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/genhd.h>
12 #include <linux/kdev_t.h>
13 #include <linux/kernel.h>
14 #include <linux/blkdev.h>
15 #include <linux/backing-dev.h>
16 #include <linux/init.h>
17 #include <linux/spinlock.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/slab.h>
21 #include <linux/kmod.h>
22 #include <linux/major.h>
23 #include <linux/mutex.h>
24 #include <linux/idr.h>
25 #include <linux/log2.h>
26 #include <linux/pm_runtime.h>
27 #include <linux/badblocks.h>
30 #include "blk-rq-qos.h"
32 static struct kobject *block_depr;
35 * Unique, monotonically increasing sequential number associated with block
36 * devices instances (i.e. incremented each time a device is attached).
37 * Associating uevents with block devices in userspace is difficult and racy:
38 * the uevent netlink socket is lossy, and on slow and overloaded systems has
39 * a very high latency.
40 * Block devices do not have exclusive owners in userspace, any process can set
41 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
42 * can be reused again and again).
43 * A userspace process setting up a block device and watching for its events
44 * cannot thus reliably tell whether an event relates to the device it just set
45 * up or another earlier instance with the same name.
46 * This sequential number allows userspace processes to solve this problem, and
47 * uniquely associate an uevent to the lifetime to a device.
49 static atomic64_t diskseq;
51 /* for extended dynamic devt allocation, currently only one major is used */
52 #define NR_EXT_DEVT (1 << MINORBITS)
53 static DEFINE_IDA(ext_devt_ida);
55 void set_capacity(struct gendisk *disk, sector_t sectors)
57 struct block_device *bdev = disk->part0;
59 spin_lock(&bdev->bd_size_lock);
60 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
61 spin_unlock(&bdev->bd_size_lock);
63 EXPORT_SYMBOL(set_capacity);
66 * Set disk capacity and notify if the size is not currently zero and will not
67 * be set to zero. Returns true if a uevent was sent, otherwise false.
69 bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
71 sector_t capacity = get_capacity(disk);
72 char *envp[] = { "RESIZE=1", NULL };
74 set_capacity(disk, size);
77 * Only print a message and send a uevent if the gendisk is user visible
78 * and alive. This avoids spamming the log and udev when setting the
79 * initial capacity during probing.
81 if (size == capacity ||
83 (disk->flags & GENHD_FL_HIDDEN))
86 pr_info("%s: detected capacity change from %lld to %lld\n",
87 disk->disk_name, capacity, size);
90 * Historically we did not send a uevent for changes to/from an empty
93 if (!capacity || !size)
95 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
98 EXPORT_SYMBOL_GPL(set_capacity_and_notify);
101 * Format the device name of the indicated block device into the supplied buffer
102 * and return a pointer to that same buffer for convenience.
104 * Note: do not use this in new code, use the %pg specifier to sprintf and
107 const char *bdevname(struct block_device *bdev, char *buf)
109 struct gendisk *hd = bdev->bd_disk;
110 int partno = bdev->bd_partno;
113 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
114 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
115 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
117 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
121 EXPORT_SYMBOL(bdevname);
123 static void part_stat_read_all(struct block_device *part,
124 struct disk_stats *stat)
128 memset(stat, 0, sizeof(struct disk_stats));
129 for_each_possible_cpu(cpu) {
130 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
133 for (group = 0; group < NR_STAT_GROUPS; group++) {
134 stat->nsecs[group] += ptr->nsecs[group];
135 stat->sectors[group] += ptr->sectors[group];
136 stat->ios[group] += ptr->ios[group];
137 stat->merges[group] += ptr->merges[group];
140 stat->io_ticks += ptr->io_ticks;
144 static unsigned int part_in_flight(struct block_device *part)
146 unsigned int inflight = 0;
149 for_each_possible_cpu(cpu) {
150 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
151 part_stat_local_read_cpu(part, in_flight[1], cpu);
153 if ((int)inflight < 0)
159 static void part_in_flight_rw(struct block_device *part,
160 unsigned int inflight[2])
166 for_each_possible_cpu(cpu) {
167 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
168 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
170 if ((int)inflight[0] < 0)
172 if ((int)inflight[1] < 0)
177 * Can be deleted altogether. Later.
