brintos

brintos / linux-shallow public Read only

0
0
Text · 37.2 KiB · 1c05dd4 Raw
1474 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 *  gendisk handling4 *5 * Portions Copyright (C) 2020 Christoph Hellwig6 */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 block40 * 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 has43 * a very high latency.44 * Block devices do not have exclusive owners in userspace, any process can set45 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop046 * can be reused again and again).47 * A userspace process setting up a block device and watching for its events48 * cannot thus reliably tell whether an event relates to the device it just set49 * up or another earlier instance with the same name.50 * This sequential number allows userspace processes to solve this problem, and51 * 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	bdev_set_nr_sectors(disk->part0, sectors);62}63EXPORT_SYMBOL(set_capacity);64 65/*66 * Set disk capacity and notify if the size is not currently zero and will not67 * be set to zero.  Returns true if a uevent was sent, otherwise false.68 */69bool set_capacity_and_notify(struct gendisk *disk, sector_t size)70{71	sector_t capacity = get_capacity(disk);72	char *envp[] = { "RESIZE=1", NULL };73 74	set_capacity(disk, size);75 76	/*77	 * Only print a message and send a uevent if the gendisk is user visible78	 * and alive.  This avoids spamming the log and udev when setting the79	 * initial capacity during probing.80	 */81	if (size == capacity ||82	    !disk_live(disk) ||83	    (disk->flags & GENHD_FL_HIDDEN))84		return false;85 86	pr_info("%s: detected capacity change from %lld to %lld\n",87		disk->disk_name, capacity, size);88 89	/*90	 * Historically we did not send a uevent for changes to/from an empty91	 * device.92	 */93	if (!capacity || !size)94		return false;95	kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);96	return true;97}98EXPORT_SYMBOL_GPL(set_capacity_and_notify);99 100static void part_stat_read_all(struct block_device *part,101		struct disk_stats *stat)102{103	int cpu;104 105	memset(stat, 0, sizeof(struct disk_stats));106	for_each_possible_cpu(cpu) {107		struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);108		int group;109 110		for (group = 0; group < NR_STAT_GROUPS; group++) {111			stat->nsecs[group] += ptr->nsecs[group];112			stat->sectors[group] += ptr->sectors[group];113			stat->ios[group] += ptr->ios[group];114			stat->merges[group] += ptr->merges[group];115		}116 117		stat->io_ticks += ptr->io_ticks;118	}119}120 121unsigned int part_in_flight(struct block_device *part)122{123	unsigned int inflight = 0;124	int cpu;125 126	for_each_possible_cpu(cpu) {127		inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +128			    part_stat_local_read_cpu(part, in_flight[1], cpu);129	}130	if ((int)inflight < 0)131		inflight = 0;132 133	return inflight;134}135 136static void part_in_flight_rw(struct block_device *part,137		unsigned int inflight[2])138{139	int cpu;140 141	inflight[0] = 0;142	inflight[1] = 0;143	for_each_possible_cpu(cpu) {144		inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);145		inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);146	}147	if ((int)inflight[0] < 0)148		inflight[0] = 0;149	if ((int)inflight[1] < 0)150		inflight[1] = 0;151}152 153/*154 * Can be deleted altogether. Later.155 *156 */157#define BLKDEV_MAJOR_HASH_SIZE 255158static struct blk_major_name {159	struct blk_major_name *next;160	int major;161	char name[16];162#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD163	void (*probe)(dev_t devt);164#endif165} *major_names[BLKDEV_MAJOR_HASH_SIZE];166static DEFINE_MUTEX(major_names_lock);167static DEFINE_SPINLOCK(major_names_spinlock);168 169/* index in the above - for now: assume no multimajor ranges */170static inline int major_to_index(unsigned major)171{172	return major % BLKDEV_MAJOR_HASH_SIZE;173}174 175#ifdef CONFIG_PROC_FS176void blkdev_show(struct seq_file *seqf, off_t offset)177{178	struct blk_major_name *dp;179 180	spin_lock(&major_names_spinlock);181	for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)182		if (dp->major == offset)183			seq_printf(seqf, "%3d %s\n", dp->major, dp->name);184	spin_unlock(&major_names_spinlock);185}186#endif /* CONFIG_PROC_FS */187 188/**189 * __register_blkdev - register a new block device190 *191 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If192 *         @major = 0, try to allocate any unused major number.193 * @name: the name of the new block device as a zero terminated string194 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their195 *	   pre-created device node is accessed. When a probe call uses196 *	   add_disk() and it fails the driver must cleanup resources. This197 *	   interface may soon be removed.198 *199 * The @name must be unique within the system.200 *201 * The return value depends on the @major input parameter:202 *203 *  - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]204 *    then the function returns zero on success, or a