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 "blk-throttle.h"
31 #include "blk-mq-sched.h"
32 #include "blk-rq-qos.h"
33 #include "blk-cgroup.h"
35 static struct kobject *block_depr;
38 * Unique, monotonically increasing sequential number associated with block
39 * devices instances (i.e. incremented each time a device is attached).
40 * Associating uevents with block devices in userspace is difficult and racy:
41 * the uevent netlink socket is lossy, and on slow and overloaded systems has
42 * a very high latency.
43 * Block devices do not have exclusive owners in userspace, any process can set
44 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
45 * can be reused again and again).
46 * A userspace process setting up a block device and watching for its events
47 * cannot thus reliably tell whether an event relates to the device it just set
48 * up or another earlier instance with the same name.
49 * This sequential number allows userspace processes to solve this problem, and
50 * uniquely associate an uevent to the lifetime to a device.
52 static atomic64_t diskseq;
54 /* for extended dynamic devt allocation, currently only one major is used */
55 #define NR_EXT_DEVT (1 << MINORBITS)
56 static DEFINE_IDA(ext_devt_ida);
58 void set_capacity(struct gendisk *disk, sector_t sectors)
60 struct block_device *bdev = disk->part0;
62 spin_lock(&bdev->bd_size_lock);
63 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
64 bdev->bd_nr_sectors = sectors;
65 spin_unlock(&bdev->bd_size_lock);
67 EXPORT_SYMBOL(set_capacity);
70 * Set disk capacity and notify if the size is not currently zero and will not
71 * be set to zero. Returns true if a uevent was sent, otherwise false.
73 bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
75 sector_t capacity = get_capacity(disk);
76 char *envp[] = { "RESIZE=1", NULL };
78 set_capacity(disk, size);
81 * Only print a message and send a uevent if the gendisk is user visible
82 * and alive. This avoids spamming the log and udev when setting the
83 * initial capacity during probing.
85 if (size == capacity ||
87 (disk->flags & GENHD_FL_HIDDEN))
90 pr_info("%s: detected capacity change from %lld to %lld\n",
91 disk->disk_name, capacity, size);
94 * Historically we did not send a uevent for changes to/from an empty
97 if (!capacity || !size)
99 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
102 EXPORT_SYMBOL_GPL(set_capacity_and_notify);
104 static void part_stat_read_all(struct block_device *part,
105 struct disk_stats *stat)
109 memset(stat, 0, sizeof(struct disk_stats));
110 for_each_possible_cpu(cpu) {
111 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
114 for (group = 0; group < NR_STAT_GROUPS; group++) {
115 stat->nsecs[group] += ptr->nsecs[group];
116 stat->sectors[group] += ptr->sectors[group];
117 stat->ios[group] += ptr->ios[group];
118 stat->merges[group] += ptr->merges[group];
121 stat->io_ticks += ptr->io_ticks;
125 static unsigned int part_in_flight(struct block_device *part)
127 unsigned int inflight = 0;
130 for_each_possible_cpu(cpu) {
131 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
132 part_stat_local_read_cpu(part, in_flight[1], cpu);
134 if ((int)inflight < 0)
140 static void part_in_flight_rw(struct block_device *part,
141 unsigned int inflight[2])
147 for_each_possible_cpu(cpu) {
148 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
149 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
151 if ((int)inflight[0] < 0)
153 if ((int)inflight[1] < 0)
158 * Can be deleted altogether. Later.
161 #define BLKDEV_MAJOR_HASH_SIZE 255
162 static struct blk_major_name {
163 struct blk_major_name *next;
166 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
167 void (*probe)(dev_t devt);
169 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
170 static DEFINE_MUTEX(major_names_lock);
171 static DEFINE_SPINLOCK(major_names_spinlock);
173 /* index in the above - for now: assume no multimajor ranges */
174 static inline int major_to_index(unsigned major)
176 return major % BLKDEV_MAJOR_HASH_SIZE;
179 #ifdef CONFIG_PROC_FS
180 void blkdev_show(struct seq_file *seqf, off_t offset)
182 struct blk_major_name *dp;
184 spin_lock(&major_names_spinlock);
185 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
186 if (dp->major == offset)
187 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
188 spin_unlock(&major_names_spinlock);
190 #endif /* CONFIG_PROC_FS */
193 * __register_blkdev - register a new block device
195 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
196 * @major = 0, try to allocate any unused major number.
