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;
363 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
365 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
367 if (disk->open_partitions)
370 set_bit(GD_NEED_PART_SCAN, &disk->state);
371 bdev = blkdev_get_by_dev(disk_devt(disk), mode, NULL);
373 return PTR_ERR(bdev);
374 blkdev_put(bdev, mode);
379 * device_add_disk - add disk information to kernel list
380 * @parent: parent device for the disk
381 * @disk: per-device partitioning information
382 * @groups: Additional per-device sysfs groups
384 * This function registers the partitioning information in @disk
387 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
388 const struct attribute_group **groups)
391 struct device *ddev = disk_to_dev(disk);
394 /* Only makes sense for bio-based to set ->poll_bio */
395 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
399 * The disk queue should now be all set with enough information about
400 * the device for the elevator code to pick an adequate default
401 * elevator if one is needed, that is, for devices requesting queue
404 elevator_init_mq(disk->queue);
407 * If the driver provides an explicit major number it also must provide
408 * the number of minors numbers supported, and those will be used to
410 * Otherwise just allocate the device numbers for both the whole device
411 * and all partitions from the extended dev_t space.
414 if (WARN_ON(!disk->minors))
417 if (disk->minors > DISK_MAX_PARTS) {
418 pr_err("block: can't allocate more than %d partitions\n",
420 disk->minors = DISK_MAX_PARTS;
422 if (disk->first_minor + disk->minors > MINORMASK + 1)
425 if (WARN_ON(disk->minors))
428 ret = blk_alloc_ext_minor();
431 disk->major = BLOCK_EXT_MAJOR;
432 disk->first_minor = ret;
435 /* delay uevents, until we scanned partition table */
436 dev_set_uevent_suppress(ddev, 1);
438 ddev->parent = parent;
439 ddev->groups = groups;
440 dev_set_name(ddev, "%s", disk->disk_name);
441 if (!(disk->flags & GENHD_FL_HIDDEN))
442 ddev->devt = MKDEV(disk->major, disk->first_minor);
443 ret = device_add(ddev);
445 goto out_free_ext_minor;
447 ret = disk_alloc_events(disk);
451 if (!sysfs_deprecated) {
452 ret = sysfs_create_link(block_depr, &ddev->kobj,
453 kobject_name(&ddev->kobj));
459 * avoid probable deadlock caused by allocating memory with
460 * GFP_KERNEL in runtime_resume callback of its all ancestor
463 pm_runtime_set_memalloc_noio(ddev, true);
465 ret = blk_integrity_add(disk);
467 goto out_del_block_link;
469 disk->part0->bd_holder_dir =
470 kobject_create_and_add("holders", &ddev->kobj);
471 if (!disk->part0->bd_holder_dir) {
473 goto out_del_integrity;
475 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
476 if (!disk->slave_dir) {
478 goto out_put_holder_dir;
481 ret = bd_register_pending_holders(disk);
483 goto out_put_slave_dir;
485 ret = blk_register_queue(disk);
487 goto out_put_slave_dir;
489 if (!(disk->flags & GENHD_FL_HIDDEN)) {
490 ret = bdi_register(disk->bdi, "%u:%u",
491 disk->major, disk->first_minor);
493 goto out_unregister_queue;
494 bdi_set_owner(disk->bdi, ddev);
495 ret = sysfs_create_link(&ddev->kobj,
496 &disk->bdi->dev->kobj, "bdi");
498 goto out_unregister_bdi;
500 bdev_add(disk->part0, ddev->devt);
501 if (get_capacity(disk))
502 disk_scan_partitions(disk, FMODE_READ);
505 * Announce the disk and partitions after all partitions are
506 * created. (for hidden disks uevents remain suppressed forever)
508 dev_set_uevent_suppress(ddev, 0);
509 disk_uevent(disk, KOBJ_ADD);
512 * Even if the block_device for a hidden gendisk is not
513 * registered, it needs to have a valid bd_dev so that the
514 * freeing of the dynamic major works.
