1 // SPDX-License-Identifier: GPL-2.0
5 * Portions Copyright (C) 2020 Christoph Hellwig
8 #include <linux/module.h>
9 #include <linux/ctype.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>
30 #include "blk-throttle.h"
32 #include "blk-mq-sched.h"
33 #include "blk-rq-qos.h"
34 #include "blk-cgroup.h"
36 static struct kobject *block_depr;
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.
53 static atomic64_t diskseq;
55 /* for extended dynamic devt allocation, currently only one major is used */
56 #define NR_EXT_DEVT (1 << MINORBITS)
57 static DEFINE_IDA(ext_devt_ida);
59 void set_capacity(struct gendisk *disk, sector_t sectors)
61 bdev_set_nr_sectors(disk->part0, sectors);
63 EXPORT_SYMBOL(set_capacity);
66 * Set disk capacity and notify if the size is not currently zero and will not
67 * be set to zero. Returns true if a uevent was sent, otherwise false.
69 bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
71 sector_t capacity = get_capacity(disk);
72 char *envp[] = { "RESIZE=1", NULL };
74 set_capacity(disk, size);
77 * Only print a message and send a uevent if the gendisk is user visible
78 * and alive. This avoids spamming the log and udev when setting the
79 * initial capacity during probing.
81 if (size == capacity ||
83 (disk->flags & GENHD_FL_HIDDEN))
86 pr_info("%s: detected capacity change from %lld to %lld\n",
87 disk->disk_name, capacity, size);
90 * Historically we did not send a uevent for changes to/from an empty
93 if (!capacity || !size)
95 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
98 EXPORT_SYMBOL_GPL(set_capacity_and_notify);
100 static void part_stat_read_all(struct block_device *part,
101 struct disk_stats *stat)
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);
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];
117 stat->io_ticks += ptr->io_ticks;
121 static unsigned int part_in_flight(struct block_device *part)
123 unsigned int inflight = 0;
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);
130 if ((int)inflight < 0)
136 static void part_in_flight_rw(struct block_device *part,
137 unsigned int inflight[2])
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);
147 if ((int)inflight[0] < 0)
149 if ((int)inflight[1] < 0)
154 * Can be deleted altogether. Later.
157 #define BLKDEV_MAJOR_HASH_SIZE 255
158 static struct blk_major_name {
159 struct blk_major_name *next;
162 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
163 void (*probe)(dev_t devt);
165 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
166 static DEFINE_MUTEX(major_names_lock);
167 static DEFINE_SPINLOCK(major_names_spinlock);
169 /* index in the above - for now: assume no multimajor ranges */
170 static inline int major_to_index(unsigned major)
172 return major % BLKDEV_MAJOR_HASH_SIZE;
175 #ifdef CONFIG_PROC_FS
176 void blkdev_show(struct seq_file *seqf, off_t offset)
178 struct blk_major_name *dp;
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);
186 #endif /* CONFIG_PROC_FS */
189 * __register_blkdev - register a new block device
191 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
192 * @major = 0, try to allocate any unused major number.
193 * @name: the name of the new block device as a zero terminated string
194 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
195 * pre-created device node is accessed. When a probe call uses
196 * add_disk() and it fails the driver must cleanup resources. This
197 * interface may soon be removed.
199 * The @name must be unique within the system.
201 * The return value depends on the @major input parameter:
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 code
205 * - if any unused major number was requested with @major = 0 parameter
206 * then the return value is the allocated major number in range
207 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
209 * See Documentation/admin-guide/devices.txt for the list of allocated
212 * Use register_blkdev instead for any new code.
