct blk_major_name *next;
 247	int major;
 248	char name[16];
 249} *major_names[BLKDEV_MAJOR_HASH_SIZE];
 250
 251/* index in the above - for now: assume no multimajor ranges */
 252static inline int major_to_index(unsigned major)
 253{
 254	return major % BLKDEV_MAJOR_HASH_SIZE;
 255}
 256
 257#ifdef CONFIG_PROC_FS
 258void blkdev_show(struct seq_file *seqf, off_t offset)
 259{
 260	struct blk_major_name *dp;
 261
 262	if (offset < BLKDEV_MAJOR_HASH_SIZE) {
 263		mutex_lock(&block_class_lock);
 264		for (dp = major_names[offset]; dp; dp = dp->next)
 265			seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
 266		mutex_unlock(&block_class_lock);
 267	}
 268}
 269#endif /* CONFIG_PROC_FS */
 270
 271/**
 272 * register_blkdev - register a new block device
 273 *
 274 * @major: the requested major device number [1..255]. If @major=0, try to
 275 *         allocate any unused major number.
 276 * @name: the name of the new block device as a zero terminated string
 277 *
 278 * The @name must be unique within the system.
 279 *
 280 * The return value depends on the @major input parameter.
 281 *  - if a major device number was requested in range [1..255] then the
 282 *    function returns zero on success, or a negative error code
 283 *  - if any unused major number was requested with @major=0 parameter
 284 *    then the return value is the allocated major number in range
 285 *    [1..255] or a negative error code otherwise
 286 */
 287int register_blkdev(unsigned int major, const char *name)
 288{
 289	struct blk_major_name **n, *p;
 290	int index, ret = 0;
 291
 292	mutex_lock(&block_class_lock);
 293
 294	/* temporary */
 295	if (major == 0) {
 296		for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
 297			if (major_names[index] == NULL)
 298				break;
 299		}
 300
 301		if (index == 0) {
 302			printk("register_blkdev: failed to get major for %s\n",
 303			       name);
 304			ret = -EBUSY;
 305			goto out;
 306		}
 307		major = index;
 308		ret = major;
 309	}
 310
 311	p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
 312	if (p == NULL) {
 313		ret = -ENOMEM;
 314		goto out;
 315	}
 316
 317	p->major = major;
 318	strlcpy(p->name, name, sizeof(p->name));
 319	p->next = NULL;
 320	index = major_to_index(major);
 321
 322	for (n = &major_names[index]; *n; n = &(*n)->next) {
 323		if ((*n)->major == major)
 324			break;
 325	}
 326	if (!*n)
 327		*n = p;
 328	else
 329		ret = -EBUSY;
 330
 331	if (ret < 0) {
 332		printk("register_blkdev: cannot get major %d for %s\n",
 333		       major, name);
 334		kfree(p);
 335	}
 336out:
 337	mutex_unlock(&block_class_lock);
 338	return ret;
 339}
 340
 341EXPORT_SYMBOL(register_blkdev);
 342
 343void unregister_blkdev(unsigned int major, const char *name)
 344{
 345	struct blk_major_name **n;
 346	struct blk_major_name *p = NULL;
 347	int index = major_to_index(major);
 348
 349	mutex_lock(&block_class_lock);
 350	for (n = &major_names[index]; *n; n = &(*n)->next)
 351		if ((*n)->major == major)
 352			break;
 353	if (!*n || strcmp((*n)->name, name)) {
 354		WARN_ON(1);
 355	} else {
 356		p = *n;
 357		*n = p->next;
 358	}
 359	mutex_unlock(&block_class_lock);
 360	kfree(p);
 361}
 362
 363EXPORT_SYMBOL(unregister_blkdev);
 364
 365static struct kobj_map *bdev_map;
 366
 367/**
 368 * blk_mangle_minor - scatter minor numbers apart
 369 * @minor: minor number to mangle
 370 *
 371 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
 372 * is enabled.  Mangling twice gives the original value.
 373 *
 374 * RETURNS:
 375 * Mangled value.
 376 *
 377 * CONTEXT:
 378 * Don't care.
 379 */
 380static int blk_mangle_minor(int minor)
 381{
 382#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
 383	int i;
 384
 385	for (i = 0; i < MINORBITS / 2; i++) {
 386		int low = minor & (1 << i);
 387		int high = minor & (1 << (MINORBITS - 1 - i));
 388		int distance = MINORBITS - 1 - 2 * i;
 389
 390		minor ^= low | high;	/* clear both bits */
 391		low <<= distance;	/* swap the positions */
 392		high >>= distance;
 393		minor |= low | high;	/* and set */
 394	}
 395#endif
 396	return minor;
 397}
 398
 399/**
 400 * blk_alloc_devt - allocate a dev_t for a partition
 401 * @part: partition to allocate dev_t for
 402 * @devt: out parameter for resulting dev_t
 403 *
 404 * Allocate a dev_t for block device.
 405 *
 406 * RETURNS:
 407 * 0 on success, allocated dev_t is returned in *@devt.  -errno on
 408 * failure.
 409 *
 410 * CONTEXT:
 411 * Might sleep.
 412 */
 413int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
 414{
 415	struct gendisk *disk = part_to_disk(part);
 416	int idx;
 417
 418	/* in consecutive minor range? */
 419	if (part->partno < disk->minors) {
 420		*devt = MKDEV(disk->major, disk->first_minor + part->partno);
 421		return 0;
 422	}
 423
 424	/* allocate ext devt */
 425	idr_preload(GFP_KERNEL);
 426
 427	spin_lock_bh(&ext_devt_lock);
 428	idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
 429	spin_unlock_bh(&ext_devt_lock);
 430
 431	idr_preload_end();
 432	if (idx < 0)
 433		return idx == -ENOSPC ? -EBUSY : idx;
 434
 435	*devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
 436	return 0;
 437}
 438
 439/**
 440 * blk_free_devt - free a dev_t
 441 * @devt: dev_t to free
 442 *
 443 * Free @devt which was allocated using blk_alloc_devt().
 444 *
 445 * CONTEXT:
 446 * Might sleep.