516 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
519 disk_update_readahead(disk);
520 disk_add_events(disk);
521 set_bit(GD_ADDED, &disk->state);
525 if (!(disk->flags & GENHD_FL_HIDDEN))
526 bdi_unregister(disk->bdi);
527 out_unregister_queue:
528 blk_unregister_queue(disk);
530 kobject_put(disk->slave_dir);
532 kobject_put(disk->part0->bd_holder_dir);
534 blk_integrity_del(disk);
536 if (!sysfs_deprecated)
537 sysfs_remove_link(block_depr, dev_name(ddev));
541 if (disk->major == BLOCK_EXT_MAJOR)
542 blk_free_ext_minor(disk->first_minor);
545 EXPORT_SYMBOL(device_add_disk);
548 * blk_mark_disk_dead - mark a disk as dead
549 * @disk: disk to mark as dead
551 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
554 void blk_mark_disk_dead(struct gendisk *disk)
556 set_bit(GD_DEAD, &disk->state);
557 blk_queue_start_drain(disk->queue);
559 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
562 * del_gendisk - remove the gendisk
563 * @disk: the struct gendisk to remove
565 * Removes the gendisk and all its associated resources. This deletes the
566 * partitions associated with the gendisk, and unregisters the associated
569 * This is the counter to the respective __device_add_disk() call.
571 * The final removal of the struct gendisk happens when its refcount reaches 0
572 * with put_disk(), which should be called after del_gendisk(), if
573 * __device_add_disk() was used.
575 * Drivers exist which depend on the release of the gendisk to be synchronous,
576 * it should not be deferred.
580 void del_gendisk(struct gendisk *disk)
582 struct request_queue *q = disk->queue;
586 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
589 blk_integrity_del(disk);
590 disk_del_events(disk);
592 mutex_lock(&disk->open_mutex);
593 remove_inode_hash(disk->part0->bd_inode);
594 blk_drop_partitions(disk);
595 mutex_unlock(&disk->open_mutex);
597 fsync_bdev(disk->part0);
598 __invalidate_device(disk->part0, true);
603 set_bit(GD_DEAD, &disk->state);
604 if (test_bit(GD_OWNS_QUEUE, &disk->state))
605 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
606 set_capacity(disk, 0);
609 * Prevent new I/O from crossing bio_queue_enter().
611 blk_queue_start_drain(q);
612 blk_mq_freeze_queue_wait(q);
614 if (!(disk->flags & GENHD_FL_HIDDEN)) {
615 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
618 * Unregister bdi before releasing device numbers (as they can
619 * get reused and we'd get clashes in sysfs).
621 bdi_unregister(disk->bdi);
624 blk_unregister_queue(disk);
626 kobject_put(disk->part0->bd_holder_dir);
627 kobject_put(disk->slave_dir);
629 part_stat_set_all(disk->part0, 0);
630 disk->part0->bd_stamp = 0;
631 if (!sysfs_deprecated)
632 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
633 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
634 device_del(disk_to_dev(disk));
636 blk_throtl_cancel_bios(disk);
639 blk_flush_integrity();
640 blk_mq_cancel_work_sync(q);
642 blk_mq_quiesce_queue(q);
644 mutex_lock(&q->sysfs_lock);
646 mutex_unlock(&q->sysfs_lock);
649 blk_mq_unquiesce_queue(q);
652 * If the disk does not own the queue, allow using passthrough requests
653 * again. Else leave the queue frozen to fail all I/O.
655 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
656 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
657 __blk_mq_unfreeze_queue(q, true);
660 blk_mq_exit_queue(q);
663 EXPORT_SYMBOL(del_gendisk);
666 * invalidate_disk - invalidate the disk
667 * @disk: the struct gendisk to invalidate
669 * A helper to invalidates the disk. It will clean the disk's associated
670 * buffer/page caches and reset its internal states so that the disk
671 * can be reused by the drivers.
675 void invalidate_disk(struct gendisk *disk)
677 struct block_device *bdev = disk->part0;
679 invalidate_bdev(bdev);
680 bdev->bd_inode->i_mapping->wb_err = 0;
681 set_capacity(disk, 0);
683 EXPORT_SYMBOL(invalidate_disk);
685 /* sysfs access to bad-blocks list. */
686 static ssize_t disk_badblocks_show(struct device *dev,
687 struct device_attribute *attr,
690 struct gendisk *disk = dev_to_disk(dev);
693 return sprintf(page, "\n");
695 return badblocks_show(disk->bb, page, 0);
698 static ssize_t disk_badblocks_store(struct device *dev,
699 struct device_attribute *attr,
700 const char *page, size_t len)
702 struct gendisk *disk = dev_to_disk(dev);
707 return badblocks_store(disk->bb, page, len, 0);
710 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
711 void blk_request_module(dev_t devt)
713 unsigned int major = MAJOR(devt);
714 struct blk_major_name **n;
716 mutex_lock(&major_names_lock);
717 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
718 if ((*n)->major == major && (*n)->probe) {
720 mutex_unlock(&major_names_lock);
724 mutex_unlock(&major_names_lock);
726 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
727 /* Make old-style 2.4 aliases work */
728 request_module("block-major-%d", MAJOR(devt));
730 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
733 * print a full list of all partitions - intended for places where the root
734 * filesystem can't be mounted and thus to give the victim some idea of what
737 void __init printk_all_partitions(void)
739 struct class_dev_iter iter;
742 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
743 while ((dev = class_dev_iter_next(&iter))) {
744 struct gendisk *disk = dev_to_disk(dev);
745 struct block_device *part;
746 char devt_buf[BDEVT_SIZE];
750 * Don't show empty devices or things that have been
753 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
757 * Note, unlike /proc/partitions, I am showing the numbers in
758 * hex - the same format as the root= option takes.