negative error code205 *  - if any unused major number was requested with @major = 0 parameter206 *    then the return value is the allocated major number in range207 *    [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise208 *209 * See Documentation/admin-guide/devices.txt for the list of allocated210 * major numbers.211 *212 * Use register_blkdev instead for any new code.213 */214int __register_blkdev(unsigned int major, const char *name,215		void (*probe)(dev_t devt))216{217	struct blk_major_name **n, *p;218	int index, ret = 0;219 220	mutex_lock(&major_names_lock);221 222	/* temporary */223	if (major == 0) {224		for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {225			if (major_names[index] == NULL)226				break;227		}228 229		if (index == 0) {230			printk("%s: failed to get major for %s\n",231			       __func__, name);232			ret = -EBUSY;233			goto out;234		}235		major = index;236		ret = major;237	}238 239	if (major >= BLKDEV_MAJOR_MAX) {240		pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",241		       __func__, major, BLKDEV_MAJOR_MAX-1, name);242 243		ret = -EINVAL;244		goto out;245	}246 247	p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);248	if (p == NULL) {249		ret = -ENOMEM;250		goto out;251	}252 253	p->major = major;254#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD255	p->probe = probe;256#endif257	strscpy(p->name, name, sizeof(p->name));258	p->next = NULL;259	index = major_to_index(major);260 261	spin_lock(&major_names_spinlock);262	for (n = &major_names[index]; *n; n = &(*n)->next) {263		if ((*n)->major == major)264			break;265	}266	if (!*n)267		*n = p;268	else269		ret = -EBUSY;270	spin_unlock(&major_names_spinlock);271 272	if (ret < 0) {273		printk("register_blkdev: cannot get major %u for %s\n",274		       major, name);275		kfree(p);276	}277out:278	mutex_unlock(&major_names_lock);279	return ret;280}281EXPORT_SYMBOL(__register_blkdev);282 283void unregister_blkdev(unsigned int major, const char *name)284{285	struct blk_major_name **n;286	struct blk_major_name *p = NULL;287	int index = major_to_index(major);288 289	mutex_lock(&major_names_lock);290	spin_lock(&major_names_spinlock);291	for (n = &major_names[index]; *n; n = &(*n)->next)292		if ((*n)->major == major)293			break;294	if (!*n || strcmp((*n)->name, name)) {295		WARN_ON(1);296	} else {297		p = *n;298		*n = p->next;299	}300	spin_unlock(&major_names_spinlock);301	mutex_unlock(&major_names_lock);302	kfree(p);303}304 305EXPORT_SYMBOL(unregister_blkdev);306 307int blk_alloc_ext_minor(void)308{309	int idx;310 311	idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);312	if (idx == -ENOSPC)313		return -EBUSY;314	return idx;315}316 317void blk_free_ext_minor(unsigned int minor)318{319	ida_free(&ext_devt_ida, minor);320}321 322void disk_uevent(struct gendisk *disk, enum kobject_action action)323{324	struct block_device *part;325	unsigned long idx;326 327	rcu_read_lock();328	xa_for_each(&disk->part_tbl, idx, part) {329		if (bdev_is_partition(part) && !bdev_nr_sectors(part))330			continue;331		if (!kobject_get_unless_zero(&part->bd_device.kobj))332			continue;333 334		rcu_read_unlock();335		kobject_uevent(bdev_kobj(part), action);336		put_device(&part->bd_device);337		rcu_read_lock();338	}339	rcu_read_unlock();340}341EXPORT_SYMBOL_GPL(disk_uevent);342 343int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode)344{345	struct file *file;346	int ret = 0;347 348	if (!disk_has_partscan(disk))349		return -EINVAL;350	if (disk->open_partitions)351		return -EBUSY;352 353	/*354	 * If the device is opened exclusively by current thread already, it's355	 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to356	 * synchronize with other exclusive openers and other partition357	 * scanners.358	 */359	if (!(mode & BLK_OPEN_EXCL)) {360		ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions,361					  NULL);362		if (ret)363			return ret;364	}365 366	set_bit(GD_NEED_PART_SCAN, &disk->state);367	file = bdev_file_open_by_dev(disk_devt(disk), mode & ~BLK_OPEN_EXCL,368				     NULL, NULL);369	if (IS_ERR(file))370		ret = PTR_ERR(file);371	else372		fput(file);373 374	/*375	 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,376	 * and this will cause that re-assemble partitioned raid device will377	 * creat partition for underlying disk.378	 */379	clear_bit(GD_NEED_PART_SCAN, &disk->state);380	if (!