genhd.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  gendisk handling
   4 *
   5 * Portions Copyright (C) 2020 Christoph Hellwig
   6 */
   7
   8#include <linux/module.h>
   9#include <linux/ctype.h>
  10#include <linux/fs.h>
  11#include <linux/kdev_t.h>
  12#include <linux/kernel.h>
  13#include <linux/blkdev.h>
  14#include <linux/backing-dev.h>
  15#include <linux/init.h>
  16#include <linux/spinlock.h>
  17#include <linux/proc_fs.h>
  18#include <linux/seq_file.h>
  19#include <linux/slab.h>
  20#include <linux/kmod.h>
  21#include <linux/major.h>
  22#include <linux/mutex.h>
  23#include <linux/idr.h>
  24#include <linux/log2.h>
  25#include <linux/pm_runtime.h>
  26#include <linux/badblocks.h>
  27#include <linux/part_stat.h>
  28#include <linux/blktrace_api.h>
  29
  30#include "blk-throttle.h"
  31#include "blk.h"
  32#include "blk-mq-sched.h"
  33#include "blk-rq-qos.h"
  34#include "blk-cgroup.h"
  35
  36static struct kobject *block_depr;
  37
  38/*
  39 * Unique, monotonically increasing sequential number associated with block
  40 * devices instances (i.e. incremented each time a device is attached).
  41 * Associating uevents with block devices in userspace is difficult and racy:
  42 * the uevent netlink socket is lossy, and on slow and overloaded systems has
  43 * a very high latency.
  44 * Block devices do not have exclusive owners in userspace, any process can set
  45 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
  46 * can be reused again and again).
  47 * A userspace process setting up a block device and watching for its events
  48 * cannot thus reliably tell whether an event relates to the device it just set
  49 * up or another earlier instance with the same name.
  50 * This sequential number allows userspace processes to solve this problem, and
  51 * uniquely associate an uevent to the lifetime to a device.
  52 */
  53static atomic64_t diskseq;
  54
  55/* for extended dynamic devt allocation, currently only one major is used */
  56#define NR_EXT_DEVT		(1 << MINORBITS)
  57static DEFINE_IDA(ext_devt_ida);
  58
  59void set_capacity(struct gendisk *disk, sector_t sectors)
  60{
  61	if (sectors > BLK_DEV_MAX_SECTORS) {
  62		pr_warn_once("%s: truncate capacity from %lld to %lld\n",
  63				disk->disk_name, sectors,
  64				BLK_DEV_MAX_SECTORS);
  65		sectors = BLK_DEV_MAX_SECTORS;
  66	}
  67
  68	bdev_set_nr_sectors(disk->part0, sectors);
  69}
  70EXPORT_SYMBOL(set_capacity);
  71
  72/*
  73 * Set disk capacity and notify if the size is not currently zero and will not
  74 * be set to zero.  Returns true if a uevent was sent, otherwise false.
  75 */
  76bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
  77{
  78	sector_t capacity = get_capacity(disk);
  79	char *envp[] = { "RESIZE=1", NULL };
  80
  81	set_capacity(disk, size);
  82
  83	/*
  84	 * Only print a message and send a uevent if the gendisk is user visible
  85	 * and alive.  This avoids spamming the log and udev when setting the
  86	 * initial capacity during probing.
  87	 */
  88	if (size == capacity ||
  89	    !disk_live(disk) ||
  90	    (disk->flags & GENHD_FL_HIDDEN))
  91		return false;
  92
  93	pr_info_ratelimited("%s: detected capacity change from %lld to %lld\n",
  94		disk->disk_name, capacity, size);
  95
  96	/*
  97	 * Historically we did not send a uevent for changes to/from an empty
  98	 * device.
  99	 */
 100	if (!capacity || !size)
 101		return false;
 102	kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
 103	return true;
 104}
 105EXPORT_SYMBOL_GPL(set_capacity_and_notify);
 106
 107static void part_stat_read_all(struct block_device *part,
 108		struct disk_stats *stat)
 109{
 110	int cpu;
 111
 112	memset(stat, 0, sizeof(struct disk_stats));
 113	for_each_possible_cpu(cpu) {
 114		struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
 115		int group;
 116
 117		for (group = 0; group < NR_STAT_GROUPS; group++) {
 118			stat->nsecs[group] += ptr->nsecs[group];
 119			stat->sectors[group] += ptr->sectors[group];
 120			stat->ios[group] += ptr->ios[group];
 121			stat->merges[group] += ptr->merges[group];
 122		}
 123
 124		stat->io_ticks += ptr->io_ticks;
 125	}
 126}
 127
 128static void bdev_count_inflight_rw(struct block_device *part,
 129		unsigned int inflight[2], bool mq_driver)
 130{
 131	int write = 0;
 132	int read = 0;
 133	int cpu;
 134
 135	if (mq_driver) {
 136		blk_mq_in_driver_rw(part, inflight);
 137		return;
 138	}
 139
 140	for_each_possible_cpu(cpu) {
 141		read += part_stat_local_read_cpu(part, in_flight[READ], cpu);
 142		write += part_stat_local_read_cpu(part, in_flight[WRITE], cpu);
 143	}
 144
 145	/*
 146	 * While iterating all CPUs, some IOs may be issued from a CPU already
 147	 * traversed and complete on a CPU that has not yet been traversed,
 148	 * causing the inflight number to be negative.
 149	 */
 150	inflight[READ] = read > 0 ? read : 0;
 151	inflight[WRITE] = write > 0 ? write : 0;
 152}
 153
 154/**
 155 * bdev_count_inflight - get the number of inflight IOs for a block device.
 156 *
 157 * @part: the block device.
 158 *
 159 * Inflight here means started IO accounting, from bdev_start_io_acct() for
 160 * bio-based block device, and from blk_account_io_start() for rq-based block
 161 * device.
 162 */
 163unsigned int bdev_count_inflight(struct block_device *part)
 164{
 165	unsigned int inflight[2] = {0};
 166
 167	bdev_count_inflight_rw(part, inflight, false);
 168
 169	return inflight[READ] + inflight[WRITE];
 170}
 171EXPORT_SYMBOL_GPL(bdev_count_inflight);
 172
 173/*
 174 * Can be deleted altogether. Later.