 447 */
 448void blk_free_devt(dev_t devt)
 449{
 
 
 450	if (devt == MKDEV(0, 0))
 451		return;
 452
 453	if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
 454		spin_lock_bh(&ext_devt_lock);
 455		idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 456		spin_unlock_bh(&ext_devt_lock);
 457	}
 458}
 459
 460static char *bdevt_str(dev_t devt, char *buf)
 461{
 462	if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
 463		char tbuf[BDEVT_SIZE];
 464		snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
 465		snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
 466	} else
 467		snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
 468
 469	return buf;
 470}
 471
 472/*
 473 * Register device numbers dev..(dev+range-1)
 474 * range must be nonzero
 475 * The hash chain is sorted on range, so that subranges can override.
 476 */
 477void blk_register_region(dev_t devt, unsigned long range, struct module *module,
 478			 struct kobject *(*probe)(dev_t, int *, void *),
 479			 int (*lock)(dev_t, void *), void *data)
 480{
 481	kobj_map(bdev_map, devt, range, module, probe, lock, data);
 482}
 483
 484EXPORT_SYMBOL(blk_register_region);
 485
 486void blk_unregister_region(dev_t devt, unsigned long range)
 487{
 488	kobj_unmap(bdev_map, devt, range);
 489}
 490
 491EXPORT_SYMBOL(blk_unregister_region);
 492
 493static struct kobject *exact_match(dev_t devt, int *partno, void *data)
 494{
 495	struct gendisk *p = data;
 496
 497	return &disk_to_dev(p)->kobj;
 498}
 499
 500static int exact_lock(dev_t devt, void *data)
 501{
 502	struct gendisk *p = data;
 503
 504	if (!get_disk(p))
 505		return -1;
 506	return 0;
 507}
 508
 509static void register_disk(struct gendisk *disk)
 510{
 511	struct device *ddev = disk_to_dev(disk);
 512	struct block_device *bdev;
 513	struct disk_part_iter piter;
 514	struct hd_struct *part;
 515	int err;
 516
 517	ddev->parent = disk->driverfs_dev;
 518
 519	dev_set_name(ddev, "%s", disk->disk_name);
 520
 521	/* delay uevents, until we scanned partition table */
 522	dev_set_uevent_suppress(ddev, 1);
 523
 524	if (device_add(ddev))
 525		return;
 526	if (!sysfs_deprecated) {
 527		err = sysfs_create_link(block_depr, &ddev->kobj,
 528					kobject_name(&ddev->kobj));
 529		if (err) {
 530			device_del(ddev);
 531			return;
 532		}
 533	}
 534
 535	/*
 536	 * avoid probable deadlock caused by allocating memory with
 537	 * GFP_KERNEL in runtime_resume callback of its all ancestor
 538	 * devices
 539	 */
 540	pm_runtime_set_memalloc_noio(ddev, true);
 541
 542	disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
 543	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
 544
 545	/* No minors to use for partitions */
 546	if (!disk_part_scan_enabled(disk))
 547		goto exit;
 548
 549	/* No such device (e.g., media were just removed) */
 550	if (!get_capacity(disk))
 551		goto exit;
 552
 553	bdev = bdget_disk(disk, 0);
 554	if (!bdev)
 555		goto exit;
 556
 557	bdev->bd_invalidated = 1;
 558	err = blkdev_get(bdev, FMODE_READ, NULL);
 559	if (err < 0)
 560		goto exit;
 561	blkdev_put(bdev, FMODE_READ);
 562
 563exit:
 564	/* announce disk after possible partitions are created */
 565	dev_set_uevent_suppress(ddev, 0);
 566	kobject_uevent(&ddev->kobj, KOBJ_ADD);
 567
 568	/* announce possible partitions */
 569	disk_part_iter_init(&piter, disk, 0);
 570	while ((part = disk_part_iter_next(&piter)))
 571		kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
 572	disk_part_iter_exit(&piter);
 573}
 574
 575/**
 576 * add_disk - add partitioning information to kernel list
 577 * @disk: per-device partitioning information
 578 *
 579 * This function registers the partitioning information in @disk
 580 * with the kernel.
 581 *
 582 * FIXME: error handling
 583 */
 584void add_disk(struct gendisk *disk)
 585{
 586	struct backing_dev_info *bdi;
 587	dev_t devt;
 588	int retval;
 589
 590	/* minors == 0 indicates to use ext devt from part0 and should
 591	 * be accompanied with EXT_DEVT flag.  Make sure all
 592	 * parameters make sense.
 593	 */
 594	WARN_ON(disk->minors && !(disk->major || disk->first_minor));
 595	WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
 596
 597	disk->flags |= GENHD_FL_UP;
 598
 599	retval = blk_alloc_devt(&disk->part0, &devt);
 600	if (retval) {
 601		WARN_ON(1);
 602		return;
 603	}
 604	disk_to_dev(disk)->devt = devt;
 605
 606	/* ->major and ->first_minor aren't supposed to be
 607	 * dereferenced from here on, but set them just in case.
 608	 */
 609	disk->major = MAJOR(devt);
 610	disk->first_minor = MINOR(devt);
 611
 612	disk_alloc_events(disk);
 613
 614	/* Register BDI before referencing it from bdev */
 615	bdi = &disk->queue->backing_dev_info;
 616	bdi_register_dev(bdi, disk_devt(disk));
 617
 618	blk_register_region(disk_devt(disk), disk->minors, NULL,
 619			    exact_match, exact_lock, disk);
 620	register_disk(disk);
 621	blk_register_queue(disk);
 622
 623	/*
 624	 * Take an extra ref on queue which will be put on disk_release()
 625	 * so that it sticks around as long as @disk is there.
