1 // SPDX-License-Identifier: GPL-2.0
6 #include <linux/module.h>
7 #include <linux/ctype.h>
9 #include <linux/genhd.h>
10 #include <linux/kdev_t.h>
11 #include <linux/kernel.h>
12 #include <linux/blkdev.h>
13 #include <linux/backing-dev.h>
14 #include <linux/init.h>
15 #include <linux/spinlock.h>
16 #include <linux/proc_fs.h>
17 #include <linux/seq_file.h>
18 #include <linux/slab.h>
19 #include <linux/kmod.h>
20 #include <linux/kobj_map.h>
21 #include <linux/mutex.h>
22 #include <linux/idr.h>
23 #include <linux/log2.h>
24 #include <linux/pm_runtime.h>
25 #include <linux/badblocks.h>
29 static DEFINE_MUTEX(block_class_lock);
30 static struct kobject *block_depr;
32 /* for extended dynamic devt allocation, currently only one major is used */
33 #define NR_EXT_DEVT (1 << MINORBITS)
35 /* For extended devt allocation. ext_devt_lock prevents look up
36 * results from going away underneath its user.
38 static DEFINE_SPINLOCK(ext_devt_lock);
39 static DEFINE_IDR(ext_devt_idr);
41 static void disk_check_events(struct disk_events *ev,
42 unsigned int *clearing_ptr);
43 static void disk_alloc_events(struct gendisk *disk);
44 static void disk_add_events(struct gendisk *disk);
45 static void disk_del_events(struct gendisk *disk);
46 static void disk_release_events(struct gendisk *disk);
49 * Set disk capacity and notify if the size is not currently
50 * zero and will not be set to zero
52 void set_capacity_revalidate_and_notify(struct gendisk *disk, sector_t size,
55 sector_t capacity = get_capacity(disk);
57 set_capacity(disk, size);
59 revalidate_disk_size(disk, true);
61 if (capacity != size && capacity != 0 && size != 0) {
62 char *envp[] = { "RESIZE=1", NULL };
64 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
68 EXPORT_SYMBOL_GPL(set_capacity_revalidate_and_notify);
71 * Format the device name of the indicated disk into the supplied buffer and
72 * return a pointer to that same buffer for convenience.
74 char *disk_name(struct gendisk *hd, int partno, char *buf)
77 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
78 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
79 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
81 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
86 const char *bdevname(struct block_device *bdev, char *buf)
88 return disk_name(bdev->bd_disk, bdev->bd_partno, buf);
90 EXPORT_SYMBOL(bdevname);
92 static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
96 memset(stat, 0, sizeof(struct disk_stats));
97 for_each_possible_cpu(cpu) {
98 struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
101 for (group = 0; group < NR_STAT_GROUPS; group++) {
102 stat->nsecs[group] += ptr->nsecs[group];
103 stat->sectors[group] += ptr->sectors[group];
104 stat->ios[group] += ptr->ios[group];
105 stat->merges[group] += ptr->merges[group];
108 stat->io_ticks += ptr->io_ticks;
112 static unsigned int part_in_flight(struct hd_struct *part)
114 unsigned int inflight = 0;
117 for_each_possible_cpu(cpu) {
118 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
119 part_stat_local_read_cpu(part, in_flight[1], cpu);
121 if ((int)inflight < 0)
127 static void part_in_flight_rw(struct hd_struct *part, unsigned int inflight[2])
133 for_each_possible_cpu(cpu) {
134 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
135 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
137 if ((int)inflight[0] < 0)
139 if ((int)inflight[1] < 0)
143 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
145 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
147 if (unlikely(partno < 0 || partno >= ptbl->len))
149 return rcu_dereference(ptbl->part[partno]);
153 * disk_get_part - get partition
154 * @disk: disk to look partition from
155 * @partno: partition number
157 * Look for partition @partno from @disk. If found, increment
158 * reference count and return it.
164 * Pointer to the found partition on success, NULL if not found.
166 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
168 struct hd_struct *part;
171 part = __disk_get_part(disk, partno);
173 get_device(part_to_dev(part));
180 * disk_part_iter_init - initialize partition iterator
181 * @piter: iterator to initialize
182 * @disk: disk to iterate over
183 * @flags: DISK_PITER_* flags
185 * Initialize @piter so that it iterates over partitions of @disk.
190 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
193 struct disk_part_tbl *ptbl;
196 ptbl = rcu_dereference(disk->part_tbl);
201 if (flags & DISK_PITER_REVERSE)
202 piter->idx = ptbl->len - 1;
203 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
208 piter->flags = flags;
212 EXPORT_SYMBOL_GPL(disk_part_iter_init);
215 * disk_part_iter_next - proceed iterator to the next partition and return it
216 * @piter: iterator of interest
218 * Proceed @piter to the next partition and return it.
223 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
225 struct disk_part_tbl *ptbl;
228 /* put the last partition */
229 disk_put_part(piter->part);
234 ptbl = rcu_dereference(piter->disk->part_tbl);
236 /* determine iteration parameters */
237 if (piter->flags & DISK_PITER_REVERSE) {
239 if (piter->flags & (DISK_PITER_INCL_PART0 |
240 DISK_PITER_INCL_EMPTY_PART0))
249 /* iterate to the next partition */
250 for (; piter->idx != end; piter->idx += inc) {
251 struct hd_struct *part;
253 part = rcu_dereference(ptbl->part[piter->idx]);
256 if (!part_nr_sects_read(part) &&
257 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
258 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
262 get_device(part_to_dev(part));
272 EXPORT_SYMBOL_GPL(disk_part_iter_next);
275 * disk_part_iter_exit - finish up partition iteration
276 * @piter: iter of interest
278 * Called when iteration is over. Cleans up @piter.
283 void disk_part_iter_exit(struct disk_part_iter *piter)
285 disk_put_part(piter->part);
288 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
290 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
292 return part->start_sect <= sector &&
293 sector < part->start_sect + part_nr_sects_read(part);
297 * disk_map_sector_rcu - map sector to partition
298 * @disk: gendisk of interest
299 * @sector: sector to map
301 * Find out which partition @sector maps to on @disk. This is
302 * primarily used for stats accounting.
305 * RCU read locked. The returned partition pointer is always valid
306 * because its refcount is grabbed except for part0, which lifetime
307 * is same with the disk.
310 * Found partition on success, part0 is returned if no partition matches
311 * or the matched partition is being deleted.