197 * @name: the name of the new block device as a zero terminated string
198 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
199 * pre-created device node is accessed. When a probe call uses
200 * add_disk() and it fails the driver must cleanup resources. This
201 * interface may soon be removed.
203 * The @name must be unique within the system.
205 * The return value depends on the @major input parameter:
207 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
208 * then the function returns zero on success, or a negative error code
209 * - if any unused major number was requested with @major = 0 parameter
210 * then the return value is the allocated major number in range
211 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
213 * See Documentation/admin-guide/devices.txt for the list of allocated
216 * Use register_blkdev instead for any new code.
218 int __register_blkdev(unsigned int major, const char *name,
219 void (*probe)(dev_t devt))
221 struct blk_major_name **n, *p;
224 mutex_lock(&major_names_lock);
228 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
229 if (major_names[index] == NULL)
234 printk("%s: failed to get major for %s\n",
243 if (major >= BLKDEV_MAJOR_MAX) {
244 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
245 __func__, major, BLKDEV_MAJOR_MAX-1, name);
251 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
258 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
261 strlcpy(p->name, name, sizeof(p->name));
263 index = major_to_index(major);
265 spin_lock(&major_names_spinlock);
266 for (n = &major_names[index]; *n; n = &(*n)->next) {
267 if ((*n)->major == major)
274 spin_unlock(&major_names_spinlock);
277 printk("register_blkdev: cannot get major %u for %s\n",
282 mutex_unlock(&major_names_lock);
285 EXPORT_SYMBOL(__register_blkdev);
287 void unregister_blkdev(unsigned int major, const char *name)
289 struct blk_major_name **n;
290 struct blk_major_name *p = NULL;
291 int index = major_to_index(major);
293 mutex_lock(&major_names_lock);
294 spin_lock(&major_names_spinlock);
295 for (n = &major_names[index]; *n; n = &(*n)->next)
296 if ((*n)->major == major)
298 if (!*n || strcmp((*n)->name, name)) {
304 spin_unlock(&major_names_spinlock);
305 mutex_unlock(&major_names_lock);
309 EXPORT_SYMBOL(unregister_blkdev);
311 int blk_alloc_ext_minor(void)
315 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
321 void blk_free_ext_minor(unsigned int minor)
323 ida_free(&ext_devt_ida, minor);
326 static char *bdevt_str(dev_t devt, char *buf)
328 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
329 char tbuf[BDEVT_SIZE];
330 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
331 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
333 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
338 void disk_uevent(struct gendisk *disk, enum kobject_action action)
340 struct block_device *part;
344 xa_for_each(&disk->part_tbl, idx, part) {
345 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
347 if (!kobject_get_unless_zero(&part->bd_device.kobj))
351 kobject_uevent(bdev_kobj(part), action);
352 put_device(&part->bd_device);
357 EXPORT_SYMBOL_GPL(disk_uevent);
359 int disk_scan_partitions(struct gendisk *disk, fmode_t mode)
361 struct block_device *bdev;
364 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
366 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
368 if (disk->open_partitions)
371 set_bit(GD_NEED_PART_SCAN, &disk->state);
373 * If the device is opened exclusively by current thread already, it's
374 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
375 * synchronize with other exclusive openers and other partition
378 if (!(mode & FMODE_EXCL)) {
379 ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions);
384 bdev = blkdev_get_by_dev(disk_devt(disk), mode & ~FMODE_EXCL, NULL);
388 blkdev_put(bdev, mode);
390 if (!(mode & FMODE_EXCL))
391 bd_abort_claiming(disk->part0, disk_scan_partitions);
396 * device_add_disk - add disk information to kernel list
397 * @parent: parent device for the disk
398 * @disk: per-device partitioning information
399 * @groups: Additional per-device sysfs groups
401 * This function registers the partitioning information in @disk
404 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
405 const struct attribute_group **groups)
408 struct device *ddev = disk_to_dev(disk);
411 /* Only makes sense for bio-based to set ->poll_bio */
412 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
416 * The disk queue should now be all set with enough information about
417 * the device for the elevator code to pick an adequate default
418 * elevator if one is needed, that is, for devices requesting queue
421 elevator_init_mq(disk->queue);
424 * If the driver provides an explicit major number it also must provide
425 * the number of minors numbers supported, and those will be used to
427 * Otherwise just allocate the device numbers for both the whole device
428 * and all partitions from the extended dev_t space.
