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 bool 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);
71 EXPORT_SYMBOL_GPL(set_capacity_revalidate_and_notify);
74 * Format the device name of the indicated disk into the supplied buffer and
75 * return a pointer to that same buffer for convenience.
77 char *disk_name(struct gendisk *hd, int partno, char *buf)
80 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
81 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
82 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
84 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
89 const char *bdevname(struct block_device *bdev, char *buf)
91 return disk_name(bdev->bd_disk, bdev->bd_partno, buf);
93 EXPORT_SYMBOL(bdevname);
95 static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
99 memset(stat, 0, sizeof(struct disk_stats));
100 for_each_possible_cpu(cpu) {
101 struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
104 for (group = 0; group < NR_STAT_GROUPS; group++) {
105 stat->nsecs[group] += ptr->nsecs[group];
106 stat->sectors[group] += ptr->sectors[group];
107 stat->ios[group] += ptr->ios[group];
108 stat->merges[group] += ptr->merges[group];
111 stat->io_ticks += ptr->io_ticks;
115 static unsigned int part_in_flight(struct hd_struct *part)
117 unsigned int inflight = 0;
120 for_each_possible_cpu(cpu) {
121 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
122 part_stat_local_read_cpu(part, in_flight[1], cpu);
124 if ((int)inflight < 0)
130 static void part_in_flight_rw(struct hd_struct *part, unsigned int inflight[2])
136 for_each_possible_cpu(cpu) {
137 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
138 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
140 if ((int)inflight[0] < 0)
142 if ((int)inflight[1] < 0)
146 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
148 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
150 if (unlikely(partno < 0 || partno >= ptbl->len))
152 return rcu_dereference(ptbl->part[partno]);
156 * disk_get_part - get partition
157 * @disk: disk to look partition from
158 * @partno: partition number
160 * Look for partition @partno from @disk. If found, increment
161 * reference count and return it.
167 * Pointer to the found partition on success, NULL if not found.
169 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
171 struct hd_struct *part;
174 part = __disk_get_part(disk, partno);
176 get_device(part_to_dev(part));
183 * disk_part_iter_init - initialize partition iterator
184 * @piter: iterator to initialize
185 * @disk: disk to iterate over
186 * @flags: DISK_PITER_* flags
188 * Initialize @piter so that it iterates over partitions of @disk.
193 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
196 struct disk_part_tbl *ptbl;
199 ptbl = rcu_dereference(disk->part_tbl);
204 if (flags & DISK_PITER_REVERSE)
205 piter->idx = ptbl->len - 1;
206 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
211 piter->flags = flags;
215 EXPORT_SYMBOL_GPL(disk_part_iter_init);
218 * disk_part_iter_next - proceed iterator to the next partition and return it
219 * @piter: iterator of interest
221 * Proceed @piter to the next partition and return it.
226 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
228 struct disk_part_tbl *ptbl;
231 /* put the last partition */
232 disk_put_part(piter->part);
237 ptbl = rcu_dereference(piter->disk->part_tbl);
239 /* determine iteration parameters */
240 if (piter->flags & DISK_PITER_REVERSE) {
242 if (piter->flags & (DISK_PITER_INCL_PART0 |
243 DISK_PITER_INCL_EMPTY_PART0))
252 /* iterate to the next partition */
253 for (; piter->idx != end; piter->idx += inc) {
254 struct hd_struct *part;
256 part = rcu_dereference(ptbl->part[piter->idx]);
259 get_device(part_to_dev(part));
261 if (!part_nr_sects_read(part) &&
262 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
263 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
265 put_device(part_to_dev(part));
278 EXPORT_SYMBOL_GPL(disk_part_iter_next);
281 * disk_part_iter_exit - finish up partition iteration
282 * @piter: iter of interest
284 * Called when iteration is over. Cleans up @piter.