313 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
315 struct disk_part_tbl *ptbl;
316 struct hd_struct *part;
320 ptbl = rcu_dereference(disk->part_tbl);
322 part = rcu_dereference(ptbl->last_lookup);
323 if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
326 for (i = 1; i < ptbl->len; i++) {
327 part = rcu_dereference(ptbl->part[i]);
329 if (part && sector_in_part(part, sector)) {
331 * only live partition can be cached for lookup,
332 * so use-after-free on cached & deleting partition
335 if (!hd_struct_try_get(part))
337 rcu_assign_pointer(ptbl->last_lookup, part);
349 * disk_has_partitions
350 * @disk: gendisk of interest
352 * Walk through the partition table and check if valid partition exists.
358 * True if the gendisk has at least one valid non-zero size partition.
361 bool disk_has_partitions(struct gendisk *disk)
363 struct disk_part_tbl *ptbl;
368 ptbl = rcu_dereference(disk->part_tbl);
370 /* Iterate partitions skipping the whole device at index 0 */
371 for (i = 1; i < ptbl->len; i++) {
372 if (rcu_dereference(ptbl->part[i])) {
382 EXPORT_SYMBOL_GPL(disk_has_partitions);
385 * Can be deleted altogether. Later.
388 #define BLKDEV_MAJOR_HASH_SIZE 255
389 static struct blk_major_name {
390 struct blk_major_name *next;
393 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
395 /* index in the above - for now: assume no multimajor ranges */
396 static inline int major_to_index(unsigned major)
398 return major % BLKDEV_MAJOR_HASH_SIZE;
401 #ifdef CONFIG_PROC_FS
402 void blkdev_show(struct seq_file *seqf, off_t offset)
404 struct blk_major_name *dp;
406 mutex_lock(&block_class_lock);
407 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
408 if (dp->major == offset)
409 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
410 mutex_unlock(&block_class_lock);
412 #endif /* CONFIG_PROC_FS */
415 * register_blkdev - register a new block device
417 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
418 * @major = 0, try to allocate any unused major number.
419 * @name: the name of the new block device as a zero terminated string
421 * The @name must be unique within the system.
423 * The return value depends on the @major input parameter:
425 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
426 * then the function returns zero on success, or a negative error code
427 * - if any unused major number was requested with @major = 0 parameter
428 * then the return value is the allocated major number in range
429 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
431 * See Documentation/admin-guide/devices.txt for the list of allocated
434 int register_blkdev(unsigned int major, const char *name)
436 struct blk_major_name **n, *p;
439 mutex_lock(&block_class_lock);
443 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
444 if (major_names[index] == NULL)
449 printk("%s: failed to get major for %s\n",
458 if (major >= BLKDEV_MAJOR_MAX) {
459 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
460 __func__, major, BLKDEV_MAJOR_MAX-1, name);
466 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
473 strlcpy(p->name, name, sizeof(p->name));
475 index = major_to_index(major);
477 for (n = &major_names[index]; *n; n = &(*n)->next) {
478 if ((*n)->major == major)
487 printk("register_blkdev: cannot get major %u for %s\n",
492 mutex_unlock(&block_class_lock);
496 EXPORT_SYMBOL(register_blkdev);
498 void unregister_blkdev(unsigned int major, const char *name)
500 struct blk_major_name **n;
501 struct blk_major_name *p = NULL;
502 int index = major_to_index(major);
504 mutex_lock(&block_class_lock);
505 for (n = &major_names[index]; *n; n = &(*n)->next)
506 if ((*n)->major == major)
508 if (!*n || strcmp((*n)->name, name)) {
514 mutex_unlock(&block_class_lock);
518 EXPORT_SYMBOL(unregister_blkdev);
520 static struct kobj_map *bdev_map;
523 * blk_mangle_minor - scatter minor numbers apart
524 * @minor: minor number to mangle
526 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
527 * is enabled. Mangling twice gives the original value.
535 static int blk_mangle_minor(int minor)
537 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
540 for (i = 0; i < MINORBITS / 2; i++) {
541 int low = minor & (1 << i);
542 int high = minor & (1 << (MINORBITS - 1 - i));
543 int distance = MINORBITS - 1 - 2 * i;
545 minor ^= low | high; /* clear both bits */
546 low <<= distance; /* swap the positions */
548 minor |= low | high; /* and set */
555 * blk_alloc_devt - allocate a dev_t for a partition
556 * @part: partition to allocate dev_t for
557 * @devt: out parameter for resulting dev_t
559 * Allocate a dev_t for block device.
562 * 0 on success, allocated dev_t is returned in *@devt. -errno on
568 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
570 struct gendisk *disk = part_to_disk(part);
573 /* in consecutive minor range? */
574 if (part->partno < disk->minors) {
575 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
579 /* allocate ext devt */
580 idr_preload(GFP_KERNEL);
582 spin_lock_bh(&ext_devt_lock);
583 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
584 spin_unlock_bh(&ext_devt_lock);
588 return idx == -ENOSPC ? -EBUSY : idx;
590 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
595 * blk_free_devt - free a dev_t
596 * @devt: dev_t to free
598 * Free @devt which was allocated using blk_alloc_devt().
603 void blk_free_devt(dev_t devt)
605 if (devt == MKDEV(0, 0))
608 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
609 spin_lock_bh(&ext_devt_lock);
610 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
611 spin_unlock_bh(&ext_devt_lock);
616 * We invalidate devt by assigning NULL pointer for devt in idr.
618 void blk_invalidate_devt(dev_t devt)
620 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
621 spin_lock_bh(&ext_devt_lock);
622 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
623 spin_unlock_bh(&ext_devt_lock);
627 static char *bdevt_str(dev_t devt, char *buf)
629 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
630 char tbuf[BDEVT_SIZE];
631 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
632 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
634 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
640 * Register device numbers dev..(dev+range-1)
641 * range must be nonzero
642 * The hash chain is sorted on range, so that subranges can override.