 626	 */
 627	WARN_ON_ONCE(!blk_get_queue(disk->queue));
 628
 629	retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
 630				   "bdi");
 631	WARN_ON(retval);
 632
 633	disk_add_events(disk);
 634	blk_integrity_add(disk);
 635}
 636EXPORT_SYMBOL(add_disk);
 637
 638void del_gendisk(struct gendisk *disk)
 639{
 640	struct disk_part_iter piter;
 641	struct hd_struct *part;
 642
 643	blk_integrity_del(disk);
 644	disk_del_events(disk);
 645
 646	/* invalidate stuff */
 647	disk_part_iter_init(&piter, disk,
 648			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
 649	while ((part = disk_part_iter_next(&piter))) {
 650		invalidate_partition(disk, part->partno);
 651		delete_partition(disk, part->partno);
 652	}
 653	disk_part_iter_exit(&piter);
 654
 655	invalidate_partition(disk, 0);
 656	set_capacity(disk, 0);
 657	disk->flags &= ~GENHD_FL_UP;
 658
 659	sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
 
 660	blk_unregister_queue(disk);
 661	blk_unregister_region(disk_devt(disk), disk->minors);
 662
 663	part_stat_set_all(&disk->part0, 0);
 664	disk->part0.stamp = 0;
 665
 666	kobject_put(disk->part0.holder_dir);
 667	kobject_put(disk->slave_dir);
 
 668	if (!sysfs_deprecated)
 669		sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
 670	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
 671	device_del(disk_to_dev(disk));
 
 672}
 673EXPORT_SYMBOL(del_gendisk);
 674
 675/* sysfs access to bad-blocks list. */
 676static ssize_t disk_badblocks_show(struct device *dev,
 677					struct device_attribute *attr,
 678					char *page)
 679{
 680	struct gendisk *disk = dev_to_disk(dev);
 681
 682	if (!disk->bb)
 683		return sprintf(page, "\n");
 684
 685	return badblocks_show(disk->bb, page, 0);
 686}
 687
 688static ssize_t disk_badblocks_store(struct device *dev,
 689					struct device_attribute *attr,
 690					const char *page, size_t len)
 691{
 692	struct gendisk *disk = dev_to_disk(dev);
 693
 694	if (!disk->bb)
 695		return -ENXIO;
 696
 697	return badblocks_store(disk->bb, page, len, 0);
 698}
 699
 700/**
 701 * get_gendisk - get partitioning information for a given device
 702 * @devt: device to get partitioning information for
 703 * @partno: returned partition index
 704 *
 705 * This function gets the structure containing partitioning
 706 * information for the given device @devt.
 707 */
 708struct gendisk *get_gendisk(dev_t devt, int *partno)
 709{
 710	struct gendisk *disk = NULL;
 711
 712	if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
 713		struct kobject *kobj;
 714
 715		kobj = kobj_lookup(bdev_map, devt, partno);
 716		if (kobj)
 717			disk = dev_to_disk(kobj_to_dev(kobj));
 718	} else {
 719		struct hd_struct *part;
 720
 721		spin_lock_bh(&ext_devt_lock);
 722		part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
 723		if (part && get_disk(part_to_disk(part))) {
 724			*partno = part->partno;
 725			disk = part_to_disk(part);
 726		}
 727		spin_unlock_bh(&ext_devt_lock);
 728	}
 729
 730	return disk;
 731}
 732EXPORT_SYMBOL(get_gendisk);
 733
 734/**
 735 * bdget_disk - do bdget() by gendisk and partition number
 736 * @disk: gendisk of interest
 737 * @partno: partition number
 738 *
 739 * Find partition @partno from @disk, do bdget() on it.
 740 *
 741 * CONTEXT:
 742 * Don't care.
 743 *
 744 * RETURNS:
 745 * Resulting block_device on success, NULL on failure.
 746 */
 747struct block_device *bdget_disk(struct gendisk *disk, int partno)
 748{
 749	struct hd_struct *part;
 750	struct block_device *bdev = NULL;
 751
 752	part = disk_get_part(disk, partno);
 753	if (part)
 754		bdev = bdget(part_devt(part));
 755	disk_put_part(part);
 756
 757	return bdev;
 758}
 759EXPORT_SYMBOL(bdget_disk);
 760
 761/*
 762 * print a full list of all partitions - intended for places where the root
 763 * filesystem can't be mounted and thus to give the victim some idea of what
 764 * went wrong
 765 */
 766void __init printk_all_partitions(void)
 767{
 768	struct class_dev_iter iter;
 769	struct device *dev;
 770
 771	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
 772	while ((dev = class_dev_iter_next(&iter))) {
 773		struct gendisk *disk = dev_to_disk(dev);
 774		struct disk_part_iter piter;
 775		struct hd_struct *part;
 776		char name_buf[BDEVNAME_SIZE];
 777		char devt_buf[BDEVT_SIZE];
 778
 779		/*
 780		 * Don't show empty devices or things that have been
 781		 * suppressed
 782		 */
 783		if (get_capacity(disk) == 0 ||
 784		    (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
 785			continue;
 786
 787		/*
 788		 * Note, unlike /proc/partitions, I am showing the
 789		 * numbers in hex - the same format as the root=
 790		 * option takes.
 791		 */
 792		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 793		while ((part = disk_part_iter_next(&piter))) {
 794			bool is_part0 = part == &disk->part0;
 795
 796			printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
 797			       bdevt_str(part_devt(part), devt_buf),
 798			       (unsigned long long)part_nr_sects_read(part) >> 1
 799			       , disk_name(disk, part->partno, name_buf),
 800			       part->info ? part->info->uuid : "");
 801			if (is_part0) {
 802				if (disk->driverfs_dev != NULL &&
 803				    disk->driverfs_dev->driver != NULL)
 804					printk(" driver: %s\n",
 805					      disk->driverfs_dev->driver->name);
 806				else
 807					printk(" (driver?)\n");
 808			} else
 809				printk("\n");
 810		}
 811		disk_part_iter_exit(&piter);
 812	}
 813	class_dev_iter_exit(&iter);
 814}
 815
 816#ifdef CONFIG_PROC_FS
 817/* iterator */
 818static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
 819{
 820	loff_t skip = *pos;
 821	struct class_dev_iter *iter;
 822	struct device *dev;
 823
 824	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 825	if (!iter)
 826		return ERR_PTR(-ENOMEM);
 827
 828	seqf->private = iter;
 829	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
 830	do {
 831		dev = class_dev_iter_next(iter);
 832		if (!dev)
 833			return NULL;
 834	} while (skip--);
 835
 836	return dev_to_disk(dev);
 837}
 838
 839static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
 840{
 841	struct device *dev;
 842
 843	(*pos)++;
 844	dev = class_dev_iter_next(seqf->private);
 845	if (dev)
 846		return dev_to_disk(dev);
 847
 848	return NULL;
 849}
 850
 851static void disk_seqf_stop(struct seq_file *seqf, void *v)
 852{
 853	struct class_dev_iter *iter = seqf->private;
 854
 855	/* stop is called even after start failed :-( */
 856	if (iter) {