289 void disk_part_iter_exit(struct disk_part_iter *piter)
291 disk_put_part(piter->part);
294 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
296 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
298 return part->start_sect <= sector &&
299 sector < part->start_sect + part_nr_sects_read(part);
303 * disk_map_sector_rcu - map sector to partition
304 * @disk: gendisk of interest
305 * @sector: sector to map
307 * Find out which partition @sector maps to on @disk. This is
308 * primarily used for stats accounting.
311 * RCU read locked. The returned partition pointer is always valid
312 * because its refcount is grabbed except for part0, which lifetime
313 * is same with the disk.
316 * Found partition on success, part0 is returned if no partition matches
317 * or the matched partition is being deleted.
319 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
321 struct disk_part_tbl *ptbl;
322 struct hd_struct *part;
326 ptbl = rcu_dereference(disk->part_tbl);
328 part = rcu_dereference(ptbl->last_lookup);
329 if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
332 for (i = 1; i < ptbl->len; i++) {
333 part = rcu_dereference(ptbl->part[i]);
335 if (part && sector_in_part(part, sector)) {
337 * only live partition can be cached for lookup,
338 * so use-after-free on cached & deleting partition
341 if (!hd_struct_try_get(part))
343 rcu_assign_pointer(ptbl->last_lookup, part);
355 * disk_has_partitions
356 * @disk: gendisk of interest
358 * Walk through the partition table and check if valid partition exists.
364 * True if the gendisk has at least one valid non-zero size partition.
367 bool disk_has_partitions(struct gendisk *disk)
369 struct disk_part_tbl *ptbl;
374 ptbl = rcu_dereference(disk->part_tbl);
376 /* Iterate partitions skipping the whole device at index 0 */
377 for (i = 1; i < ptbl->len; i++) {
378 if (rcu_dereference(ptbl->part[i])) {
388 EXPORT_SYMBOL_GPL(disk_has_partitions);
391 * Can be deleted altogether. Later.
394 #define BLKDEV_MAJOR_HASH_SIZE 255
395 static struct blk_major_name {
396 struct blk_major_name *next;
399 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
401 /* index in the above - for now: assume no multimajor ranges */
402 static inline int major_to_index(unsigned major)
404 return major % BLKDEV_MAJOR_HASH_SIZE;
407 #ifdef CONFIG_PROC_FS
408 void blkdev_show(struct seq_file *seqf, off_t offset)
410 struct blk_major_name *dp;
412 mutex_lock(&block_class_lock);
413 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
414 if (dp->major == offset)
415 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
416 mutex_unlock(&block_class_lock);
418 #endif /* CONFIG_PROC_FS */
421 * register_blkdev - register a new block device
423 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
424 * @major = 0, try to allocate any unused major number.
425 * @name: the name of the new block device as a zero terminated string
427 * The @name must be unique within the system.
429 * The return value depends on the @major input parameter:
431 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
432 * then the function returns zero on success, or a negative error code
433 * - if any unused major number was requested with @major = 0 parameter
434 * then the return value is the allocated major number in range
435 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
437 * See Documentation/admin-guide/devices.txt for the list of allocated
440 int register_blkdev(unsigned int major, const char *name)
442 struct blk_major_name **n, *p;
445 mutex_lock(&block_class_lock);
449 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
450 if (major_names[index] == NULL)
455 printk("%s: failed to get major for %s\n",
464 if (major >= BLKDEV_MAJOR_MAX) {
465 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
466 __func__, major, BLKDEV_MAJOR_MAX-1, name);
472 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
479 strlcpy(p->name, name, sizeof(p->name));
481 index = major_to_index(major);
483 for (n = &major_names[index]; *n; n = &(*n)->next) {
484 if ((*n)->major == major)
493 printk("register_blkdev: cannot get major %u for %s\n",
498 mutex_unlock(&block_class_lock);
502 EXPORT_SYMBOL(register_blkdev);
504 void unregister_blkdev(unsigned int major, const char *name)
506 struct blk_major_name **n;
507 struct blk_major_name *p = NULL;
508 int index = major_to_index(major);
510 mutex_lock(&block_class_lock);
511 for (n = &major_names[index]; *n; n = &(*n)->next)
512 if ((*n)->major == major)
514 if (!*n || strcmp((*n)->name, name)) {
520 mutex_unlock(&block_class_lock);
524 EXPORT_SYMBOL(unregister_blkdev);
526 static struct kobj_map *bdev_map;
529 * blk_mangle_minor - scatter minor numbers apart
530 * @minor: minor number to mangle
532 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
533 * is enabled. Mangling twice gives the original value.