761 xa_for_each(&disk->part_tbl, idx, part) {
762 if (!bdev_nr_sectors(part))
764 printk("%s%s %10llu %pg %s",
765 bdev_is_partition(part) ? " " : "",
766 bdevt_str(part->bd_dev, devt_buf),
767 bdev_nr_sectors(part) >> 1, part,
769 part->bd_meta_info->uuid : "");
770 if (bdev_is_partition(part))
772 else if (dev->parent && dev->parent->driver)
773 printk(" driver: %s\n",
774 dev->parent->driver->name);
776 printk(" (driver?)\n");
780 class_dev_iter_exit(&iter);
783 #ifdef CONFIG_PROC_FS
785 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
788 struct class_dev_iter *iter;
791 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
793 return ERR_PTR(-ENOMEM);
795 seqf->private = iter;
796 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
798 dev = class_dev_iter_next(iter);
803 return dev_to_disk(dev);
806 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
811 dev = class_dev_iter_next(seqf->private);
813 return dev_to_disk(dev);
818 static void disk_seqf_stop(struct seq_file *seqf, void *v)
820 struct class_dev_iter *iter = seqf->private;
822 /* stop is called even after start failed :-( */
824 class_dev_iter_exit(iter);
826 seqf->private = NULL;
830 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
834 p = disk_seqf_start(seqf, pos);
835 if (!IS_ERR_OR_NULL(p) && !*pos)
836 seq_puts(seqf, "major minor #blocks name\n\n");
840 static int show_partition(struct seq_file *seqf, void *v)
842 struct gendisk *sgp = v;
843 struct block_device *part;
846 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
850 xa_for_each(&sgp->part_tbl, idx, part) {
851 if (!bdev_nr_sectors(part))
853 seq_printf(seqf, "%4d %7d %10llu %pg\n",
854 MAJOR(part->bd_dev), MINOR(part->bd_dev),
855 bdev_nr_sectors(part) >> 1, part);
861 static const struct seq_operations partitions_op = {
862 .start = show_partition_start,
863 .next = disk_seqf_next,
864 .stop = disk_seqf_stop,
865 .show = show_partition
869 static int __init genhd_device_init(void)
873 block_class.dev_kobj = sysfs_dev_block_kobj;
874 error = class_register(&block_class);
879 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
881 /* create top-level block dir */
882 if (!sysfs_deprecated)
883 block_depr = kobject_create_and_add("block", NULL);
887 subsys_initcall(genhd_device_init);
889 static ssize_t disk_range_show(struct device *dev,
890 struct device_attribute *attr, char *buf)
892 struct gendisk *disk = dev_to_disk(dev);
894 return sprintf(buf, "%d\n", disk->minors);
897 static ssize_t disk_ext_range_show(struct device *dev,
898 struct device_attribute *attr, char *buf)
900 struct gendisk *disk = dev_to_disk(dev);
902 return sprintf(buf, "%d\n",
903 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
906 static ssize_t disk_removable_show(struct device *dev,
907 struct device_attribute *attr, char *buf)
909 struct gendisk *disk = dev_to_disk(dev);
911 return sprintf(buf, "%d\n",
912 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
915 static ssize_t disk_hidden_show(struct device *dev,
916 struct device_attribute *attr, char *buf)
918 struct gendisk *disk = dev_to_disk(dev);
920 return sprintf(buf, "%d\n",
921 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
924 static ssize_t disk_ro_show(struct device *dev,
925 struct device_attribute *attr, char *buf)
927 struct gendisk *disk = dev_to_disk(dev);
929 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
932 ssize_t part_size_show(struct device *dev,
933 struct device_attribute *attr, char *buf)
935 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
938 ssize_t part_stat_show(struct device *dev,
939 struct device_attribute *attr, char *buf)
941 struct block_device *bdev = dev_to_bdev(dev);
942 struct request_queue *q = bdev_get_queue(bdev);
943 struct disk_stats stat;
944 unsigned int inflight;
947 inflight = blk_mq_in_flight(q, bdev);
949 inflight = part_in_flight(bdev);
953 update_io_ticks(bdev, jiffies, true);
956 part_stat_read_all(bdev, &stat);
958 "%8lu %8lu %8llu %8u "
959 "%8lu %8lu %8llu %8u "
961 "%8lu %8lu %8llu %8u "
965 stat.merges[STAT_READ],
966 (unsigned long long)stat.sectors[STAT_READ],
967 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
968 stat.ios[STAT_WRITE],
969 stat.merges[STAT_WRITE],
970 (unsigned long long)stat.sectors[STAT_WRITE],
971 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