644 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
645 struct kobject *(*probe)(dev_t, int *, void *),
646 int (*lock)(dev_t, void *), void *data)
648 kobj_map(bdev_map, devt, range, module, probe, lock, data);
651 EXPORT_SYMBOL(blk_register_region);
653 void blk_unregister_region(dev_t devt, unsigned long range)
655 kobj_unmap(bdev_map, devt, range);
658 EXPORT_SYMBOL(blk_unregister_region);
660 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
662 struct gendisk *p = data;
664 return &disk_to_dev(p)->kobj;
667 static int exact_lock(dev_t devt, void *data)
669 struct gendisk *p = data;
671 if (!get_disk_and_module(p))
676 static void disk_scan_partitions(struct gendisk *disk)
678 struct block_device *bdev;
680 if (!get_capacity(disk) || !disk_part_scan_enabled(disk))
683 set_bit(GD_NEED_PART_SCAN, &disk->state);
684 bdev = blkdev_get_by_dev(disk_devt(disk), FMODE_READ, NULL);
686 blkdev_put(bdev, FMODE_READ);
689 static void register_disk(struct device *parent, struct gendisk *disk,
690 const struct attribute_group **groups)
692 struct device *ddev = disk_to_dev(disk);
693 struct disk_part_iter piter;
694 struct hd_struct *part;
697 ddev->parent = parent;
699 dev_set_name(ddev, "%s", disk->disk_name);
701 /* delay uevents, until we scanned partition table */
702 dev_set_uevent_suppress(ddev, 1);
705 WARN_ON(ddev->groups);
706 ddev->groups = groups;
708 if (device_add(ddev))
710 if (!sysfs_deprecated) {
711 err = sysfs_create_link(block_depr, &ddev->kobj,
712 kobject_name(&ddev->kobj));
720 * avoid probable deadlock caused by allocating memory with
721 * GFP_KERNEL in runtime_resume callback of its all ancestor
724 pm_runtime_set_memalloc_noio(ddev, true);
726 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
727 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
729 if (disk->flags & GENHD_FL_HIDDEN) {
730 dev_set_uevent_suppress(ddev, 0);
734 disk_scan_partitions(disk);
736 /* announce disk after possible partitions are created */
737 dev_set_uevent_suppress(ddev, 0);
738 kobject_uevent(&ddev->kobj, KOBJ_ADD);
740 /* announce possible partitions */
741 disk_part_iter_init(&piter, disk, 0);
742 while ((part = disk_part_iter_next(&piter)))
743 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
744 disk_part_iter_exit(&piter);
746 if (disk->queue->backing_dev_info->dev) {
747 err = sysfs_create_link(&ddev->kobj,
748 &disk->queue->backing_dev_info->dev->kobj,
755 * __device_add_disk - add disk information to kernel list
756 * @parent: parent device for the disk
757 * @disk: per-device partitioning information
758 * @groups: Additional per-device sysfs groups
759 * @register_queue: register the queue if set to true
761 * This function registers the partitioning information in @disk
764 * FIXME: error handling
766 static void __device_add_disk(struct device *parent, struct gendisk *disk,
767 const struct attribute_group **groups,
774 * The disk queue should now be all set with enough information about
775 * the device for the elevator code to pick an adequate default
776 * elevator if one is needed, that is, for devices requesting queue
780 elevator_init_mq(disk->queue);
782 /* minors == 0 indicates to use ext devt from part0 and should
783 * be accompanied with EXT_DEVT flag. Make sure all
784 * parameters make sense.
786 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
787 WARN_ON(!disk->minors &&
788 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
790 disk->flags |= GENHD_FL_UP;
792 retval = blk_alloc_devt(&disk->part0, &devt);
797 disk->major = MAJOR(devt);
798 disk->first_minor = MINOR(devt);
800 disk_alloc_events(disk);
802 if (disk->flags & GENHD_FL_HIDDEN) {
804 * Don't let hidden disks show up in /proc/partitions,
805 * and don't bother scanning for partitions either.
807 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
808 disk->flags |= GENHD_FL_NO_PART_SCAN;
810 struct backing_dev_info *bdi = disk->queue->backing_dev_info;
811 struct device *dev = disk_to_dev(disk);
814 /* Register BDI before referencing it from bdev */
816 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
818 bdi_set_owner(bdi, dev);
819 blk_register_region(disk_devt(disk), disk->minors, NULL,
820 exact_match, exact_lock, disk);
822 register_disk(parent, disk, groups);
824 blk_register_queue(disk);
827 * Take an extra ref on queue which will be put on disk_release()
828 * so that it sticks around as long as @disk is there.
830 WARN_ON_ONCE(!blk_get_queue(disk->queue));
832 disk_add_events(disk);
833 blk_integrity_add(disk);
836 void device_add_disk(struct device *parent, struct gendisk *disk,
837 const struct attribute_group **groups)
840 __device_add_disk(parent, disk, groups, true);
842 EXPORT_SYMBOL(device_add_disk);
844 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
846 __device_add_disk(parent, disk, NULL, false);
848 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
850 static void invalidate_partition(struct gendisk *disk, int partno)
852 struct block_device *bdev;
854 bdev = bdget_disk(disk, partno);
859 __invalidate_device(bdev, true);
862 * Unhash the bdev inode for this device so that it gets evicted as soon
863 * as last inode reference is dropped.
865 remove_inode_hash(bdev->bd_inode);
870 * del_gendisk - remove the gendisk
871 * @disk: the struct gendisk to remove
873 * Removes the gendisk and all its associated resources. This deletes the
874 * partitions associated with the gendisk, and unregisters the associated
877 * This is the counter to the respective __device_add_disk() call.
879 * The final removal of the struct gendisk happens when its refcount reaches 0
880 * with put_disk(), which should be called after del_gendisk(), if
881 * __device_add_disk() was used.
883 * Drivers exist which depend on the release of the gendisk to be synchronous,
884 * it should not be deferred.
888 void del_gendisk(struct gendisk *disk)
890 struct disk_part_iter piter;
891 struct hd_struct *part;
895 blk_integrity_del(disk);
896 disk_del_events(disk);
899 * Block lookups of the disk until all bdevs are unhashed and the
900 * disk is marked as dead (GENHD_FL_UP cleared).
902 down_write(&disk->lookup_sem);
903 /* invalidate stuff */
904 disk_part_iter_init(&piter, disk,
905 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
906 while ((part = disk_part_iter_next(&piter))) {
907 invalidate_partition(disk, part->partno);
908 delete_partition(part);
910 disk_part_iter_exit(&piter);
912 invalidate_partition(disk, 0);
913 set_capacity(disk, 0);
914 disk->flags &= ~GENHD_FL_UP;
915 up_write(&disk->lookup_sem);
917 if (!(disk->flags & GENHD_FL_HIDDEN))
918 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
921 * Unregister bdi before releasing device numbers (as they can
922 * get reused and we'd get clashes in sysfs).
924 if (!(disk->flags & GENHD_FL_HIDDEN))
925 bdi_unregister(disk->queue->backing_dev_info);
926 blk_unregister_queue(disk);
931 if (!(disk->flags & GENHD_FL_HIDDEN))
932 blk_unregister_region(disk_devt(disk), disk->minors);
934 * Remove gendisk pointer from idr so that it cannot be looked up
935 * while RCU period before freeing gendisk is running to prevent
936 * use-after-free issues. Note that the device number stays
937 * "in-use" until we really free the gendisk.