214 int __register_blkdev(unsigned int major, const char *name,
215 void (*probe)(dev_t devt))
217 struct blk_major_name **n, *p;
220 mutex_lock(&major_names_lock);
224 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
225 if (major_names[index] == NULL)
230 printk("%s: failed to get major for %s\n",
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);
247 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
254 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
257 strscpy(p->name, name, sizeof(p->name));
259 index = major_to_index(major);
261 spin_lock(&major_names_spinlock);
262 for (n = &major_names[index]; *n; n = &(*n)->next) {
263 if ((*n)->major == major)
270 spin_unlock(&major_names_spinlock);
273 printk("register_blkdev: cannot get major %u for %s\n",
278 mutex_unlock(&major_names_lock);
281 EXPORT_SYMBOL(__register_blkdev);
283 void unregister_blkdev(unsigned int major, const char *name)
285 struct blk_major_name **n;
286 struct blk_major_name *p = NULL;
287 int index = major_to_index(major);
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)
294 if (!*n || strcmp((*n)->name, name)) {
300 spin_unlock(&major_names_spinlock);
301 mutex_unlock(&major_names_lock);
305 EXPORT_SYMBOL(unregister_blkdev);
307 int blk_alloc_ext_minor(void)
311 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
317 void blk_free_ext_minor(unsigned int minor)
319 ida_free(&ext_devt_ida, minor);
322 void disk_uevent(struct gendisk *disk, enum kobject_action action)
324 struct block_device *part;
328 xa_for_each(&disk->part_tbl, idx, part) {
329 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
331 if (!kobject_get_unless_zero(&part->bd_device.kobj))
335 kobject_uevent(bdev_kobj(part), action);
336 put_device(&part->bd_device);
341 EXPORT_SYMBOL_GPL(disk_uevent);
343 int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode)
345 struct block_device *bdev;
348 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
350 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
352 if (disk->open_partitions)
356 * If the device is opened exclusively by current thread already, it's
357 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
358 * synchronize with other exclusive openers and other partition
361 if (!(mode & BLK_OPEN_EXCL)) {
362 ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions,
368 set_bit(GD_NEED_PART_SCAN, &disk->state);
369 bdev = blkdev_get_by_dev(disk_devt(disk), mode & ~BLK_OPEN_EXCL, NULL,
374 blkdev_put(bdev, NULL);
377 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
378 * and this will cause that re-assemble partitioned raid device will
379 * creat partition for underlying disk.
381 clear_bit(GD_NEED_PART_SCAN, &disk->state);
382 if (!(mode & BLK_OPEN_EXCL))
383 bd_abort_claiming(disk->part0, disk_scan_partitions);
388 * device_add_disk - add disk information to kernel list
389 * @parent: parent device for the disk
390 * @disk: per-device partitioning information
391 * @groups: Additional per-device sysfs groups
393 * This function registers the partitioning information in @disk
396 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
397 const struct attribute_group **groups)
400 struct device *ddev = disk_to_dev(disk);
403 /* Only makes sense for bio-based to set ->poll_bio */
404 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
408 * The disk queue should now be all set with enough information about
409 * the device for the elevator code to pick an adequate default
410 * elevator if one is needed, that is, for devices requesting queue
413 elevator_init_mq(disk->queue);
415 /* Mark bdev as having a submit_bio, if needed */
416 disk->part0->bd_has_submit_bio = disk->fops->submit_bio != NULL;
419 * If the driver provides an explicit major number it also must provide
420 * the number of minors numbers supported, and those will be used to
422 * Otherwise just allocate the device numbers for both the whole device
423 * and all partitions from the extended dev_t space.
