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);
105 * Format the device name of the indicated block device into the supplied buffer
106 * and return a pointer to that same buffer for convenience.
108 * Note: do not use this in new code, use the %pg specifier to sprintf and
111 const char *bdevname(struct block_device *bdev, char *buf)
113 struct gendisk *hd = bdev->bd_disk;
114 int partno = bdev->bd_partno;
117 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
118 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
119 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
121 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
125 EXPORT_SYMBOL(bdevname);
127 static void part_stat_read_all(struct block_device *part,
128 struct disk_stats *stat)
132 memset(stat, 0, sizeof(struct disk_stats));
133 for_each_possible_cpu(cpu) {
134 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
137 for (group = 0; group < NR_STAT_GROUPS; group++) {
138 stat->nsecs[group] += ptr->nsecs[group];
139 stat->sectors[group] += ptr->sectors[group];
140 stat->ios[group] += ptr->ios[group];
141 stat->merges[group] += ptr->merges[group];
144 stat->io_ticks += ptr->io_ticks;
148 static unsigned int part_in_flight(struct block_device *part)
150 unsigned int inflight = 0;
153 for_each_possible_cpu(cpu) {
154 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
155 part_stat_local_read_cpu(part, in_flight[1], cpu);
157 if ((int)inflight < 0)
163 static void part_in_flight_rw(struct block_device *part,
164 unsigned int inflight[2])
170 for_each_possible_cpu(cpu) {
171 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
172 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
174 if ((int)inflight[0] < 0)
176 if ((int)inflight[1] < 0)
181 * Can be deleted altogether. Later.
184 #define BLKDEV_MAJOR_HASH_SIZE 255
185 static struct blk_major_name {
186 struct blk_major_name *next;
189 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
190 void (*probe)(dev_t devt);
192 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
193 static DEFINE_MUTEX(major_names_lock);
194 static DEFINE_SPINLOCK(major_names_spinlock);
196 /* index in the above - for now: assume no multimajor ranges */
197 static inline int major_to_index(unsigned major)
199 return major % BLKDEV_MAJOR_HASH_SIZE;
202 #ifdef CONFIG_PROC_FS
203 void blkdev_show(struct seq_file *seqf, off_t offset)
205 struct blk_major_name *dp;
207 spin_lock(&major_names_spinlock);
208 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
209 if (dp->major == offset)
210 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
211 spin_unlock(&major_names_spinlock);
213 #endif /* CONFIG_PROC_FS */
216 * __register_blkdev - register a new block device
218 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
219 * @major = 0, try to allocate any unused major number.
220 * @name: the name of the new block device as a zero terminated string
221 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
222 * pre-created device node is accessed. When a probe call uses
223 * add_disk() and it fails the driver must cleanup resources. This
224 * interface may soon be removed.
226 * The @name must be unique within the system.
228 * The return value depends on the @major input parameter:
230 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
231 * then the function returns zero on success, or a negative error code
232 * - if any unused major number was requested with @major = 0 parameter
233 * then the return value is the allocated major number in range
234 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
236 * See Documentation/admin-guide/devices.txt for the list of allocated
239 * Use register_blkdev instead for any new code.