 857		class_dev_iter_exit(iter);
 858		kfree(iter);
 859	}
 860}
 861
 862static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
 863{
 864	void *p;
 865
 866	p = disk_seqf_start(seqf, pos);
 867	if (!IS_ERR_OR_NULL(p) && !*pos)
 868		seq_puts(seqf, "major minor  #blocks  name\n\n");
 869	return p;
 870}
 871
 872static int show_partition(struct seq_file *seqf, void *v)
 873{
 874	struct gendisk *sgp = v;
 875	struct disk_part_iter piter;
 876	struct hd_struct *part;
 877	char buf[BDEVNAME_SIZE];
 878
 879	/* Don't show non-partitionable removeable devices or empty devices */
 880	if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
 881				   (sgp->flags & GENHD_FL_REMOVABLE)))
 882		return 0;
 883	if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
 884		return 0;
 885
 886	/* show the full disk and all non-0 size partitions of it */
 887	disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
 888	while ((part = disk_part_iter_next(&piter)))
 889		seq_printf(seqf, "%4d  %7d %10llu %s\n",
 890			   MAJOR(part_devt(part)), MINOR(part_devt(part)),
 891			   (unsigned long long)part_nr_sects_read(part) >> 1,
 892			   disk_name(sgp, part->partno, buf));
 893	disk_part_iter_exit(&piter);
 894
 895	return 0;
 896}
 897
 898static const struct seq_operations partitions_op = {
 899	.start	= show_partition_start,
 900	.next	= disk_seqf_next,
 901	.stop	= disk_seqf_stop,
 902	.show	= show_partition
 903};
 904
 905static int partitions_open(struct inode *inode, struct file *file)
 906{
 907	return seq_open(file, &partitions_op);
 908}
 909
 910static const struct file_operations proc_partitions_operations = {
 911	.open		= partitions_open,
 912	.read		= seq_read,
 913	.llseek		= seq_lseek,
 914	.release	= seq_release,
 915};
 916#endif
 917
 918
 919static struct kobject *base_probe(dev_t devt, int *partno, void *data)
 920{
 921	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
 922		/* Make old-style 2.4 aliases work */
 923		request_module("block-major-%d", MAJOR(devt));
 924	return NULL;
 925}
 926
 927static int __init genhd_device_init(void)
 928{
 929	int error;
 930
 931	block_class.dev_kobj = sysfs_dev_block_kobj;
 932	error = class_register(&block_class);
 933	if (unlikely(error))
 934		return error;
 935	bdev_map = kobj_map_init(base_probe, &block_class_lock);
 936	blk_dev_init();
 937
 938	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
 939
 940	/* create top-level block dir */
 941	if (!sysfs_deprecated)
 942		block_depr = kobject_create_and_add("block", NULL);
 943	return 0;
 944}
 945
 946subsys_initcall(genhd_device_init);
 947
 948static ssize_t disk_range_show(struct device *dev,
 949			       struct device_attribute *attr, char *buf)
 950{
 951	struct gendisk *disk = dev_to_disk(dev);
 952
 953	return sprintf(buf, "%d\n", disk->minors);
 954}
 955
 956static ssize_t disk_ext_range_show(struct device *dev,
 957				   struct device_attribute *attr, char *buf)
 958{
 959	struct gendisk *disk = dev_to_disk(dev);
 960
 961	return sprintf(buf, "%d\n", disk_max_parts(disk));
 962}
 963
 964static ssize_t disk_removable_show(struct device *dev,
 965				   struct device_attribute *attr, char *buf)
 966{
 967	struct gendisk *disk = dev_to_disk(dev);
 968
 969	return sprintf(buf, "%d\n",
 970		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
 971}
 972
 973static ssize_t disk_ro_show(struct device *dev,
 974				   struct device_attribute *attr, char *buf)
 975{
 976	struct gendisk *disk = dev_to_disk(dev);
 977
 978	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
 979}
 980
 981static ssize_t disk_capability_show(struct device *dev,
 982				    struct device_attribute *attr, char *buf)
 983{
 984	struct gendisk *disk = dev_to_disk(dev);
 985
 986	return sprintf(buf, "%x\n", disk->flags);
 987}
 988
 989static ssize_t disk_alignment_offset_show(struct device *dev,
 990					  struct device_attribute *attr,
 991					  char *buf)
 992{
 993	struct gendisk *disk = dev_to_disk(dev);
 994
 995	return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
 996}
 997
 998static ssize_t disk_discard_alignment_show(struct device *dev,
 999					   struct device_attribute *attr,
1000					   char *buf)
1001{
1002	struct gendisk *disk = dev_to_disk(dev);
1003
1004	return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1005}
1006
1007static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
1008static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
1009static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
1010static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
1011static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
1012static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
1013static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
1014		   NULL);
1015static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
1016static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
1017static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
1018static DEVICE_ATTR(badblocks, S_IRUGO | S_IWUSR, disk_badblocks_show,
1019		disk_badblocks_store);
1020#ifdef CONFIG_FAIL_MAKE_REQUEST
1021static struct device_attribute dev_attr_fail =
1022	__ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
1023#endif
1024#ifdef CONFIG_FAIL_IO_TIMEOUT
1025static struct device_attribute dev_attr_fail_timeout =
1026	__ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
1027		part_timeout_store);
1028#endif
1029
1030static struct attribute *disk_attrs[] = {
1031	&dev_attr_range.attr,
1032	&dev_attr_ext_range.attr,
1033	&dev_attr_removable.attr,
1034	&dev_attr_ro.attr,
1035	&dev_attr_size.attr,
1036	&dev_attr_alignment_offset.attr,
1037	&dev_attr_discard_alignment.attr,
1038	&dev_attr_capability.attr,
1039	&dev_attr_stat.attr,
1040	&dev_attr_inflight.attr,
1041	&dev_attr_badblocks.attr,
1042#ifdef CONFIG_FAIL_MAKE_REQUEST
1043	&dev_attr_fail.attr,
1044#endif
1045#ifdef CONFIG_FAIL_IO_TIMEOUT
1046	&dev_attr_fail_timeout.attr,
1047#endif
1048	NULL
1049};
1050
1051static struct attribute_group disk_attr_group = {
1052	.attrs = disk_attrs,
1053};
1054
1055static const struct attribute_group *disk_attr_groups[] = {
1056	&disk_attr_group,
1057	NULL
1058};
1059
1060/**
1061 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1062 * @disk: disk to replace part_tbl for
1063 * @new_ptbl: new part_tbl to install
1064 *
1065 * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
1066 * original ptbl is freed using RCU callback.
1067 *
1068 * LOCKING:
1069 * Matching bd_mutx locked.