427 if (WARN_ON(!disk->minors))
428 goto out_exit_elevator;
430 if (disk->minors > DISK_MAX_PARTS) {
431 pr_err("block: can't allocate more than %d partitions\n",
433 disk->minors = DISK_MAX_PARTS;
435 if (disk->first_minor + disk->minors > MINORMASK + 1)
436 goto out_exit_elevator;
438 if (WARN_ON(disk->minors))
439 goto out_exit_elevator;
441 ret = blk_alloc_ext_minor();
443 goto out_exit_elevator;
444 disk->major = BLOCK_EXT_MAJOR;
445 disk->first_minor = ret;
448 /* delay uevents, until we scanned partition table */
449 dev_set_uevent_suppress(ddev, 1);
451 ddev->parent = parent;
452 ddev->groups = groups;
453 dev_set_name(ddev, "%s", disk->disk_name);
454 if (!(disk->flags & GENHD_FL_HIDDEN))
455 ddev->devt = MKDEV(disk->major, disk->first_minor);
456 ret = device_add(ddev);
458 goto out_free_ext_minor;
460 ret = disk_alloc_events(disk);
464 ret = sysfs_create_link(block_depr, &ddev->kobj,
465 kobject_name(&ddev->kobj));
470 * avoid probable deadlock caused by allocating memory with
471 * GFP_KERNEL in runtime_resume callback of its all ancestor
474 pm_runtime_set_memalloc_noio(ddev, true);
476 disk->part0->bd_holder_dir =
477 kobject_create_and_add("holders", &ddev->kobj);
478 if (!disk->part0->bd_holder_dir) {
480 goto out_del_block_link;
482 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
483 if (!disk->slave_dir) {
485 goto out_put_holder_dir;
488 ret = blk_register_queue(disk);
490 goto out_put_slave_dir;
492 if (!(disk->flags & GENHD_FL_HIDDEN)) {
493 ret = bdi_register(disk->bdi, "%u:%u",
494 disk->major, disk->first_minor);
496 goto out_unregister_queue;
497 bdi_set_owner(disk->bdi, ddev);
498 ret = sysfs_create_link(&ddev->kobj,
499 &disk->bdi->dev->kobj, "bdi");
501 goto out_unregister_bdi;
503 /* Make sure the first partition scan will be proceed */
504 if (get_capacity(disk) && !(disk->flags & GENHD_FL_NO_PART) &&
505 !test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
506 set_bit(GD_NEED_PART_SCAN, &disk->state);
508 bdev_add(disk->part0, ddev->devt);
509 if (get_capacity(disk))
510 disk_scan_partitions(disk, BLK_OPEN_READ);
513 * Announce the disk and partitions after all partitions are
514 * created. (for hidden disks uevents remain suppressed forever)
516 dev_set_uevent_suppress(ddev, 0);
517 disk_uevent(disk, KOBJ_ADD);
520 * Even if the block_device for a hidden gendisk is not
521 * registered, it needs to have a valid bd_dev so that the
522 * freeing of the dynamic major works.
524 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
527 disk_update_readahead(disk);
528 disk_add_events(disk);
529 set_bit(GD_ADDED, &disk->state);
533 if (!(disk->flags & GENHD_FL_HIDDEN))
534 bdi_unregister(disk->bdi);
535 out_unregister_queue:
536 blk_unregister_queue(disk);
537 rq_qos_exit(disk->queue);
539 kobject_put(disk->slave_dir);
540 disk->slave_dir = NULL;
542 kobject_put(disk->part0->bd_holder_dir);
544 sysfs_remove_link(block_depr, dev_name(ddev));
548 if (disk->major == BLOCK_EXT_MAJOR)
549 blk_free_ext_minor(disk->first_minor);
551 if (disk->queue->elevator)
552 elevator_exit(disk->queue);
555 EXPORT_SYMBOL(device_add_disk);
557 static void blk_report_disk_dead(struct gendisk *disk, bool surprise)
559 struct block_device *bdev;
563 xa_for_each(&disk->part_tbl, idx, bdev) {
564 if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
568 bdev_mark_dead(bdev, surprise);
570 put_device(&bdev->bd_device);
576 static void __blk_mark_disk_dead(struct gendisk *disk)
581 if (test_and_set_bit(GD_DEAD, &disk->state))
584 if (test_bit(GD_OWNS_QUEUE, &disk->state))
585 blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);
588 * Stop buffered writers from dirtying pages that can't be written out.
590 set_capacity(disk, 0);
593 * Prevent new I/O from crossing bio_queue_enter().