432 if (WARN_ON(!disk->minors))
433 goto out_exit_elevator;
435 if (disk->minors > DISK_MAX_PARTS) {
436 pr_err("block: can't allocate more than %d partitions\n",
438 disk->minors = DISK_MAX_PARTS;
440 if (disk->first_minor + disk->minors > MINORMASK + 1)
441 goto out_exit_elevator;
443 if (WARN_ON(disk->minors))
444 goto out_exit_elevator;
446 ret = blk_alloc_ext_minor();
448 goto out_exit_elevator;
449 disk->major = BLOCK_EXT_MAJOR;
450 disk->first_minor = ret;
453 /* delay uevents, until we scanned partition table */
454 dev_set_uevent_suppress(ddev, 1);
456 ddev->parent = parent;
457 ddev->groups = groups;
458 dev_set_name(ddev, "%s", disk->disk_name);
459 if (!(disk->flags & GENHD_FL_HIDDEN))
460 ddev->devt = MKDEV(disk->major, disk->first_minor);
461 ret = device_add(ddev);
463 goto out_free_ext_minor;
465 ret = disk_alloc_events(disk);
469 if (!sysfs_deprecated) {
470 ret = sysfs_create_link(block_depr, &ddev->kobj,
471 kobject_name(&ddev->kobj));
477 * avoid probable deadlock caused by allocating memory with
478 * GFP_KERNEL in runtime_resume callback of its all ancestor
481 pm_runtime_set_memalloc_noio(ddev, true);
483 ret = blk_integrity_add(disk);
485 goto out_del_block_link;
487 disk->part0->bd_holder_dir =
488 kobject_create_and_add("holders", &ddev->kobj);
489 if (!disk->part0->bd_holder_dir) {
491 goto out_del_integrity;
493 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
494 if (!disk->slave_dir) {
496 goto out_put_holder_dir;
499 ret = blk_register_queue(disk);
501 goto out_put_slave_dir;
503 if (!(disk->flags & GENHD_FL_HIDDEN)) {
504 ret = bdi_register(disk->bdi, "%u:%u",
505 disk->major, disk->first_minor);
507 goto out_unregister_queue;
508 bdi_set_owner(disk->bdi, ddev);
509 ret = sysfs_create_link(&ddev->kobj,
510 &disk->bdi->dev->kobj, "bdi");
512 goto out_unregister_bdi;
514 /* Make sure the first partition scan will be proceed */
515 if (get_capacity(disk) && !(disk->flags & GENHD_FL_NO_PART) &&
516 !test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
517 set_bit(GD_NEED_PART_SCAN, &disk->state);
519 bdev_add(disk->part0, ddev->devt);
520 if (get_capacity(disk))
521 disk_scan_partitions(disk, FMODE_READ);
524 * Announce the disk and partitions after all partitions are
525 * created. (for hidden disks uevents remain suppressed forever)
527 dev_set_uevent_suppress(ddev, 0);
528 disk_uevent(disk, KOBJ_ADD);
531 * Even if the block_device for a hidden gendisk is not
532 * registered, it needs to have a valid bd_dev so that the
533 * freeing of the dynamic major works.
535 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
538 disk_update_readahead(disk);
539 disk_add_events(disk);
540 set_bit(GD_ADDED, &disk->state);
544 if (!(disk->flags & GENHD_FL_HIDDEN))
545 bdi_unregister(disk->bdi);
546 out_unregister_queue:
547 blk_unregister_queue(disk);
548 rq_qos_exit(disk->queue);
550 kobject_put(disk->slave_dir);
551 disk->slave_dir = NULL;
553 kobject_put(disk->part0->bd_holder_dir);
555 blk_integrity_del(disk);
557 if (!sysfs_deprecated)
558 sysfs_remove_link(block_depr, dev_name(ddev));
562 if (disk->major == BLOCK_EXT_MAJOR)
563 blk_free_ext_minor(disk->first_minor);
565 if (disk->queue->elevator)
566 elevator_exit(disk->queue);
569 EXPORT_SYMBOL(device_add_disk);
572 * blk_mark_disk_dead - mark a disk as dead
573 * @disk: disk to mark as dead
575 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
578 void blk_mark_disk_dead(struct gendisk *disk)
580 set_bit(GD_DEAD, &disk->state);
581 blk_queue_start_drain(disk->queue);
584 * Stop buffered writers from dirtying pages that can't be written out.