(mode & BLK_OPEN_EXCL))381		bd_abort_claiming(disk->part0, disk_scan_partitions);382	return ret;383}384 385/**386 * device_add_disk - add disk information to kernel list387 * @parent: parent device for the disk388 * @disk: per-device partitioning information389 * @groups: Additional per-device sysfs groups390 *391 * This function registers the partitioning information in @disk392 * with the kernel.393 */394int __must_check device_add_disk(struct device *parent, struct gendisk *disk,395				 const struct attribute_group **groups)396 397{398	struct device *ddev = disk_to_dev(disk);399	int ret;400 401	/* Only makes sense for bio-based to set ->poll_bio */402	if (queue_is_mq(disk->queue) && disk->fops->poll_bio)403		return -EINVAL;404 405	/*406	 * The disk queue should now be all set with enough information about407	 * the device for the elevator code to pick an adequate default408	 * elevator if one is needed, that is, for devices requesting queue409	 * registration.410	 */411	elevator_init_mq(disk->queue);412 413	/* Mark bdev as having a submit_bio, if needed */414	if (disk->fops->submit_bio)415		bdev_set_flag(disk->part0, BD_HAS_SUBMIT_BIO);416 417	/*418	 * If the driver provides an explicit major number it also must provide419	 * the number of minors numbers supported, and those will be used to420	 * setup the gendisk.421	 * Otherwise just allocate the device numbers for both the whole device422	 * and all partitions from the extended dev_t space.423	 */424	ret = -EINVAL;425	if (disk->major) {426		if (WARN_ON(!disk->minors))427			goto out_exit_elevator;428 429		if (disk->minors > DISK_MAX_PARTS) {430			pr_err("block: can't allocate more than %d partitions\n",431				DISK_MAX_PARTS);432			disk->minors = DISK_MAX_PARTS;433		}434		if (disk->first_minor > MINORMASK ||435		    disk->minors > MINORMASK + 1 ||436		    disk->first_minor + disk->minors > MINORMASK + 1)437			goto out_exit_elevator;438	} else {439		if (WARN_ON(disk->minors))440			goto out_exit_elevator;441 442		ret = blk_alloc_ext_minor();443		if (ret < 0)444			goto out_exit_elevator;445		disk->major = BLOCK_EXT_MAJOR;446		disk->first_minor = ret;447	}448 449	/* delay uevents, until we scanned partition table */450	dev_set_uevent_suppress(ddev, 1);451 452	ddev->parent = parent;453	ddev->groups = groups;454	dev_set_name(ddev, "%s", disk->disk_name);455	if (!(disk->flags & GENHD_FL_HIDDEN))456		ddev->devt = MKDEV(disk->major, disk->first_minor);457	ret = device_add(ddev);458	if (ret)459		goto out_free_ext_minor;460 461	ret = disk_alloc_events(disk);462	if (ret)463		goto out_device_del;464 465	ret = sysfs_create_link(block_depr, &ddev->kobj,466				kobject_name(&ddev->kobj));467	if (ret)468		goto out_device_del;469 470	/*471	 * avoid probable deadlock caused by allocating memory with472	 * GFP_KERNEL in runtime_resume callback of its all ancestor473	 * devices474	 */475	pm_runtime_set_memalloc_noio(ddev, true);476 477	disk->part0->bd_holder_dir =478		kobject_create_and_add("holders", &ddev->kobj);479	if (!disk->part0->bd_holder_dir) {480		ret = -ENOMEM;481		goto out_del_block_link;482	}483	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);484	if (!disk->slave_dir) {485		ret = -ENOMEM;486		goto out_put_holder_dir;487	}488 489	ret = blk_register_queue(disk);490	if (ret)491		goto out_put_slave_dir;492 493	if (!(disk->flags & GENHD_FL_HIDDEN)) {494		ret = bdi_register(disk->bdi, "%u:%u",495				   disk->major, disk->first_minor);496		if (ret)497			goto out_unregister_queue;498		bdi_set_owner(disk->bdi, ddev);499		ret = sysfs_create_link(&ddev->kobj,500					&disk->bdi->dev->kobj, "bdi");501		if (ret)502			goto out_unregister_bdi;503 504		/* Make sure the first partition scan will be proceed */505		if (get_capacity(disk) && disk_has_partscan(disk))506			set_bit(GD_NEED_PART_SCAN, &disk->state);507 508		bdev_add(disk->part0, ddev->devt);509		if (get_capacity(disk))510			disk_scan_partitions(disk, BLK_OPEN_READ);511 512		/*513		 * Announce the disk and partitions after all partitions are514		 * created. (for hidden disks uevents remain suppressed forever)515		 */516		dev_set_uevent_suppress(ddev, 0);517		disk_uevent(disk, KOBJ_ADD);518	} else {519		/*520		 * Even if the block_device for a hidden gendisk is not521		 * registered, it needs to have a valid bd_dev so that the522		 * freeing of the dynamic major works.523		 */524		disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);525	}526 527	blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);528	disk_add_events(disk);529	set_bit(GD_ADDED, &disk->state);530	return 0;531 532out_unregister_bdi:533	if (!(disk->flags & GENHD_FL_HIDDEN))534		bdi_unregister(disk->bdi);535out_unregister_queue:536	blk_unregister_queue(disk);537	rq_qos_exit(disk->queue);538out_put_slave_dir:539	kobject_put(disk->slave_dir);540	disk->slave_dir = NULL;541out_put_holder_dir:542	kobject_put(disk->part0->bd_holder_dir);543out_del_block_link:544	sysfs_remove_link(block_depr, dev_name(ddev));545	pm_runtime_set_memalloc_noio(ddev, false);546out_device_del:547	device_del(ddev);548out_free_ext_minor:549	if (disk->major == BLOCK_EXT_MAJOR)550		blk_free_ext_minor(disk->first_minor);551out_exit_elevator:552	if (disk->queue->elevator)553		elevator_exit(disk->queue);554	return ret;555}556EXPORT_SYMBOL(device_add_disk);557 558static void blk_report_disk_dead(struct gendisk *disk, bool surprise)559{560	struct block_device *bdev;561	unsigned long idx;562 563	/*564	 * On surprise disk removal, bdev_mark_dead() may call into file565	 * systems below. Make it clear that we're expecting to not hold566	 * disk->open_mutex.567	 */568	lockdep_assert_not_held(&disk->open_mutex);569 570	rcu_read_lock();571	xa_for_each(&disk->part_tbl, idx, bdev) {572		if (!kobject_get_unless_zero(&bdev->bd_device.kobj))573			continue;574		rcu_read_unlock();575 576		bdev_mark_dead(bdev, surprise);577 578		put_device(&bdev->bd_device);579		rcu_read_lock();580	}581	rcu_read_unlock();582}583 584static void __blk_mark_disk_dead(struct gendisk *disk)585{586	/*587	 * Fail any new I/O.588	 */589	if (test_and_set_bit(GD_DEAD, &disk->state))590		return;591 592	if (test_bit(GD_OWNS_QUEUE, &disk->state))593		blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);594 595	/*596	 * Stop buffered writers from dirtying pages that can't be written out.597	 */598	set_capacity(disk, 0);599 600	/*601	 * Prevent new I/O from crossing bio_queue_enter().602	 */603	blk_queue_start_drain(disk->queue);604}605 606/**607 * blk_mark_disk_dead - mark a disk as dead608 * @disk: disk to mark as dead609 *610 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O611 * to this disk.612 */613void blk_mark_disk_dead(struct gendisk *disk)614{615	__blk_mark_disk_dead(disk);616	blk_report_disk_dead(disk, true);617}618EXPORT_SYMBOL_GPL(blk_mark_disk_dead);619 620/**621 * del_gendisk - remove the gendisk622 * @disk: the struct gendisk to remove623 *624 * Removes the gendisk and all its associated resources. This deletes the625 * partitions associated with the gendisk, and unregisters the associated626 * request_queue.627 *628 * This is the counter to the respective __device_add_disk() call.629 *630 * The final removal of the struct gendisk happens when its refcount reaches 0631 * with put_disk(), which should be called after del_gendisk(), if632 * __device_add_disk() was used.633 *634 * Drivers exist which depend on the release of the gendisk to be synchronous,635 * it should not be deferred.636 *637 * Context: can sleep638 */639void del_gendisk(struct gendisk *disk)640{641	struct request_queue *q = disk->queue;642	struct block_device *part;643	unsigned long idx;644 645	might_sleep();646 647	if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))648		return;649 650	disk_del_events(disk);651 652	/*653	 * Prevent new openers by unlinked the bdev inode.654	 */655	mutex_lock(&disk->open_mutex);656	xa_for_each(&disk->part_tbl, idx, part)657		bdev_unhash(part);658	mutex_unlock(&disk->open_mutex);659 660	/*661	 * Tell the file system to write back all dirty data and shut down if662	 * it hasn't been notified earlier.663	 */664	if (!test_bit(GD_DEAD, &disk->state))665		blk_report_disk_dead(disk, false);666 667	/*668	 * Drop all partitions now that the disk is marked dead.669	 */670	mutex_lock(&disk->open_mutex);671	__blk_mark_disk_dead(disk);672	xa_for_each_start(&disk->part_tbl, idx, part, 1)673		drop_partition(part);674	mutex_unlock(&disk->open_mutex);675 676	if (!(disk->flags & GENHD_FL_HIDDEN)) {677		sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");678 679		/*680		 * Unregister bdi before releasing device numbers (as they can681		 * get reused and we'd get clashes in sysfs).682		 */683		bdi_unregister(disk->bdi);684	}685 686	blk_unregister_queue(disk);687 688	kobject_put(disk->part0->bd_holder_dir);689	kobject_put(disk->slave_dir);690	disk->slave_dir = NULL;691 692	part_stat_set_all(disk->part0, 0);693	disk->part0->bd_stamp = 0;694	sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));695	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);696	device_del(disk_to_dev(disk));697 698	blk_mq_freeze_queue_wait(q);699 700	blk_throtl_cancel_bios(disk);701 702	blk_sync_queue(q);703	blk_flush_integrity();704 705	if (queue_is_mq(q))706		blk_mq_cancel_work_sync(q);707 708	blk_mq_quiesce_queue(q);709	if (q->elevator) {710		mutex_lock(&q->sysfs_lock);711		elevator_exit(q);712		mutex_unlock(&q->sysfs_lock);713	}714	rq_qos_exit(q);715	blk_mq_unquiesce_queue(q);716 717	/*718	 * If the disk does not own the queue, allow using passthrough requests719	 * again.  Else leave the queue frozen to fail all I/O.720	 */721	if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {722		blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);723		__blk_mq_unfreeze_queue(q, true);724	} else {725		if (queue_is_mq(q))726			blk_mq_exit_queue(q);727	}728}729EXPORT_SYMBOL(del_gendisk);730 731/**732 * invalidate_disk - invalidate the disk733 * @disk: the struct gendisk to invalidate734 *735 * A helper to invalidates the disk. It will clean the disk's associated736 * buffer/page caches and reset its internal states so that the disk737 * can be reused by the drivers.738 *739 * Context: can sleep740 */741void invalidate_disk(struct gendisk *disk)742{743	struct block_device *bdev = disk->part0;744 745	invalidate_bdev(bdev);746	bdev->bd_mapping->wb_err = 0;747	set_capacity(disk, 0);748}749EXPORT_SYMBOL(invalidate_disk);750 751/* sysfs access to bad-blocks list. */752static ssize_t disk_badblocks_show(struct device *dev,753					struct device_attribute *attr,754					char *page)755{756	struct gendisk *disk = dev_to_disk(dev);757 758	if (!disk->bb)759		return sprintf(page, "\n");760 761	return badblocks_show(disk->bb, page, 0);762}763 764static ssize_t disk_badblocks_store(struct device *dev,765					