939 blk_invalidate_devt(disk_devt(disk));
941 kobject_put(disk->part0.holder_dir);
942 kobject_put(disk->slave_dir);
944 part_stat_set_all(&disk->part0, 0);
945 disk->part0.stamp = 0;
946 if (!sysfs_deprecated)
947 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
948 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
949 device_del(disk_to_dev(disk));
951 EXPORT_SYMBOL(del_gendisk);
953 /* sysfs access to bad-blocks list. */
954 static ssize_t disk_badblocks_show(struct device *dev,
955 struct device_attribute *attr,
958 struct gendisk *disk = dev_to_disk(dev);
961 return sprintf(page, "\n");
963 return badblocks_show(disk->bb, page, 0);
966 static ssize_t disk_badblocks_store(struct device *dev,
967 struct device_attribute *attr,
968 const char *page, size_t len)
970 struct gendisk *disk = dev_to_disk(dev);
975 return badblocks_store(disk->bb, page, len, 0);
979 * get_gendisk - get partitioning information for a given device
980 * @devt: device to get partitioning information for
981 * @partno: returned partition index
983 * This function gets the structure containing partitioning
984 * information for the given device @devt.
988 struct gendisk *get_gendisk(dev_t devt, int *partno)
990 struct gendisk *disk = NULL;
994 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
995 struct kobject *kobj;
997 kobj = kobj_lookup(bdev_map, devt, partno);
999 disk = dev_to_disk(kobj_to_dev(kobj));
1001 struct hd_struct *part;
1003 spin_lock_bh(&ext_devt_lock);
1004 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
1005 if (part && get_disk_and_module(part_to_disk(part))) {
1006 *partno = part->partno;
1007 disk = part_to_disk(part);
1009 spin_unlock_bh(&ext_devt_lock);
1016 * Synchronize with del_gendisk() to not return disk that is being
1019 down_read(&disk->lookup_sem);
1020 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1021 !(disk->flags & GENHD_FL_UP))) {
1022 up_read(&disk->lookup_sem);
1023 put_disk_and_module(disk);
1026 up_read(&disk->lookup_sem);
1032 * bdget_disk - do bdget() by gendisk and partition number
1033 * @disk: gendisk of interest
1034 * @partno: partition number
1036 * Find partition @partno from @disk, do bdget() on it.
1042 * Resulting block_device on success, NULL on failure.
1044 struct block_device *bdget_disk(struct gendisk *disk, int partno)
1046 struct hd_struct *part;
1047 struct block_device *bdev = NULL;
1049 part = disk_get_part(disk, partno);
1051 bdev = bdget_part(part);
1052 disk_put_part(part);
1056 EXPORT_SYMBOL(bdget_disk);
1059 * print a full list of all partitions - intended for places where the root
1060 * filesystem can't be mounted and thus to give the victim some idea of what
1063 void __init printk_all_partitions(void)
1065 struct class_dev_iter iter;
1068 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1069 while ((dev = class_dev_iter_next(&iter))) {
1070 struct gendisk *disk = dev_to_disk(dev);
1071 struct disk_part_iter piter;
1072 struct hd_struct *part;
1073 char name_buf[BDEVNAME_SIZE];
1074 char devt_buf[BDEVT_SIZE];
1077 * Don't show empty devices or things that have been
1080 if (get_capacity(disk) == 0 ||
1081 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1085 * Note, unlike /proc/partitions, I am showing the
1086 * numbers in hex - the same format as the root=
1089 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1090 while ((part = disk_part_iter_next(&piter))) {
1091 bool is_part0 = part == &disk->part0;
1093 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1094 bdevt_str(part_devt(part), devt_buf),
1095 (unsigned long long)part_nr_sects_read(part) >> 1
1096 , disk_name(disk, part->partno, name_buf),
1097 part->info ? part->info->uuid : "");
1099 if (dev->parent && dev->parent->driver)
1100 printk(" driver: %s\n",
1101 dev->parent->driver->name);
1103 printk(" (driver?)\n");
1107 disk_part_iter_exit(&piter);
1109 class_dev_iter_exit(&iter);
1112 #ifdef CONFIG_PROC_FS
1114 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1117 struct class_dev_iter *iter;
1120 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1122 return ERR_PTR(-ENOMEM);
1124 seqf->private = iter;
1125 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1127 dev = class_dev_iter_next(iter);
1132 return dev_to_disk(dev);
1135 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1140 dev = class_dev_iter_next(seqf->private);
1142 return dev_to_disk(dev);
1147 static void disk_seqf_stop(struct seq_file *seqf, void *v)
1149 struct class_dev_iter *iter = seqf->private;
1151 /* stop is called even after start failed :-( */
1153 class_dev_iter_exit(iter);
1155 seqf->private = NULL;
1159 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1163 p = disk_seqf_start(seqf, pos);
1164 if (!IS_ERR_OR_NULL(p) && !*pos)
1165 seq_puts(seqf, "major minor #blocks name\n\n");
1169 static int show_partition(struct seq_file *seqf, void *v)
1171 struct gendisk *sgp = v;
1172 struct disk_part_iter piter;
1173 struct hd_struct *part;
1174 char buf[BDEVNAME_SIZE];
1176 /* Don't show non-partitionable removeable devices or empty devices */
1177 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1178 (sgp->flags & GENHD_FL_REMOVABLE)))
1180 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1183 /* show the full disk and all non-0 size partitions of it */
1184 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1185 while ((part = disk_part_iter_next(&piter)))
1186 seq_printf(seqf, "%4d %7d %10llu %s\n",
1187 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1188 (unsigned long long)part_nr_sects_read(part) >> 1,
1189 disk_name(sgp, part->partno, buf));
1190 disk_part_iter_exit(&piter);
1195 static const struct seq_operations partitions_op = {
1196 .start = show_partition_start,
1197 .next = disk_seqf_next,
1198 .stop = disk_seqf_stop,
1199 .show = show_partition
1204 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1206 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1207 /* Make old-style 2.4 aliases work */
1208 request_module("block-major-%d", MAJOR(devt));
1212 static int __init genhd_device_init(void)
1216 block_class.dev_kobj = sysfs_dev_block_kobj;
1217 error = class_register(&block_class);
1218 if (unlikely(error))
1220 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1223 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1225 /* create top-level block dir */
1226 if (!sysfs_deprecated)
1227 block_depr = kobject_create_and_add("block", NULL);
1231 subsys_initcall(genhd_device_init);
1233 static ssize_t disk_range_show(struct device *dev,
1234 struct device_attribute *attr, char *buf)
1236 struct gendisk *disk = dev_to_disk(dev);
1238 return sprintf(buf, "%d\n", disk->minors);
1241 static ssize_t disk_ext_range_show(struct device *dev,
1242 struct device_attribute *attr, char *buf)
1244 struct gendisk *disk = dev_to_disk(dev);
1246 return sprintf(buf, "%d\n", disk_max_parts(disk));
1249 static ssize_t disk_removable_show(struct device *dev,
1250 struct device_attribute *attr, char *buf)
1252 struct gendisk *disk = dev_to_disk(dev);
1254 return sprintf(buf, "%d\n",
1255 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1258 static ssize_t disk_hidden_show(struct device *dev,