1070 */
1071static void disk_replace_part_tbl(struct gendisk *disk,
1072				  struct disk_part_tbl *new_ptbl)
1073{
1074	struct disk_part_tbl *old_ptbl = disk->part_tbl;
1075
1076	rcu_assign_pointer(disk->part_tbl, new_ptbl);
1077
1078	if (old_ptbl) {
1079		rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1080		kfree_rcu(old_ptbl, rcu_head);
1081	}
1082}
1083
1084/**
1085 * disk_expand_part_tbl - expand disk->part_tbl
1086 * @disk: disk to expand part_tbl for
1087 * @partno: expand such that this partno can fit in
1088 *
1089 * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
1090 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1091 *
1092 * LOCKING:
1093 * Matching bd_mutex locked, might sleep.
1094 *
1095 * RETURNS:
1096 * 0 on success, -errno on failure.
1097 */
1098int disk_expand_part_tbl(struct gendisk *disk, int partno)
1099{
1100	struct disk_part_tbl *old_ptbl = disk->part_tbl;
1101	struct disk_part_tbl *new_ptbl;
1102	int len = old_ptbl ? old_ptbl->len : 0;
1103	int i, target;
1104	size_t size;
1105
1106	/*
1107	 * check for int overflow, since we can get here from blkpg_ioctl()
1108	 * with a user passed 'partno'.
1109	 */
1110	target = partno + 1;
1111	if (target < 0)
1112		return -EINVAL;
1113
1114	/* disk_max_parts() is zero during initialization, ignore if so */
1115	if (disk_max_parts(disk) && target > disk_max_parts(disk))
1116		return -EINVAL;
1117
1118	if (target <= len)
1119		return 0;
1120
1121	size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1122	new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1123	if (!new_ptbl)
1124		return -ENOMEM;
1125
1126	new_ptbl->len = target;
1127
1128	for (i = 0; i < len; i++)
1129		rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1130
1131	disk_replace_part_tbl(disk, new_ptbl);
1132	return 0;
1133}
1134
1135static void disk_release(struct device *dev)
1136{
1137	struct gendisk *disk = dev_to_disk(dev);
1138
1139	blk_free_devt(dev->devt);
1140	disk_release_events(disk);
1141	kfree(disk->random);
1142	disk_replace_part_tbl(disk, NULL);
1143	hd_free_part(&disk->part0);
 
1144	if (disk->queue)
1145		blk_put_queue(disk->queue);
1146	kfree(disk);
1147}
1148struct class block_class = {
1149	.name		= "block",
1150};
1151
1152static char *block_devnode(struct device *dev, umode_t *mode,
1153			   kuid_t *uid, kgid_t *gid)
1154{
1155	struct gendisk *disk = dev_to_disk(dev);
1156
1157	if (disk->devnode)
1158		return disk->devnode(disk, mode);
1159	return NULL;
1160}
1161
1162static struct device_type disk_type = {
1163	.name		= "disk",
1164	.groups		= disk_attr_groups,
1165	.release	= disk_release,
1166	.devnode	= block_devnode,
1167};
1168
1169#ifdef CONFIG_PROC_FS
1170/*
1171 * aggregate disk stat collector.  Uses the same stats that the sysfs
1172 * entries do, above, but makes them available through one seq_file.
1173 *
1174 * The output looks suspiciously like /proc/partitions with a bunch of
1175 * extra fields.
1176 */
1177static int diskstats_show(struct seq_file *seqf, void *v)
1178{
1179	struct gendisk *gp = v;
1180	struct disk_part_iter piter;
1181	struct hd_struct *hd;
1182	char buf[BDEVNAME_SIZE];
1183	int cpu;
1184
1185	/*
1186	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1187		seq_puts(seqf,	"major minor name"
1188				"     rio rmerge rsect ruse wio wmerge "
1189				"wsect wuse running use aveq"
1190				"\n\n");
1191	*/
1192
1193	disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1194	while ((hd = disk_part_iter_next(&piter))) {
1195		cpu = part_stat_lock();
1196		part_round_stats(cpu, hd);
1197		part_stat_unlock();
1198		seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1199			   "%u %lu %lu %lu %u %u %u %u\n",
1200			   MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1201			   disk_name(gp, hd->partno, buf),
1202			   part_stat_read(hd, ios[READ]),
1203			   part_stat_read(hd, merges[READ]),
1204			   part_stat_read(hd, sectors[READ]),
1205			   jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1206			   part_stat_read(hd, ios[WRITE]),
1207			   part_stat_read(hd, merges[WRITE]),
1208			   part_stat_read(hd, sectors[WRITE]),
1209			   jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1210			   part_in_flight(hd),
1211			   jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1212			   jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1213			);
1214	}
1215	disk_part_iter_exit(&piter);
1216
1217	return 0;
1218}
1219
1220static const struct seq_operations diskstats_op = {
1221	.start	= disk_seqf_start,
1222	.next	= disk_seqf_next,
1223	.stop	= disk_seqf_stop,
1224	.show	= diskstats_show
1225};
1226
1227static int diskstats_open(struct inode *inode, struct file *file)
1228{
1229	return seq_open(file, &diskstats_op);
1230}
1231
1232static const struct file_operations proc_diskstats_operations = {
1233	.open		= diskstats_open,
1234	.read		= seq_read,
1235	.llseek		= seq_lseek,
1236	.release	= seq_release,
1237};
1238
1239static int __init proc_genhd_init(void)
1240{
1241	proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1242	proc_create("partitions", 0, NULL, &proc_partitions_operations);
1243	return 0;
1244}
1245module_init(proc_genhd_init);
1246#endif /* CONFIG_PROC_FS */
1247
1248dev_t blk_lookup_devt(const char *name, int partno)
1249{
1250	dev_t devt = MKDEV(0, 0);
1251	struct class_dev_iter iter;
1252	struct device *dev;
1253
1254	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1255	while ((dev = class_dev_iter_next(&iter))) {
1256		struct gendisk *disk = dev_to_disk(dev);
1257		struct hd_struct *part;
1258
1259		if (strcmp(dev_name(dev), name))
1260			continue;
1261
1262		if (partno < disk->minors) {
1263			/* We need to return the right devno, even
1264			 * if the partition doesn't exist yet.