595 blk_queue_start_drain(disk->queue);
599 * blk_mark_disk_dead - mark a disk as dead
600 * @disk: disk to mark as dead
602 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
605 void blk_mark_disk_dead(struct gendisk *disk)
607 __blk_mark_disk_dead(disk);
608 blk_report_disk_dead(disk, true);
610 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
613 * del_gendisk - remove the gendisk
614 * @disk: the struct gendisk to remove
616 * Removes the gendisk and all its associated resources. This deletes the
617 * partitions associated with the gendisk, and unregisters the associated
620 * This is the counter to the respective __device_add_disk() call.
622 * The final removal of the struct gendisk happens when its refcount reaches 0
623 * with put_disk(), which should be called after del_gendisk(), if
624 * __device_add_disk() was used.
626 * Drivers exist which depend on the release of the gendisk to be synchronous,
627 * it should not be deferred.
631 void del_gendisk(struct gendisk *disk)
633 struct request_queue *q = disk->queue;
634 struct block_device *part;
639 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
642 disk_del_events(disk);
645 * Prevent new openers by unlinked the bdev inode.
647 mutex_lock(&disk->open_mutex);
648 xa_for_each(&disk->part_tbl, idx, part)
649 remove_inode_hash(part->bd_inode);
650 mutex_unlock(&disk->open_mutex);
653 * Tell the file system to write back all dirty data and shut down if
654 * it hasn't been notified earlier.
656 if (!test_bit(GD_DEAD, &disk->state))
657 blk_report_disk_dead(disk, false);
658 __blk_mark_disk_dead(disk);
661 * Drop all partitions now that the disk is marked dead.
663 mutex_lock(&disk->open_mutex);
664 xa_for_each_start(&disk->part_tbl, idx, part, 1)
665 drop_partition(part);
666 mutex_unlock(&disk->open_mutex);
668 if (!(disk->flags & GENHD_FL_HIDDEN)) {
669 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
672 * Unregister bdi before releasing device numbers (as they can
673 * get reused and we'd get clashes in sysfs).
675 bdi_unregister(disk->bdi);
678 blk_unregister_queue(disk);
680 kobject_put(disk->part0->bd_holder_dir);
681 kobject_put(disk->slave_dir);
682 disk->slave_dir = NULL;
684 part_stat_set_all(disk->part0, 0);
685 disk->part0->bd_stamp = 0;
686 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
687 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
688 device_del(disk_to_dev(disk));
690 blk_mq_freeze_queue_wait(q);
692 blk_throtl_cancel_bios(disk);
695 blk_flush_integrity();
698 blk_mq_cancel_work_sync(q);
700 blk_mq_quiesce_queue(q);
702 mutex_lock(&q->sysfs_lock);
704 mutex_unlock(&q->sysfs_lock);
707 blk_mq_unquiesce_queue(q);
710 * If the disk does not own the queue, allow using passthrough requests
711 * again. Else leave the queue frozen to fail all I/O.
713 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
714 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
715 __blk_mq_unfreeze_queue(q, true);
718 blk_mq_exit_queue(q);
721 EXPORT_SYMBOL(del_gendisk);
724 * invalidate_disk - invalidate the disk
725 * @disk: the struct gendisk to invalidate
727 * A helper to invalidates the disk. It will clean the disk's associated
728 * buffer/page caches and reset its internal states so that the disk
729 * can be reused by the drivers.