1259 struct device_attribute *attr, char *buf)
1261 struct gendisk *disk = dev_to_disk(dev);
1263 return sprintf(buf, "%d\n",
1264 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1267 static ssize_t disk_ro_show(struct device *dev,
1268 struct device_attribute *attr, char *buf)
1270 struct gendisk *disk = dev_to_disk(dev);
1272 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1275 ssize_t part_size_show(struct device *dev,
1276 struct device_attribute *attr, char *buf)
1278 struct hd_struct *p = dev_to_part(dev);
1280 return sprintf(buf, "%llu\n",
1281 (unsigned long long)part_nr_sects_read(p));
1284 ssize_t part_stat_show(struct device *dev,
1285 struct device_attribute *attr, char *buf)
1287 struct hd_struct *p = dev_to_part(dev);
1288 struct request_queue *q = part_to_disk(p)->queue;
1289 struct disk_stats stat;
1290 unsigned int inflight;
1292 part_stat_read_all(p, &stat);
1294 inflight = blk_mq_in_flight(q, p);
1296 inflight = part_in_flight(p);
1299 "%8lu %8lu %8llu %8u "
1300 "%8lu %8lu %8llu %8u "
1302 "%8lu %8lu %8llu %8u "
1305 stat.ios[STAT_READ],
1306 stat.merges[STAT_READ],
1307 (unsigned long long)stat.sectors[STAT_READ],
1308 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1309 stat.ios[STAT_WRITE],
1310 stat.merges[STAT_WRITE],
1311 (unsigned long long)stat.sectors[STAT_WRITE],
1312 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1314 jiffies_to_msecs(stat.io_ticks),
1315 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1316 stat.nsecs[STAT_WRITE] +
1317 stat.nsecs[STAT_DISCARD] +
1318 stat.nsecs[STAT_FLUSH],
1320 stat.ios[STAT_DISCARD],
1321 stat.merges[STAT_DISCARD],
1322 (unsigned long long)stat.sectors[STAT_DISCARD],
1323 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1324 stat.ios[STAT_FLUSH],
1325 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1328 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1331 struct hd_struct *p = dev_to_part(dev);
1332 struct request_queue *q = part_to_disk(p)->queue;
1333 unsigned int inflight[2];
1336 blk_mq_in_flight_rw(q, p, inflight);
1338 part_in_flight_rw(p, inflight);
1340 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1343 static ssize_t disk_capability_show(struct device *dev,
1344 struct device_attribute *attr, char *buf)
1346 struct gendisk *disk = dev_to_disk(dev);
1348 return sprintf(buf, "%x\n", disk->flags);
1351 static ssize_t disk_alignment_offset_show(struct device *dev,
1352 struct device_attribute *attr,
1355 struct gendisk *disk = dev_to_disk(dev);
1357 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1360 static ssize_t disk_discard_alignment_show(struct device *dev,
1361 struct device_attribute *attr,
1364 struct gendisk *disk = dev_to_disk(dev);
1366 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1369 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1370 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1371 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1372 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1373 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1374 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1375 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1376 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1377 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1378 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1379 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1380 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1382 #ifdef CONFIG_FAIL_MAKE_REQUEST
1383 ssize_t part_fail_show(struct device *dev,
1384 struct device_attribute *attr, char *buf)
1386 struct hd_struct *p = dev_to_part(dev);
1388 return sprintf(buf, "%d\n", p->make_it_fail);
1391 ssize_t part_fail_store(struct device *dev,
1392 struct device_attribute *attr,
1393 const char *buf, size_t count)
1395 struct hd_struct *p = dev_to_part(dev);
1398 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1399 p->make_it_fail = (i == 0) ? 0 : 1;
1404 static struct device_attribute dev_attr_fail =
1405 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1406 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1408 #ifdef CONFIG_FAIL_IO_TIMEOUT
1409 static struct device_attribute dev_attr_fail_timeout =
1410 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1413 static struct attribute *disk_attrs[] = {
1414 &dev_attr_range.attr,
1415 &dev_attr_ext_range.attr,
1416 &dev_attr_removable.attr,
1417 &dev_attr_hidden.attr,
1419 &dev_attr_size.attr,
1420 &dev_attr_alignment_offset.attr,
1421 &dev_attr_discard_alignment.attr,
1422 &dev_attr_capability.attr,
1423 &dev_attr_stat.attr,
1424 &dev_attr_inflight.attr,
1425 &dev_attr_badblocks.attr,
1426 #ifdef CONFIG_FAIL_MAKE_REQUEST
1427 &dev_attr_fail.attr,
1429 #ifdef CONFIG_FAIL_IO_TIMEOUT
1430 &dev_attr_fail_timeout.attr,
1435 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1437 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1438 struct gendisk *disk = dev_to_disk(dev);
1440 if (a == &dev_attr_badblocks.attr && !disk->bb)
1445 static struct attribute_group disk_attr_group = {
1446 .attrs = disk_attrs,
1447 .is_visible = disk_visible,
1450 static const struct attribute_group *disk_attr_groups[] = {
1456 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1457 * @disk: disk to replace part_tbl for
1458 * @new_ptbl: new part_tbl to install
1460 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1461 * original ptbl is freed using RCU callback.
1464 * Matching bd_mutex locked or the caller is the only user of @disk.
1466 static void disk_replace_part_tbl(struct gendisk *disk,
1467 struct disk_part_tbl *new_ptbl)
1469 struct disk_part_tbl *old_ptbl =
1470 rcu_dereference_protected(disk->part_tbl, 1);
1472 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1475 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1476 kfree_rcu(old_ptbl, rcu_head);
1481 * disk_expand_part_tbl - expand disk->part_tbl
1482 * @disk: disk to expand part_tbl for
1483 * @partno: expand such that this partno can fit in
1485 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1486 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1489 * Matching bd_mutex locked or the caller is the only user of @disk.
1493 * 0 on success, -errno on failure.
1495 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1497 struct disk_part_tbl *old_ptbl =
1498 rcu_dereference_protected(disk->part_tbl, 1);
1499 struct disk_part_tbl *new_ptbl;
1500 int len = old_ptbl ? old_ptbl->len : 0;
1504 * check for int overflow, since we can get here from blkpg_ioctl()
1505 * with a user passed 'partno'.
1507 target = partno + 1;
1511 /* disk_max_parts() is zero during initialization, ignore if so */
1512 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1518 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1523 new_ptbl->len = target;
1525 for (i = 0; i < len; i++)
1526 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1528 disk_replace_part_tbl(disk, new_ptbl);
1533 * disk_release - releases all allocated resources of the gendisk
1534 * @dev: the device representing this disk
1536 * This function releases all allocated resources of the gendisk.