973 jiffies_to_msecs(stat.io_ticks),
974 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
975 stat.nsecs[STAT_WRITE] +
976 stat.nsecs[STAT_DISCARD] +
977 stat.nsecs[STAT_FLUSH],
979 stat.ios[STAT_DISCARD],
980 stat.merges[STAT_DISCARD],
981 (unsigned long long)stat.sectors[STAT_DISCARD],
982 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
983 stat.ios[STAT_FLUSH],
984 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
987 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
990 struct block_device *bdev = dev_to_bdev(dev);
991 struct request_queue *q = bdev_get_queue(bdev);
992 unsigned int inflight[2];
995 blk_mq_in_flight_rw(q, bdev, inflight);
997 part_in_flight_rw(bdev, inflight);
999 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1002 static ssize_t disk_capability_show(struct device *dev,
1003 struct device_attribute *attr, char *buf)
1005 struct gendisk *disk = dev_to_disk(dev);
1007 return sprintf(buf, "%x\n", disk->flags);
1010 static ssize_t disk_alignment_offset_show(struct device *dev,
1011 struct device_attribute *attr,
1014 struct gendisk *disk = dev_to_disk(dev);
1016 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1019 static ssize_t disk_discard_alignment_show(struct device *dev,
1020 struct device_attribute *attr,
1023 struct gendisk *disk = dev_to_disk(dev);
1025 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1028 static ssize_t diskseq_show(struct device *dev,
1029 struct device_attribute *attr, char *buf)
1031 struct gendisk *disk = dev_to_disk(dev);
1033 return sprintf(buf, "%llu\n", disk->diskseq);
1036 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1037 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1038 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1039 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1040 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1041 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1042 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1043 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1044 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1045 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1046 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1047 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1048 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1050 #ifdef CONFIG_FAIL_MAKE_REQUEST
1051 ssize_t part_fail_show(struct device *dev,
1052 struct device_attribute *attr, char *buf)
1054 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1057 ssize_t part_fail_store(struct device *dev,
1058 struct device_attribute *attr,
1059 const char *buf, size_t count)
1063 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1064 dev_to_bdev(dev)->bd_make_it_fail = i;
1069 static struct device_attribute dev_attr_fail =
1070 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1071 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1073 #ifdef CONFIG_FAIL_IO_TIMEOUT
1074 static struct device_attribute dev_attr_fail_timeout =
1075 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1078 static struct attribute *disk_attrs[] = {
1079 &dev_attr_range.attr,
1080 &dev_attr_ext_range.attr,
1081 &dev_attr_removable.attr,
1082 &dev_attr_hidden.attr,
1084 &dev_attr_size.attr,
1085 &dev_attr_alignment_offset.attr,
1086 &dev_attr_discard_alignment.attr,
1087 &dev_attr_capability.attr,
1088 &dev_attr_stat.attr,
1089 &dev_attr_inflight.attr,
1090 &dev_attr_badblocks.attr,
1091 &dev_attr_events.attr,
1092 &dev_attr_events_async.attr,
1093 &dev_attr_events_poll_msecs.attr,
1094 &dev_attr_diskseq.attr,
1095 #ifdef CONFIG_FAIL_MAKE_REQUEST
1096 &dev_attr_fail.attr,
1098 #ifdef CONFIG_FAIL_IO_TIMEOUT
1099 &dev_attr_fail_timeout.attr,
1104 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1106 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1107 struct gendisk *disk = dev_to_disk(dev);
1109 if (a == &dev_attr_badblocks.attr && !disk->bb)
1114 static struct attribute_group disk_attr_group = {
1115 .attrs = disk_attrs,
1116 .is_visible = disk_visible,
1119 static const struct attribute_group *disk_attr_groups[] = {
1121 #ifdef CONFIG_BLK_DEV_IO_TRACE
1122 &blk_trace_attr_group,
1128 * disk_release - releases all allocated resources of the gendisk
1129 * @dev: the device representing this disk
1131 * This function releases all allocated resources of the gendisk.