 175 *
 176 */
 177#define BLKDEV_MAJOR_HASH_SIZE 255
 178static struct blk_major_name {
 179	struct blk_major_name *next;
 180	int major;
 181	char name[16];
 182#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
 183	void (*probe)(dev_t devt);
 184#endif
 185} *major_names[BLKDEV_MAJOR_HASH_SIZE];
 186static DEFINE_MUTEX(major_names_lock);
 187static DEFINE_SPINLOCK(major_names_spinlock);
 188
 189/* index in the above - for now: assume no multimajor ranges */
 190static inline int major_to_index(unsigned major)
 191{
 192	return major % BLKDEV_MAJOR_HASH_SIZE;
 193}
 194
 195#ifdef CONFIG_PROC_FS
 196void blkdev_show(struct seq_file *seqf, off_t offset)
 197{
 198	struct blk_major_name *dp;
 199
 200	spin_lock(&major_names_spinlock);
 201	for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
 202		if (dp->major == offset)
 203			seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
 204	spin_unlock(&major_names_spinlock);
 205}
 206#endif /* CONFIG_PROC_FS */
 207
 208/**
 209 * __register_blkdev - register a new block device
 210 *
 211 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
 212 *         @major = 0, try to allocate any unused major number.
 213 * @name: the name of the new block device as a zero terminated string
 214 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
 215 *	   pre-created device node is accessed. When a probe call uses
 216 *	   add_disk() and it fails the driver must cleanup resources. This
 217 *	   interface may soon be removed.
 218 *
 219 * The @name must be unique within the system.
 220 *
 221 * The return value depends on the @major input parameter:
 222 *
 223 *  - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
 224 *    then the function returns zero on success, or a negative error code
 225 *  - if any unused major number was requested with @major = 0 parameter
 226 *    then the return value is the allocated major number in range
 227 *    [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
 228 *
 229 * See Documentation/admin-guide/devices.txt for the list of allocated
 230 * major numbers.
 231 *
 232 * Use register_blkdev instead for any new code.
 233 */
 234int __register_blkdev(unsigned int major, const char *name,
 235		void (*probe)(dev_t devt))
 236{
 237	struct blk_major_name **n, *p;
 238	int index, ret = 0;
 239
 240	mutex_lock(&major_names_lock);
 241
 242	/* temporary */
 243	if (major == 0) {
 244		for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
 245			if (major_names[index] == NULL)
 246				break;
 247		}
 248
 249		if (index == 0) {
 250			printk("%s: failed to get major for %s\n",
 251			       __func__, name);
 252			ret = -EBUSY;
 253			goto out;
 254		}
 255		major = index;
 256		ret = major;
 257	}
 258
 259	if (major >= BLKDEV_MAJOR_MAX) {
 260		pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
 261		       __func__, major, BLKDEV_MAJOR_MAX-1, name);
 262
 263		ret = -EINVAL;
 264		goto out;
 265	}
 266
 267	p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
 268	if (p == NULL) {
 269		ret = -ENOMEM;
 270		goto out;
 271	}
 272
 273	p->major = major;
 274#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
 275	p->probe = probe;
 276#endif
 277	strscpy(p->name, name, sizeof(p->name));
 278	p->next = NULL;
 279	index = major_to_index(major);
 280
 281	spin_lock(&major_names_spinlock);
 282	for (n = &major_names[index]; *n; n = &(*n)->next) {
 283		if ((*n)->major == major)
 284			break;
 285	}
 286	if (!*n)
 287		*n = p;
 288	else
 289		ret = -EBUSY;
 290	spin_unlock(&major_names_spinlock);
 291
 292	if (ret < 0) {
 293		printk("register_blkdev: cannot get major %u for %s\n",
 294		       major, name);
 295		kfree(p);
 296	}
 297out:
 298	mutex_unlock(&major_names_lock);
 299	return ret;
 300}
 301EXPORT_SYMBOL(__register_blkdev);
 302
 303void unregister_blkdev(unsigned int major, const char *name)
 304{
 305	struct blk_major_name **n;
 306	struct blk_major_name *p = NULL;
 307	int index = major_to_index(major);
 308
 309	mutex_lock(&major_names_lock);
 310	spin_lock(&major_names_spinlock);
 311	for (n = &major_names[index]; *n; n = &(*n)->next)
 312		if ((*n)->major == major)
 313			break;
 314	if (!*n || strcmp((*n)->name, name)) {
 315		WARN_ON(1);
 316	} else {
 317		p = *n;
 318		*n = p->next;
 319	}
 320	spin_unlock(&major_names_spinlock);
 321	mutex_unlock(&major_names_lock);
 322	kfree(p);
 323}
 324
 325EXPORT_SYMBOL(unregister_blkdev);
 326
 327int blk_alloc_ext_minor(void)
 328{
 329	int idx;
 330
 331	idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
 332	if (idx == -ENOSPC)
 333		return -EBUSY;
 334	return idx;
 335}
 336
 337void blk_free_ext_minor(unsigned int minor)
 338{
 339	ida_free(&ext_devt_ida, minor);
 340}
 341
 342void disk_uevent(struct gendisk *disk, enum kobject_action action)
 343{
 344	struct block_device *part;
 345	unsigned long idx;
 346
 347	rcu_read_lock();
 348	xa_for_each(&disk->part_tbl, idx, part) {
 349		if (bdev_is_partition(part) && !bdev_nr_sectors(part))
 350			continue;
 351		if (!kobject_get_unless_zero(&part->bd_device.kobj))
 352			continue;
 353
 354		rcu_read_unlock();
 355		kobject_uevent(bdev_kobj(part), action);
 356		put_device(&part->bd_device);
 357		rcu_read_lock();
 358	}
 359	rcu_read_unlock();
 360}
 361EXPORT_SYMBOL_GPL(disk_uevent);
 362
 363int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode)
 364{
 365	struct file *file;
 366	int ret = 0;
 367
 368	if (!disk_has_partscan(disk))
 369		return -EINVAL;
 370	if (disk->open_partitions)
 371		return -EBUSY;
 372
 373	/*
 374	 * If the device is opened exclusively by current thread already, it's
 375	 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
 376	 * synchronize with other exclusive openers and other partition
 377	 * scanners.
 378	 */
 379	if (!(mode & BLK_OPEN_EXCL)) {
 380		ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions,
 381					  NULL);
 382		if (ret)
 383			return ret;
 384	}
 385
 386	set_bit(GD_NEED_PART_SCAN, &disk->state);
 387	file = bdev_file_open_by_dev(disk_devt(disk), mode & ~BLK_OPEN_EXCL,
 388				     NULL, NULL);
 389	if (IS_ERR(file))
 390		ret = PTR_ERR(file);
 391	else
 392		fput(file);
 393
 394	/*
 395	 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
 396	 * and this will cause that re-assemble partitioned raid device will
 397	 * creat partition for underlying disk.