1538 * The struct gendisk refcount is incremented with get_gendisk() or
1539 * get_disk_and_module(), and its refcount is decremented with
1540 * put_disk_and_module() or put_disk(). Once the refcount reaches 0 this
1541 * function is called.
1543 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1544 * assigned. Since the request_queue sits on top of the gendisk for these
1545 * drivers we also call blk_put_queue() for them, and we expect the
1546 * request_queue refcount to reach 0 at this point, and so the request_queue
1547 * will also be freed prior to the disk.
1549 * Context: can sleep
1551 static void disk_release(struct device *dev)
1553 struct gendisk *disk = dev_to_disk(dev);
1557 blk_free_devt(dev->devt);
1558 disk_release_events(disk);
1559 kfree(disk->random);
1560 disk_replace_part_tbl(disk, NULL);
1561 hd_free_part(&disk->part0);
1563 blk_put_queue(disk->queue);
1566 struct class block_class = {
1570 static char *block_devnode(struct device *dev, umode_t *mode,
1571 kuid_t *uid, kgid_t *gid)
1573 struct gendisk *disk = dev_to_disk(dev);
1575 if (disk->fops->devnode)
1576 return disk->fops->devnode(disk, mode);
1580 const struct device_type disk_type = {
1582 .groups = disk_attr_groups,
1583 .release = disk_release,
1584 .devnode = block_devnode,
1587 #ifdef CONFIG_PROC_FS
1589 * aggregate disk stat collector. Uses the same stats that the sysfs
1590 * entries do, above, but makes them available through one seq_file.
1592 * The output looks suspiciously like /proc/partitions with a bunch of
1595 static int diskstats_show(struct seq_file *seqf, void *v)
1597 struct gendisk *gp = v;
1598 struct disk_part_iter piter;
1599 struct hd_struct *hd;
1600 char buf[BDEVNAME_SIZE];
1601 unsigned int inflight;
1602 struct disk_stats stat;
1605 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1606 seq_puts(seqf, "major minor name"
1607 " rio rmerge rsect ruse wio wmerge "
1608 "wsect wuse running use aveq"
1612 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1613 while ((hd = disk_part_iter_next(&piter))) {
1614 part_stat_read_all(hd, &stat);
1615 if (queue_is_mq(gp->queue))
1616 inflight = blk_mq_in_flight(gp->queue, hd);
1618 inflight = part_in_flight(hd);
1620 seq_printf(seqf, "%4d %7d %s "
1627 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1628 disk_name(gp, hd->partno, buf),
1629 stat.ios[STAT_READ],
1630 stat.merges[STAT_READ],
1631 stat.sectors[STAT_READ],
1632 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1634 stat.ios[STAT_WRITE],
1635 stat.merges[STAT_WRITE],
1636 stat.sectors[STAT_WRITE],
1637 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1640 jiffies_to_msecs(stat.io_ticks),
1641 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1642 stat.nsecs[STAT_WRITE] +
1643 stat.nsecs[STAT_DISCARD] +
1644 stat.nsecs[STAT_FLUSH],
1646 stat.ios[STAT_DISCARD],
1647 stat.merges[STAT_DISCARD],
1648 stat.sectors[STAT_DISCARD],
1649 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1651 stat.ios[STAT_FLUSH],
1652 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1656 disk_part_iter_exit(&piter);
1661 static const struct seq_operations diskstats_op = {
1662 .start = disk_seqf_start,
1663 .next = disk_seqf_next,
1664 .stop = disk_seqf_stop,
1665 .show = diskstats_show
1668 static int __init proc_genhd_init(void)
1670 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1671 proc_create_seq("partitions", 0, NULL, &partitions_op);
1674 module_init(proc_genhd_init);
1675 #endif /* CONFIG_PROC_FS */
1677 dev_t blk_lookup_devt(const char *name, int partno)
1679 dev_t devt = MKDEV(0, 0);
1680 struct class_dev_iter iter;
1683 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1684 while ((dev = class_dev_iter_next(&iter))) {
1685 struct gendisk *disk = dev_to_disk(dev);
1686 struct hd_struct *part;
1688 if (strcmp(dev_name(dev), name))
1691 if (partno < disk->minors) {
1692 /* We need to return the right devno, even
1693 * if the partition doesn't exist yet.
1695 devt = MKDEV(MAJOR(dev->devt),
1696 MINOR(dev->devt) + partno);
1699 part = disk_get_part(disk, partno);
1701 devt = part_devt(part);
1702 disk_put_part(part);
1705 disk_put_part(part);
1707 class_dev_iter_exit(&iter);
1711 struct gendisk *__alloc_disk_node(int minors, int node_id)
1713 struct gendisk *disk;
1714 struct disk_part_tbl *ptbl;
1716 if (minors > DISK_MAX_PARTS) {
1718 "block: can't allocate more than %d partitions\n",
1720 minors = DISK_MAX_PARTS;
1723 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1727 disk->part0.dkstats = alloc_percpu(struct disk_stats);
1728 if (!disk->part0.dkstats)
1731 init_rwsem(&disk->lookup_sem);
1732 disk->node_id = node_id;
1733 if (disk_expand_part_tbl(disk, 0)) {
1734 free_percpu(disk->part0.dkstats);
1738 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1739 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1742 * set_capacity() and get_capacity() currently don't use
1743 * seqcounter to read/update the part0->nr_sects. Still init
1744 * the counter as we can read the sectors in IO submission
1745 * patch using seqence counters.
1747 * TODO: Ideally set_capacity() and get_capacity() should be
1748 * converted to make use of bd_mutex and sequence counters.
1750 hd_sects_seq_init(&disk->part0);
1751 if (hd_ref_init(&disk->part0))
1752 goto out_free_part0;
1754 disk->minors = minors;
1755 rand_initialize_disk(disk);
1756 disk_to_dev(disk)->class = &block_class;
1757 disk_to_dev(disk)->type = &disk_type;
1758 device_initialize(disk_to_dev(disk));
1762 hd_free_part(&disk->part0);
1767 EXPORT_SYMBOL(__alloc_disk_node);
1770 * get_disk_and_module - increments the gendisk and gendisk fops module refcount
1771 * @disk: the struct gendisk to increment the refcount for
1773 * This increments the refcount for the struct gendisk, and the gendisk's
1774 * fops module owner.
1776 * Context: Any context.
1778 struct kobject *get_disk_and_module(struct gendisk *disk)
1780 struct module *owner;
1781 struct kobject *kobj;
1785 owner = disk->fops->owner;
1786 if (owner && !try_module_get(owner))
1788 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1796 EXPORT_SYMBOL(get_disk_and_module);
1799 * put_disk - decrements the gendisk refcount
1800 * @disk: the struct gendisk to decrement the refcount for
1802 * This decrements the refcount for the struct gendisk. When this reaches 0
1803 * we'll have disk_release() called.