241 int __register_blkdev(unsigned int major, const char *name,
242 void (*probe)(dev_t devt))
244 struct blk_major_name **n, *p;
247 mutex_lock(&major_names_lock);
251 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
252 if (major_names[index] == NULL)
257 printk("%s: failed to get major for %s\n",
266 if (major >= BLKDEV_MAJOR_MAX) {
267 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
268 __func__, major, BLKDEV_MAJOR_MAX-1, name);
274 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
281 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
284 strlcpy(p->name, name, sizeof(p->name));
286 index = major_to_index(major);
288 spin_lock(&major_names_spinlock);
289 for (n = &major_names[index]; *n; n = &(*n)->next) {
290 if ((*n)->major == major)
297 spin_unlock(&major_names_spinlock);
300 printk("register_blkdev: cannot get major %u for %s\n",
305 mutex_unlock(&major_names_lock);
308 EXPORT_SYMBOL(__register_blkdev);
310 void unregister_blkdev(unsigned int major, const char *name)
312 struct blk_major_name **n;
313 struct blk_major_name *p = NULL;
314 int index = major_to_index(major);
316 mutex_lock(&major_names_lock);
317 spin_lock(&major_names_spinlock);
318 for (n = &major_names[index]; *n; n = &(*n)->next)
319 if ((*n)->major == major)
321 if (!*n || strcmp((*n)->name, name)) {
327 spin_unlock(&major_names_spinlock);
328 mutex_unlock(&major_names_lock);
332 EXPORT_SYMBOL(unregister_blkdev);
334 int blk_alloc_ext_minor(void)
338 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
344 void blk_free_ext_minor(unsigned int minor)
346 ida_free(&ext_devt_ida, minor);
349 static char *bdevt_str(dev_t devt, char *buf)
351 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
352 char tbuf[BDEVT_SIZE];
353 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
354 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
356 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
361 void disk_uevent(struct gendisk *disk, enum kobject_action action)
363 struct block_device *part;
367 xa_for_each(&disk->part_tbl, idx, part) {
368 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
370 if (!kobject_get_unless_zero(&part->bd_device.kobj))
374 kobject_uevent(bdev_kobj(part), action);
375 put_device(&part->bd_device);
380 EXPORT_SYMBOL_GPL(disk_uevent);
382 int disk_scan_partitions(struct gendisk *disk, fmode_t mode)
384 struct block_device *bdev;
386 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
388 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
390 if (disk->open_partitions)
393 set_bit(GD_NEED_PART_SCAN, &disk->state);
394 bdev = blkdev_get_by_dev(disk_devt(disk), mode, NULL);
396 return PTR_ERR(bdev);
397 blkdev_put(bdev, mode);
402 * device_add_disk - add disk information to kernel list
403 * @parent: parent device for the disk
404 * @disk: per-device partitioning information
405 * @groups: Additional per-device sysfs groups
407 * This function registers the partitioning information in @disk
410 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
411 const struct attribute_group **groups)
414 struct device *ddev = disk_to_dev(disk);
417 /* Only makes sense for bio-based to set ->poll_bio */
418 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
422 * The disk queue should now be all set with enough information about
423 * the device for the elevator code to pick an adequate default
424 * elevator if one is needed, that is, for devices requesting queue
427 elevator_init_mq(disk->queue);
430 * If the driver provides an explicit major number it also must provide
431 * the number of minors numbers supported, and those will be used to
433 * Otherwise just allocate the device numbers for both the whole device
434 * and all partitions from the extended dev_t space.
437 if (WARN_ON(!disk->minors))
440 if (disk->minors > DISK_MAX_PARTS) {
441 pr_err("block: can't allocate more than %d partitions\n",
443 disk->minors = DISK_MAX_PARTS;
445 if (disk->first_minor + disk->minors > MINORMASK + 1)
448 if (WARN_ON(disk->minors))
451 ret = blk_alloc_ext_minor();
454 disk->major = BLOCK_EXT_MAJOR;
455 disk->first_minor = ret;
458 /* delay uevents, until we scanned partition table */
459 dev_set_uevent_suppress(ddev, 1);
461 ddev->parent = parent;
462 ddev->groups = groups;
463 dev_set_name(ddev, "%s", disk->disk_name);
464 if (!(disk->flags & GENHD_FL_HIDDEN))