 398	 */
 399	clear_bit(GD_NEED_PART_SCAN, &disk->state);
 400	if (!(mode & BLK_OPEN_EXCL))
 401		bd_abort_claiming(disk->part0, disk_scan_partitions);
 402	return ret;
 403}
 404
 405static void add_disk_final(struct gendisk *disk)
 406{
 407	struct device *ddev = disk_to_dev(disk);
 408
 409	if (!(disk->flags & GENHD_FL_HIDDEN)) {
 410		/* Make sure the first partition scan will be proceed */
 411		if (get_capacity(disk) && disk_has_partscan(disk))
 412			set_bit(GD_NEED_PART_SCAN, &disk->state);
 413
 414		bdev_add(disk->part0, ddev->devt);
 415		if (get_capacity(disk))
 416			disk_scan_partitions(disk, BLK_OPEN_READ);
 417
 418		/*
 419		 * Announce the disk and partitions after all partitions are
 420		 * created. (for hidden disks uevents remain suppressed forever)
 421		 */
 422		dev_set_uevent_suppress(ddev, 0);
 423		disk_uevent(disk, KOBJ_ADD);
 424	}
 425
 426	blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);
 427	disk_add_events(disk);
 428	set_bit(GD_ADDED, &disk->state);
 429}
 430
 431static int __add_disk(struct device *parent, struct gendisk *disk,
 432		      const struct attribute_group **groups,
 433		      struct fwnode_handle *fwnode)
 434
 435{
 436	struct device *ddev = disk_to_dev(disk);
 437	int ret;
 438
 439	if (WARN_ON_ONCE(bdev_nr_sectors(disk->part0) > BLK_DEV_MAX_SECTORS))
 440		return -EINVAL;
 441
 442	if (queue_is_mq(disk->queue)) {
 443		/*
 444		 * ->submit_bio and ->poll_bio are bypassed for blk-mq drivers.
 445		 */
 446		if (disk->fops->submit_bio || disk->fops->poll_bio)
 447			return -EINVAL;
 448	} else {
 449		if (!disk->fops->submit_bio)
 450			return -EINVAL;
 451		bdev_set_flag(disk->part0, BD_HAS_SUBMIT_BIO);
 452	}
 453
 454	/*
 455	 * If the driver provides an explicit major number it also must provide
 456	 * the number of minors numbers supported, and those will be used to
 457	 * setup the gendisk.
 458	 * Otherwise just allocate the device numbers for both the whole device
 459	 * and all partitions from the extended dev_t space.
 460	 */
 461	ret = -EINVAL;
 462	if (disk->major) {
 463		if (WARN_ON(!disk->minors))
 464			goto out;
 465
 466		if (disk->minors > DISK_MAX_PARTS) {
 467			pr_err("block: can't allocate more than %d partitions\n",
 468				DISK_MAX_PARTS);
 469			disk->minors = DISK_MAX_PARTS;
 470		}
 471		if (disk->first_minor > MINORMASK ||
 472		    disk->minors > MINORMASK + 1 ||
 473		    disk->first_minor + disk->minors > MINORMASK + 1)
 474			goto out;
 475	} else {
 476		if (WARN_ON(disk->minors))
 477			goto out;
 478
 479		ret = blk_alloc_ext_minor();
 480		if (ret < 0)
 481			goto out;
 482		disk->major = BLOCK_EXT_MAJOR;
 483		disk->first_minor = ret;
 484	}
 485
 486	/* delay uevents, until we scanned partition table */
 487	dev_set_uevent_suppress(ddev, 1);
 488
 489	ddev->parent = parent;
 490	ddev->groups = groups;
 491	dev_set_name(ddev, "%s", disk->disk_name);
 492	if (fwnode)
 493		device_set_node(ddev, fwnode);
 494	if (!(disk->flags & GENHD_FL_HIDDEN))
 495		ddev->devt = MKDEV(disk->major, disk->first_minor);
 496	ret = device_add(ddev);
 497	if (ret)
 498		goto out_free_ext_minor;
 499
 500	ret = disk_alloc_events(disk);
 501	if (ret)
 502		goto out_device_del;
 503
 504	ret = sysfs_create_link(block_depr, &ddev->kobj,
 505				kobject_name(&ddev->kobj));
 506	if (ret)
 507		goto out_device_del;
 508
 509	/*
 510	 * avoid probable deadlock caused by allocating memory with
 511	 * GFP_KERNEL in runtime_resume callback of its all ancestor
 512	 * devices
 513	 */
 514	pm_runtime_set_memalloc_noio(ddev, true);
 515
 516	disk->part0->bd_holder_dir =
 517		kobject_create_and_add("holders", &ddev->kobj);
 518	if (!disk->part0->bd_holder_dir) {
 519		ret = -ENOMEM;
 520		goto out_del_block_link;
 521	}
 522	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
 523	if (!disk->slave_dir) {
 524		ret = -ENOMEM;
 525		goto out_put_holder_dir;
 526	}
 527
 528	ret = blk_register_queue(disk);
 529	if (ret)
 530		goto out_put_slave_dir;
 531
 532	if (!(disk->flags & GENHD_FL_HIDDEN)) {
 533		ret = bdi_register(disk->bdi, "%u:%u",
 534				   disk->major, disk->first_minor);
 535		if (ret)
 536			goto out_unregister_queue;
 537		bdi_set_owner(disk->bdi, ddev);
 538		ret = sysfs_create_link(&ddev->kobj,
 539					&disk->bdi->dev->kobj, "bdi");
 540		if (ret)
 541			goto out_unregister_bdi;
 542	} else {
 543		/*
 544		 * Even if the block_device for a hidden gendisk is not
 545		 * registered, it needs to have a valid bd_dev so that the
 546		 * freeing of the dynamic major works.
 547		 */
 548		disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
 549	}
 550	return 0;
 551
 552out_unregister_bdi:
 553	if (!(disk->flags & GENHD_FL_HIDDEN))
 554		bdi_unregister(disk->bdi);
 555out_unregister_queue:
 556	blk_unregister_queue(disk);
 557	rq_qos_exit(disk->queue);
 558out_put_slave_dir:
 559	kobject_put(disk->slave_dir);
 560	disk->slave_dir = NULL;
 561out_put_holder_dir:
 562	kobject_put(disk->part0->bd_holder_dir);
 563out_del_block_link:
 564	sysfs_remove_link(block_depr, dev_name(ddev));
 565	pm_runtime_set_memalloc_noio(ddev, false);
 566out_device_del:
 567	device_del(ddev);
 568out_free_ext_minor:
 569	if (disk->major == BLOCK_EXT_MAJOR)
 570		blk_free_ext_minor(disk->first_minor);
 571out:
 572	return ret;
 573}
 574
 575/**
 576 * add_disk_fwnode - add disk information to kernel list with fwnode
 577 * @parent: parent device for the disk
 578 * @disk: per-device partitioning information
 579 * @groups: Additional per-device sysfs groups
 580 * @fwnode: attached disk fwnode
 581 *
 582 * This function registers the partitioning information in @disk
 583 * with the kernel. Also attach a fwnode to the disk device.