180 #define BLKDEV_MAJOR_HASH_SIZE 255
181 static struct blk_major_name {
182 struct blk_major_name *next;
185 void (*probe)(dev_t devt);
186 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
187 static DEFINE_MUTEX(major_names_lock);
188 static DEFINE_SPINLOCK(major_names_spinlock);
190 /* index in the above - for now: assume no multimajor ranges */
191 static inline int major_to_index(unsigned major)
193 return major % BLKDEV_MAJOR_HASH_SIZE;
196 #ifdef CONFIG_PROC_FS
197 void blkdev_show(struct seq_file *seqf, off_t offset)
199 struct blk_major_name *dp;
201 spin_lock(&major_names_spinlock);
202 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
203 if (dp->major == offset)
204 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
205 spin_unlock(&major_names_spinlock);
207 #endif /* CONFIG_PROC_FS */
210 * __register_blkdev - register a new block device
212 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
213 * @major = 0, try to allocate any unused major number.
214 * @name: the name of the new block device as a zero terminated string
215 * @probe: allback that is called on access to any minor number of @major
217 * The @name must be unique within the system.
219 * The return value depends on the @major input parameter:
221 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
222 * then the function returns zero on success, or a negative error code
223 * - if any unused major number was requested with @major = 0 parameter
224 * then the return value is the allocated major number in range
225 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
227 * See Documentation/admin-guide/devices.txt for the list of allocated
230 * Use register_blkdev instead for any new code.
232 int __register_blkdev(unsigned int major, const char *name,
233 void (*probe)(dev_t devt))
235 struct blk_major_name **n, *p;
238 mutex_lock(&major_names_lock);
242 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
243 if (major_names[index] == NULL)
248 printk("%s: failed to get major for %s\n",
257 if (major >= BLKDEV_MAJOR_MAX) {
258 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
259 __func__, major, BLKDEV_MAJOR_MAX-1, name);
265 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
273 strlcpy(p->name, name, sizeof(p->name));
275 index = major_to_index(major);
277 spin_lock(&major_names_spinlock);
278 for (n = &major_names[index]; *n; n = &(*n)->next) {
279 if ((*n)->major == major)
286 spin_unlock(&major_names_spinlock);
289 printk("register_blkdev: cannot get major %u for %s\n",
294 mutex_unlock(&major_names_lock);
297 EXPORT_SYMBOL(__register_blkdev);
299 void unregister_blkdev(unsigned int major, const char *name)
301 struct blk_major_name **n;
302 struct blk_major_name *p = NULL;
303 int index = major_to_index(major);
305 mutex_lock(&major_names_lock);
306 spin_lock(&major_names_spinlock);
307 for (n = &major_names[index]; *n; n = &(*n)->next)
308 if ((*n)->major == major)
310 if (!*n || strcmp((*n)->name, name)) {
316 spin_unlock(&major_names_spinlock);
317 mutex_unlock(&major_names_lock);
321 EXPORT_SYMBOL(unregister_blkdev);
323 int blk_alloc_ext_minor(void)
327 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
333 void blk_free_ext_minor(unsigned int minor)
335 ida_free(&ext_devt_ida, minor);
338 static char *bdevt_str(dev_t devt, char *buf)
340 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
341 char tbuf[BDEVT_SIZE];
342 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
343 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
345 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
350 void disk_uevent(struct gendisk *disk, enum kobject_action action)
352 struct block_device *part;
356 xa_for_each(&disk->part_tbl, idx, part) {
357 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
359 if (!kobject_get_unless_zero(&part->bd_device.kobj))
363 kobject_uevent(bdev_kobj(part), action);
364 put_device(&part->bd_device);
369 EXPORT_SYMBOL_GPL(disk_uevent);
371 static void disk_scan_partitions(struct gendisk *disk)
373 struct block_device *bdev;
375 if (!get_capacity(disk) || !disk_part_scan_enabled(disk))
378 set_bit(GD_NEED_PART_SCAN, &disk->state);
379 bdev = blkdev_get_by_dev(disk_devt(disk), FMODE_READ, NULL);
381 blkdev_put(bdev, FMODE_READ);
385 * device_add_disk - add disk information to kernel list
386 * @parent: parent device for the disk
387 * @disk: per-device partitioning information
388 * @groups: Additional per-device sysfs groups
390 * This function registers the partitioning information in @disk
393 int device_add_disk(struct device *parent, struct gendisk *disk,
394 const struct attribute_group **groups)
397 struct device *ddev = disk_to_dev(disk);
401 * The disk queue should now be all set with enough information about
402 * the device for the elevator code to pick an adequate default
403 * elevator if one is needed, that is, for devices requesting queue
406 elevator_init_mq(disk->queue);
409 * If the driver provides an explicit major number it also must provide
410 * the number of minors numbers supported, and those will be used to
412 * Otherwise just allocate the device numbers for both the whole device
413 * and all partitions from the extended dev_t space.