586 set_capacity_and_notify(disk, 0);
588 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
591 * del_gendisk - remove the gendisk
592 * @disk: the struct gendisk to remove
594 * Removes the gendisk and all its associated resources. This deletes the
595 * partitions associated with the gendisk, and unregisters the associated
598 * This is the counter to the respective __device_add_disk() call.
600 * The final removal of the struct gendisk happens when its refcount reaches 0
601 * with put_disk(), which should be called after del_gendisk(), if
602 * __device_add_disk() was used.
604 * Drivers exist which depend on the release of the gendisk to be synchronous,
605 * it should not be deferred.
609 void del_gendisk(struct gendisk *disk)
611 struct request_queue *q = disk->queue;
615 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
618 blk_integrity_del(disk);
619 disk_del_events(disk);
621 mutex_lock(&disk->open_mutex);
622 remove_inode_hash(disk->part0->bd_inode);
623 blk_drop_partitions(disk);
624 mutex_unlock(&disk->open_mutex);
626 fsync_bdev(disk->part0);
627 __invalidate_device(disk->part0, true);
632 set_bit(GD_DEAD, &disk->state);
633 if (test_bit(GD_OWNS_QUEUE, &disk->state))
634 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
635 set_capacity(disk, 0);
638 * Prevent new I/O from crossing bio_queue_enter().
640 blk_queue_start_drain(q);
642 if (!(disk->flags & GENHD_FL_HIDDEN)) {
643 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
646 * Unregister bdi before releasing device numbers (as they can
647 * get reused and we'd get clashes in sysfs).
649 bdi_unregister(disk->bdi);
652 blk_unregister_queue(disk);
654 kobject_put(disk->part0->bd_holder_dir);
655 kobject_put(disk->slave_dir);
656 disk->slave_dir = NULL;
658 part_stat_set_all(disk->part0, 0);
659 disk->part0->bd_stamp = 0;
660 if (!sysfs_deprecated)
661 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
662 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
663 device_del(disk_to_dev(disk));
665 blk_mq_freeze_queue_wait(q);
667 blk_throtl_cancel_bios(disk);
670 blk_flush_integrity();
673 blk_mq_cancel_work_sync(q);
675 blk_mq_quiesce_queue(q);
677 mutex_lock(&q->sysfs_lock);
679 mutex_unlock(&q->sysfs_lock);
682 blk_mq_unquiesce_queue(q);
685 * If the disk does not own the queue, allow using passthrough requests
686 * again. Else leave the queue frozen to fail all I/O.
688 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
689 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
690 __blk_mq_unfreeze_queue(q, true);
693 blk_mq_exit_queue(q);
696 EXPORT_SYMBOL(del_gendisk);
699 * invalidate_disk - invalidate the disk
700 * @disk: the struct gendisk to invalidate
702 * A helper to invalidates the disk. It will clean the disk's associated
703 * buffer/page caches and reset its internal states so that the disk
704 * can be reused by the drivers.
708 void invalidate_disk(struct gendisk *disk)
710 struct block_device *bdev = disk->part0;
712 invalidate_bdev(bdev);
713 bdev->bd_inode->i_mapping->wb_err = 0;
714 set_capacity(disk, 0);
716 EXPORT_SYMBOL(invalidate_disk);
718 /* sysfs access to bad-blocks list. */
719 static ssize_t disk_badblocks_show(struct device *dev,
720 struct device_attribute *attr,
723 struct gendisk *disk = dev_to_disk(dev);
726 return sprintf(page, "\n");
728 return badblocks_show(disk->bb, page, 0);
731 static ssize_t disk_badblocks_store(struct device *dev,
732 struct device_attribute *attr,
733 const char *page, size_t len)
735 struct gendisk *disk = dev_to_disk(dev);
740 return badblocks_store(disk->bb, page, len, 0);
743 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
744 void blk_request_module(dev_t devt)
746 unsigned int major = MAJOR(devt);
747 struct blk_major_name **n;
749 mutex_lock(&major_names_lock);
750 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
751 if ((*n)->major == major && (*n)->probe) {
753 mutex_unlock(&major_names_lock);
757 mutex_unlock(&major_names_lock);
759 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
760 /* Make old-style 2.4 aliases work */
761 request_module("block-major-%d", MAJOR(devt));
763 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
766 * print a full list of all partitions - intended for places where the root
767 * filesystem can't be mounted and thus to give the victim some idea of what
770 void __init printk_all_partitions(void)
772 struct class_dev_iter iter;
775 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
776 while ((dev = class_dev_iter_next(&iter))) {
777 struct gendisk *disk = dev_to_disk(dev);
778 struct block_device *part;
779 char devt_buf[BDEVT_SIZE];
783 * Don't show empty devices or things that have been
786 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
790 * Note, unlike /proc/partitions, I am showing the numbers in
791 * hex - the same format as the root= option takes.