 584 */
 585int __must_check add_disk_fwnode(struct device *parent, struct gendisk *disk,
 586				 const struct attribute_group **groups,
 587				 struct fwnode_handle *fwnode)
 588{
 589	struct blk_mq_tag_set *set;
 590	unsigned int memflags;
 591	int ret;
 592
 593	if (queue_is_mq(disk->queue)) {
 594		set = disk->queue->tag_set;
 595		memflags = memalloc_noio_save();
 596		down_read(&set->update_nr_hwq_lock);
 597		ret = __add_disk(parent, disk, groups, fwnode);
 598		up_read(&set->update_nr_hwq_lock);
 599		memalloc_noio_restore(memflags);
 600	} else {
 601		ret = __add_disk(parent, disk, groups, fwnode);
 602	}
 603
 604	/*
 605	 * add_disk_final() needn't to read `nr_hw_queues`, so move it out
 606	 * of read lock `set->update_nr_hwq_lock` for avoiding unnecessary
 607	 * lock dependency on `disk->open_mutex` from scanning partition.
 608	 */
 609	if (!ret)
 610		add_disk_final(disk);
 611	return ret;
 612}
 613EXPORT_SYMBOL_GPL(add_disk_fwnode);
 614
 615/**
 616 * device_add_disk - add disk information to kernel list
 617 * @parent: parent device for the disk
 618 * @disk: per-device partitioning information
 619 * @groups: Additional per-device sysfs groups
 620 *
 621 * This function registers the partitioning information in @disk
 622 * with the kernel.
 623 */
 624int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
 625				 const struct attribute_group **groups)
 626{
 627	return add_disk_fwnode(parent, disk, groups, NULL);
 628}
 629EXPORT_SYMBOL(device_add_disk);
 630
 631static void blk_report_disk_dead(struct gendisk *disk, bool surprise)
 632{
 633	struct block_device *bdev;
 634	unsigned long idx;
 635
 636	/*
 637	 * On surprise disk removal, bdev_mark_dead() may call into file
 638	 * systems below. Make it clear that we're expecting to not hold
 639	 * disk->open_mutex.
 640	 */
 641	lockdep_assert_not_held(&disk->open_mutex);
 642
 643	rcu_read_lock();
 644	xa_for_each(&disk->part_tbl, idx, bdev) {
 645		if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
 646			continue;
 647		rcu_read_unlock();
 648
 649		bdev_mark_dead(bdev, surprise);
 650
 651		put_device(&bdev->bd_device);
 652		rcu_read_lock();
 653	}
 654	rcu_read_unlock();
 655}
 656
 657static bool __blk_mark_disk_dead(struct gendisk *disk)
 658{
 659	/*
 660	 * Fail any new I/O.
 661	 */
 662	if (test_and_set_bit(GD_DEAD, &disk->state))
 663		return false;
 664
 665	if (test_bit(GD_OWNS_QUEUE, &disk->state))
 666		blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);
 667
 668	/*
 669	 * Stop buffered writers from dirtying pages that can't be written out.
 670	 */
 671	set_capacity(disk, 0);
 672
 673	/*
 674	 * Prevent new I/O from crossing bio_queue_enter().
 675	 */
 676	return blk_queue_start_drain(disk->queue);
 677}
 678
 679/**
 680 * blk_mark_disk_dead - mark a disk as dead
 681 * @disk: disk to mark as dead
 682 *
 683 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
 684 * to this disk.
 685 */
 686void blk_mark_disk_dead(struct gendisk *disk)
 687{
 688	__blk_mark_disk_dead(disk);
 689	blk_report_disk_dead(disk, true);
 690}
 691EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
 692
 693static void __del_gendisk(struct gendisk *disk)
 694{
 695	struct request_queue *q = disk->queue;
 696	struct block_device *part;
 697	unsigned long idx;
 698	bool start_drain;
 699
 700	might_sleep();
 701
 702	if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
 703		return;
 704
 705	disk_del_events(disk);
 706
 707	/*
 708	 * Prevent new openers by unlinked the bdev inode.
 709	 */
 710	mutex_lock(&disk->open_mutex);
 711	xa_for_each(&disk->part_tbl, idx, part)
 712		bdev_unhash(part);
 713	mutex_unlock(&disk->open_mutex);
 714
 715	/*
 716	 * Tell the file system to write back all dirty data and shut down if
 717	 * it hasn't been notified earlier.
 718	 */
 719	if (!test_bit(GD_DEAD, &disk->state))
 720		blk_report_disk_dead(disk, false);
 721
 722	/*
 723	 * Drop all partitions now that the disk is marked dead.
 724	 */
 725	mutex_lock(&disk->open_mutex);
 726	start_drain = __blk_mark_disk_dead(disk);
 727	if (start_drain)
 728		blk_freeze_acquire_lock(q);
 729	xa_for_each_start(&disk->part_tbl, idx, part, 1)
 730		drop_partition(part);
 731	mutex_unlock(&disk->open_mutex);
 732
 733	if (!(disk->flags & GENHD_FL_HIDDEN)) {
 734		sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
 735
 736		/*
 737		 * Unregister bdi before releasing device numbers (as they can
 738		 * get reused and we'd get clashes in sysfs).
 739		 */
 740		bdi_unregister(disk->bdi);
 741	}
 742
 743	blk_unregister_queue(disk);
 744
 745	kobject_put(disk->part0->bd_holder_dir);
 746	kobject_put(disk->slave_dir);
 747	disk->slave_dir = NULL;
 748
 749	part_stat_set_all(disk->part0, 0);
 750	disk->part0->bd_stamp = 0;
 751	sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
 752	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
 753	device_del(disk_to_dev(disk));
 754
 755	blk_mq_freeze_queue_wait(q);
 756
 757	blk_throtl_cancel_bios(disk);
 758
 759	blk_sync_queue(q);
 760	blk_flush_integrity();
 761
 762	if (queue_is_mq(q))
 763		blk_mq_cancel_work_sync(q);
 764
 765	rq_qos_exit(q);
 766
 767	/*
 768	 * If the disk does not own the queue, allow using passthrough requests
 769	 * again.  Else leave the queue frozen to fail all I/O.