541 static int blk_mangle_minor(int minor)
543 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
546 for (i = 0; i < MINORBITS / 2; i++) {
547 int low = minor & (1 << i);
548 int high = minor & (1 << (MINORBITS - 1 - i));
549 int distance = MINORBITS - 1 - 2 * i;
551 minor ^= low | high; /* clear both bits */
552 low <<= distance; /* swap the positions */
554 minor |= low | high; /* and set */
561 * blk_alloc_devt - allocate a dev_t for a partition
562 * @part: partition to allocate dev_t for
563 * @devt: out parameter for resulting dev_t
565 * Allocate a dev_t for block device.
568 * 0 on success, allocated dev_t is returned in *@devt. -errno on
574 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
576 struct gendisk *disk = part_to_disk(part);
579 /* in consecutive minor range? */
580 if (part->partno < disk->minors) {
581 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
585 /* allocate ext devt */
586 idr_preload(GFP_KERNEL);
588 spin_lock_bh(&ext_devt_lock);
589 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
590 spin_unlock_bh(&ext_devt_lock);
594 return idx == -ENOSPC ? -EBUSY : idx;
596 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
601 * blk_free_devt - free a dev_t
602 * @devt: dev_t to free
604 * Free @devt which was allocated using blk_alloc_devt().
609 void blk_free_devt(dev_t devt)
611 if (devt == MKDEV(0, 0))
614 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
615 spin_lock_bh(&ext_devt_lock);
616 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
617 spin_unlock_bh(&ext_devt_lock);
622 * We invalidate devt by assigning NULL pointer for devt in idr.
624 void blk_invalidate_devt(dev_t devt)
626 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
627 spin_lock_bh(&ext_devt_lock);
628 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
629 spin_unlock_bh(&ext_devt_lock);
633 static char *bdevt_str(dev_t devt, char *buf)
635 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
636 char tbuf[BDEVT_SIZE];
637 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
638 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
640 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
646 * Register device numbers dev..(dev+range-1)
647 * range must be nonzero
648 * The hash chain is sorted on range, so that subranges can override.
650 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
651 struct kobject *(*probe)(dev_t, int *, void *),
652 int (*lock)(dev_t, void *), void *data)
654 kobj_map(bdev_map, devt, range, module, probe, lock, data);
657 EXPORT_SYMBOL(blk_register_region);
659 void blk_unregister_region(dev_t devt, unsigned long range)
661 kobj_unmap(bdev_map, devt, range);
664 EXPORT_SYMBOL(blk_unregister_region);
666 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
668 struct gendisk *p = data;
670 return &disk_to_dev(p)->kobj;
673 static int exact_lock(dev_t devt, void *data)
675 struct gendisk *p = data;
677 if (!get_disk_and_module(p))
682 static void disk_scan_partitions(struct gendisk *disk)
684 struct block_device *bdev;
686 if (!get_capacity(disk) || !disk_part_scan_enabled(disk))
689 set_bit(GD_NEED_PART_SCAN, &disk->state);
690 bdev = blkdev_get_by_dev(disk_devt(disk), FMODE_READ, NULL);
692 blkdev_put(bdev, FMODE_READ);
695 static void register_disk(struct device *parent, struct gendisk *disk,
696 const struct attribute_group **groups)
698 struct device *ddev = disk_to_dev(disk);
699 struct disk_part_iter piter;
700 struct hd_struct *part;
703 ddev->parent = parent;
705 dev_set_name(ddev, "%s", disk->disk_name);
707 /* delay uevents, until we scanned partition table */
708 dev_set_uevent_suppress(ddev, 1);
711 WARN_ON(ddev->groups);
712 ddev->groups = groups;
714 if (device_add(ddev))