1133 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1134 * assigned. Since the request_queue sits on top of the gendisk for these
1135 * drivers we also call blk_put_queue() for them, and we expect the
1136 * request_queue refcount to reach 0 at this point, and so the request_queue
1137 * will also be freed prior to the disk.
1139 * Context: can sleep
1141 static void disk_release(struct device *dev)
1143 struct gendisk *disk = dev_to_disk(dev);
1146 WARN_ON_ONCE(disk_live(disk));
1149 * To undo the all initialization from blk_mq_init_allocated_queue in
1150 * case of a probe failure where add_disk is never called we have to
1151 * call blk_mq_exit_queue here. We can't do this for the more common
1152 * teardown case (yet) as the tagset can be gone by the time the disk
1153 * is released once it was added.
1155 if (queue_is_mq(disk->queue) &&
1156 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1157 !test_bit(GD_ADDED, &disk->state))
1158 blk_mq_exit_queue(disk->queue);
1160 blkcg_exit_disk(disk);
1162 bioset_exit(&disk->bio_split);
1164 disk_release_events(disk);
1165 kfree(disk->random);
1166 disk_free_zone_bitmaps(disk);
1167 xa_destroy(&disk->part_tbl);
1169 disk->queue->disk = NULL;
1170 blk_put_queue(disk->queue);
1172 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1173 disk->fops->free_disk(disk);
1175 iput(disk->part0->bd_inode); /* frees the disk */
1178 static int block_uevent(struct device *dev, struct kobj_uevent_env *env)
1180 struct gendisk *disk = dev_to_disk(dev);
1182 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1185 struct class block_class = {
1187 .dev_uevent = block_uevent,
1190 static char *block_devnode(struct device *dev, umode_t *mode,
1191 kuid_t *uid, kgid_t *gid)
1193 struct gendisk *disk = dev_to_disk(dev);
1195 if (disk->fops->devnode)
1196 return disk->fops->devnode(disk, mode);
1200 const struct device_type disk_type = {
1202 .groups = disk_attr_groups,
1203 .release = disk_release,
1204 .devnode = block_devnode,
1207 #ifdef CONFIG_PROC_FS
1209 * aggregate disk stat collector. Uses the same stats that the sysfs
1210 * entries do, above, but makes them available through one seq_file.
1212 * The output looks suspiciously like /proc/partitions with a bunch of
1215 static int diskstats_show(struct seq_file *seqf, void *v)
1217 struct gendisk *gp = v;
1218 struct block_device *hd;
1219 unsigned int inflight;
1220 struct disk_stats stat;
1224 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1225 seq_puts(seqf, "major minor name"
1226 " rio rmerge rsect ruse wio wmerge "
1227 "wsect wuse running use aveq"
1232 xa_for_each(&gp->part_tbl, idx, hd) {
1233 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1235 if (queue_is_mq(gp->queue))
1236 inflight = blk_mq_in_flight(gp->queue, hd);
1238 inflight = part_in_flight(hd);
1242 update_io_ticks(hd, jiffies, true);
1245 part_stat_read_all(hd, &stat);
1246 seq_printf(seqf, "%4d %7d %pg "
1253 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1254 stat.ios[STAT_READ],
1255 stat.merges[STAT_READ],
1256 stat.sectors[STAT_READ],
1257 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1259 stat.ios[STAT_WRITE],
1260 stat.merges[STAT_WRITE],
1261 stat.sectors[STAT_WRITE],
1262 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1265 jiffies_to_msecs(stat.io_ticks),
1266 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1267 stat.nsecs[STAT_WRITE] +
1268 stat.nsecs[STAT_DISCARD] +
1269 stat.nsecs[STAT_FLUSH],
1271 stat.ios[STAT_DISCARD],
1272 stat.merges[STAT_DISCARD],
1273 stat.sectors[STAT_DISCARD],
1274 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1276 stat.ios[STAT_FLUSH],
1277 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1286 static const struct seq_operations diskstats_op = {
1287 .start = disk_seqf_start,
1288 .next = disk_seqf_next,
1289 .stop = disk_seqf_stop,
1290 .show = diskstats_show