 770	 */
 771	if (!test_bit(GD_OWNS_QUEUE, &disk->state))
 772		__blk_mq_unfreeze_queue(q, true);
 773	else if (queue_is_mq(q))
 774		blk_mq_exit_queue(q);
 775
 776	if (start_drain)
 777		blk_unfreeze_release_lock(q);
 778}
 779
 780static void disable_elv_switch(struct request_queue *q)
 781{
 782	struct blk_mq_tag_set *set = q->tag_set;
 783	WARN_ON_ONCE(!queue_is_mq(q));
 784
 785	down_write(&set->update_nr_hwq_lock);
 786	blk_queue_flag_set(QUEUE_FLAG_NO_ELV_SWITCH, q);
 787	up_write(&set->update_nr_hwq_lock);
 788}
 789
 790/**
 791 * del_gendisk - remove the gendisk
 792 * @disk: the struct gendisk to remove
 793 *
 794 * Removes the gendisk and all its associated resources. This deletes the
 795 * partitions associated with the gendisk, and unregisters the associated
 796 * request_queue.
 797 *
 798 * This is the counter to the respective device_add_disk() call.
 799 *
 800 * The final removal of the struct gendisk happens when its refcount reaches 0
 801 * with put_disk(), which should be called after del_gendisk(), if
 802 * device_add_disk() was used.
 803 *
 804 * Drivers exist which depend on the release of the gendisk to be synchronous,
 805 * it should not be deferred.
 806 *
 807 * Context: can sleep
 808 */
 809void del_gendisk(struct gendisk *disk)
 810{
 811	struct blk_mq_tag_set *set;
 812	unsigned int memflags;
 813
 814	if (!queue_is_mq(disk->queue)) {
 815		__del_gendisk(disk);
 816	} else {
 817		set = disk->queue->tag_set;
 818
 819		disable_elv_switch(disk->queue);
 820
 821		memflags = memalloc_noio_save();
 822		down_read(&set->update_nr_hwq_lock);
 823		__del_gendisk(disk);
 824		up_read(&set->update_nr_hwq_lock);
 825		memalloc_noio_restore(memflags);
 826	}
 827}
 828EXPORT_SYMBOL(del_gendisk);
 829
 830/**
 831 * invalidate_disk - invalidate the disk
 832 * @disk: the struct gendisk to invalidate
 833 *
 834 * A helper to invalidates the disk. It will clean the disk's associated
 835 * buffer/page caches and reset its internal states so that the disk
 836 * can be reused by the drivers.
 837 *
 838 * Context: can sleep
 839 */
 840void invalidate_disk(struct gendisk *disk)
 841{
 842	struct block_device *bdev = disk->part0;
 843
 844	invalidate_bdev(bdev);
 845	bdev->bd_mapping->wb_err = 0;
 846	set_capacity(disk, 0);
 847}
 848EXPORT_SYMBOL(invalidate_disk);
 849
 850/* sysfs access to bad-blocks list. */
 851static ssize_t disk_badblocks_show(struct device *dev,
 852					struct device_attribute *attr,
 853					char *page)
 854{
 855	struct gendisk *disk = dev_to_disk(dev);
 856
 857	if (!disk->bb)
 858		return sysfs_emit(page, "\n");
 859
 860	return badblocks_show(disk->bb, page, 0);
 861}
 862
 863static ssize_t disk_badblocks_store(struct device *dev,
 864					struct device_attribute *attr,
 865					const char *page, size_t len)
 866{
 867	struct gendisk *disk = dev_to_disk(dev);
 868
 869	if (!disk->bb)
 870		return -ENXIO;
 871
 872	return badblocks_store(disk->bb, page, len, 0);
 873}
 874
 875#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
 876static bool blk_probe_dev(dev_t devt)
 877{
 878	unsigned int major = MAJOR(devt);
 879	struct blk_major_name **n;
 880
 881	mutex_lock(&major_names_lock);
 882	for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
 883		if ((*n)->major == major && (*n)->probe) {
 884			(*n)->probe(devt);
 885			mutex_unlock(&major_names_lock);
 886			return true;
 887		}
 888	}
 889	mutex_unlock(&major_names_lock);
 890	return false;
 891}
 892
 893void blk_request_module(dev_t devt)
 894{
 895	int error;
 896
 897	if (blk_probe_dev(devt))
 898		return;
 899
 900	error = request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt));
 901	/* Make old-style 2.4 aliases work */
 902	if (error > 0)
 903		error = request_module("block-major-%d", MAJOR(devt));
 904	if (!error)
 905		blk_probe_dev(devt);
 906}
 907#endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
 908
 909#ifdef CONFIG_PROC_FS
 910/* iterator */
 911static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
 912{
 913	loff_t skip = *pos;
 914	struct class_dev_iter *iter;
 915	struct device *dev;
 916
 917	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
 918	if (!iter)
 919		return ERR_PTR(-ENOMEM);
 920
 921	seqf->private = iter;
 922	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
 923	do {
 924		dev = class_dev_iter_next(iter);
 925		if (!dev)
 926			return NULL;
 927	} while (skip--);
 928
 929	return dev_to_disk(dev);
 930}
 931
 932static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
 933{
 934	struct device *dev;
 935
 936	(*pos)++;
 937	dev = class_dev_iter_next(seqf->private);
 938	if (dev)
 939		return dev_to_disk(dev);
 940
 941	return NULL;
 942}
 943
 944static void disk_seqf_stop(struct seq_file *seqf, void *v)
 945{
 946	struct class_dev_iter *iter = seqf->private;
 947
 948	/* stop is called even after start failed :-( */
 949	if (iter) {
 950		class_dev_iter_exit(iter);
 951		kfree(iter);
 952		seqf->private = NULL;
 953	}
 954}
 955
 956static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
 957{
 958	void *p;
 959
 960	p = disk_seqf_start(seqf, pos);
 961	if (!IS_ERR_OR_NULL(p) && !*pos)
 962		seq_puts(seqf, "major minor  #blocks  name\n\n");
 963	return p;
 964}
 965
 966static int show_partition(struct seq_file *seqf, void *v)
 967{
 968	struct gendisk *sgp = v;
 969	struct block_device *part;
 970	unsigned long idx;
 971
 972	if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
 973		return 0;
 974
 975	rcu_read_lock();
 976	xa_for_each(&sgp->part_tbl, idx, part) {
 977		if (!bdev_nr_sectors(part))
 978			continue;
 979		seq_printf(seqf, "%4d  %7d %10llu %pg\n",
 980			   MAJOR(part->bd_dev), MINOR(part->bd_dev),
 981			   bdev_nr_sectors(part) >> 1, part);
 982	}