716 if (!sysfs_deprecated) {
717 err = sysfs_create_link(block_depr, &ddev->kobj,
718 kobject_name(&ddev->kobj));
726 * avoid probable deadlock caused by allocating memory with
727 * GFP_KERNEL in runtime_resume callback of its all ancestor
730 pm_runtime_set_memalloc_noio(ddev, true);
732 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
733 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
735 if (disk->flags & GENHD_FL_HIDDEN)
738 disk_scan_partitions(disk);
740 /* announce disk after possible partitions are created */
741 dev_set_uevent_suppress(ddev, 0);
742 kobject_uevent(&ddev->kobj, KOBJ_ADD);
744 /* announce possible partitions */
745 disk_part_iter_init(&piter, disk, 0);
746 while ((part = disk_part_iter_next(&piter)))
747 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
748 disk_part_iter_exit(&piter);
750 if (disk->queue->backing_dev_info->dev) {
751 err = sysfs_create_link(&ddev->kobj,
752 &disk->queue->backing_dev_info->dev->kobj,
759 * __device_add_disk - add disk information to kernel list
760 * @parent: parent device for the disk
761 * @disk: per-device partitioning information
762 * @groups: Additional per-device sysfs groups
763 * @register_queue: register the queue if set to true
765 * This function registers the partitioning information in @disk
768 * FIXME: error handling
770 static void __device_add_disk(struct device *parent, struct gendisk *disk,
771 const struct attribute_group **groups,
778 * The disk queue should now be all set with enough information about
779 * the device for the elevator code to pick an adequate default
780 * elevator if one is needed, that is, for devices requesting queue
784 elevator_init_mq(disk->queue);
786 /* minors == 0 indicates to use ext devt from part0 and should
787 * be accompanied with EXT_DEVT flag. Make sure all
788 * parameters make sense.
790 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
791 WARN_ON(!disk->minors &&
792 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
794 disk->flags |= GENHD_FL_UP;
796 retval = blk_alloc_devt(&disk->part0, &devt);
801 disk->major = MAJOR(devt);
802 disk->first_minor = MINOR(devt);
804 disk_alloc_events(disk);
806 if (disk->flags & GENHD_FL_HIDDEN) {
808 * Don't let hidden disks show up in /proc/partitions,
809 * and don't bother scanning for partitions either.
811 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
812 disk->flags |= GENHD_FL_NO_PART_SCAN;
814 struct backing_dev_info *bdi = disk->queue->backing_dev_info;
815 struct device *dev = disk_to_dev(disk);
818 /* Register BDI before referencing it from bdev */
820 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
822 bdi_set_owner(bdi, dev);
823 blk_register_region(disk_devt(disk), disk->minors, NULL,
824 exact_match, exact_lock, disk);
826 register_disk(parent, disk, groups);
828 blk_register_queue(disk);
831 * Take an extra ref on queue which will be put on disk_release()
832 * so that it sticks around as long as @disk is there.
834 WARN_ON_ONCE(!blk_get_queue(disk->queue));
836 disk_add_events(disk);
837 blk_integrity_add(disk);
840 void device_add_disk(struct device *parent, struct gendisk *disk,
841 const struct attribute_group **groups)
844 __device_add_disk(parent, disk, groups, true);
846 EXPORT_SYMBOL(device_add_disk);
848 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
850 __device_add_disk(parent, disk, NULL, false);
852 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
854 static void invalidate_partition(struct gendisk *disk, int partno)
856 struct block_device *bdev;
858 bdev = bdget_disk(disk, partno);
863 __invalidate_device(bdev, true);
866 * Unhash the bdev inode for this device so that it gets evicted as soon
867 * as last inode reference is dropped.