465 ddev->devt = MKDEV(disk->major, disk->first_minor);
466 ret = device_add(ddev);
468 goto out_free_ext_minor;
470 ret = disk_alloc_events(disk);
474 if (!sysfs_deprecated) {
475 ret = sysfs_create_link(block_depr, &ddev->kobj,
476 kobject_name(&ddev->kobj));
482 * avoid probable deadlock caused by allocating memory with
483 * GFP_KERNEL in runtime_resume callback of its all ancestor
486 pm_runtime_set_memalloc_noio(ddev, true);
488 ret = blk_integrity_add(disk);
490 goto out_del_block_link;
492 disk->part0->bd_holder_dir =
493 kobject_create_and_add("holders", &ddev->kobj);
494 if (!disk->part0->bd_holder_dir) {
496 goto out_del_integrity;
498 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
499 if (!disk->slave_dir) {
501 goto out_put_holder_dir;
504 ret = bd_register_pending_holders(disk);
506 goto out_put_slave_dir;
508 ret = blk_register_queue(disk);
510 goto out_put_slave_dir;
512 if (!(disk->flags & GENHD_FL_HIDDEN)) {
513 ret = bdi_register(disk->bdi, "%u:%u",
514 disk->major, disk->first_minor);
516 goto out_unregister_queue;
517 bdi_set_owner(disk->bdi, ddev);
518 ret = sysfs_create_link(&ddev->kobj,
519 &disk->bdi->dev->kobj, "bdi");
521 goto out_unregister_bdi;
523 bdev_add(disk->part0, ddev->devt);
524 if (get_capacity(disk))
525 disk_scan_partitions(disk, FMODE_READ);
528 * Announce the disk and partitions after all partitions are
529 * created. (for hidden disks uevents remain suppressed forever)
531 dev_set_uevent_suppress(ddev, 0);
532 disk_uevent(disk, KOBJ_ADD);
535 disk_update_readahead(disk);
536 disk_add_events(disk);
537 set_bit(GD_ADDED, &disk->state);
541 if (!(disk->flags & GENHD_FL_HIDDEN))
542 bdi_unregister(disk->bdi);
543 out_unregister_queue:
544 blk_unregister_queue(disk);
546 kobject_put(disk->slave_dir);
548 kobject_put(disk->part0->bd_holder_dir);
550 blk_integrity_del(disk);
552 if (!sysfs_deprecated)
553 sysfs_remove_link(block_depr, dev_name(ddev));
557 if (disk->major == BLOCK_EXT_MAJOR)
558 blk_free_ext_minor(disk->first_minor);
561 EXPORT_SYMBOL(device_add_disk);
564 * blk_mark_disk_dead - mark a disk as dead
565 * @disk: disk to mark as dead
567 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
570 void blk_mark_disk_dead(struct gendisk *disk)
572 set_bit(GD_DEAD, &disk->state);
573 blk_queue_start_drain(disk->queue);
575 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
578 * del_gendisk - remove the gendisk
579 * @disk: the struct gendisk to remove
581 * Removes the gendisk and all its associated resources. This deletes the
582 * partitions associated with the gendisk, and unregisters the associated
585 * This is the counter to the respective __device_add_disk() call.
587 * The final removal of the struct gendisk happens when its refcount reaches 0
588 * with put_disk(), which should be called after del_gendisk(), if
589 * __device_add_disk() was used.
591 * Drivers exist which depend on the release of the gendisk to be synchronous,
592 * it should not be deferred.
596 void del_gendisk(struct gendisk *disk)
598 struct request_queue *q = disk->queue;
602 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
605 blk_integrity_del(disk);
606 disk_del_events(disk);
608 mutex_lock(&disk->open_mutex);
609 remove_inode_hash(disk->part0->bd_inode);
610 blk_drop_partitions(disk);
611 mutex_unlock(&disk->open_mutex);
613 fsync_bdev(disk->part0);
614 __invalidate_device(disk->part0, true);
619 set_bit(GD_DEAD, &disk->state);
620 set_capacity(disk, 0);
623 * Prevent new I/O from crossing bio_queue_enter().
625 blk_queue_start_drain(q);
627 if (!(disk->flags & GENHD_FL_HIDDEN)) {
628 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
631 * Unregister bdi before releasing device numbers (as they can
632 * get reused and we'd get clashes in sysfs).
634 bdi_unregister(disk->bdi);
637 blk_unregister_queue(disk);
639 kobject_put(disk->part0->bd_holder_dir);
640 kobject_put(disk->slave_dir);
642 part_stat_set_all(disk->part0, 0);
643 disk->part0->bd_stamp = 0;
644 if (!sysfs_deprecated)
645 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
646 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
647 device_del(disk_to_dev(disk));
649 blk_mq_freeze_queue_wait(q);
651 blk_throtl_cancel_bios(disk->queue);
654 blk_flush_integrity();
656 * Allow using passthrough request again after the queue is torn down.