 983	rcu_read_unlock();
 984	return 0;
 985}
 986
 987static const struct seq_operations partitions_op = {
 988	.start	= show_partition_start,
 989	.next	= disk_seqf_next,
 990	.stop	= disk_seqf_stop,
 991	.show	= show_partition
 992};
 993#endif
 994
 995static int __init genhd_device_init(void)
 996{
 997	int error;
 998
 999	error = class_register(&block_class);
1000	if (unlikely(error))
1001		return error;
1002	blk_dev_init();
1003
1004	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1005
1006	/* create top-level block dir */
1007	block_depr = kobject_create_and_add("block", NULL);
1008	return 0;
1009}
1010
1011subsys_initcall(genhd_device_init);
1012
1013static ssize_t disk_range_show(struct device *dev,
1014			       struct device_attribute *attr, char *buf)
1015{
1016	struct gendisk *disk = dev_to_disk(dev);
1017
1018	return sysfs_emit(buf, "%d\n", disk->minors);
1019}
1020
1021static ssize_t disk_ext_range_show(struct device *dev,
1022				   struct device_attribute *attr, char *buf)
1023{
1024	struct gendisk *disk = dev_to_disk(dev);
1025
1026	return sysfs_emit(buf, "%d\n",
1027		(disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
1028}
1029
1030static ssize_t disk_removable_show(struct device *dev,
1031				   struct device_attribute *attr, char *buf)
1032{
1033	struct gendisk *disk = dev_to_disk(dev);
1034
1035	return sysfs_emit(buf, "%d\n",
1036		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1037}
1038
1039static ssize_t disk_hidden_show(struct device *dev,
1040				   struct device_attribute *attr, char *buf)
1041{
1042	struct gendisk *disk = dev_to_disk(dev);
1043
1044	return sysfs_emit(buf, "%d\n",
1045		       (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1046}
1047
1048static ssize_t disk_ro_show(struct device *dev,
1049				   struct device_attribute *attr, char *buf)
1050{
1051	struct gendisk *disk = dev_to_disk(dev);
1052
1053	return sysfs_emit(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1054}
1055
1056ssize_t part_size_show(struct device *dev,
1057		       struct device_attribute *attr, char *buf)
1058{
1059	return sysfs_emit(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
1060}
1061
1062ssize_t part_stat_show(struct device *dev,
1063		       struct device_attribute *attr, char *buf)
1064{
1065	struct block_device *bdev = dev_to_bdev(dev);
1066	struct disk_stats stat;
1067	unsigned int inflight;
1068
1069	inflight = bdev_count_inflight(bdev);
1070	if (inflight) {
1071		part_stat_lock();
1072		update_io_ticks(bdev, jiffies, true);
1073		part_stat_unlock();
1074	}
1075	part_stat_read_all(bdev, &stat);
1076	return sysfs_emit(buf,
1077		"%8lu %8lu %8llu %8u "
1078		"%8lu %8lu %8llu %8u "
1079		"%8u %8u %8u "
1080		"%8lu %8lu %8llu %8u "
1081		"%8lu %8u"
1082		"\n",
1083		stat.ios[STAT_READ],
1084		stat.merges[STAT_READ],
1085		(unsigned long long)stat.sectors[STAT_READ],
1086		(unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1087		stat.ios[STAT_WRITE],
1088		stat.merges[STAT_WRITE],
1089		(unsigned long long)stat.sectors[STAT_WRITE],
1090		(unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1091		inflight,
1092		jiffies_to_msecs(stat.io_ticks),
1093		(unsigned int)div_u64(stat.nsecs[STAT_READ] +
1094				      stat.nsecs[STAT_WRITE] +
1095				      stat.nsecs[STAT_DISCARD] +
1096				      stat.nsecs[STAT_FLUSH],
1097						NSEC_PER_MSEC),
1098		stat.ios[STAT_DISCARD],
1099		stat.merges[STAT_DISCARD],
1100		(unsigned long long)stat.sectors[STAT_DISCARD],
1101		(unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1102		stat.ios[STAT_FLUSH],
1103		(unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1104}
1105
1106/*
1107 * Show the number of IOs issued to driver.
1108 * For bio-based device, started from bdev_start_io_acct();
1109 * For rq-based device, started from blk_mq_start_request();
1110 */
1111ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1112			   char *buf)
1113{
1114	struct block_device *bdev = dev_to_bdev(dev);
1115	struct request_queue *q = bdev_get_queue(bdev);
1116	unsigned int inflight[2] = {0};
1117
1118	bdev_count_inflight_rw(bdev, inflight, queue_is_mq(q));
1119
1120	return sysfs_emit(buf, "%8u %8u\n", inflight[READ], inflight[WRITE]);
1121}
1122
1123static ssize_t disk_capability_show(struct device *dev,
1124				    struct device_attribute *attr, char *buf)
1125{
1126	dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1127	return sysfs_emit(buf, "0\n");
1128}
1129
1130static ssize_t disk_alignment_offset_show(struct device *dev,
1131					  struct device_attribute *attr,
1132					  char *buf)
1133{
1134	struct gendisk *disk = dev_to_disk(dev);
1135
1136	return sysfs_emit(buf, "%d\n", bdev_alignment_offset(disk->part0));
1137}
1138
1139static ssize_t disk_discard_alignment_show(struct device *dev,
1140					   struct device_attribute *attr,
1141					   char *buf)
1142{
1143	struct gendisk *disk = dev_to_disk(dev);
1144
1145	return sysfs_emit(buf, "%d\n", bdev_alignment_offset(disk->part0));
1146}
1147
1148static ssize_t diskseq_show(struct device *dev,
1149			    struct device_attribute *attr, char *buf)
1150{
1151	struct gendisk *disk = dev_to_disk(dev);
1152
1153	return sysfs_emit(buf, "%llu\n", disk->diskseq);
1154}
1155
1156static ssize_t partscan_show(struct device *dev,
1157		struct device_attribute *attr, char *buf)
1158{
1159	return sysfs_emit(buf, "%u\n", disk_has_partscan(dev_to_disk(dev)));
1160}
1161