869 remove_inode_hash(bdev->bd_inode);
874 * del_gendisk - remove the gendisk
875 * @disk: the struct gendisk to remove
877 * Removes the gendisk and all its associated resources. This deletes the
878 * partitions associated with the gendisk, and unregisters the associated
881 * This is the counter to the respective __device_add_disk() call.
883 * The final removal of the struct gendisk happens when its refcount reaches 0
884 * with put_disk(), which should be called after del_gendisk(), if
885 * __device_add_disk() was used.
887 * Drivers exist which depend on the release of the gendisk to be synchronous,
888 * it should not be deferred.
892 void del_gendisk(struct gendisk *disk)
894 struct disk_part_iter piter;
895 struct hd_struct *part;
899 blk_integrity_del(disk);
900 disk_del_events(disk);
903 * Block lookups of the disk until all bdevs are unhashed and the
904 * disk is marked as dead (GENHD_FL_UP cleared).
906 down_write(&disk->lookup_sem);
907 /* invalidate stuff */
908 disk_part_iter_init(&piter, disk,
909 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
910 while ((part = disk_part_iter_next(&piter))) {
911 invalidate_partition(disk, part->partno);
912 delete_partition(part);
914 disk_part_iter_exit(&piter);
916 invalidate_partition(disk, 0);
917 set_capacity(disk, 0);
918 disk->flags &= ~GENHD_FL_UP;
919 up_write(&disk->lookup_sem);
921 if (!(disk->flags & GENHD_FL_HIDDEN))
922 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
925 * Unregister bdi before releasing device numbers (as they can
926 * get reused and we'd get clashes in sysfs).
928 if (!(disk->flags & GENHD_FL_HIDDEN))
929 bdi_unregister(disk->queue->backing_dev_info);
930 blk_unregister_queue(disk);
935 if (!(disk->flags & GENHD_FL_HIDDEN))
936 blk_unregister_region(disk_devt(disk), disk->minors);
938 * Remove gendisk pointer from idr so that it cannot be looked up
939 * while RCU period before freeing gendisk is running to prevent
940 * use-after-free issues. Note that the device number stays
941 * "in-use" until we really free the gendisk.
943 blk_invalidate_devt(disk_devt(disk));
945 kobject_put(disk->part0.holder_dir);
946 kobject_put(disk->slave_dir);
948 part_stat_set_all(&disk->part0, 0);
949 disk->part0.stamp = 0;
950 if (!sysfs_deprecated)
951 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
952 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
953 device_del(disk_to_dev(disk));
955 EXPORT_SYMBOL(del_gendisk);
957 /* sysfs access to bad-blocks list. */
958 static ssize_t disk_badblocks_show(struct device *dev,
959 struct device_attribute *attr,
962 struct gendisk *disk = dev_to_disk(dev);
965 return sprintf(page, "\n");
967 return badblocks_show(disk->bb, page, 0);
970 static ssize_t disk_badblocks_store(struct device *dev,
971 struct device_attribute *attr,
972 const char *page, size_t len)
974 struct gendisk *disk = dev_to_disk(dev);
979 return badblocks_store(disk->bb, page, len, 0);
983 * get_gendisk - get partitioning information for a given device
984 * @devt: device to get partitioning information for
985 * @partno: returned partition index
987 * This function gets the structure containing partitioning
988 * information for the given device @devt.