733 void invalidate_disk(struct gendisk *disk)
735 struct block_device *bdev = disk->part0;
737 invalidate_bdev(bdev);
738 bdev->bd_inode->i_mapping->wb_err = 0;
739 set_capacity(disk, 0);
741 EXPORT_SYMBOL(invalidate_disk);
743 /* sysfs access to bad-blocks list. */
744 static ssize_t disk_badblocks_show(struct device *dev,
745 struct device_attribute *attr,
748 struct gendisk *disk = dev_to_disk(dev);
751 return sprintf(page, "\n");
753 return badblocks_show(disk->bb, page, 0);
756 static ssize_t disk_badblocks_store(struct device *dev,
757 struct device_attribute *attr,
758 const char *page, size_t len)
760 struct gendisk *disk = dev_to_disk(dev);
765 return badblocks_store(disk->bb, page, len, 0);
768 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
769 void blk_request_module(dev_t devt)
771 unsigned int major = MAJOR(devt);
772 struct blk_major_name **n;
774 mutex_lock(&major_names_lock);
775 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
776 if ((*n)->major == major && (*n)->probe) {
778 mutex_unlock(&major_names_lock);
782 mutex_unlock(&major_names_lock);
784 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
785 /* Make old-style 2.4 aliases work */
786 request_module("block-major-%d", MAJOR(devt));
788 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
790 #ifdef CONFIG_PROC_FS
792 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
795 struct class_dev_iter *iter;
798 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
800 return ERR_PTR(-ENOMEM);
802 seqf->private = iter;
803 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
805 dev = class_dev_iter_next(iter);
810 return dev_to_disk(dev);
813 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
818 dev = class_dev_iter_next(seqf->private);
820 return dev_to_disk(dev);
825 static void disk_seqf_stop(struct seq_file *seqf, void *v)
827 struct class_dev_iter *iter = seqf->private;
829 /* stop is called even after start failed :-( */
831 class_dev_iter_exit(iter);
833 seqf->private = NULL;
837 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
841 p = disk_seqf_start(seqf, pos);
842 if (!IS_ERR_OR_NULL(p) && !*pos)
843 seq_puts(seqf, "major minor #blocks name\n\n");
847 static int show_partition(struct seq_file *seqf, void *v)
849 struct gendisk *sgp = v;
850 struct block_device *part;
853 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
857 xa_for_each(&sgp->part_tbl, idx, part) {
858 if (!bdev_nr_sectors(part))
860 seq_printf(seqf, "%4d %7d %10llu %pg\n",
861 MAJOR(part->bd_dev), MINOR(part->bd_dev),
862 bdev_nr_sectors(part) >> 1, part);
868 static const struct seq_operations partitions_op = {
869 .start = show_partition_start,
870 .next = disk_seqf_next,
871 .stop = disk_seqf_stop,
872 .show = show_partition
876 static int __init genhd_device_init(void)
880 error = class_register(&block_class);
885 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
887 /* create top-level block dir */
888 block_depr = kobject_create_and_add("block", NULL);
892 subsys_initcall(genhd_device_init);
894 static ssize_t disk_range_show(struct device *dev,
895 struct device_attribute *attr, char *buf)
897 struct gendisk *disk = dev_to_disk(dev);
899 return sprintf(buf, "%d\n", disk->minors);
902 static ssize_t disk_ext_range_show(struct device *dev,
903 struct device_attribute *attr, char *buf)
905 struct gendisk *disk = dev_to_disk(dev);
907 return sprintf(buf, "%d\n",
908 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
911 static ssize_t disk_removable_show(struct device *dev,
912 struct device_attribute *attr, char *buf)
914 struct gendisk *disk = dev_to_disk(dev);
916 return sprintf(buf, "%d\n",
917 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
920 static ssize_t disk_hidden_show(struct device *dev,
921 struct device_attribute *attr, char *buf)
923 struct gendisk *disk = dev_to_disk(dev);
925 return sprintf(buf, "%d\n",
926 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
929 static ssize_t disk_ro_show(struct device *dev,
930 struct device_attribute *attr, char *buf)
932 struct gendisk *disk = dev_to_disk(dev);
934 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
937 ssize_t part_size_show(struct device *dev,
938 struct device_attribute *attr, char *buf)
940 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
943 ssize_t part_stat_show(struct device *dev,
944 struct device_attribute *attr, char *buf)
946 struct block_device *bdev = dev_to_bdev(dev);
947 struct request_queue *q = bdev_get_queue(bdev);
948 struct disk_stats stat;
949 unsigned int inflight;
952 inflight = blk_mq_in_flight(q, bdev);
954 inflight = part_in_flight(bdev);
958 update_io_ticks(bdev, jiffies, true);
961 part_stat_read_all(bdev, &stat);
963 "%8lu %8lu %8llu %8u "
964 "%8lu %8lu %8llu %8u "