1265			 */
1266			devt = MKDEV(MAJOR(dev->devt),
1267				     MINOR(dev->devt) + partno);
1268			break;
1269		}
1270		part = disk_get_part(disk, partno);
1271		if (part) {
1272			devt = part_devt(part);
1273			disk_put_part(part);
1274			break;
1275		}
1276		disk_put_part(part);
1277	}
1278	class_dev_iter_exit(&iter);
1279	return devt;
1280}
1281EXPORT_SYMBOL(blk_lookup_devt);
1282
1283struct gendisk *alloc_disk(int minors)
1284{
1285	return alloc_disk_node(minors, NUMA_NO_NODE);
1286}
1287EXPORT_SYMBOL(alloc_disk);
1288
1289struct gendisk *alloc_disk_node(int minors, int node_id)
1290{
1291	struct gendisk *disk;
1292
1293	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1294	if (disk) {
1295		if (!init_part_stats(&disk->part0)) {
1296			kfree(disk);
1297			return NULL;
1298		}
1299		disk->node_id = node_id;
1300		if (disk_expand_part_tbl(disk, 0)) {
1301			free_part_stats(&disk->part0);
1302			kfree(disk);
1303			return NULL;
1304		}
1305		disk->part_tbl->part[0] = &disk->part0;
1306
1307		/*
1308		 * set_capacity() and get_capacity() currently don't use
1309		 * seqcounter to read/update the part0->nr_sects. Still init
1310		 * the counter as we can read the sectors in IO submission
1311		 * patch using seqence counters.
1312		 *
1313		 * TODO: Ideally set_capacity() and get_capacity() should be
1314		 * converted to make use of bd_mutex and sequence counters.
1315		 */
1316		seqcount_init(&disk->part0.nr_sects_seq);
1317		if (hd_ref_init(&disk->part0)) {
1318			hd_free_part(&disk->part0);
1319			kfree(disk);
1320			return NULL;
1321		}
1322
1323		disk->minors = minors;
1324		rand_initialize_disk(disk);
1325		disk_to_dev(disk)->class = &block_class;
1326		disk_to_dev(disk)->type = &disk_type;
1327		device_initialize(disk_to_dev(disk));
1328	}
1329	return disk;
1330}
1331EXPORT_SYMBOL(alloc_disk_node);
1332
1333struct kobject *get_disk(struct gendisk *disk)
1334{
1335	struct module *owner;
1336	struct kobject *kobj;
1337
1338	if (!disk->fops)
1339		return NULL;
1340	owner = disk->fops->owner;
1341	if (owner && !try_module_get(owner))
1342		return NULL;
1343	kobj = kobject_get(&disk_to_dev(disk)->kobj);
1344	if (kobj == NULL) {
1345		module_put(owner);
1346		return NULL;
1347	}
1348	return kobj;
1349
1350}
1351
1352EXPORT_SYMBOL(get_disk);
1353
1354void put_disk(struct gendisk *disk)
1355{
1356	if (disk)
1357		kobject_put(&disk_to_dev(disk)->kobj);
1358}
1359
1360EXPORT_SYMBOL(put_disk);
1361
1362static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1363{
1364	char event[] = "DISK_RO=1";
1365	char *envp[] = { event, NULL };
1366
1367	if (!ro)
1368		event[8] = '0';
1369	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1370}
1371
1372void set_device_ro(struct block_device *bdev, int flag)
1373{
1374	bdev->bd_part->policy = flag;
1375}
1376
1377EXPORT_SYMBOL(set_device_ro);
1378
1379void set_disk_ro(struct gendisk *disk, int flag)
1380{
1381	struct disk_part_iter piter;
1382	struct hd_struct *part;
1383
1384	if (disk->part0.policy != flag) {
1385		set_disk_ro_uevent(disk, flag);
1386		disk->part0.policy = flag;
1387	}
1388
1389	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1390	while ((part = disk_part_iter_next(&piter)))
1391		part->policy = flag;
1392	disk_part_iter_exit(&piter);
1393}
1394
1395EXPORT_SYMBOL(set_disk_ro);
1396
1397int bdev_read_only(struct block_device *bdev)
1398{
1399	if (!bdev)
1400		return 0;
1401	return bdev->bd_part->policy;
1402}
1403
1404EXPORT_SYMBOL(bdev_read_only);
1405
1406int invalidate_partition(struct gendisk *disk, int partno)
1407{
1408	int res = 0;
1409	struct block_device *bdev = bdget_disk(disk, partno);
1410	if (bdev) {
1411		fsync_bdev(bdev);
1412		res = __invalidate_device(bdev, true);
1413		bdput(bdev);
1414	}
1415	return res;
1416}
1417
1418EXPORT_SYMBOL(invalidate_partition);
1419
1420/*
1421 * Disk events - monitor disk events like media change and eject request.
1422 */
1423struct disk_events {
1424	struct list_head	node;		/* all disk_event's */
1425	struct gendisk		*disk;		/* the associated disk */
1426	spinlock_t		lock;
1427
1428	struct mutex		block_mutex;	/* protects blocking */
1429	int			block;		/* event blocking depth */
1430	unsigned int		pending;	/* events already sent out */
1431	unsigned int		clearing;	/* events being cleared */
1432
1433	long			poll_msecs;	/* interval, -1 for default */
1434	struct delayed_work	dwork;
1435};
1436
1437static const char *disk_events_strs[] = {
1438	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "media_change",
1439	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "eject_request",
1440};
1441
1442static char *disk_uevents[] = {
1443	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "DISK_MEDIA_CHANGE=1",
1444	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "DISK_EJECT_REQUEST=1",
1445};
1446
1447/* list of all disk_events */
1448static DEFINE_MUTEX(disk_events_mutex);
1449static LIST_HEAD(disk_events);
1450
1451/* disable in-kernel polling by default */
1452static unsigned long disk_events_dfl_poll_msecs;
1453
1454static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1455{
1456	struct disk_events *ev = disk->ev;
1457	long intv_msecs = 0;
1458
1459	/*
1460	 * If device-specific poll interval is set, always use it.  If
1461	 * the default is being used, poll iff there are events which
1462	 * can't be monitored asynchronously.
1463	 */
1464	if (ev->poll_msecs >= 0)
1465		intv_msecs = ev->poll_msecs;
1466	else if (disk->events & ~disk->async_events)
1467		intv_msecs = disk_events_dfl_poll_msecs;
1468
1469	return msecs_to_jiffies(intv_msecs);
1470}
1471
1472/**
1473 * disk_block_events - block and flush disk event checking
1474 * @disk: disk to block events for
1475 *
1476 * On return from this function, it is guaranteed that event checking
1477 * isn't in progress and won't happen until unblocked by
1478 * disk_unblock_events().  Events blocking is counted and the actual
1479 * unblocking happens after the matching number of unblocks are done.