992 struct gendisk *get_gendisk(dev_t devt, int *partno)
994 struct gendisk *disk = NULL;
998 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
999 struct kobject *kobj;
1001 kobj = kobj_lookup(bdev_map, devt, partno);
1003 disk = dev_to_disk(kobj_to_dev(kobj));
1005 struct hd_struct *part;
1007 spin_lock_bh(&ext_devt_lock);
1008 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
1009 if (part && get_disk_and_module(part_to_disk(part))) {
1010 *partno = part->partno;
1011 disk = part_to_disk(part);
1013 spin_unlock_bh(&ext_devt_lock);
1020 * Synchronize with del_gendisk() to not return disk that is being
1023 down_read(&disk->lookup_sem);
1024 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1025 !(disk->flags & GENHD_FL_UP))) {
1026 up_read(&disk->lookup_sem);
1027 put_disk_and_module(disk);
1030 up_read(&disk->lookup_sem);
1036 * bdget_disk - do bdget() by gendisk and partition number
1037 * @disk: gendisk of interest
1038 * @partno: partition number
1040 * Find partition @partno from @disk, do bdget() on it.
1046 * Resulting block_device on success, NULL on failure.
1048 struct block_device *bdget_disk(struct gendisk *disk, int partno)
1050 struct hd_struct *part;
1051 struct block_device *bdev = NULL;
1053 part = disk_get_part(disk, partno);
1055 bdev = bdget_part(part);
1056 disk_put_part(part);
1060 EXPORT_SYMBOL(bdget_disk);
1063 * print a full list of all partitions - intended for places where the root
1064 * filesystem can't be mounted and thus to give the victim some idea of what
1067 void __init printk_all_partitions(void)
1069 struct class_dev_iter iter;
1072 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1073 while ((dev = class_dev_iter_next(&iter))) {
1074 struct gendisk *disk = dev_to_disk(dev);
1075 struct disk_part_iter piter;
1076 struct hd_struct *part;
1077 char name_buf[BDEVNAME_SIZE];
1078 char devt_buf[BDEVT_SIZE];
1081 * Don't show empty devices or things that have been
1084 if (get_capacity(disk) == 0 ||
1085 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1089 * Note, unlike /proc/partitions, I am showing the
1090 * numbers in hex - the same format as the root=
1093 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1094 while ((part = disk_part_iter_next(&piter))) {
1095 bool is_part0 = part == &disk->part0;
1097 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1098 bdevt_str(part_devt(part), devt_buf),
1099 (unsigned long long)part_nr_sects_read(part) >> 1
1100 , disk_name(disk, part->partno, name_buf),
1101 part->info ? part->info->uuid : "");
1103 if (dev->parent && dev->parent->driver)
1104 printk(" driver: %s\n",
1105 dev->parent->driver->name);
1107 printk(" (driver?)\n");
1111 disk_part_iter_exit(&piter);
1113 class_dev_iter_exit(&iter);
1116 #ifdef CONFIG_PROC_FS
1118 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1121 struct class_dev_iter *iter;
1124 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1126 return ERR_PTR(-ENOMEM);
1128 seqf->private = iter;
1129 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1131 dev = class_dev_iter_next(iter);
1136 return dev_to_disk(dev);
1139 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1144 dev = class_dev_iter_next(seqf->private);
1146 return dev_to_disk(dev);
1151 static void disk_seqf_stop(struct seq_file *seqf, void *v)
1153 struct class_dev_iter *iter = seqf->private;
1155 /* stop is called even after start failed :-( */
1157 class_dev_iter_exit(iter);
1159 seqf->private = NULL;
1163 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1167 p = disk_seqf_start(seqf, pos);
1168 if (!IS_ERR_OR_NULL(p) && !*pos)
1169 seq_puts(seqf, "major minor #blocks name\n\n");
1173 static int show_partition(struct seq_file *seqf, void *v)
1175 struct gendisk *sgp = v;
1176 struct disk_part_iter piter;
1177 struct hd_struct *part;
1178 char buf[BDEVNAME_SIZE];
1180 /* Don't show non-partitionable removeable devices or empty devices */
1181 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1182 (sgp->flags & GENHD_FL_REMOVABLE)))
1184 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1187 /* show the full disk and all non-0 size partitions of it */