416 if (WARN_ON(!disk->minors))
419 if (disk->minors > DISK_MAX_PARTS) {
420 pr_err("block: can't allocate more than %d partitions\n",
422 disk->minors = DISK_MAX_PARTS;
424 if (disk->first_minor + disk->minors > MINORMASK + 1)
427 if (WARN_ON(disk->minors))
430 ret = blk_alloc_ext_minor();
433 disk->major = BLOCK_EXT_MAJOR;
434 disk->first_minor = ret;
435 disk->flags |= GENHD_FL_EXT_DEVT;
438 /* delay uevents, until we scanned partition table */
439 dev_set_uevent_suppress(ddev, 1);
441 ddev->parent = parent;
442 ddev->groups = groups;
443 dev_set_name(ddev, "%s", disk->disk_name);
444 if (!(disk->flags & GENHD_FL_HIDDEN))
445 ddev->devt = MKDEV(disk->major, disk->first_minor);
446 ret = device_add(ddev);
448 goto out_free_ext_minor;
450 ret = disk_alloc_events(disk);
454 if (!sysfs_deprecated) {
455 ret = sysfs_create_link(block_depr, &ddev->kobj,
456 kobject_name(&ddev->kobj));
462 * avoid probable deadlock caused by allocating memory with
463 * GFP_KERNEL in runtime_resume callback of its all ancestor
466 pm_runtime_set_memalloc_noio(ddev, true);
468 ret = blk_integrity_add(disk);
470 goto out_del_block_link;
472 disk->part0->bd_holder_dir =
473 kobject_create_and_add("holders", &ddev->kobj);
474 if (!disk->part0->bd_holder_dir) {
476 goto out_del_integrity;
478 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
479 if (!disk->slave_dir) {
481 goto out_put_holder_dir;
484 ret = bd_register_pending_holders(disk);
486 goto out_put_slave_dir;
488 ret = blk_register_queue(disk);
490 goto out_put_slave_dir;
492 if (disk->flags & GENHD_FL_HIDDEN) {
494 * Don't let hidden disks show up in /proc/partitions,
495 * and don't bother scanning for partitions either.
497 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
498 disk->flags |= GENHD_FL_NO_PART_SCAN;
500 ret = bdi_register(disk->bdi, "%u:%u",
501 disk->major, disk->first_minor);
503 goto out_unregister_queue;
504 bdi_set_owner(disk->bdi, ddev);
505 ret = sysfs_create_link(&ddev->kobj,
506 &disk->bdi->dev->kobj, "bdi");
508 goto out_unregister_bdi;
510 bdev_add(disk->part0, ddev->devt);
511 disk_scan_partitions(disk);
514 * Announce the disk and partitions after all partitions are
515 * created. (for hidden disks uevents remain suppressed forever)
517 dev_set_uevent_suppress(ddev, 0);
518 disk_uevent(disk, KOBJ_ADD);
521 disk_update_readahead(disk);
522 disk_add_events(disk);
526 if (!(disk->flags & GENHD_FL_HIDDEN))
527 bdi_unregister(disk->bdi);
528 out_unregister_queue:
529 blk_unregister_queue(disk);
531 kobject_put(disk->slave_dir);
533 kobject_put(disk->part0->bd_holder_dir);
535 blk_integrity_del(disk);
537 if (!sysfs_deprecated)
538 sysfs_remove_link(block_depr, dev_name(ddev));
542 if (disk->major == BLOCK_EXT_MAJOR)
543 blk_free_ext_minor(disk->first_minor);
544 return WARN_ON_ONCE(ret); /* keep until all callers handle errors */
546 EXPORT_SYMBOL(device_add_disk);
549 * blk_mark_disk_dead - mark a disk as dead
550 * @disk: disk to mark as dead
552 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
555 void blk_mark_disk_dead(struct gendisk *disk)
557 set_bit(GD_DEAD, &disk->state);
558 blk_queue_start_drain(disk->queue);
560 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
563 * del_gendisk - remove the gendisk
564 * @disk: the struct gendisk to remove
566 * Removes the gendisk and all its associated resources. This deletes the
567 * partitions associated with the gendisk, and unregisters the associated
570 * This is the counter to the respective __device_add_disk() call.