794 xa_for_each(&disk->part_tbl, idx, part) {
795 if (!bdev_nr_sectors(part))
797 printk("%s%s %10llu %pg %s",
798 bdev_is_partition(part) ? " " : "",
799 bdevt_str(part->bd_dev, devt_buf),
800 bdev_nr_sectors(part) >> 1, part,
802 part->bd_meta_info->uuid : "");
803 if (bdev_is_partition(part))
805 else if (dev->parent && dev->parent->driver)
806 printk(" driver: %s\n",
807 dev->parent->driver->name);
809 printk(" (driver?)\n");
813 class_dev_iter_exit(&iter);
816 #ifdef CONFIG_PROC_FS
818 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
821 struct class_dev_iter *iter;
824 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
826 return ERR_PTR(-ENOMEM);
828 seqf->private = iter;
829 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
831 dev = class_dev_iter_next(iter);
836 return dev_to_disk(dev);
839 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
844 dev = class_dev_iter_next(seqf->private);
846 return dev_to_disk(dev);
851 static void disk_seqf_stop(struct seq_file *seqf, void *v)
853 struct class_dev_iter *iter = seqf->private;
855 /* stop is called even after start failed :-( */
857 class_dev_iter_exit(iter);
859 seqf->private = NULL;
863 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
867 p = disk_seqf_start(seqf, pos);
868 if (!IS_ERR_OR_NULL(p) && !*pos)
869 seq_puts(seqf, "major minor #blocks name\n\n");
873 static int show_partition(struct seq_file *seqf, void *v)
875 struct gendisk *sgp = v;
876 struct block_device *part;
879 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
883 xa_for_each(&sgp->part_tbl, idx, part) {
884 if (!bdev_nr_sectors(part))
886 seq_printf(seqf, "%4d %7d %10llu %pg\n",
887 MAJOR(part->bd_dev), MINOR(part->bd_dev),
888 bdev_nr_sectors(part) >> 1, part);
894 static const struct seq_operations partitions_op = {
895 .start = show_partition_start,
896 .next = disk_seqf_next,
897 .stop = disk_seqf_stop,
898 .show = show_partition
902 static int __init genhd_device_init(void)
906 block_class.dev_kobj = sysfs_dev_block_kobj;
907 error = class_register(&block_class);
912 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
914 /* create top-level block dir */
915 if (!sysfs_deprecated)
916 block_depr = kobject_create_and_add("block", NULL);
920 subsys_initcall(genhd_device_init);
922 static ssize_t disk_range_show(struct device *dev,
923 struct device_attribute *attr, char *buf)
925 struct gendisk *disk = dev_to_disk(dev);
927 return sprintf(buf, "%d\n", disk->minors);
930 static ssize_t disk_ext_range_show(struct device *dev,
931 struct device_attribute *attr, char *buf)
933 struct gendisk *disk = dev_to_disk(dev);
935 return sprintf(buf, "%d\n",
936 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
939 static ssize_t disk_removable_show(struct device *dev,
940 struct device_attribute *attr, char *buf)
942 struct gendisk *disk = dev_to_disk(dev);
944 return sprintf(buf, "%d\n",
945 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
948 static ssize_t disk_hidden_show(struct device *dev,
949 struct device_attribute *attr, char *buf)
951 struct gendisk *disk = dev_to_disk(dev);
953 return sprintf(buf, "%d\n",
954 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
957 static ssize_t disk_ro_show(struct device *dev,
958 struct device_attribute *attr, char *buf)
960 struct gendisk *disk = dev_to_disk(dev);
962 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
965 ssize_t part_size_show(struct device *dev,
966 struct device_attribute *attr, char *buf)
968 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
971 ssize_t part_stat_show(struct device *dev,
972 struct device_attribute *attr, char *buf)
974 struct block_device *bdev = dev_to_bdev(dev);
975 struct request_queue *q = bdev_get_queue(bdev);
976 struct disk_stats stat;
977 unsigned int inflight;
980 inflight = blk_mq_in_flight(q, bdev);
982 inflight = part_in_flight(bdev);
986 update_io_ticks(bdev, jiffies, true);
989 part_stat_read_all(bdev, &stat);
991 "%8lu %8lu %8llu %8u "
992 "%8lu %8lu %8llu %8u "
994 "%8lu %8lu %8llu %8u "
998 stat.merges[STAT_READ],
999 (unsigned long long)stat.sectors[STAT_READ],
1000 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1001 stat.ios[STAT_WRITE],
1002 stat.merges[STAT_WRITE],
1003 (unsigned long long)stat.sectors[STAT_WRITE],
1004 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1006 jiffies_to_msecs(stat.io_ticks),