1188 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1189 while ((part = disk_part_iter_next(&piter)))
1190 seq_printf(seqf, "%4d %7d %10llu %s\n",
1191 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1192 (unsigned long long)part_nr_sects_read(part) >> 1,
1193 disk_name(sgp, part->partno, buf));
1194 disk_part_iter_exit(&piter);
1199 static const struct seq_operations partitions_op = {
1200 .start = show_partition_start,
1201 .next = disk_seqf_next,
1202 .stop = disk_seqf_stop,
1203 .show = show_partition
1208 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1210 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1211 /* Make old-style 2.4 aliases work */
1212 request_module("block-major-%d", MAJOR(devt));
1216 static int __init genhd_device_init(void)
1220 block_class.dev_kobj = sysfs_dev_block_kobj;
1221 error = class_register(&block_class);
1222 if (unlikely(error))
1224 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1227 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1229 /* create top-level block dir */
1230 if (!sysfs_deprecated)
1231 block_depr = kobject_create_and_add("block", NULL);
1235 subsys_initcall(genhd_device_init);
1237 static ssize_t disk_range_show(struct device *dev,
1238 struct device_attribute *attr, char *buf)
1240 struct gendisk *disk = dev_to_disk(dev);
1242 return sprintf(buf, "%d\n", disk->minors);
1245 static ssize_t disk_ext_range_show(struct device *dev,
1246 struct device_attribute *attr, char *buf)
1248 struct gendisk *disk = dev_to_disk(dev);
1250 return sprintf(buf, "%d\n", disk_max_parts(disk));
1253 static ssize_t disk_removable_show(struct device *dev,
1254 struct device_attribute *attr, char *buf)
1256 struct gendisk *disk = dev_to_disk(dev);
1258 return sprintf(buf, "%d\n",
1259 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1262 static ssize_t disk_hidden_show(struct device *dev,
1263 struct device_attribute *attr, char *buf)
1265 struct gendisk *disk = dev_to_disk(dev);
1267 return sprintf(buf, "%d\n",
1268 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1271 static ssize_t disk_ro_show(struct device *dev,
1272 struct device_attribute *attr, char *buf)
1274 struct gendisk *disk = dev_to_disk(dev);
1276 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1279 ssize_t part_size_show(struct device *dev,
1280 struct device_attribute *attr, char *buf)
1282 struct hd_struct *p = dev_to_part(dev);
1284 return sprintf(buf, "%llu\n",
1285 (unsigned long long)part_nr_sects_read(p));
1288 ssize_t part_stat_show(struct device *dev,
1289 struct device_attribute *attr, char *buf)
1291 struct hd_struct *p = dev_to_part(dev);
1292 struct request_queue *q = part_to_disk(p)->queue;
1293 struct disk_stats stat;
1294 unsigned int inflight;
1296 part_stat_read_all(p, &stat);
1298 inflight = blk_mq_in_flight(q, p);
1300 inflight = part_in_flight(p);
1303 "%8lu %8lu %8llu %8u "
1304 "%8lu %8lu %8llu %8u "
1306 "%8lu %8lu %8llu %8u "
1309 stat.ios[STAT_READ],
1310 stat.merges[STAT_READ],
1311 (unsigned long long)stat.sectors[STAT_READ],
1312 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1313 stat.ios[STAT_WRITE],
1314 stat.merges[STAT_WRITE],
1315 (unsigned long long)stat.sectors[STAT_WRITE],
1316 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1318 jiffies_to_msecs(stat.io_ticks),
1319 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1320 stat.nsecs[STAT_WRITE] +
1321 stat.nsecs[STAT_DISCARD] +
1322 stat.nsecs[STAT_FLUSH],
1324 stat.ios[STAT_DISCARD],
1325 stat.merges[STAT_DISCARD],
1326 (unsigned long long)stat.sectors[STAT_DISCARD],
1327 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1328 stat.ios[STAT_FLUSH],
1329 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1332 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1335 struct hd_struct *p = dev_to_part(dev);
1336 struct request_queue *q = part_to_disk(p)->queue;
1337 unsigned int inflight[2];
1340 blk_mq_in_flight_rw(q, p, inflight);
1342 part_in_flight_rw(p, inflight);
1344 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1347 static ssize_t disk_capability_show(struct device *dev,