572 * The final removal of the struct gendisk happens when its refcount reaches 0
573 * with put_disk(), which should be called after del_gendisk(), if
574 * __device_add_disk() was used.
576 * Drivers exist which depend on the release of the gendisk to be synchronous,
577 * it should not be deferred.
581 void del_gendisk(struct gendisk *disk)
583 struct request_queue *q = disk->queue;
587 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
590 blk_integrity_del(disk);
591 disk_del_events(disk);
593 mutex_lock(&disk->open_mutex);
594 remove_inode_hash(disk->part0->bd_inode);
595 blk_drop_partitions(disk);
596 mutex_unlock(&disk->open_mutex);
598 fsync_bdev(disk->part0);
599 __invalidate_device(disk->part0, true);
604 set_bit(GD_DEAD, &disk->state);
605 set_capacity(disk, 0);
608 * Prevent new I/O from crossing bio_queue_enter().
610 blk_queue_start_drain(q);
612 if (!(disk->flags & GENHD_FL_HIDDEN)) {
613 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
616 * Unregister bdi before releasing device numbers (as they can
617 * get reused and we'd get clashes in sysfs).
619 bdi_unregister(disk->bdi);
622 blk_unregister_queue(disk);
624 kobject_put(disk->part0->bd_holder_dir);
625 kobject_put(disk->slave_dir);
627 part_stat_set_all(disk->part0, 0);
628 disk->part0->bd_stamp = 0;
629 if (!sysfs_deprecated)
630 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
631 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
632 device_del(disk_to_dev(disk));
634 blk_mq_freeze_queue_wait(q);
638 blk_flush_integrity();
640 * Allow using passthrough request again after the queue is torn down.
642 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
643 __blk_mq_unfreeze_queue(q, true);
646 EXPORT_SYMBOL(del_gendisk);
648 /* sysfs access to bad-blocks list. */
649 static ssize_t disk_badblocks_show(struct device *dev,
650 struct device_attribute *attr,
653 struct gendisk *disk = dev_to_disk(dev);
656 return sprintf(page, "\n");
658 return badblocks_show(disk->bb, page, 0);
661 static ssize_t disk_badblocks_store(struct device *dev,
662 struct device_attribute *attr,
663 const char *page, size_t len)
665 struct gendisk *disk = dev_to_disk(dev);
670 return badblocks_store(disk->bb, page, len, 0);
673 void blk_request_module(dev_t devt)
675 unsigned int major = MAJOR(devt);
676 struct blk_major_name **n;
678 mutex_lock(&major_names_lock);
679 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
680 if ((*n)->major == major && (*n)->probe) {
682 mutex_unlock(&major_names_lock);
686 mutex_unlock(&major_names_lock);
688 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
689 /* Make old-style 2.4 aliases work */
690 request_module("block-major-%d", MAJOR(devt));
694 * print a full list of all partitions - intended for places where the root
695 * filesystem can't be mounted and thus to give the victim some idea of what
698 void __init printk_all_partitions(void)
700 struct class_dev_iter iter;
703 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
704 while ((dev = class_dev_iter_next(&iter))) {
705 struct gendisk *disk = dev_to_disk(dev);
706 struct block_device *part;
707 char devt_buf[BDEVT_SIZE];
711 * Don't show empty devices or things that have been
714 if (get_capacity(disk) == 0 ||
715 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
719 * Note, unlike /proc/partitions, I am showing the numbers in
720 * hex - the same format as the root= option takes.