658 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
659 __blk_mq_unfreeze_queue(q, true);
662 EXPORT_SYMBOL(del_gendisk);
665 * invalidate_disk - invalidate the disk
666 * @disk: the struct gendisk to invalidate
668 * A helper to invalidates the disk. It will clean the disk's associated
669 * buffer/page caches and reset its internal states so that the disk
670 * can be reused by the drivers.
674 void invalidate_disk(struct gendisk *disk)
676 struct block_device *bdev = disk->part0;
678 invalidate_bdev(bdev);
679 bdev->bd_inode->i_mapping->wb_err = 0;
680 set_capacity(disk, 0);
682 EXPORT_SYMBOL(invalidate_disk);
684 /* sysfs access to bad-blocks list. */
685 static ssize_t disk_badblocks_show(struct device *dev,
686 struct device_attribute *attr,
689 struct gendisk *disk = dev_to_disk(dev);
692 return sprintf(page, "\n");
694 return badblocks_show(disk->bb, page, 0);
697 static ssize_t disk_badblocks_store(struct device *dev,
698 struct device_attribute *attr,
699 const char *page, size_t len)
701 struct gendisk *disk = dev_to_disk(dev);
706 return badblocks_store(disk->bb, page, len, 0);
709 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
710 void blk_request_module(dev_t devt)
712 unsigned int major = MAJOR(devt);
713 struct blk_major_name **n;
715 mutex_lock(&major_names_lock);
716 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
717 if ((*n)->major == major && (*n)->probe) {
719 mutex_unlock(&major_names_lock);
723 mutex_unlock(&major_names_lock);
725 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
726 /* Make old-style 2.4 aliases work */
727 request_module("block-major-%d", MAJOR(devt));
729 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
732 * print a full list of all partitions - intended for places where the root
733 * filesystem can't be mounted and thus to give the victim some idea of what
736 void __init printk_all_partitions(void)
738 struct class_dev_iter iter;
741 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
742 while ((dev = class_dev_iter_next(&iter))) {
743 struct gendisk *disk = dev_to_disk(dev);
744 struct block_device *part;
745 char devt_buf[BDEVT_SIZE];
749 * Don't show empty devices or things that have been
752 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
756 * Note, unlike /proc/partitions, I am showing the numbers in
757 * hex - the same format as the root= option takes.
760 xa_for_each(&disk->part_tbl, idx, part) {
761 if (!bdev_nr_sectors(part))
763 printk("%s%s %10llu %pg %s",
764 bdev_is_partition(part) ? " " : "",
765 bdevt_str(part->bd_dev, devt_buf),
766 bdev_nr_sectors(part) >> 1, part,
768 part->bd_meta_info->uuid : "");
769 if (bdev_is_partition(part))
771 else if (dev->parent && dev->parent->driver)
772 printk(" driver: %s\n",
773 dev->parent->driver->name);
775 printk(" (driver?)\n");
779 class_dev_iter_exit(&iter);
782 #ifdef CONFIG_PROC_FS
784 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
787 struct class_dev_iter *iter;
790 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
792 return ERR_PTR(-ENOMEM);
794 seqf->private = iter;
795 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
797 dev = class_dev_iter_next(iter);
802 return dev_to_disk(dev);
805 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
810 dev = class_dev_iter_next(seqf->private);
812 return dev_to_disk(dev);
817 static void disk_seqf_stop(struct seq_file *seqf, void *v)
819 struct class_dev_iter *iter = seqf->private;
821 /* stop is called even after start failed :-( */
823 class_dev_iter_exit(iter);
825 seqf->private = NULL;
829 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
833 p = disk_seqf_start(seqf, pos);
834 if (!IS_ERR_OR_NULL(p) && !*pos)
835 seq_puts(seqf, "major minor #blocks name\n\n");
839 static int show_partition(struct seq_file *seqf, void *v)
841 struct gendisk *sgp = v;
842 struct block_device *part;
845 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
849 xa_for_each(&sgp->part_tbl, idx, part) {