966 "%8lu %8lu %8llu %8u "
970 stat.merges[STAT_READ],
971 (unsigned long long)stat.sectors[STAT_READ],
972 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
973 stat.ios[STAT_WRITE],
974 stat.merges[STAT_WRITE],
975 (unsigned long long)stat.sectors[STAT_WRITE],
976 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
978 jiffies_to_msecs(stat.io_ticks),
979 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
980 stat.nsecs[STAT_WRITE] +
981 stat.nsecs[STAT_DISCARD] +
982 stat.nsecs[STAT_FLUSH],
984 stat.ios[STAT_DISCARD],
985 stat.merges[STAT_DISCARD],
986 (unsigned long long)stat.sectors[STAT_DISCARD],
987 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
988 stat.ios[STAT_FLUSH],
989 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
992 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
995 struct block_device *bdev = dev_to_bdev(dev);
996 struct request_queue *q = bdev_get_queue(bdev);
997 unsigned int inflight[2];
1000 blk_mq_in_flight_rw(q, bdev, inflight);
1002 part_in_flight_rw(bdev, inflight);
1004 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1007 static ssize_t disk_capability_show(struct device *dev,
1008 struct device_attribute *attr, char *buf)
1010 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1011 return sprintf(buf, "0\n");
1014 static ssize_t disk_alignment_offset_show(struct device *dev,
1015 struct device_attribute *attr,
1018 struct gendisk *disk = dev_to_disk(dev);
1020 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1023 static ssize_t disk_discard_alignment_show(struct device *dev,
1024 struct device_attribute *attr,
1027 struct gendisk *disk = dev_to_disk(dev);
1029 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1032 static ssize_t diskseq_show(struct device *dev,
1033 struct device_attribute *attr, char *buf)
1035 struct gendisk *disk = dev_to_disk(dev);
1037 return sprintf(buf, "%llu\n", disk->diskseq);
1040 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1041 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1042 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1043 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1044 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1045 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1046 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1047 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1048 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1049 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1050 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1051 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1052 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1054 #ifdef CONFIG_FAIL_MAKE_REQUEST
1055 ssize_t part_fail_show(struct device *dev,
1056 struct device_attribute *attr, char *buf)
1058 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1061 ssize_t part_fail_store(struct device *dev,
1062 struct device_attribute *attr,
1063 const char *buf, size_t count)
1067 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1068 dev_to_bdev(dev)->bd_make_it_fail = i;
1073 static struct device_attribute dev_attr_fail =
1074 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1075 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1077 #ifdef CONFIG_FAIL_IO_TIMEOUT
1078 static struct device_attribute dev_attr_fail_timeout =
1079 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1082 static struct attribute *disk_attrs[] = {
1083 &dev_attr_range.attr,
1084 &dev_attr_ext_range.attr,
1085 &dev_attr_removable.attr,
1086 &dev_attr_hidden.attr,
1088 &dev_attr_size.attr,
1089 &dev_attr_alignment_offset.attr,
1090 &dev_attr_discard_alignment.attr,
1091 &dev_attr_capability.attr,
1092 &dev_attr_stat.attr,
1093 &dev_attr_inflight.attr,
1094 &dev_attr_badblocks.attr,
1095 &dev_attr_events.attr,
1096 &dev_attr_events_async.attr,
1097 &dev_attr_events_poll_msecs.attr,
1098 &dev_attr_diskseq.attr,
1099 #ifdef CONFIG_FAIL_MAKE_REQUEST
1100 &dev_attr_fail.attr,
1102 #ifdef CONFIG_FAIL_IO_TIMEOUT
1103 &dev_attr_fail_timeout.attr,
1108 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1110 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1111 struct gendisk *disk = dev_to_disk(dev);
1113 if (a == &dev_attr_badblocks.attr && !disk->bb)
1118 static struct attribute_group disk_attr_group = {
1119 .attrs = disk_attrs,
1120 .is_visible = disk_visible,
1123 static const struct attribute_group *disk_attr_groups[] = {
1125 #ifdef CONFIG_BLK_DEV_IO_TRACE
1126 &blk_trace_attr_group,
1128 #ifdef CONFIG_BLK_DEV_INTEGRITY
1129 &blk_integrity_attr_group,
1135 * disk_release - releases all allocated resources of the gendisk
1136 * @dev: the device representing this disk
1138 * This function releases all allocated resources of the gendisk.