723 xa_for_each(&disk->part_tbl, idx, part) {
724 if (!bdev_nr_sectors(part))
726 printk("%s%s %10llu %pg %s",
727 bdev_is_partition(part) ? " " : "",
728 bdevt_str(part->bd_dev, devt_buf),
729 bdev_nr_sectors(part) >> 1, part,
731 part->bd_meta_info->uuid : "");
732 if (bdev_is_partition(part))
734 else if (dev->parent && dev->parent->driver)
735 printk(" driver: %s\n",
736 dev->parent->driver->name);
738 printk(" (driver?)\n");
742 class_dev_iter_exit(&iter);
745 #ifdef CONFIG_PROC_FS
747 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
750 struct class_dev_iter *iter;
753 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
755 return ERR_PTR(-ENOMEM);
757 seqf->private = iter;
758 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
760 dev = class_dev_iter_next(iter);
765 return dev_to_disk(dev);
768 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
773 dev = class_dev_iter_next(seqf->private);
775 return dev_to_disk(dev);
780 static void disk_seqf_stop(struct seq_file *seqf, void *v)
782 struct class_dev_iter *iter = seqf->private;
784 /* stop is called even after start failed :-( */
786 class_dev_iter_exit(iter);
788 seqf->private = NULL;
792 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
796 p = disk_seqf_start(seqf, pos);
797 if (!IS_ERR_OR_NULL(p) && !*pos)
798 seq_puts(seqf, "major minor #blocks name\n\n");
802 static int show_partition(struct seq_file *seqf, void *v)
804 struct gendisk *sgp = v;
805 struct block_device *part;
808 /* Don't show non-partitionable removeable devices or empty devices */
809 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
810 (sgp->flags & GENHD_FL_REMOVABLE)))
812 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
816 xa_for_each(&sgp->part_tbl, idx, part) {
817 if (!bdev_nr_sectors(part))
819 seq_printf(seqf, "%4d %7d %10llu %pg\n",
820 MAJOR(part->bd_dev), MINOR(part->bd_dev),
821 bdev_nr_sectors(part) >> 1, part);
827 static const struct seq_operations partitions_op = {
828 .start = show_partition_start,
829 .next = disk_seqf_next,
830 .stop = disk_seqf_stop,
831 .show = show_partition
835 static int __init genhd_device_init(void)
839 block_class.dev_kobj = sysfs_dev_block_kobj;
840 error = class_register(&block_class);
845 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
847 /* create top-level block dir */
848 if (!sysfs_deprecated)
849 block_depr = kobject_create_and_add("block", NULL);
853 subsys_initcall(genhd_device_init);
855 static ssize_t disk_range_show(struct device *dev,
856 struct device_attribute *attr, char *buf)
858 struct gendisk *disk = dev_to_disk(dev);
860 return sprintf(buf, "%d\n", disk->minors);
863 static ssize_t disk_ext_range_show(struct device *dev,
864 struct device_attribute *attr, char *buf)
866 struct gendisk *disk = dev_to_disk(dev);
868 return sprintf(buf, "%d\n", disk_max_parts(disk));
871 static ssize_t disk_removable_show(struct device *dev,
872 struct device_attribute *attr, char *buf)
874 struct gendisk *disk = dev_to_disk(dev);
876 return sprintf(buf, "%d\n",
877 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
880 static ssize_t disk_hidden_show(struct device *dev,
881 struct device_attribute *attr, char *buf)
883 struct gendisk *disk = dev_to_disk(dev);
885 return sprintf(buf, "%d\n",
886 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
889 static ssize_t disk_ro_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", get_disk_ro(disk) ? 1 : 0);
897 ssize_t part_size_show(struct device *dev,
898 struct device_attribute *attr, char *buf)
900 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
903 ssize_t part_stat_show(struct device *dev,
904 struct device_attribute *attr, char *buf)
906 struct block_device *bdev = dev_to_bdev(dev);
907 struct request_queue *q = bdev->bd_disk->queue;
908 struct disk_stats stat;
909 unsigned int inflight;
911 part_stat_read_all(bdev, &stat);
913 inflight = blk_mq_in_flight(q, bdev);
915 inflight = part_in_flight(bdev);
918 "%8lu %8lu %8llu %8u "
919 "%8lu %8lu %8llu %8u "
921 "%8lu %8lu %8llu %8u "
925 stat.merges[STAT_READ],
926 (unsigned long long)stat.sectors[STAT_READ],
927 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
928 stat.ios[STAT_WRITE],
929 stat.merges[STAT_WRITE],
930 (unsigned long long)stat.sectors[STAT_WRITE],
931 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