1293 static int __init proc_genhd_init(void)
1295 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1296 proc_create_seq("partitions", 0, NULL, &partitions_op);
1299 module_init(proc_genhd_init);
1300 #endif /* CONFIG_PROC_FS */
1302 dev_t part_devt(struct gendisk *disk, u8 partno)
1304 struct block_device *part;
1308 part = xa_load(&disk->part_tbl, partno);
1310 devt = part->bd_dev;
1316 dev_t blk_lookup_devt(const char *name, int partno)
1318 dev_t devt = MKDEV(0, 0);
1319 struct class_dev_iter iter;
1322 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1323 while ((dev = class_dev_iter_next(&iter))) {
1324 struct gendisk *disk = dev_to_disk(dev);
1326 if (strcmp(dev_name(dev), name))
1329 if (partno < disk->minors) {
1330 /* We need to return the right devno, even
1331 * if the partition doesn't exist yet.
1333 devt = MKDEV(MAJOR(dev->devt),
1334 MINOR(dev->devt) + partno);
1336 devt = part_devt(disk, partno);
1341 class_dev_iter_exit(&iter);
1345 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1346 struct lock_class_key *lkclass)
1348 struct gendisk *disk;
1350 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1354 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1357 disk->bdi = bdi_alloc(node_id);
1359 goto out_free_bioset;
1361 /* bdev_alloc() might need the queue, set before the first call */
1364 disk->part0 = bdev_alloc(disk, 0);
1368 disk->node_id = node_id;
1369 mutex_init(&disk->open_mutex);
1370 xa_init(&disk->part_tbl);
1371 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1372 goto out_destroy_part_tbl;
1374 if (blkcg_init_disk(disk))
1375 goto out_erase_part0;
1377 rand_initialize_disk(disk);
1378 disk_to_dev(disk)->class = &block_class;
1379 disk_to_dev(disk)->type = &disk_type;
1380 device_initialize(disk_to_dev(disk));
1383 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1384 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1385 INIT_LIST_HEAD(&disk->slave_bdevs);
1390 xa_erase(&disk->part_tbl, 0);
1391 out_destroy_part_tbl:
1392 xa_destroy(&disk->part_tbl);
1393 disk->part0->bd_disk = NULL;
1394 iput(disk->part0->bd_inode);
1398 bioset_exit(&disk->bio_split);
1404 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1406 struct request_queue *q;
1407 struct gendisk *disk;
1409 q = blk_alloc_queue(node, false);
1413 disk = __alloc_disk_node(q, node, lkclass);
1418 set_bit(GD_OWNS_QUEUE, &disk->state);
1421 EXPORT_SYMBOL(__blk_alloc_disk);
1424 * put_disk - decrements the gendisk refcount
1425 * @disk: the struct gendisk to decrement the refcount for
1427 * This decrements the refcount for the struct gendisk. When this reaches 0
1428 * we'll have disk_release() called.
1430 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1431 * when handling probe errors (that is before add_disk() is called).
1433 * Context: Any context, but the last reference must not be dropped from
1436 void put_disk(struct gendisk *disk)
1439 put_device(disk_to_dev(disk));
1441 EXPORT_SYMBOL(put_disk);
1443 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1445 char event[] = "DISK_RO=1";
1446 char *envp[] = { event, NULL };
1450 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1454 * set_disk_ro - set a gendisk read-only
1455 * @disk: gendisk to operate on
1456 * @read_only: %true to set the disk read-only, %false set the disk read/write
1458 * This function is used to indicate whether a given disk device should have its
1459 * read-only flag set. set_disk_ro() is typically used by device drivers to
1460 * indicate whether the underlying physical device is write-protected.
1462 void set_disk_ro(struct gendisk *disk, bool read_only)
1465 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1468 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1471 set_disk_ro_uevent(disk, read_only);
1473 EXPORT_SYMBOL(set_disk_ro);
1475 void inc_diskseq(struct gendisk *disk)
1477 disk->diskseq = atomic64_inc_return(&diskseq);