1480 *
1481 * Note that this intentionally does not block event checking from
1482 * disk_clear_events().
1483 *
1484 * CONTEXT:
1485 * Might sleep.
1486 */
1487void disk_block_events(struct gendisk *disk)
1488{
1489	struct disk_events *ev = disk->ev;
1490	unsigned long flags;
1491	bool cancel;
1492
1493	if (!ev)
1494		return;
1495
1496	/*
1497	 * Outer mutex ensures that the first blocker completes canceling
1498	 * the event work before further blockers are allowed to finish.
1499	 */
1500	mutex_lock(&ev->block_mutex);
1501
1502	spin_lock_irqsave(&ev->lock, flags);
1503	cancel = !ev->block++;
1504	spin_unlock_irqrestore(&ev->lock, flags);
1505
1506	if (cancel)
1507		cancel_delayed_work_sync(&disk->ev->dwork);
1508
1509	mutex_unlock(&ev->block_mutex);
1510}
1511
1512static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1513{
1514	struct disk_events *ev = disk->ev;
1515	unsigned long intv;
1516	unsigned long flags;
1517
1518	spin_lock_irqsave(&ev->lock, flags);
1519
1520	if (WARN_ON_ONCE(ev->block <= 0))
1521		goto out_unlock;
1522
1523	if (--ev->block)
1524		goto out_unlock;
1525
1526	/*
1527	 * Not exactly a latency critical operation, set poll timer
1528	 * slack to 25% and kick event check.
1529	 */
1530	intv = disk_events_poll_jiffies(disk);
1531	set_timer_slack(&ev->dwork.timer, intv / 4);
1532	if (check_now)
1533		queue_delayed_work(system_freezable_power_efficient_wq,
1534				&ev->dwork, 0);
1535	else if (intv)
1536		queue_delayed_work(system_freezable_power_efficient_wq,
1537				&ev->dwork, intv);
1538out_unlock:
1539	spin_unlock_irqrestore(&ev->lock, flags);
1540}
1541
1542/**
1543 * disk_unblock_events - unblock disk event checking
1544 * @disk: disk to unblock events for
1545 *
1546 * Undo disk_block_events().  When the block count reaches zero, it
1547 * starts events polling if configured.
1548 *
1549 * CONTEXT:
1550 * Don't care.  Safe to call from irq context.
1551 */
1552void disk_unblock_events(struct gendisk *disk)
1553{
1554	if (disk->ev)
1555		__disk_unblock_events(disk, false);
1556}
1557
1558/**
1559 * disk_flush_events - schedule immediate event checking and flushing
1560 * @disk: disk to check and flush events for
1561 * @mask: events to flush
1562 *
1563 * Schedule immediate event checking on @disk if not blocked.  Events in
1564 * @mask are scheduled to be cleared from the driver.  Note that this
1565 * doesn't clear the events from @disk->ev.
1566 *
1567 * CONTEXT:
1568 * If @mask is non-zero must be called with bdev->bd_mutex held.
1569 */
1570void disk_flush_events(struct gendisk *disk, unsigned int mask)
1571{
1572	struct disk_events *ev = disk->ev;
1573
1574	if (!ev)
1575		return;
1576
1577	spin_lock_irq(&ev->lock);
1578	ev->clearing |= mask;
1579	if (!ev->block)
1580		mod_delayed_work(system_freezable_power_efficient_wq,
1581				&ev->dwork, 0);
1582	spin_unlock_irq(&ev->lock);
1583}
1584
1585/**
1586 * disk_clear_events - synchronously check, clear and return pending events
1587 * @disk: disk to fetch and clear events from
1588 * @mask: mask of events to be fetched and cleared
1589 *
1590 * Disk events are synchronously checked and pending events in @mask
1591 * are cleared and returned.  This ignores the block count.
1592 *
1593 * CONTEXT:
1594 * Might sleep.
1595 */
1596unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1597{
1598	const struct block_device_operations *bdops = disk->fops;
1599	struct disk_events *ev = disk->ev;
1600	unsigned int pending;
1601	unsigned int clearing = mask;
1602
1603	if (!ev) {
1604		/* for drivers still using the old ->media_changed method */
1605		if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1606		    bdops->media_changed && bdops->media_changed(disk))
1607			return DISK_EVENT_MEDIA_CHANGE;
1608		return 0;
1609	}
1610
1611	disk_block_events(disk);
1612
1613	/*
1614	 * store the union of mask and ev->clearing on the stack so that the
1615	 * race with disk_flush_events does not cause ambiguity (ev->clearing
1616	 * can still be modified even if events are blocked).
1617	 */
1618	spin_lock_irq(&ev->lock);
1619	clearing |= ev->clearing;
1620	ev->clearing = 0;
1621	spin_unlock_irq(&ev->lock);
1622
1623	disk_check_events(ev, &clearing);
1624	/*
1625	 * if ev->clearing is not 0, the disk_flush_events got called in the
1626	 * middle of this function, so we want to run the workfn without delay.
1627	 */
1628	__disk_unblock_events(disk, ev->clearing ? true : false);
1629
1630	/* then, fetch and clear pending events */
1631	spin_lock_irq(&ev->lock);
1632	pending = ev->pending & mask;
1633	ev->pending &= ~mask;
1634	spin_unlock_irq(&ev->lock);
1635	WARN_ON_ONCE(clearing & mask);
1636
1637	return pending;
1638}
1639
1640/*
1641 * Separate this part out so that a different pointer for clearing_ptr can be
1642 * passed in for disk_clear_events.
1643 */
1644static void disk_events_workfn(struct work_struct *work)
1645{
1646	struct delayed_work *dwork = to_delayed_work(work);
1647	struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1648
1649	disk_check_events(ev, &ev->clearing);
1650}
1651
1652static void disk_check_events(struct disk_events *ev,
1653			      unsigned int *clearing_ptr)
1654{
1655	struct gendisk *disk = ev->disk;
1656	char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1657	unsigned int clearing = *clearing_ptr;
1658	unsigned int events;
1659	unsigned long intv;
1660	int nr_events = 0, i;
1661
1662	/* check events */
1663	events = disk->fops->check_events(disk, clearing);
1664
1665	/* accumulate pending events and schedule next poll if necessary */
1666	spin_lock_irq(&ev->lock);
1667
1668	events &= ~ev->pending;
1669	ev->pending |= events;
1670	*clearing_ptr &= ~clearing;
1671
1672	intv = disk_events_poll_jiffies(disk);
1673	if (!ev->block && intv)
1674		queue_delayed_work(system_freezable_power_efficient_wq,
1675				&ev->dwork, intv);
1676
1677	spin_unlock_irq(&ev->lock);
1678
1679	/*
1680	 * Tell userland about new events.  Only the events listed in
1681	 * @disk->events are reported.  Unlisted events are processed the
1682	 * same internally but never get reported to userland.