933 jiffies_to_msecs(stat.io_ticks),
934 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
935 stat.nsecs[STAT_WRITE] +
936 stat.nsecs[STAT_DISCARD] +
937 stat.nsecs[STAT_FLUSH],
939 stat.ios[STAT_DISCARD],
940 stat.merges[STAT_DISCARD],
941 (unsigned long long)stat.sectors[STAT_DISCARD],
942 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
943 stat.ios[STAT_FLUSH],
944 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
947 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
950 struct block_device *bdev = dev_to_bdev(dev);
951 struct request_queue *q = bdev->bd_disk->queue;
952 unsigned int inflight[2];
955 blk_mq_in_flight_rw(q, bdev, inflight);
957 part_in_flight_rw(bdev, inflight);
959 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
962 static ssize_t disk_capability_show(struct device *dev,
963 struct device_attribute *attr, char *buf)
965 struct gendisk *disk = dev_to_disk(dev);
967 return sprintf(buf, "%x\n", disk->flags);
970 static ssize_t disk_alignment_offset_show(struct device *dev,
971 struct device_attribute *attr,
974 struct gendisk *disk = dev_to_disk(dev);
976 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
979 static ssize_t disk_discard_alignment_show(struct device *dev,
980 struct device_attribute *attr,
983 struct gendisk *disk = dev_to_disk(dev);
985 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
988 static ssize_t diskseq_show(struct device *dev,
989 struct device_attribute *attr, char *buf)
991 struct gendisk *disk = dev_to_disk(dev);
993 return sprintf(buf, "%llu\n", disk->diskseq);
996 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
997 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
998 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
999 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1000 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1001 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1002 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1003 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1004 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1005 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1006 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1007 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1008 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1010 #ifdef CONFIG_FAIL_MAKE_REQUEST
1011 ssize_t part_fail_show(struct device *dev,
1012 struct device_attribute *attr, char *buf)
1014 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1017 ssize_t part_fail_store(struct device *dev,
1018 struct device_attribute *attr,
1019 const char *buf, size_t count)
1023 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1024 dev_to_bdev(dev)->bd_make_it_fail = i;
1029 static struct device_attribute dev_attr_fail =
1030 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1031 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1033 #ifdef CONFIG_FAIL_IO_TIMEOUT
1034 static struct device_attribute dev_attr_fail_timeout =
1035 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1038 static struct attribute *disk_attrs[] = {
1039 &dev_attr_range.attr,
1040 &dev_attr_ext_range.attr,
1041 &dev_attr_removable.attr,
1042 &dev_attr_hidden.attr,
1044 &dev_attr_size.attr,
1045 &dev_attr_alignment_offset.attr,
1046 &dev_attr_discard_alignment.attr,
1047 &dev_attr_capability.attr,
1048 &dev_attr_stat.attr,
1049 &dev_attr_inflight.attr,
1050 &dev_attr_badblocks.attr,
1051 &dev_attr_events.attr,
1052 &dev_attr_events_async.attr,
1053 &dev_attr_events_poll_msecs.attr,
1054 &dev_attr_diskseq.attr,
1055 #ifdef CONFIG_FAIL_MAKE_REQUEST
1056 &dev_attr_fail.attr,
1058 #ifdef CONFIG_FAIL_IO_TIMEOUT
1059 &dev_attr_fail_timeout.attr,
1064 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1066 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1067 struct gendisk *disk = dev_to_disk(dev);
1069 if (a == &dev_attr_badblocks.attr && !disk->bb)
1074 static struct attribute_group disk_attr_group = {
1075 .attrs = disk_attrs,
1076 .is_visible = disk_visible,
1079 static const struct attribute_group *disk_attr_groups[] = {
1085 * disk_release - releases all allocated resources of the gendisk
1086 * @dev: the device representing this disk
1088 * This function releases all allocated resources of the gendisk.