struct device_attribute *attr,766					const char *page, size_t len)767{768	struct gendisk *disk = dev_to_disk(dev);769 770	if (!disk->bb)771		return -ENXIO;772 773	return badblocks_store(disk->bb, page, len, 0);774}775 776#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD777void blk_request_module(dev_t devt)778{779	unsigned int major = MAJOR(devt);780	struct blk_major_name **n;781 782	mutex_lock(&major_names_lock);783	for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {784		if ((*n)->major == major && (*n)->probe) {785			(*n)->probe(devt);786			mutex_unlock(&major_names_lock);787			return;788		}789	}790	mutex_unlock(&major_names_lock);791 792	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)793		/* Make old-style 2.4 aliases work */794		request_module("block-major-%d", MAJOR(devt));795}796#endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */797 798#ifdef CONFIG_PROC_FS799/* iterator */800static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)801{802	loff_t skip = *pos;803	struct class_dev_iter *iter;804	struct device *dev;805 806	iter = kmalloc(sizeof(*iter), GFP_KERNEL);807	if (!iter)808		return ERR_PTR(-ENOMEM);809 810	seqf->private = iter;811	class_dev_iter_init(iter, &block_class, NULL, &disk_type);812	do {813		dev = class_dev_iter_next(iter);814		if (!dev)815			return NULL;816	} while (skip--);817 818	return dev_to_disk(dev);819}820 821static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)822{823	struct device *dev;824 825	(*pos)++;826	dev = class_dev_iter_next(seqf->private);827	if (dev)828		return dev_to_disk(dev);829 830	return NULL;831}832 833static void disk_seqf_stop(struct seq_file *seqf, void *v)834{835	struct class_dev_iter *iter = seqf->private;836 837	/* stop is called even after start failed :-( */838	if (iter) {839		class_dev_iter_exit(iter);840		kfree(iter);841		seqf->private = NULL;842	}843}844 845static void *show_partition_start(struct seq_file *seqf, loff_t *pos)846{847	void *p;848 849	p = disk_seqf_start(seqf, pos);850	if (!IS_ERR_OR_NULL(p) && !*pos)851		seq_puts(seqf, "major minor  #blocks  name\n\n");852	return p;853}854 855static int show_partition(struct seq_file *seqf, void *v)856{857	struct gendisk *sgp = v;858	struct block_device *part;859	unsigned long idx;860 861	if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))862		return 0;863 864	rcu_read_lock();865	xa_for_each(&sgp->part_tbl, idx, part) {866		if (!bdev_nr_sectors(part))867			continue;868		seq_printf(seqf, "%4d  %7d %10llu %pg\n",869			   MAJOR(part->bd_dev), MINOR(part->bd_dev),870			   bdev_nr_sectors(part) >> 1, part);871	}872	rcu_read_unlock();873	return 0;874}875 876static const struct seq_operations partitions_op = {877	.start	= show_partition_start,878	.next	= disk_seqf_next,879	.stop	= disk_seqf_stop,880	.show	= show_partition881};882#endif883 884static int __init genhd_device_init(void)885{886	int error;887 888	error = class_register(&block_class);889	if (unlikely(error))890		return error;891	blk_dev_init();892 893	register_blkdev(BLOCK_EXT_MAJOR, "blkext");894 895	/* create top-level block dir */896	block_depr = kobject_create_and_add("block", NULL);897	return 0;898}899 900subsys_initcall(genhd_device_init);901 902static ssize_t disk_range_show(struct device *dev,903			       struct device_attribute *attr, char *buf)904{905	struct gendisk *disk = dev_to_disk(dev);906 907	return sprintf(buf, "%d\n", disk->minors);908}909 910static ssize_t disk_ext_range_show(struct device *dev,911				   struct device_attribute *attr, char *buf)912{913	struct gendisk *disk = dev_to_disk(dev);914 915	return sprintf(buf, "%d\n",916		(disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);917}918 919static ssize_t disk_removable_show(struct device *dev,920				   struct device_attribute *attr, char *buf)921{922	struct gendisk *disk = dev_to_disk(dev);923 924	return sprintf(buf, "%d\n",925		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));926}927 928static ssize_t disk_hidden_show(struct device *dev,929				   struct device_attribute *attr, char *buf)930{931	struct gendisk *disk = dev_to_disk(dev);932 933	return sprintf(buf, "%d\n",934		       (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));935}936 937static ssize_t disk_ro_show(struct device *dev,938				   struct device_attribute *attr, char *buf)939{940	struct gendisk *disk = dev_to_disk(dev);941 942	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);943}944 945ssize_t part_size_show(struct device *dev,946		       struct device_attribute *attr, char *buf)947{948	return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));949}950 951ssize_t part_stat_show(struct device *dev,952		       struct device_attribute *attr, char *buf)953{954	struct block_device *bdev = dev_to_bdev(dev);955	struct disk_stats stat;956	unsigned int inflight;957 958	inflight = part_in_flight(bdev);959	if (inflight) {960		part_stat_lock();961		update_io_ticks(bdev, jiffies, true);962		part_stat_unlock();963	}964	part_stat_read_all(bdev, &stat);965	return sprintf(buf,966		"%8lu %8lu %8llu %8u "967		"%8lu %8lu %8llu %8u "968		"%8u %8u %8u "969		"%8lu %8lu %8llu %8u "970		"%8lu %8u"971		"\n",972		stat.ios[STAT_READ],973		stat.merges[STAT_READ],974		(unsigned long long)stat.sectors[STAT_READ],975		(unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),976		stat.ios[STAT_WRITE],977		stat.merges[STAT_WRITE],978		(unsigned long long)stat.sectors[STAT_WRITE],979		(unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),980		inflight,981		jiffies_to_msecs(stat.io_ticks),982		(unsigned int)div_u64(stat.nsecs[STAT_READ] +983				      stat.nsecs[STAT_WRITE] +984				      stat.nsecs[STAT_DISCARD] +985				      stat.nsecs[STAT_FLUSH],986						NSEC_PER_MSEC),987		stat.ios[STAT_DISCARD],988		stat.merges[STAT_DISCARD],989		(unsigned long long)stat.sectors[STAT_DISCARD],990		(unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),991		stat.ios[STAT_FLUSH],992		(unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));993}994 995ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,996			   char *buf)997{998	struct block_device *bdev = dev_to_bdev(dev);999	struct request_queue *q = bdev_get_queue(bdev);1000	unsigned int inflight[2];1001 1002	if (queue_is_mq(q))1003		blk_mq_in_flight_rw(q, bdev, inflight);1004	else1005		part_in_flight_rw(bdev, inflight);1006 1007	return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);1008}1009 1010static ssize_t disk_capability_show(struct device *dev,1011				    struct device_attribute *attr, char *buf)1012{1013	dev_warn_once(dev, "the capability attribute has been deprecated.\n");1014	return sprintf(buf, "0\n");1015}1016 1017static ssize_t disk_alignment_offset_show(struct device *dev,1018					  struct device_attribute *attr,1019					  char *buf)1020{1021	struct gendisk *disk = dev_to_disk(dev);1022 1023	return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));1024}1025 1026static ssize_t disk_discard_alignment_show(struct device *dev,1027					   struct device_attribute *attr,1028					   char *buf)1029{1030	struct gendisk *disk = dev_to_disk(dev);1031 1032	return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));1033}1034 1035static ssize_t diskseq_show(struct device *dev,1036			    struct device_attribute *attr, char *buf)1037{1038	struct gendisk *disk = dev_to_disk(dev);1039 1040	return sprintf(buf, "%llu\n", disk->diskseq);1041}1042 1043static ssize_t partscan_show(struct device *dev,1044		struct device_attribute *attr, char *buf)1045{1046	return sprintf(buf, "%u\n", disk_has_partscan(dev_to_disk(dev)));1047}1048 1049static DEVICE_ATTR(range, 0444, disk_range_show, NULL);1050static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);1051static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);1052static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);1053static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);1054static DEVICE_ATTR(size, 0444, part_size_show, NULL);1055static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);1056static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);1057static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);1058static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);1059static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);1060static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);1061static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);1062static DEVICE_ATTR(partscan, 0444, partscan_show, NULL);1063 1064#ifdef CONFIG_FAIL_MAKE_REQUEST1065ssize_t part_fail_show(struct device *dev,1066		       struct device_attribute *attr, char *buf)1067{1068	return sprintf(buf, "%d\n",1069		       bdev_test_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL));1070}1071 1072ssize_t part_fail_store(struct device *dev,1073			struct device_attribute *attr,1074			const char *buf, size_t count)1075{1076	int i;1077 1078	if (count > 0 && sscanf(buf, "%d", &i) > 0) {1079		if (i)1080			bdev_set_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL);1081		else1082			bdev_clear_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL);1083	}1084	return count;1085}1086 1087static struct device_attribute dev_attr_fail =1088	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);1089#endif /* CONFIG_FAIL_MAKE_REQUEST */1090 1091#ifdef CONFIG_FAIL_IO_TIMEOUT1092static struct device_attribute dev_attr_fail_timeout =1093	__ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);1094#endif1095 1096static struct attribute *disk_attrs[] = {1097	&dev_attr_range.attr,1098	&dev_attr_ext_range.attr,1099	&dev_attr_removable.attr,1100	&dev_attr_hidden.attr,1101	&dev_attr_ro.attr,1102	&dev_attr_size.attr,1103	&dev_attr_alignment_offset.attr,1104	&dev_attr_discard_alignment.attr,1105	&dev_attr_capability.attr,1106	&dev_attr_stat.attr,1107	&dev_attr_inflight.attr,1108	&dev_attr_badblocks.attr,1109	&dev_attr_events.attr,1110	&dev_attr_events_async.attr,1111	&dev_attr_events_poll_msecs.attr,1112	&dev_attr_diskseq.attr,1113	&dev_attr_partscan.attr,1114#ifdef CONFIG_FAIL_MAKE_REQUEST1115	&dev_attr_fail.attr,1116#endif1117#ifdef CONFIG_FAIL_IO_TIMEOUT1118	