1683	 */
1684	for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1685		if (events & disk->events & (1 << i))
1686			envp[nr_events++] = disk_uevents[i];
1687
1688	if (nr_events)
1689		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1690}
1691
1692/*
1693 * A disk events enabled device has the following sysfs nodes under
1694 * its /sys/block/X/ directory.
1695 *
1696 * events		: list of all supported events
1697 * events_async		: list of events which can be detected w/o polling
1698 * events_poll_msecs	: polling interval, 0: disable, -1: system default
1699 */
1700static ssize_t __disk_events_show(unsigned int events, char *buf)
1701{
1702	const char *delim = "";
1703	ssize_t pos = 0;
1704	int i;
1705
1706	for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1707		if (events & (1 << i)) {
1708			pos += sprintf(buf + pos, "%s%s",
1709				       delim, disk_events_strs[i]);
1710			delim = " ";
1711		}
1712	if (pos)
1713		pos += sprintf(buf + pos, "\n");
1714	return pos;
1715}
1716
1717static ssize_t disk_events_show(struct device *dev,
1718				struct device_attribute *attr, char *buf)
1719{
1720	struct gendisk *disk = dev_to_disk(dev);
1721
1722	return __disk_events_show(disk->events, buf);
1723}
1724
1725static ssize_t disk_events_async_show(struct device *dev,
1726				      struct device_attribute *attr, char *buf)
1727{
1728	struct gendisk *disk = dev_to_disk(dev);
1729
1730	return __disk_events_show(disk->async_events, buf);
1731}
1732
1733static ssize_t disk_events_poll_msecs_show(struct device *dev,
1734					   struct device_attribute *attr,
1735					   char *buf)
1736{
1737	struct gendisk *disk = dev_to_disk(dev);
1738
1739	return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1740}
1741
1742static ssize_t disk_events_poll_msecs_store(struct device *dev,
1743					    struct device_attribute *attr,
1744					    const char *buf, size_t count)
1745{
1746	struct gendisk *disk = dev_to_disk(dev);
1747	long intv;
1748
1749	if (!count || !sscanf(buf, "%ld", &intv))
1750		return -EINVAL;
1751
1752	if (intv < 0 && intv != -1)
1753		return -EINVAL;
1754
1755	disk_block_events(disk);
1756	disk->ev->poll_msecs = intv;
1757	__disk_unblock_events(disk, true);
1758
1759	return count;
1760}
1761
1762static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1763static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1764static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1765			 disk_events_poll_msecs_show,
1766			 disk_events_poll_msecs_store);
1767
1768static const struct attribute *disk_events_attrs[] = {
1769	&dev_attr_events.attr,
1770	&dev_attr_events_async.attr,
1771	&dev_attr_events_poll_msecs.attr,
1772	NULL,
1773};
1774
1775/*
1776 * The default polling interval can be specified by the kernel
1777 * parameter block.events_dfl_poll_msecs which defaults to 0
1778 * (disable).  This can also be modified runtime by writing to
1779 * /sys/module/block/events_dfl_poll_msecs.
1780 */
1781static int disk_events_set_dfl_poll_msecs(const char *val,
1782					  const struct kernel_param *kp)
1783{
1784	struct disk_events *ev;
1785	int ret;
1786
1787	ret = param_set_ulong(val, kp);
1788	if (ret < 0)
1789		return ret;
1790
1791	mutex_lock(&disk_events_mutex);
1792
1793	list_for_each_entry(ev, &disk_events, node)
1794		disk_flush_events(ev->disk, 0);
1795
1796	mutex_unlock(&disk_events_mutex);
1797
1798	return 0;
1799}
1800
1801static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1802	.set	= disk_events_set_dfl_poll_msecs,
1803	.get	= param_get_ulong,
1804};
1805
1806#undef MODULE_PARAM_PREFIX
1807#define MODULE_PARAM_PREFIX	"block."
1808
1809module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1810		&disk_events_dfl_poll_msecs, 0644);
1811
1812/*
1813 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1814 */
1815static void disk_alloc_events(struct gendisk *disk)
1816{
1817	struct disk_events *ev;
1818
1819	if (!disk->fops->check_events)
1820		return;
1821
1822	ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1823	if (!ev) {
1824		pr_warn("%s: failed to initialize events\n", disk->disk_name);
1825		return;
1826	}
1827
1828	INIT_LIST_HEAD(&ev->node);
1829	ev->disk = disk;
1830	spin_lock_init(&ev->lock);
1831	mutex_init(&ev->block_mutex);
1832	ev->block = 1;
1833	ev->poll_msecs = -1;
1834	INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1835
1836	disk->ev = ev;
1837}
1838
1839static void disk_add_events(struct gendisk *disk)
1840{
1841	if (!disk->ev)
1842		return;
1843
1844	/* FIXME: error handling */
1845	if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1846		pr_warn("%s: failed to create sysfs files for events\n",
1847			disk->disk_name);
1848
1849	mutex_lock(&disk_events_mutex);
1850	list_add_tail(&disk->ev->node, &disk_events);
1851	mutex_unlock(&disk_events_mutex);
1852
1853	/*
1854	 * Block count is initialized to 1 and the following initial
1855	 * unblock kicks it into action.
1856	 */
1857	__disk_unblock_events(disk, true);
1858}
1859
1860static void disk_del_events(struct gendisk *disk)
1861{
1862	if (!disk->ev)
1863		return;
1864
1865	disk_block_events(disk);
1866
1867	mutex_lock(&disk_events_mutex);
1868	list_del_init(&disk->ev->node);
1869	mutex_unlock(&disk_events_mutex);
1870
1871	sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1872}
1873
1874static void disk_release_events(struct gendisk *disk)
1875{
1876	/* the block count should be 1 from disk_del_events() */
1877	WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1878	kfree(disk->ev);
1879}