1140 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1141 * assigned. Since the request_queue sits on top of the gendisk for these
1142 * drivers we also call blk_put_queue() for them, and we expect the
1143 * request_queue refcount to reach 0 at this point, and so the request_queue
1144 * will also be freed prior to the disk.
1146 * Context: can sleep
1148 static void disk_release(struct device *dev)
1150 struct gendisk *disk = dev_to_disk(dev);
1153 WARN_ON_ONCE(disk_live(disk));
1155 blk_trace_remove(disk->queue);
1158 * To undo the all initialization from blk_mq_init_allocated_queue in
1159 * case of a probe failure where add_disk is never called we have to
1160 * call blk_mq_exit_queue here. We can't do this for the more common
1161 * teardown case (yet) as the tagset can be gone by the time the disk
1162 * is released once it was added.
1164 if (queue_is_mq(disk->queue) &&
1165 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1166 !test_bit(GD_ADDED, &disk->state))
1167 blk_mq_exit_queue(disk->queue);
1169 blkcg_exit_disk(disk);
1171 bioset_exit(&disk->bio_split);
1173 disk_release_events(disk);
1174 kfree(disk->random);
1175 disk_free_zone_bitmaps(disk);
1176 xa_destroy(&disk->part_tbl);
1178 disk->queue->disk = NULL;
1179 blk_put_queue(disk->queue);
1181 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1182 disk->fops->free_disk(disk);
1184 iput(disk->part0->bd_inode); /* frees the disk */
1187 static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1189 const struct gendisk *disk = dev_to_disk(dev);
1191 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1194 struct class block_class = {
1196 .dev_uevent = block_uevent,
1199 static char *block_devnode(const struct device *dev, umode_t *mode,
1200 kuid_t *uid, kgid_t *gid)
1202 struct gendisk *disk = dev_to_disk(dev);
1204 if (disk->fops->devnode)
1205 return disk->fops->devnode(disk, mode);
1209 const struct device_type disk_type = {
1211 .groups = disk_attr_groups,
1212 .release = disk_release,
1213 .devnode = block_devnode,
1216 #ifdef CONFIG_PROC_FS
1218 * aggregate disk stat collector. Uses the same stats that the sysfs
1219 * entries do, above, but makes them available through one seq_file.