1007 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1008 stat.nsecs[STAT_WRITE] +
1009 stat.nsecs[STAT_DISCARD] +
1010 stat.nsecs[STAT_FLUSH],
1012 stat.ios[STAT_DISCARD],
1013 stat.merges[STAT_DISCARD],
1014 (unsigned long long)stat.sectors[STAT_DISCARD],
1015 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1016 stat.ios[STAT_FLUSH],
1017 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1020 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1023 struct block_device *bdev = dev_to_bdev(dev);
1024 struct request_queue *q = bdev_get_queue(bdev);
1025 unsigned int inflight[2];
1028 blk_mq_in_flight_rw(q, bdev, inflight);
1030 part_in_flight_rw(bdev, inflight);
1032 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1035 static ssize_t disk_capability_show(struct device *dev,
1036 struct device_attribute *attr, char *buf)
1038 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1039 return sprintf(buf, "0\n");
1042 static ssize_t disk_alignment_offset_show(struct device *dev,
1043 struct device_attribute *attr,
1046 struct gendisk *disk = dev_to_disk(dev);
1048 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1051 static ssize_t disk_discard_alignment_show(struct device *dev,
1052 struct device_attribute *attr,
1055 struct gendisk *disk = dev_to_disk(dev);
1057 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1060 static ssize_t diskseq_show(struct device *dev,
1061 struct device_attribute *attr, char *buf)
1063 struct gendisk *disk = dev_to_disk(dev);
1065 return sprintf(buf, "%llu\n", disk->diskseq);
1068 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1069 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1070 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1071 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1072 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1073 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1074 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1075 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1076 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1077 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1078 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1079 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1080 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1082 #ifdef CONFIG_FAIL_MAKE_REQUEST
1083 ssize_t part_fail_show(struct device *dev,
1084 struct device_attribute *attr, char *buf)
1086 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1089 ssize_t part_fail_store(struct device *dev,
1090 struct device_attribute *attr,
1091 const char *buf, size_t count)
1095 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1096 dev_to_bdev(dev)->bd_make_it_fail = i;
1101 static struct device_attribute dev_attr_fail =
1102 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1103 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1105 #ifdef CONFIG_FAIL_IO_TIMEOUT
1106 static struct device_attribute dev_attr_fail_timeout =
1107 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1110 static struct attribute *disk_attrs[] = {
1111 &dev_attr_range.attr,
1112 &dev_attr_ext_range.attr,
1113 &dev_attr_removable.attr,
1114 &dev_attr_hidden.attr,
1116 &dev_attr_size.attr,
1117 &dev_attr_alignment_offset.attr,
1118 &dev_attr_discard_alignment.attr,
1119 &dev_attr_capability.attr,
1120 &dev_attr_stat.attr,
1121 &dev_attr_inflight.attr,
1122 &dev_attr_badblocks.attr,
1123 &dev_attr_events.attr,
1124 &dev_attr_events_async.attr,
1125 &dev_attr_events_poll_msecs.attr,
1126 &dev_attr_diskseq.attr,
1127 #ifdef CONFIG_FAIL_MAKE_REQUEST
1128 &dev_attr_fail.attr,
1130 #ifdef CONFIG_FAIL_IO_TIMEOUT
1131 &dev_attr_fail_timeout.attr,
1136 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1138 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1139 struct gendisk *disk = dev_to_disk(dev);
1141 if (a == &dev_attr_badblocks.attr && !disk->bb)
1146 static struct attribute_group disk_attr_group = {
1147 .attrs = disk_attrs,
1148 .is_visible = disk_visible,
1151 static const struct attribute_group *disk_attr_groups[] = {
1153 #ifdef CONFIG_BLK_DEV_IO_TRACE
1154 &blk_trace_attr_group,
1160 * disk_release - releases all allocated resources of the gendisk
1161 * @dev: the device representing this disk
1163 * This function releases all allocated resources of the gendisk.