850 if (!bdev_nr_sectors(part))
852 seq_printf(seqf, "%4d %7d %10llu %pg\n",
853 MAJOR(part->bd_dev), MINOR(part->bd_dev),
854 bdev_nr_sectors(part) >> 1, part);
860 static const struct seq_operations partitions_op = {
861 .start = show_partition_start,
862 .next = disk_seqf_next,
863 .stop = disk_seqf_stop,
864 .show = show_partition
868 static int __init genhd_device_init(void)
872 block_class.dev_kobj = sysfs_dev_block_kobj;
873 error = class_register(&block_class);
878 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
880 /* create top-level block dir */
881 if (!sysfs_deprecated)
882 block_depr = kobject_create_and_add("block", NULL);
886 subsys_initcall(genhd_device_init);
888 static ssize_t disk_range_show(struct device *dev,
889 struct device_attribute *attr, char *buf)
891 struct gendisk *disk = dev_to_disk(dev);
893 return sprintf(buf, "%d\n", disk->minors);
896 static ssize_t disk_ext_range_show(struct device *dev,
897 struct device_attribute *attr, char *buf)
899 struct gendisk *disk = dev_to_disk(dev);
901 return sprintf(buf, "%d\n",
902 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
905 static ssize_t disk_removable_show(struct device *dev,
906 struct device_attribute *attr, char *buf)
908 struct gendisk *disk = dev_to_disk(dev);
910 return sprintf(buf, "%d\n",
911 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
914 static ssize_t disk_hidden_show(struct device *dev,
915 struct device_attribute *attr, char *buf)
917 struct gendisk *disk = dev_to_disk(dev);
919 return sprintf(buf, "%d\n",
920 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
923 static ssize_t disk_ro_show(struct device *dev,
924 struct device_attribute *attr, char *buf)
926 struct gendisk *disk = dev_to_disk(dev);
928 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
931 ssize_t part_size_show(struct device *dev,
932 struct device_attribute *attr, char *buf)
934 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
937 ssize_t part_stat_show(struct device *dev,
938 struct device_attribute *attr, char *buf)
940 struct block_device *bdev = dev_to_bdev(dev);
941 struct request_queue *q = bdev_get_queue(bdev);
942 struct disk_stats stat;
943 unsigned int inflight;
946 inflight = blk_mq_in_flight(q, bdev);
948 inflight = part_in_flight(bdev);
952 update_io_ticks(bdev, jiffies, true);
955 part_stat_read_all(bdev, &stat);
957 "%8lu %8lu %8llu %8u "
958 "%8lu %8lu %8llu %8u "
960 "%8lu %8lu %8llu %8u "
964 stat.merges[STAT_READ],
965 (unsigned long long)stat.sectors[STAT_READ],
966 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
967 stat.ios[STAT_WRITE],
968 stat.merges[STAT_WRITE],
969 (unsigned long long)stat.sectors[STAT_WRITE],
970 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
972 jiffies_to_msecs(stat.io_ticks),
973 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
974 stat.nsecs[STAT_WRITE] +
975 stat.nsecs[STAT_DISCARD] +
976 stat.nsecs[STAT_FLUSH],
978 stat.ios[STAT_DISCARD],
979 stat.merges[STAT_DISCARD],
980 (unsigned long long)stat.sectors[STAT_DISCARD],
981 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
982 stat.ios[STAT_FLUSH],
983 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
986 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
989 struct block_device *bdev = dev_to_bdev(dev);
990 struct request_queue *q = bdev_get_queue(bdev);
991 unsigned int inflight[2];
994 blk_mq_in_flight_rw(q, bdev, inflight);
996 part_in_flight_rw(bdev, inflight);
998 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1001 static ssize_t disk_capability_show(struct device *dev,
1002 struct device_attribute *attr, char *buf)
1004 struct gendisk *disk = dev_to_disk(dev);
1006 return sprintf(buf, "%x\n", disk->flags);
1009 static ssize_t disk_alignment_offset_show(struct device *dev,
1010 struct device_attribute *attr,
1013 struct gendisk *disk = dev_to_disk(dev);
1015 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1018 static ssize_t disk_discard_alignment_show(struct device *dev,