&dev_attr_fail_timeout.attr,1119#endif1120	NULL1121};1122 1123static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)1124{1125	struct device *dev = container_of(kobj, typeof(*dev), kobj);1126	struct gendisk *disk = dev_to_disk(dev);1127 1128	if (a == &dev_attr_badblocks.attr && !disk->bb)1129		return 0;1130	return a->mode;1131}1132 1133static struct attribute_group disk_attr_group = {1134	.attrs = disk_attrs,1135	.is_visible = disk_visible,1136};1137 1138static const struct attribute_group *disk_attr_groups[] = {1139	&disk_attr_group,1140#ifdef CONFIG_BLK_DEV_IO_TRACE1141	&blk_trace_attr_group,1142#endif1143#ifdef CONFIG_BLK_DEV_INTEGRITY1144	&blk_integrity_attr_group,1145#endif1146	NULL1147};1148 1149/**1150 * disk_release - releases all allocated resources of the gendisk1151 * @dev: the device representing this disk1152 *1153 * This function releases all allocated resources of the gendisk.1154 *1155 * Drivers which used __device_add_disk() have a gendisk with a request_queue1156 * assigned. Since the request_queue sits on top of the gendisk for these1157 * drivers we also call blk_put_queue() for them, and we expect the1158 * request_queue refcount to reach 0 at this point, and so the request_queue1159 * will also be freed prior to the disk.1160 *1161 * Context: can sleep1162 */1163static void disk_release(struct device *dev)1164{1165	struct gendisk *disk = dev_to_disk(dev);1166 1167	might_sleep();1168	WARN_ON_ONCE(disk_live(disk));1169 1170	blk_trace_remove(disk->queue);1171 1172	/*1173	 * To undo the all initialization from blk_mq_init_allocated_queue in1174	 * case of a probe failure where add_disk is never called we have to1175	 * call blk_mq_exit_queue here. We can't do this for the more common1176	 * teardown case (yet) as the tagset can be gone by the time the disk1177	 * is released once it was added.1178	 */1179	if (queue_is_mq(disk->queue) &&1180	    test_bit(GD_OWNS_QUEUE, &disk->state) &&1181	    !test_bit(GD_ADDED, &disk->state))1182		blk_mq_exit_queue(disk->queue);1183 1184	blkcg_exit_disk(disk);1185 1186	bioset_exit(&disk->bio_split);1187 1188	disk_release_events(disk);1189	kfree(disk->random);1190	disk_free_zone_resources(disk);1191	xa_destroy(&disk->part_tbl);1192 1193	disk->queue->disk = NULL;1194	blk_put_queue(disk->queue);1195 1196	if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)1197		disk->fops->free_disk(disk);1198 1199	bdev_drop(disk->part0);	/* frees the disk */1200}1201 1202static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)1203{1204	const struct gendisk *disk = dev_to_disk(dev);1205 1206	return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);1207}1208 1209const struct class block_class = {1210	.name		= "block",1211	.dev_uevent	= block_uevent,1212};1213 1214static char *block_devnode(const struct device *dev, umode_t *mode,1215			   kuid_t *uid, kgid_t *gid)1216{1217	struct gendisk *disk = dev_to_disk(dev);1218 1219	if (disk->fops->devnode)1220		return disk->fops->devnode(disk, mode);1221	return NULL;1222}1223 1224const struct device_type disk_type = {1225	.name		= "disk",1226	.groups		= disk_attr_groups,1227	.release	= disk_release,1228	.devnode	= block_devnode,1229};1230 1231#ifdef CONFIG_PROC_FS1232/*1233 * aggregate disk stat collector.  Uses the same stats that the sysfs1234 * entries do, above, but makes them available through one seq_file.1235 *1236 * The output looks suspiciously like /proc/partitions with a bunch of1237 * extra fields.1238 */1239static int diskstats_show(struct seq_file *seqf, void *v)1240{1241	struct gendisk *gp = v;1242	struct block_device *hd;1243	unsigned int inflight;1244	struct disk_stats stat;1245	unsigned long idx;1246 1247	/*1248	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)1249		seq_puts(seqf,	"major minor name"1250				"     rio rmerge rsect ruse wio wmerge "1251				"wsect wuse running use aveq"1252				"\n\n");1253	*/1254 1255	rcu_read_lock();1256	xa_for_each(&gp->part_tbl, idx, hd) {1257		if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))1258			continue;1259 1260		inflight = part_in_flight(hd);1261		if (inflight) {1262			part_stat_lock();1263			update_io_ticks(hd, jiffies, true);1264			part_stat_unlock();1265		}1266		part_stat_read_all(hd, &stat);1267		seq_printf(seqf, "%4d %7d %pg "1268			   "%lu %lu %lu %u "1269			   "%lu %lu %lu %u "1270			   "%u %u %u "1271			   "%lu %lu %lu %u "1272			   "%lu %u"1273			   "\n",1274			   MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,1275			   stat.ios[STAT_READ],1276			   stat.merges[STAT_READ],1277			   stat.sectors[STAT_READ],1278			   (unsigned int)div_u64(stat.nsecs[STAT_READ],1279							NSEC_PER_MSEC),1280			   stat.ios[STAT_WRITE],1281			   stat.merges[STAT_WRITE],1282			   stat.sectors[STAT_WRITE],1283			   (unsigned int)div_u64(stat.nsecs[STAT_WRITE],1284							NSEC_PER_MSEC),1285			   inflight,1286			   jiffies_to_msecs(stat.io_ticks),1287			   (unsigned int)div_u64(stat.nsecs[STAT_READ] +1288						 stat.nsecs[STAT_WRITE] +1289						 stat.nsecs[STAT_DISCARD] +1290						 