1805 * Context: Any context, but the last reference must not be dropped from
1808 void put_disk(struct gendisk *disk)
1811 kobject_put(&disk_to_dev(disk)->kobj);
1813 EXPORT_SYMBOL(put_disk);
1816 * put_disk_and_module - decrements the module and gendisk refcount
1817 * @disk: the struct gendisk to decrement the refcount for
1819 * This is a counterpart of get_disk_and_module() and thus also of
1822 * Context: Any context, but the last reference must not be dropped from
1825 void put_disk_and_module(struct gendisk *disk)
1828 struct module *owner = disk->fops->owner;
1834 EXPORT_SYMBOL(put_disk_and_module);
1836 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1838 char event[] = "DISK_RO=1";
1839 char *envp[] = { event, NULL };
1843 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1846 void set_device_ro(struct block_device *bdev, int flag)
1848 bdev->bd_part->policy = flag;
1851 EXPORT_SYMBOL(set_device_ro);
1853 void set_disk_ro(struct gendisk *disk, int flag)
1855 struct disk_part_iter piter;
1856 struct hd_struct *part;
1858 if (disk->part0.policy != flag) {
1859 set_disk_ro_uevent(disk, flag);
1860 disk->part0.policy = flag;
1863 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1864 while ((part = disk_part_iter_next(&piter)))
1865 part->policy = flag;
1866 disk_part_iter_exit(&piter);
1869 EXPORT_SYMBOL(set_disk_ro);
1871 int bdev_read_only(struct block_device *bdev)
1875 return bdev->bd_part->policy;
1878 EXPORT_SYMBOL(bdev_read_only);
1881 * Disk events - monitor disk events like media change and eject request.
1883 struct disk_events {
1884 struct list_head node; /* all disk_event's */
1885 struct gendisk *disk; /* the associated disk */
1888 struct mutex block_mutex; /* protects blocking */
1889 int block; /* event blocking depth */
1890 unsigned int pending; /* events already sent out */
1891 unsigned int clearing; /* events being cleared */
1893 long poll_msecs; /* interval, -1 for default */
1894 struct delayed_work dwork;
1897 static const char *disk_events_strs[] = {
1898 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1899 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1902 static char *disk_uevents[] = {
1903 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1904 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1907 /* list of all disk_events */
1908 static DEFINE_MUTEX(disk_events_mutex);
1909 static LIST_HEAD(disk_events);
1911 /* disable in-kernel polling by default */
1912 static unsigned long disk_events_dfl_poll_msecs;
1914 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1916 struct disk_events *ev = disk->ev;
1917 long intv_msecs = 0;
1920 * If device-specific poll interval is set, always use it. If
1921 * the default is being used, poll if the POLL flag is set.
1923 if (ev->poll_msecs >= 0)
1924 intv_msecs = ev->poll_msecs;
1925 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1926 intv_msecs = disk_events_dfl_poll_msecs;
1928 return msecs_to_jiffies(intv_msecs);
1932 * disk_block_events - block and flush disk event checking
1933 * @disk: disk to block events for
1935 * On return from this function, it is guaranteed that event checking
1936 * isn't in progress and won't happen until unblocked by
1937 * disk_unblock_events(). Events blocking is counted and the actual
1938 * unblocking happens after the matching number of unblocks are done.
1940 * Note that this intentionally does not block event checking from
1941 * disk_clear_events().
1946 void disk_block_events(struct gendisk *disk)
1948 struct disk_events *ev = disk->ev;
1949 unsigned long flags;
1956 * Outer mutex ensures that the first blocker completes canceling
1957 * the event work before further blockers are allowed to finish.
1959 mutex_lock(&ev->block_mutex);
1961 spin_lock_irqsave(&ev->lock, flags);
1962 cancel = !ev->block++;
1963 spin_unlock_irqrestore(&ev->lock, flags);
1966 cancel_delayed_work_sync(&disk->ev->dwork);
1968 mutex_unlock(&ev->block_mutex);
1971 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1973 struct disk_events *ev = disk->ev;
1975 unsigned long flags;
1977 spin_lock_irqsave(&ev->lock, flags);
1979 if (WARN_ON_ONCE(ev->block <= 0))
1985 intv = disk_events_poll_jiffies(disk);
1987 queue_delayed_work(system_freezable_power_efficient_wq,
1990 queue_delayed_work(system_freezable_power_efficient_wq,
1993 spin_unlock_irqrestore(&ev->lock, flags);
1997 * disk_unblock_events - unblock disk event checking
1998 * @disk: disk to unblock events for
2000 * Undo disk_block_events(). When the block count reaches zero, it
2001 * starts events polling if configured.
2004 * Don't care. Safe to call from irq context.
2006 void disk_unblock_events(struct gendisk *disk)
2009 __disk_unblock_events(disk, false);
2013 * disk_flush_events - schedule immediate event checking and flushing
2014 * @disk: disk to check and flush events for
2015 * @mask: events to flush
2017 * Schedule immediate event checking on @disk if not blocked. Events in
2018 * @mask are scheduled to be cleared from the driver. Note that this
2019 * doesn't clear the events from @disk->ev.
2022 * If @mask is non-zero must be called with bdev->bd_mutex held.
2024 void disk_flush_events(struct gendisk *disk, unsigned int mask)
2026 struct disk_events *ev = disk->ev;
2031 spin_lock_irq(&ev->lock);
2032 ev->clearing |= mask;
2034 mod_delayed_work(system_freezable_power_efficient_wq,
2036 spin_unlock_irq(&ev->lock);
2040 * disk_clear_events - synchronously check, clear and return pending events
2041 * @disk: disk to fetch and clear events from
2042 * @mask: mask of events to be fetched and cleared
2044 * Disk events are synchronously checked and pending events in @mask
2045 * are cleared and returned. This ignores the block count.
2050 static unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
2052 struct disk_events *ev = disk->ev;
2053 unsigned int pending;
2054 unsigned int clearing = mask;
2059 disk_block_events(disk);
2062 * store the union of mask and ev->clearing on the stack so that the
2063 * race with disk_flush_events does not cause ambiguity (ev->clearing
2064 * can still be modified even if events are blocked).
2066 spin_lock_irq(&ev->lock);
2067 clearing |= ev->clearing;
2069 spin_unlock_irq(&ev->lock);
2071 disk_check_events(ev, &clearing);
2073 * if ev->clearing is not 0, the disk_flush_events got called in the
2074 * middle of this function, so we want to run the workfn without delay.