1019 struct device_attribute *attr,
1022 struct gendisk *disk = dev_to_disk(dev);
1024 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1027 static ssize_t diskseq_show(struct device *dev,
1028 struct device_attribute *attr, char *buf)
1030 struct gendisk *disk = dev_to_disk(dev);
1032 return sprintf(buf, "%llu\n", disk->diskseq);
1035 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1036 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1037 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1038 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1039 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1040 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1041 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1042 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1043 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1044 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1045 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1046 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1047 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1049 #ifdef CONFIG_FAIL_MAKE_REQUEST
1050 ssize_t part_fail_show(struct device *dev,
1051 struct device_attribute *attr, char *buf)
1053 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1056 ssize_t part_fail_store(struct device *dev,
1057 struct device_attribute *attr,
1058 const char *buf, size_t count)
1062 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1063 dev_to_bdev(dev)->bd_make_it_fail = i;
1068 static struct device_attribute dev_attr_fail =
1069 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1070 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1072 #ifdef CONFIG_FAIL_IO_TIMEOUT
1073 static struct device_attribute dev_attr_fail_timeout =
1074 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1077 static struct attribute *disk_attrs[] = {
1078 &dev_attr_range.attr,
1079 &dev_attr_ext_range.attr,
1080 &dev_attr_removable.attr,
1081 &dev_attr_hidden.attr,
1083 &dev_attr_size.attr,
1084 &dev_attr_alignment_offset.attr,
1085 &dev_attr_discard_alignment.attr,
1086 &dev_attr_capability.attr,
1087 &dev_attr_stat.attr,
1088 &dev_attr_inflight.attr,
1089 &dev_attr_badblocks.attr,
1090 &dev_attr_events.attr,
1091 &dev_attr_events_async.attr,
1092 &dev_attr_events_poll_msecs.attr,
1093 &dev_attr_diskseq.attr,
1094 #ifdef CONFIG_FAIL_MAKE_REQUEST
1095 &dev_attr_fail.attr,
1097 #ifdef CONFIG_FAIL_IO_TIMEOUT
1098 &dev_attr_fail_timeout.attr,
1103 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1105 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1106 struct gendisk *disk = dev_to_disk(dev);
1108 if (a == &dev_attr_badblocks.attr && !disk->bb)
1113 static struct attribute_group disk_attr_group = {
1114 .attrs = disk_attrs,
1115 .is_visible = disk_visible,
1118 static const struct attribute_group *disk_attr_groups[] = {
1123 static void disk_release_mq(struct request_queue *q)
1125 blk_mq_cancel_work_sync(q);
1128 * There can't be any non non-passthrough bios in flight here, but
1129 * requests stay around longer, including passthrough ones so we
1130 * still need to freeze the queue here.
1132 blk_mq_freeze_queue(q);
1135 * Since the I/O scheduler exit code may access cgroup information,
1136 * perform I/O scheduler exit before disassociating from the block
1137 * cgroup controller.
1140 mutex_lock(&q->sysfs_lock);
1142 mutex_unlock(&q->sysfs_lock);
1145 __blk_mq_unfreeze_queue(q, true);
1149 * disk_release - releases all allocated resources of the gendisk
1150 * @dev: the device representing this disk
1152 * This function releases all allocated resources of the gendisk.
1154 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1155 * assigned. Since the request_queue sits on top of the gendisk for these
1156 * drivers we also call blk_put_queue() for them, and we expect the
1157 * request_queue refcount to reach 0 at this point, and so the request_queue
1158 * will also be freed prior to the disk.