1090 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1091 * assigned. Since the request_queue sits on top of the gendisk for these
1092 * drivers we also call blk_put_queue() for them, and we expect the
1093 * request_queue refcount to reach 0 at this point, and so the request_queue
1094 * will also be freed prior to the disk.
1096 * Context: can sleep
1098 static void disk_release(struct device *dev)
1100 struct gendisk *disk = dev_to_disk(dev);
1103 WARN_ON_ONCE(disk_live(disk));
1105 blk_mq_cancel_work_sync(disk->queue);
1107 disk_release_events(disk);
1108 kfree(disk->random);
1109 xa_destroy(&disk->part_tbl);
1110 disk->queue->disk = NULL;
1111 blk_put_queue(disk->queue);
1112 iput(disk->part0->bd_inode); /* frees the disk */
1115 static int block_uevent(struct device *dev, struct kobj_uevent_env *env)
1117 struct gendisk *disk = dev_to_disk(dev);
1119 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1122 struct class block_class = {
1124 .dev_uevent = block_uevent,
1127 static char *block_devnode(struct device *dev, umode_t *mode,
1128 kuid_t *uid, kgid_t *gid)
1130 struct gendisk *disk = dev_to_disk(dev);
1132 if (disk->fops->devnode)
1133 return disk->fops->devnode(disk, mode);
1137 const struct device_type disk_type = {
1139 .groups = disk_attr_groups,
1140 .release = disk_release,
1141 .devnode = block_devnode,
1144 #ifdef CONFIG_PROC_FS
1146 * aggregate disk stat collector. Uses the same stats that the sysfs
1147 * entries do, above, but makes them available through one seq_file.
1149 * The output looks suspiciously like /proc/partitions with a bunch of
1152 static int diskstats_show(struct seq_file *seqf, void *v)
1154 struct gendisk *gp = v;
1155 struct block_device *hd;
1156 unsigned int inflight;
1157 struct disk_stats stat;
1161 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1162 seq_puts(seqf, "major minor name"
1163 " rio rmerge rsect ruse wio wmerge "
1164 "wsect wuse running use aveq"
1169 xa_for_each(&gp->part_tbl, idx, hd) {
1170 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1172 part_stat_read_all(hd, &stat);
1173 if (queue_is_mq(gp->queue))
1174 inflight = blk_mq_in_flight(gp->queue, hd);
1176 inflight = part_in_flight(hd);
1178 seq_printf(seqf, "%4d %7d %pg "
1185 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1186 stat.ios[STAT_READ],
1187 stat.merges[STAT_READ],
1188 stat.sectors[STAT_READ],
1189 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1191 stat.ios[STAT_WRITE],
1192 stat.merges[STAT_WRITE],
1193 stat.sectors[STAT_WRITE],
1194 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1197 jiffies_to_msecs(stat.io_ticks),
1198 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1199 stat.nsecs[STAT_WRITE] +
1200 stat.nsecs[STAT_DISCARD] +
1201 stat.nsecs[STAT_FLUSH],
1203 stat.ios[STAT_DISCARD],
1204 stat.merges[STAT_DISCARD],
1205 stat.sectors[STAT_DISCARD],
1206 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1208 stat.ios[STAT_FLUSH],
1209 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1218 static const struct seq_operations diskstats_op = {
1219 .start = disk_seqf_start,
1220 .next = disk_seqf_next,
1221 .stop = disk_seqf_stop,
1222 .show = diskstats_show
1225 static int __init proc_genhd_init(void)
1227 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1228 proc_create_seq("partitions", 0, NULL, &partitions_op);
1231 module_init(proc_genhd_init);
1232 #endif /* CONFIG_PROC_FS */
1234 dev_t part_devt(struct gendisk *disk, u8 partno)
1236 struct block_device *part;
1240 part = xa_load(&disk->part_tbl, partno);
1242 devt = part->bd_dev;
1248 dev_t blk_lookup_devt(const char *name, int partno)
1250 dev_t devt = MKDEV(0, 0);
1251 struct class_dev_iter iter;
1254 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1255 while ((dev = class_dev_iter_next(&iter))) {
1256 struct gendisk *disk = dev_to_disk(dev);
1258 if (strcmp(dev_name(dev), name))
1261 if (partno < disk->minors) {
1262 /* We need to return the right devno, even
1263 * if the partition doesn't exist yet.