1221 * The output looks suspiciously like /proc/partitions with a bunch of
1224 static int diskstats_show(struct seq_file *seqf, void *v)
1226 struct gendisk *gp = v;
1227 struct block_device *hd;
1228 unsigned int inflight;
1229 struct disk_stats stat;
1233 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1234 seq_puts(seqf, "major minor name"
1235 " rio rmerge rsect ruse wio wmerge "
1236 "wsect wuse running use aveq"
1241 xa_for_each(&gp->part_tbl, idx, hd) {
1242 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1244 if (queue_is_mq(gp->queue))
1245 inflight = blk_mq_in_flight(gp->queue, hd);
1247 inflight = part_in_flight(hd);
1251 update_io_ticks(hd, jiffies, true);
1254 part_stat_read_all(hd, &stat);
1255 seq_printf(seqf, "%4d %7d %pg "
1262 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1263 stat.ios[STAT_READ],
1264 stat.merges[STAT_READ],
1265 stat.sectors[STAT_READ],
1266 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1268 stat.ios[STAT_WRITE],
1269 stat.merges[STAT_WRITE],
1270 stat.sectors[STAT_WRITE],
1271 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1274 jiffies_to_msecs(stat.io_ticks),
1275 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1276 stat.nsecs[STAT_WRITE] +
1277 stat.nsecs[STAT_DISCARD] +
1278 stat.nsecs[STAT_FLUSH],
1280 stat.ios[STAT_DISCARD],
1281 stat.merges[STAT_DISCARD],
1282 stat.sectors[STAT_DISCARD],
1283 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1285 stat.ios[STAT_FLUSH],
1286 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1295 static const struct seq_operations diskstats_op = {
1296 .start = disk_seqf_start,
1297 .next = disk_seqf_next,
1298 .stop = disk_seqf_stop,
1299 .show = diskstats_show
1302 static int __init proc_genhd_init(void)
1304 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1305 proc_create_seq("partitions", 0, NULL, &partitions_op);
1308 module_init(proc_genhd_init);
1309 #endif /* CONFIG_PROC_FS */
1311 dev_t part_devt(struct gendisk *disk, u8 partno)
1313 struct block_device *part;
1317 part = xa_load(&disk->part_tbl, partno);
1319 devt = part->bd_dev;
1325 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1326 struct lock_class_key *lkclass)
1328 struct gendisk *disk;
1330 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1334 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1337 disk->bdi = bdi_alloc(node_id);
1339 goto out_free_bioset;
1341 /* bdev_alloc() might need the queue, set before the first call */
1344 disk->part0 = bdev_alloc(disk, 0);
1348 disk->node_id = node_id;
1349 mutex_init(&disk->open_mutex);
1350 xa_init(&disk->part_tbl);
1351 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1352 goto out_destroy_part_tbl;
1354 if (blkcg_init_disk(disk))
1355 goto out_erase_part0;
1357 rand_initialize_disk(disk);
1358 disk_to_dev(disk)->class = &block_class;
1359 disk_to_dev(disk)->type = &disk_type;
1360 device_initialize(disk_to_dev(disk));
1363 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1364 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1365 INIT_LIST_HEAD(&disk->slave_bdevs);
1370 xa_erase(&disk->part_tbl, 0);
1371 out_destroy_part_tbl:
1372 xa_destroy(&disk->part_tbl);
1373 disk->part0->bd_disk = NULL;
1374 iput(disk->part0->bd_inode);
1378 bioset_exit(&disk->bio_split);
1384 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1386 struct request_queue *q;
1387 struct gendisk *disk;
1389 q = blk_alloc_queue(node);
1393 disk = __alloc_disk_node(q, node, lkclass);
1398 set_bit(GD_OWNS_QUEUE, &disk->state);
1401 EXPORT_SYMBOL(__blk_alloc_disk);
1404 * put_disk - decrements the gendisk refcount
1405 * @disk: the struct gendisk to decrement the refcount for
1407 * This decrements the refcount for the struct gendisk. When this reaches 0
1408 * we'll have disk_release() called.
1410 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1411 * when handling probe errors (that is before add_disk() is called).
1413 * Context: Any context, but the last reference must not be dropped from
1416 void put_disk(struct gendisk *disk)
1419 put_device(disk_to_dev(disk));
1421 EXPORT_SYMBOL(put_disk);
1423 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1425 char event[] = "DISK_RO=1";
1426 char *envp[] = { event, NULL };
1430 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1434 * set_disk_ro - set a gendisk read-only
1435 * @disk: gendisk to operate on
1436 * @read_only: %true to set the disk read-only, %false set the disk read/write
1438 * This function is used to indicate whether a given disk device should have its
1439 * read-only flag set. set_disk_ro() is typically used by device drivers to
1440 * indicate whether the underlying physical device is write-protected.
1442 void set_disk_ro(struct gendisk *disk, bool read_only)
1445 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1448 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1451 set_disk_ro_uevent(disk, read_only);
1453 EXPORT_SYMBOL(set_disk_ro);
1455 void inc_diskseq(struct gendisk *disk)
1457 disk->diskseq = atomic64_inc_return(&diskseq);