1160 * Context: can sleep
1162 static void disk_release(struct device *dev)
1164 struct gendisk *disk = dev_to_disk(dev);
1167 WARN_ON_ONCE(disk_live(disk));
1169 if (queue_is_mq(disk->queue))
1170 disk_release_mq(disk->queue);
1172 blkcg_exit_queue(disk->queue);
1174 disk_release_events(disk);
1175 kfree(disk->random);
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(struct device *dev, struct kobj_uevent_env *env)
1189 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(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 dev_t blk_lookup_devt(const char *name, int partno)
1327 dev_t devt = MKDEV(0, 0);
1328 struct class_dev_iter iter;
1331 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1332 while ((dev = class_dev_iter_next(&iter))) {
1333 struct gendisk *disk = dev_to_disk(dev);
1335 if (strcmp(dev_name(dev), name))
1338 if (partno < disk->minors) {
1339 /* We need to return the right devno, even
1340 * if the partition doesn't exist yet.
1342 devt = MKDEV(MAJOR(dev->devt),
1343 MINOR(dev->devt) + partno);
1345 devt = part_devt(disk, partno);
1350 class_dev_iter_exit(&iter);
1354 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1355 struct lock_class_key *lkclass)
1357 struct gendisk *disk;
1359 if (!blk_get_queue(q))
1362 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1366 disk->bdi = bdi_alloc(node_id);
1370 /* bdev_alloc() might need the queue, set before the first call */
1373 disk->part0 = bdev_alloc(disk, 0);
1377 disk->node_id = node_id;
1378 mutex_init(&disk->open_mutex);
1379 xa_init(&disk->part_tbl);
1380 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1381 goto out_destroy_part_tbl;
1383 if (blkcg_init_queue(q))
1384 goto out_erase_part0;
1386 rand_initialize_disk(disk);
1387 disk_to_dev(disk)->class = &block_class;
1388 disk_to_dev(disk)->type = &disk_type;
1389 device_initialize(disk_to_dev(disk));
1392 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1393 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1394 INIT_LIST_HEAD(&disk->slave_bdevs);
1399 xa_erase(&disk->part_tbl, 0);
1400 out_destroy_part_tbl:
1401 xa_destroy(&disk->part_tbl);
1402 disk->part0->bd_disk = NULL;
1403 iput(disk->part0->bd_inode);
1412 EXPORT_SYMBOL(__alloc_disk_node);
1414 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1416 struct request_queue *q;
1417 struct gendisk *disk;
1419 q = blk_alloc_queue(node, false);
1423 disk = __alloc_disk_node(q, node, lkclass);
1425 blk_cleanup_queue(q);
1430 EXPORT_SYMBOL(__blk_alloc_disk);
1433 * put_disk - decrements the gendisk refcount
1434 * @disk: the struct gendisk to decrement the refcount for
1436 * This decrements the refcount for the struct gendisk. When this reaches 0
1437 * we'll have disk_release() called.
1439 * Context: Any context, but the last reference must not be dropped from
1442 void put_disk(struct gendisk *disk)
1445 put_device(disk_to_dev(disk));
1447 EXPORT_SYMBOL(put_disk);
1450 * blk_cleanup_disk - shutdown a gendisk allocated by blk_alloc_disk
1451 * @disk: gendisk to shutdown
1453 * Mark the queue hanging off @disk DYING, drain all pending requests, then mark
1454 * the queue DEAD, destroy and put it and the gendisk structure.
1456 * Context: can sleep
1458 void blk_cleanup_disk(struct gendisk *disk)
1460 blk_cleanup_queue(disk->queue);
1463 EXPORT_SYMBOL(blk_cleanup_disk);
1465 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1467 char event[] = "DISK_RO=1";
1468 char *envp[] = { event, NULL };
1472 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1476 * set_disk_ro - set a gendisk read-only
1477 * @disk: gendisk to operate on
1478 * @read_only: %true to set the disk read-only, %false set the disk read/write
1480 * This function is used to indicate whether a given disk device should have its
1481 * read-only flag set. set_disk_ro() is typically used by device drivers to
1482 * indicate whether the underlying physical device is write-protected.
1484 void set_disk_ro(struct gendisk *disk, bool read_only)
1487 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1490 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1493 set_disk_ro_uevent(disk, read_only);
1495 EXPORT_SYMBOL(set_disk_ro);
1497 void inc_diskseq(struct gendisk *disk)
1499 disk->diskseq = atomic64_inc_return(&diskseq);