1162static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1163static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1164static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1165static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1166static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1167static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1168static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1169static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1170static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1171static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1172static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1173static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1174static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1175static DEVICE_ATTR(partscan, 0444, partscan_show, NULL);
1176
1177#ifdef CONFIG_FAIL_MAKE_REQUEST
1178ssize_t part_fail_show(struct device *dev,
1179		       struct device_attribute *attr, char *buf)
1180{
1181	return sysfs_emit(buf, "%d\n",
1182		       bdev_test_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL));
1183}
1184
1185ssize_t part_fail_store(struct device *dev,
1186			struct device_attribute *attr,
1187			const char *buf, size_t count)
1188{
1189	int i;
1190
1191	if (count > 0 && sscanf(buf, "%d", &i) > 0) {
1192		if (i)
1193			bdev_set_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL);
1194		else
1195			bdev_clear_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL);
1196	}
1197	return count;
1198}
1199
1200static struct device_attribute dev_attr_fail =
1201	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1202#endif /* CONFIG_FAIL_MAKE_REQUEST */
1203
1204#ifdef CONFIG_FAIL_IO_TIMEOUT
1205static struct device_attribute dev_attr_fail_timeout =
1206	__ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1207#endif
1208
1209static struct attribute *disk_attrs[] = {
1210	&dev_attr_range.attr,
1211	&dev_attr_ext_range.attr,
1212	&dev_attr_removable.attr,
1213	&dev_attr_hidden.attr,
1214	&dev_attr_ro.attr,
1215	&dev_attr_size.attr,
1216	&dev_attr_alignment_offset.attr,
1217	&dev_attr_discard_alignment.attr,
1218	&dev_attr_capability.attr,
1219	&dev_attr_stat.attr,
1220	&dev_attr_inflight.attr,
1221	&dev_attr_badblocks.attr,
1222	&dev_attr_events.attr,
1223	&dev_attr_events_async.attr,
1224	&dev_attr_events_poll_msecs.attr,
1225	&dev_attr_diskseq.attr,
1226	&dev_attr_partscan.attr,
1227#ifdef CONFIG_FAIL_MAKE_REQUEST
1228	&dev_attr_fail.attr,
1229#endif
1230#ifdef CONFIG_FAIL_IO_TIMEOUT
1231	&dev_attr_fail_timeout.attr,
1232#endif
1233	NULL
1234};
1235
1236static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1237{
1238	struct device *dev = container_of(kobj, typeof(*dev), kobj);
1239	struct gendisk *disk = dev_to_disk(dev);
1240
1241	if (a == &dev_attr_badblocks.attr && !disk->bb)
1242		return 0;
1243	return a->mode;
1244}
1245
1246static struct attribute_group disk_attr_group = {
1247	.attrs = disk_attrs,
1248	.is_visible = disk_visible,
1249};
1250
1251static const struct attribute_group *disk_attr_groups[] = {
1252	&disk_attr_group,
1253#ifdef CONFIG_BLK_DEV_IO_TRACE
1254	&blk_trace_attr_group,
1255#endif
1256#ifdef CONFIG_BLK_DEV_INTEGRITY
1257	&blk_integrity_attr_group,
1258#endif
1259	NULL
1260};
1261
1262/**
1263 * disk_release - releases all allocated resources of the gendisk
1264 * @dev: the device representing this disk
1265 *
1266 * This function releases all allocated resources of the gendisk.
1267 *
1268 * Drivers which used device_add_disk() have a gendisk with a request_queue
1269 * assigned. Since the request_queue sits on top of the gendisk for these
1270 * drivers we also call blk_put_queue() for them, and we expect the
1271 * request_queue refcount to reach 0 at this point, and so the request_queue
1272 * will also be freed prior to the disk.
1273 *
1274 * Context: can sleep
1275 */
1276static void disk_release(struct device *dev)
1277{
1278	struct gendisk *disk = dev_to_disk(dev);
1279
1280	might_sleep();
1281	WARN_ON_ONCE(disk_live(disk));
1282
1283	blk_trace_remove(disk->queue);
1284
1285	/*
1286	 * To undo the all initialization from blk_mq_init_allocated_queue in
1287	 * case of a probe failure where add_disk is never called we have to
1288	 * call blk_mq_exit_queue here. We can't do this for the more common
1289	 * teardown case (yet) as the tagset can be gone by the time the disk
1290	 * is released once it was added.
1291	 */
1292	if (queue_is_mq(disk->queue) &&
1293	    test_bit(GD_OWNS_QUEUE, &disk->state) &&
1294	    !test_bit(GD_ADDED, &disk->state))
1295		blk_mq_exit_queue(disk->queue);
1296
1297	blkcg_exit_disk(disk);
1298
1299	bioset_exit(&disk->bio_split);
1300
1301	disk_release_events(disk);
1302	kfree(disk->random);
1303	disk_free_zone_resources(disk);
1304	xa_destroy(&disk->part_tbl);
1305
1306	kobject_put(&disk->queue_kobj);
1307	disk->queue->disk = NULL;
1308	blk_put_queue(disk->queue);
1309
1310	if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1311		disk->fops->free_disk(disk);
1312
1313	bdev_drop(disk->part0);	/* frees the disk */
1314}
1315
1316static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1317{
1318	const struct gendisk *disk = dev_to_disk(dev);
1319
1320	return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1321}
1322
1323const struct class block_class = {
1324	.name		= "block",
1325	.dev_uevent	= block_uevent,
1326};
1327
1328static char *block_devnode(const struct device *dev, umode_t *mode,
1329			   kuid_t *uid, kgid_t *gid)
1330{
1331	struct gendisk *disk = dev_to_disk(dev);
1332
1333	if (disk->fops->devnode)
1334		return disk->fops->devnode(disk, mode);
1335	return NULL;
1336}
1337
1338const struct device_type disk_type = {
1339	.name		= "disk",
1340	.groups		= disk_attr_groups,
1341	.release	= disk_release,
1342	.devnode	= block_devnode,
1343};
1344
1345#ifdef CONFIG_PROC_FS
1346/*
1347 * aggregate disk stat collector.  Uses the same stats that the sysfs
1348 * entries do, above, but makes them available through one seq_file.
1349 *
1350 * The output looks suspiciously like /proc/partitions with a bunch of
1351 * extra fields.