stat.nsecs[STAT_FLUSH],1291							NSEC_PER_MSEC),1292			   stat.ios[STAT_DISCARD],1293			   stat.merges[STAT_DISCARD],1294			   stat.sectors[STAT_DISCARD],1295			   (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],1296						 NSEC_PER_MSEC),1297			   stat.ios[STAT_FLUSH],1298			   (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],1299						 NSEC_PER_MSEC)1300			);1301	}1302	rcu_read_unlock();1303 1304	return 0;1305}1306 1307static const struct seq_operations diskstats_op = {1308	.start	= disk_seqf_start,1309	.next	= disk_seqf_next,1310	.stop	= disk_seqf_stop,1311	.show	= diskstats_show1312};1313 1314static int __init proc_genhd_init(void)1315{1316	proc_create_seq("diskstats", 0, NULL, &diskstats_op);1317	proc_create_seq("partitions", 0, NULL, &partitions_op);1318	return 0;1319}1320module_init(proc_genhd_init);1321#endif /* CONFIG_PROC_FS */1322 1323dev_t part_devt(struct gendisk *disk, u8 partno)1324{1325	struct block_device *part;1326	dev_t devt = 0;1327 1328	rcu_read_lock();1329	part = xa_load(&disk->part_tbl, partno);1330	if (part)1331		devt = part->bd_dev;1332	rcu_read_unlock();1333 1334	return devt;1335}1336 1337struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,1338		struct lock_class_key *lkclass)1339{1340	struct gendisk *disk;1341 1342	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);1343	if (!disk)1344		return NULL;1345 1346	if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))1347		goto out_free_disk;1348 1349	disk->bdi = bdi_alloc(node_id);1350	if (!disk->bdi)1351		goto out_free_bioset;1352 1353	/* bdev_alloc() might need the queue, set before the first call */1354	disk->queue = q;1355 1356	disk->part0 = bdev_alloc(disk, 0);1357	if (!disk->part0)1358		goto out_free_bdi;1359 1360	disk->node_id = node_id;1361	mutex_init(&disk->open_mutex);1362	xa_init(&disk->part_tbl);1363	if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))1364		goto out_destroy_part_tbl;1365 1366	if (blkcg_init_disk(disk))1367		goto out_erase_part0;1368 1369	disk_init_zone_resources(disk);1370	rand_initialize_disk(disk);1371	disk_to_dev(disk)->class = &block_class;1372	disk_to_dev(disk)->type = &disk_type;1373	device_initialize(disk_to_dev(disk));1374	inc_diskseq(disk);1375	q->disk = disk;1376	lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);1377#ifdef CONFIG_BLOCK_HOLDER_DEPRECATED1378	INIT_LIST_HEAD(&disk->slave_bdevs);1379#endif1380	return disk;1381 1382out_erase_part0:1383	xa_erase(&disk->part_tbl, 0);1384out_destroy_part_tbl:1385	xa_destroy(&disk->part_tbl);1386	disk->part0->bd_disk = NULL;1387	bdev_drop(disk->part0);1388out_free_bdi:1389	bdi_put(disk->bdi);1390out_free_bioset:1391	bioset_exit(&disk->bio_split);1392out_free_disk:1393	kfree(disk);1394	return NULL;1395}1396 1397struct gendisk *__blk_alloc_disk(struct queue_limits *lim, int node,1398		struct lock_class_key *lkclass)1399{1400	struct queue_limits default_lim = { };1401	struct request_queue *q;1402	struct gendisk *disk;1403 1404	q = blk_alloc_queue(lim ? lim : &default_lim, node);1405	if (IS_ERR(q))1406		return ERR_CAST(q);1407 1408	disk = __alloc_disk_node(q, node, lkclass);1409	if (!disk) {1410		blk_put_queue(q);1411		return ERR_PTR(-ENOMEM);1412	}1413	set_bit(GD_OWNS_QUEUE, &disk->state);1414	return disk;1415}1416EXPORT_SYMBOL(__blk_alloc_disk);1417 1418/**1419 * put_disk - decrements the gendisk refcount1420 * @disk: the struct gendisk to decrement the refcount for1421 *1422 * This decrements the refcount for the struct gendisk. When this reaches 01423 * we'll have disk_release() called.1424 *1425 * Note: for blk-mq disk put_disk must be called before freeing the tag_set1426 * when handling probe errors (that is before add_disk() is called).1427 *1428 * Context: Any context, but the last reference must not be dropped from1429 *          atomic context.1430 */1431void put_disk(struct gendisk *disk)1432{1433	if (disk)1434		put_device(disk_to_dev(disk));1435}1436EXPORT_SYMBOL(put_disk);1437 1438static void set_disk_ro_uevent(struct gendisk *gd, int ro)1439{1440	char event[] = "DISK_RO=1";1441	char *envp[] = { event, NULL };1442 1443	if (!ro)1444		event[8] = '0';1445	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);1446}1447 1448/**1449 * set_disk_ro - set a gendisk read-only1450 * @disk:	gendisk to operate on1451 * @read_only:	%true to set the disk read-only, %false set the disk read/write1452 *1453 * This function is used to indicate whether a given disk device should have its1454 * read-only flag set. set_disk_ro() is typically used by device drivers to1455 * indicate whether the underlying physical device is write-protected.1456 */1457void set_disk_ro(struct gendisk *disk, bool read_only)1458{1459	if (read_only) {1460		if (test_and_set_bit(GD_READ_ONLY, &disk->state))1461			return;1462	} else {1463		if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))1464			return;1465	}1466	set_disk_ro_uevent(disk, read_only);1467}1468EXPORT_SYMBOL(set_disk_ro);1469 1470void inc_diskseq(struct gendisk *disk)1471{1472	disk->diskseq = atomic64_inc_return(&diskseq);1473}1474