1165 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1166 * assigned. Since the request_queue sits on top of the gendisk for these
1167 * drivers we also call blk_put_queue() for them, and we expect the
1168 * request_queue refcount to reach 0 at this point, and so the request_queue
1169 * will also be freed prior to the disk.
1171 * Context: can sleep
1173 static void disk_release(struct device *dev)
1175 struct gendisk *disk = dev_to_disk(dev);
1178 WARN_ON_ONCE(disk_live(disk));
1181 * To undo the all initialization from blk_mq_init_allocated_queue in
1182 * case of a probe failure where add_disk is never called we have to
1183 * call blk_mq_exit_queue here. We can't do this for the more common
1184 * teardown case (yet) as the tagset can be gone by the time the disk
1185 * is released once it was added.
1187 if (queue_is_mq(disk->queue) &&
1188 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1189 !test_bit(GD_ADDED, &disk->state))
1190 blk_mq_exit_queue(disk->queue);
1192 blkcg_exit_disk(disk);
1194 bioset_exit(&disk->bio_split);
1196 disk_release_events(disk);
1197 kfree(disk->random);
1198 disk_free_zone_bitmaps(disk);
1199 xa_destroy(&disk->part_tbl);
1201 disk->queue->disk = NULL;
1202 blk_put_queue(disk->queue);
1204 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1205 disk->fops->free_disk(disk);
1207 iput(disk->part0->bd_inode); /* frees the disk */
1210 static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1212 const struct gendisk *disk = dev_to_disk(dev);
1214 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1217 struct class block_class = {
1219 .dev_uevent = block_uevent,
1222 static char *block_devnode(const struct device *dev, umode_t *mode,
1223 kuid_t *uid, kgid_t *gid)
1225 struct gendisk *disk = dev_to_disk(dev);
1227 if (disk->fops->devnode)
1228 return disk->fops->devnode(disk, mode);
1232 const struct device_type disk_type = {
1234 .groups = disk_attr_groups,
1235 .release = disk_release,
1236 .devnode = block_devnode,
1239 #ifdef CONFIG_PROC_FS
1241 * aggregate disk stat collector. Uses the same stats that the sysfs
1242 * entries do, above, but makes them available through one seq_file.
1244 * The output looks suspiciously like /proc/partitions with a bunch of
1247 static int diskstats_show(struct seq_file *seqf, void *v)
1249 struct gendisk *gp = v;
1250 struct block_device *hd;
1251 unsigned int inflight;
1252 struct disk_stats stat;
1256 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1257 seq_puts(seqf, "major minor name"
1258 " rio rmerge rsect ruse wio wmerge "
1259 "wsect wuse running use aveq"
1264 xa_for_each(&gp->part_tbl, idx, hd) {
1265 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1267 if (queue_is_mq(gp->queue))
1268 inflight = blk_mq_in_flight(gp->queue, hd);
1270 inflight = part_in_flight(hd);
1274 update_io_ticks(hd, jiffies, true);
1277 part_stat_read_all(hd, &stat);
1278 seq_printf(seqf, "%4d %7d %pg "
1285 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1286 stat.ios[STAT_READ],
1287 stat.merges[STAT_READ],
1288 stat.sectors[STAT_READ],
1289 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1291 stat.ios[STAT_WRITE],
1292 stat.merges[STAT_WRITE],
1293 stat.sectors[STAT_WRITE],
1294 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1297 jiffies_to_msecs(stat.io_ticks),
1298 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1299 stat.nsecs[STAT_WRITE] +
1300 stat.nsecs[STAT_DISCARD] +
1301 stat.nsecs[STAT_FLUSH],
1303 stat.ios[STAT_DISCARD],
1304 stat.merges[STAT_DISCARD],
1305 stat.sectors[STAT_DISCARD],
1306 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1308 stat.ios[STAT_FLUSH],
1309 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1318 static const struct seq_operations diskstats_op = {
1319 .start = disk_seqf_start,
1320 .next = disk_seqf_next,
1321 .stop = disk_seqf_stop,
1322 .show = diskstats_show
1325 static int __init proc_genhd_init(void)
1327 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1328 proc_create_seq("partitions", 0, NULL, &partitions_op);
1331 module_init(proc_genhd_init);
1332 #endif /* CONFIG_PROC_FS */
1334 dev_t part_devt(struct gendisk *disk, u8 partno)
1336 struct block_device *part;
1340 part = xa_load(&disk->part_tbl, partno);
1342 devt = part->bd_dev;
1348 dev_t blk_lookup_devt(const char *name, int partno)
1350 dev_t devt = MKDEV(0, 0);
1351 struct class_dev_iter iter;
1354 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1355 while ((dev = class_dev_iter_next(&iter))) {
1356 struct gendisk *disk = dev_to_disk(dev);
1358 if (strcmp(dev_name(dev), name))
1361 if (partno < disk->minors) {
1362 /* We need to return the right devno, even
1363 * if the partition doesn't exist yet.