2076 __disk_unblock_events(disk, ev->clearing ? true : false);
2078 /* then, fetch and clear pending events */
2079 spin_lock_irq(&ev->lock);
2080 pending = ev->pending & mask;
2081 ev->pending &= ~mask;
2082 spin_unlock_irq(&ev->lock);
2083 WARN_ON_ONCE(clearing & mask);
2089 * bdev_check_media_change - check if a removable media has been changed
2090 * @bdev: block device to check
2092 * Check whether a removable media has been changed, and attempt to free all
2093 * dentries and inodes and invalidates all block device page cache entries in
2096 * Returns %true if the block device changed, or %false if not.
2098 bool bdev_check_media_change(struct block_device *bdev)
2100 unsigned int events;
2102 events = disk_clear_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE |
2103 DISK_EVENT_EJECT_REQUEST);
2104 if (!(events & DISK_EVENT_MEDIA_CHANGE))
2107 if (__invalidate_device(bdev, true))
2108 pr_warn("VFS: busy inodes on changed media %s\n",
2109 bdev->bd_disk->disk_name);
2110 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
2113 EXPORT_SYMBOL(bdev_check_media_change);
2116 * Separate this part out so that a different pointer for clearing_ptr can be
2117 * passed in for disk_clear_events.
2119 static void disk_events_workfn(struct work_struct *work)
2121 struct delayed_work *dwork = to_delayed_work(work);
2122 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2124 disk_check_events(ev, &ev->clearing);
2127 static void disk_check_events(struct disk_events *ev,
2128 unsigned int *clearing_ptr)
2130 struct gendisk *disk = ev->disk;
2131 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2132 unsigned int clearing = *clearing_ptr;
2133 unsigned int events;
2135 int nr_events = 0, i;
2138 events = disk->fops->check_events(disk, clearing);
2140 /* accumulate pending events and schedule next poll if necessary */
2141 spin_lock_irq(&ev->lock);
2143 events &= ~ev->pending;
2144 ev->pending |= events;
2145 *clearing_ptr &= ~clearing;
2147 intv = disk_events_poll_jiffies(disk);
2148 if (!ev->block && intv)
2149 queue_delayed_work(system_freezable_power_efficient_wq,
2152 spin_unlock_irq(&ev->lock);
2155 * Tell userland about new events. Only the events listed in
2156 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2157 * is set. Otherwise, events are processed internally but never
2158 * get reported to userland.
2160 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2161 if ((events & disk->events & (1 << i)) &&
2162 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2163 envp[nr_events++] = disk_uevents[i];
2166 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2170 * A disk events enabled device has the following sysfs nodes under
2171 * its /sys/block/X/ directory.
2173 * events : list of all supported events
2174 * events_async : list of events which can be detected w/o polling
2175 * (always empty, only for backwards compatibility)
2176 * events_poll_msecs : polling interval, 0: disable, -1: system default
2178 static ssize_t __disk_events_show(unsigned int events, char *buf)
2180 const char *delim = "";
2184 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2185 if (events & (1 << i)) {
2186 pos += sprintf(buf + pos, "%s%s",
2187 delim, disk_events_strs[i]);
2191 pos += sprintf(buf + pos, "\n");
2195 static ssize_t disk_events_show(struct device *dev,
2196 struct device_attribute *attr, char *buf)
2198 struct gendisk *disk = dev_to_disk(dev);
2200 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2203 return __disk_events_show(disk->events, buf);
2206 static ssize_t disk_events_async_show(struct device *dev,
2207 struct device_attribute *attr, char *buf)
2212 static ssize_t disk_events_poll_msecs_show(struct device *dev,
2213 struct device_attribute *attr,
2216 struct gendisk *disk = dev_to_disk(dev);
2219 return sprintf(buf, "-1\n");
2221 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2224 static ssize_t disk_events_poll_msecs_store(struct device *dev,
2225 struct device_attribute *attr,
2226 const char *buf, size_t count)
2228 struct gendisk *disk = dev_to_disk(dev);
2231 if (!count || !sscanf(buf, "%ld", &intv))
2234 if (intv < 0 && intv != -1)
2240 disk_block_events(disk);
2241 disk->ev->poll_msecs = intv;
2242 __disk_unblock_events(disk, true);
2247 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2248 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2249 static const DEVICE_ATTR(events_poll_msecs, 0644,
2250 disk_events_poll_msecs_show,
2251 disk_events_poll_msecs_store);
2253 static const struct attribute *disk_events_attrs[] = {
2254 &dev_attr_events.attr,
2255 &dev_attr_events_async.attr,
2256 &dev_attr_events_poll_msecs.attr,
2261 * The default polling interval can be specified by the kernel
2262 * parameter block.events_dfl_poll_msecs which defaults to 0
2263 * (disable). This can also be modified runtime by writing to
2264 * /sys/module/block/parameters/events_dfl_poll_msecs.
2266 static int disk_events_set_dfl_poll_msecs(const char *val,
2267 const struct kernel_param *kp)
2269 struct disk_events *ev;
2272 ret = param_set_ulong(val, kp);
2276 mutex_lock(&disk_events_mutex);
2278 list_for_each_entry(ev, &disk_events, node)
2279 disk_flush_events(ev->disk, 0);
2281 mutex_unlock(&disk_events_mutex);
2286 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2287 .set = disk_events_set_dfl_poll_msecs,
2288 .get = param_get_ulong,
2291 #undef MODULE_PARAM_PREFIX
2292 #define MODULE_PARAM_PREFIX "block."
2294 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2295 &disk_events_dfl_poll_msecs, 0644);
2298 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2300 static void disk_alloc_events(struct gendisk *disk)
2302 struct disk_events *ev;
2304 if (!disk->fops->check_events || !disk->events)
2307 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2309 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2313 INIT_LIST_HEAD(&ev->node);
2315 spin_lock_init(&ev->lock);
2316 mutex_init(&ev->block_mutex);
2318 ev->poll_msecs = -1;
2319 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2324 static void disk_add_events(struct gendisk *disk)
2326 /* FIXME: error handling */
2327 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2328 pr_warn("%s: failed to create sysfs files for events\n",
2334 mutex_lock(&disk_events_mutex);
2335 list_add_tail(&disk->ev->node, &disk_events);
2336 mutex_unlock(&disk_events_mutex);
2339 * Block count is initialized to 1 and the following initial
2340 * unblock kicks it into action.
2342 __disk_unblock_events(disk, true);
2345 static void disk_del_events(struct gendisk *disk)
2348 disk_block_events(disk);
2350 mutex_lock(&disk_events_mutex);
2351 list_del_init(&disk->ev->node);
2352 mutex_unlock(&disk_events_mutex);
2355 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2358 static void disk_release_events(struct gendisk *disk)
2360 /* the block count should be 1 from disk_del_events() */
2361 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);