1265 devt = MKDEV(MAJOR(dev->devt),
1266 MINOR(dev->devt) + partno);
1268 devt = part_devt(disk, partno);
1273 class_dev_iter_exit(&iter);
1277 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1278 struct lock_class_key *lkclass)
1280 struct gendisk *disk;
1282 if (!blk_get_queue(q))
1285 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1289 disk->bdi = bdi_alloc(node_id);
1293 disk->part0 = bdev_alloc(disk, 0);
1297 disk->node_id = node_id;
1298 mutex_init(&disk->open_mutex);
1299 xa_init(&disk->part_tbl);
1300 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1301 goto out_destroy_part_tbl;
1303 rand_initialize_disk(disk);
1304 disk_to_dev(disk)->class = &block_class;
1305 disk_to_dev(disk)->type = &disk_type;
1306 device_initialize(disk_to_dev(disk));
1310 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1311 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1312 INIT_LIST_HEAD(&disk->slave_bdevs);
1316 out_destroy_part_tbl:
1317 xa_destroy(&disk->part_tbl);
1318 disk->part0->bd_disk = NULL;
1319 iput(disk->part0->bd_inode);
1328 EXPORT_SYMBOL(__alloc_disk_node);
1330 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1332 struct request_queue *q;
1333 struct gendisk *disk;
1335 q = blk_alloc_queue(node);
1339 disk = __alloc_disk_node(q, node, lkclass);
1341 blk_cleanup_queue(q);
1346 EXPORT_SYMBOL(__blk_alloc_disk);
1349 * put_disk - decrements the gendisk refcount
1350 * @disk: the struct gendisk to decrement the refcount for
1352 * This decrements the refcount for the struct gendisk. When this reaches 0
1353 * we'll have disk_release() called.
1355 * Context: Any context, but the last reference must not be dropped from
1358 void put_disk(struct gendisk *disk)
1361 put_device(disk_to_dev(disk));
1363 EXPORT_SYMBOL(put_disk);
1366 * blk_cleanup_disk - shutdown a gendisk allocated by blk_alloc_disk
1367 * @disk: gendisk to shutdown
1369 * Mark the queue hanging off @disk DYING, drain all pending requests, then mark
1370 * the queue DEAD, destroy and put it and the gendisk structure.
1372 * Context: can sleep
1374 void blk_cleanup_disk(struct gendisk *disk)
1376 blk_cleanup_queue(disk->queue);
1379 EXPORT_SYMBOL(blk_cleanup_disk);
1381 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1383 char event[] = "DISK_RO=1";
1384 char *envp[] = { event, NULL };
1388 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1392 * set_disk_ro - set a gendisk read-only
1393 * @disk: gendisk to operate on
1394 * @read_only: %true to set the disk read-only, %false set the disk read/write
1396 * This function is used to indicate whether a given disk device should have its
1397 * read-only flag set. set_disk_ro() is typically used by device drivers to
1398 * indicate whether the underlying physical device is write-protected.
1400 void set_disk_ro(struct gendisk *disk, bool read_only)
1403 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1406 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1409 set_disk_ro_uevent(disk, read_only);
1411 EXPORT_SYMBOL(set_disk_ro);
1413 int bdev_read_only(struct block_device *bdev)
1415 return bdev->bd_read_only || get_disk_ro(bdev->bd_disk);
1417 EXPORT_SYMBOL(bdev_read_only);
1419 void inc_diskseq(struct gendisk *disk)
1421 disk->diskseq = atomic64_inc_return(&diskseq);