1365 devt = MKDEV(MAJOR(dev->devt),
1366 MINOR(dev->devt) + partno);
1368 devt = part_devt(disk, partno);
1373 class_dev_iter_exit(&iter);
1377 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1378 struct lock_class_key *lkclass)
1380 struct gendisk *disk;
1382 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1386 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1389 disk->bdi = bdi_alloc(node_id);
1391 goto out_free_bioset;
1393 /* bdev_alloc() might need the queue, set before the first call */
1396 disk->part0 = bdev_alloc(disk, 0);
1400 disk->node_id = node_id;
1401 mutex_init(&disk->open_mutex);
1402 xa_init(&disk->part_tbl);
1403 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1404 goto out_destroy_part_tbl;
1406 if (blkcg_init_disk(disk))
1407 goto out_erase_part0;
1409 rand_initialize_disk(disk);
1410 disk_to_dev(disk)->class = &block_class;
1411 disk_to_dev(disk)->type = &disk_type;
1412 device_initialize(disk_to_dev(disk));
1415 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1416 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1417 INIT_LIST_HEAD(&disk->slave_bdevs);
1422 xa_erase(&disk->part_tbl, 0);
1423 out_destroy_part_tbl:
1424 xa_destroy(&disk->part_tbl);
1425 disk->part0->bd_disk = NULL;
1426 iput(disk->part0->bd_inode);
1430 bioset_exit(&disk->bio_split);
1436 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1438 struct request_queue *q;
1439 struct gendisk *disk;
1441 q = blk_alloc_queue(node);
1445 disk = __alloc_disk_node(q, node, lkclass);
1450 set_bit(GD_OWNS_QUEUE, &disk->state);
1453 EXPORT_SYMBOL(__blk_alloc_disk);
1456 * put_disk - decrements the gendisk refcount
1457 * @disk: the struct gendisk to decrement the refcount for
1459 * This decrements the refcount for the struct gendisk. When this reaches 0
1460 * we'll have disk_release() called.
1462 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1463 * when handling probe errors (that is before add_disk() is called).
1465 * Context: Any context, but the last reference must not be dropped from
1468 void put_disk(struct gendisk *disk)
1471 put_device(disk_to_dev(disk));
1473 EXPORT_SYMBOL(put_disk);
1475 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1477 char event[] = "DISK_RO=1";
1478 char *envp[] = { event, NULL };
1482 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1486 * set_disk_ro - set a gendisk read-only
1487 * @disk: gendisk to operate on
1488 * @read_only: %true to set the disk read-only, %false set the disk read/write
1490 * This function is used to indicate whether a given disk device should have its
1491 * read-only flag set. set_disk_ro() is typically used by device drivers to
1492 * indicate whether the underlying physical device is write-protected.
1494 void set_disk_ro(struct gendisk *disk, bool read_only)
1497 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1500 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1503 set_disk_ro_uevent(disk, read_only);
1505 EXPORT_SYMBOL(set_disk_ro);
1507 void inc_diskseq(struct gendisk *disk)
1509 disk->diskseq = atomic64_inc_return(&diskseq);