1352 */
1353static int diskstats_show(struct seq_file *seqf, void *v)
1354{
1355	struct gendisk *gp = v;
1356	struct block_device *hd;
1357	unsigned int inflight;
1358	struct disk_stats stat;
1359	unsigned long idx;
1360
1361	/*
1362	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1363		seq_puts(seqf,	"major minor name"
1364				"     rio rmerge rsect ruse wio wmerge "
1365				"wsect wuse running use aveq"
1366				"\n\n");
1367	*/
1368
1369	rcu_read_lock();
1370	xa_for_each(&gp->part_tbl, idx, hd) {
1371		if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1372			continue;
1373
1374		inflight = bdev_count_inflight(hd);
1375		if (inflight) {
1376			part_stat_lock();
1377			update_io_ticks(hd, jiffies, true);
1378			part_stat_unlock();
1379		}
1380		part_stat_read_all(hd, &stat);
1381		seq_put_decimal_ull_width(seqf, "",  MAJOR(hd->bd_dev), 4);
1382		seq_put_decimal_ull_width(seqf, " ", MINOR(hd->bd_dev), 7);
1383		seq_printf(seqf, " %pg", hd);
1384		seq_put_decimal_ull(seqf, " ", stat.ios[STAT_READ]);
1385		seq_put_decimal_ull(seqf, " ", stat.merges[STAT_READ]);
1386		seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_READ]);
1387		seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_READ],
1388								     NSEC_PER_MSEC));
1389		seq_put_decimal_ull(seqf, " ", stat.ios[STAT_WRITE]);
1390		seq_put_decimal_ull(seqf, " ", stat.merges[STAT_WRITE]);
1391		seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_WRITE]);
1392		seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1393								     NSEC_PER_MSEC));
1394		seq_put_decimal_ull(seqf, " ", inflight);
1395		seq_put_decimal_ull(seqf, " ", jiffies_to_msecs(stat.io_ticks));
1396		seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1397								     stat.nsecs[STAT_WRITE] +
1398								     stat.nsecs[STAT_DISCARD] +
1399								     stat.nsecs[STAT_FLUSH],
1400								     NSEC_PER_MSEC));
1401		seq_put_decimal_ull(seqf, " ", stat.ios[STAT_DISCARD]);
1402		seq_put_decimal_ull(seqf, " ", stat.merges[STAT_DISCARD]);
1403		seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_DISCARD]);
1404		seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1405								     NSEC_PER_MSEC));
1406		seq_put_decimal_ull(seqf, " ", stat.ios[STAT_FLUSH]);
1407		seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1408								     NSEC_PER_MSEC));
1409		seq_putc(seqf, '\n');
1410	}
1411	rcu_read_unlock();
1412
1413	return 0;
1414}
1415
1416static const struct seq_operations diskstats_op = {
1417	.start	= disk_seqf_start,
1418	.next	= disk_seqf_next,
1419	.stop	= disk_seqf_stop,
1420	.show	= diskstats_show
1421};
1422
1423static int __init proc_genhd_init(void)
1424{
1425	proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1426	proc_create_seq("partitions", 0, NULL, &partitions_op);
1427	return 0;
1428}
1429module_init(proc_genhd_init);
1430#endif /* CONFIG_PROC_FS */
1431
1432dev_t part_devt(struct gendisk *disk, u8 partno)
1433{
1434	struct block_device *part;
1435	dev_t devt = 0;
1436
1437	rcu_read_lock();
1438	part = xa_load(&disk->part_tbl, partno);
1439	if (part)
1440		devt = part->bd_dev;
1441	rcu_read_unlock();
1442
1443	return devt;
1444}
1445
1446struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1447		struct lock_class_key *lkclass)
1448{
1449	struct gendisk *disk;
1450
1451	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1452	if (!disk)
1453		return NULL;
1454
1455	if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1456		goto out_free_disk;
1457
1458	disk->bdi = bdi_alloc(node_id);
1459	if (!disk->bdi)
1460		goto out_free_bioset;
1461
1462	/* bdev_alloc() might need the queue, set before the first call */
1463	disk->queue = q;
1464
1465	disk->part0 = bdev_alloc(disk, 0);
1466	if (!disk->part0)
1467		goto out_free_bdi;
1468
1469	disk->node_id = node_id;
1470	mutex_init(&disk->open_mutex);
1471	xa_init(&disk->part_tbl);
1472	if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1473		goto out_destroy_part_tbl;
1474
1475	if (blkcg_init_disk(disk))
1476		goto out_erase_part0;
1477
1478	disk_init_zone_resources(disk);
1479	rand_initialize_disk(disk);
1480	disk_to_dev(disk)->class = &block_class;
1481	disk_to_dev(disk)->type = &disk_type;
1482	device_initialize(disk_to_dev(disk));
1483	inc_diskseq(disk);
1484	q->disk = disk;
1485	lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1486#ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1487	INIT_LIST_HEAD(&disk->slave_bdevs);
1488#endif
1489	mutex_init(&disk->rqos_state_mutex);
1490	kobject_init(&disk->queue_kobj, &blk_queue_ktype);
1491	return disk;
1492
1493out_erase_part0:
1494	xa_erase(&disk->part_tbl, 0);
1495out_destroy_part_tbl:
1496	xa_destroy(&disk->part_tbl);
1497	disk->part0->bd_disk = NULL;
1498	bdev_drop(disk->part0);
1499out_free_bdi:
1500	bdi_put(disk->bdi);
1501out_free_bioset:
1502	bioset_exit(&disk->bio_split);
1503out_free_disk:
1504	kfree(disk);
1505	return NULL;
1506}
1507
1508struct gendisk *__blk_alloc_disk(struct queue_limits *lim, int node,
1509		struct lock_class_key *lkclass)
1510{
1511	struct queue_limits default_lim = { };
1512	struct request_queue *q;
1513	struct gendisk *disk;
1514
1515	q = blk_alloc_queue(lim ? lim : &default_lim, node);
1516	if (IS_ERR(q))
1517		return ERR_CAST(q);
1518
1519	disk = __alloc_disk_node(q, node, lkclass);
1520	if (!disk) {
1521		blk_put_queue(q);
1522		return ERR_PTR(-ENOMEM);
1523	}
1524	set_bit(GD_OWNS_QUEUE, &disk->state);
1525	return disk;
1526}
1527EXPORT_SYMBOL(__blk_alloc_disk);
1528
1529/**
1530 * put_disk - decrements the gendisk refcount
1531 * @disk: the struct gendisk to decrement the refcount for
1532 *
1533 * This decrements the refcount for the struct gendisk. When this reaches 0
1534 * we'll have disk_release() called.
1535 *
1536 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1537 * when handling probe errors (that is before add_disk() is called).
1538 *
1539 * Context: Any context, but the last reference must not be dropped from
1540 *          atomic context.
1541 */
1542void put_disk(struct gendisk *disk)
1543{
1544	if (disk)
1545		put_device(disk_to_dev(disk));
1546}
1547EXPORT_SYMBOL(put_disk);
1548
1549static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1550{
1551	char event[] = "DISK_RO=1";
1552	char *envp[] = { event, NULL };
1553
1554	if (!ro)
1555		event[8] = '0';
1556	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1557}
1558
1559/**
1560 * set_disk_ro - set a gendisk read-only
1561 * @disk:	gendisk to operate on
1562 * @read_only:	%true to set the disk read-only, %false set the disk read/write
1563 *
1564 * This function is used to indicate whether a given disk device should have its
1565 * read-only flag set. set_disk_ro() is typically used by device drivers to
1566 * indicate whether the underlying physical device is write-protected.
1567 */
1568void set_disk_ro(struct gendisk *disk, bool read_only)
1569{
1570	if (read_only) {
1571		if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1572			return;
1573	} else {
1574		if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1575			return;
1576	}
1577	set_disk_ro_uevent(disk, read_only);
1578}
1579EXPORT_SYMBOL(set_disk_ro);
1580
1581void inc_diskseq(struct gendisk *disk)
1582{
1583	disk->diskseq = atomic64_inc_return(&diskseq);
1584}