1348 struct device_attribute *attr, char *buf)
1350 struct gendisk *disk = dev_to_disk(dev);
1352 return sprintf(buf, "%x\n", disk->flags);
1355 static ssize_t disk_alignment_offset_show(struct device *dev,
1356 struct device_attribute *attr,
1359 struct gendisk *disk = dev_to_disk(dev);
1361 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1364 static ssize_t disk_discard_alignment_show(struct device *dev,
1365 struct device_attribute *attr,
1368 struct gendisk *disk = dev_to_disk(dev);
1370 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1373 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1374 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1375 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1376 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1377 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1378 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1379 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1380 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1381 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1382 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1383 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1384 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1386 #ifdef CONFIG_FAIL_MAKE_REQUEST
1387 ssize_t part_fail_show(struct device *dev,
1388 struct device_attribute *attr, char *buf)
1390 struct hd_struct *p = dev_to_part(dev);
1392 return sprintf(buf, "%d\n", p->make_it_fail);
1395 ssize_t part_fail_store(struct device *dev,
1396 struct device_attribute *attr,
1397 const char *buf, size_t count)
1399 struct hd_struct *p = dev_to_part(dev);
1402 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1403 p->make_it_fail = (i == 0) ? 0 : 1;
1408 static struct device_attribute dev_attr_fail =
1409 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1410 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1412 #ifdef CONFIG_FAIL_IO_TIMEOUT
1413 static struct device_attribute dev_attr_fail_timeout =
1414 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1417 static struct attribute *disk_attrs[] = {
1418 &dev_attr_range.attr,
1419 &dev_attr_ext_range.attr,
1420 &dev_attr_removable.attr,
1421 &dev_attr_hidden.attr,
1423 &dev_attr_size.attr,
1424 &dev_attr_alignment_offset.attr,
1425 &dev_attr_discard_alignment.attr,
1426 &dev_attr_capability.attr,
1427 &dev_attr_stat.attr,
1428 &dev_attr_inflight.attr,
1429 &dev_attr_badblocks.attr,
1430 #ifdef CONFIG_FAIL_MAKE_REQUEST
1431 &dev_attr_fail.attr,
1433 #ifdef CONFIG_FAIL_IO_TIMEOUT
1434 &dev_attr_fail_timeout.attr,
1439 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1441 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1442 struct gendisk *disk = dev_to_disk(dev);
1444 if (a == &dev_attr_badblocks.attr && !disk->bb)
1449 static struct attribute_group disk_attr_group = {
1450 .attrs = disk_attrs,
1451 .is_visible = disk_visible,
1454 static const struct attribute_group *disk_attr_groups[] = {
1460 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1461 * @disk: disk to replace part_tbl for
1462 * @new_ptbl: new part_tbl to install
1464 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1465 * original ptbl is freed using RCU callback.
1468 * Matching bd_mutex locked or the caller is the only user of @disk.
1470 static void disk_replace_part_tbl(struct gendisk *disk,
1471 struct disk_part_tbl *new_ptbl)
1473 struct disk_part_tbl *old_ptbl =
1474 rcu_dereference_protected(disk->part_tbl, 1);
1476 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1479 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1480 kfree_rcu(old_ptbl, rcu_head);
1485 * disk_expand_part_tbl - expand disk->part_tbl
1486 * @disk: disk to expand part_tbl for
1487 * @partno: expand such that this partno can fit in
1489 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1490 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1493 * Matching bd_mutex locked or the caller is the only user of @disk.
1497 * 0 on success, -errno on failure.
1499 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1501 struct disk_part_tbl *old_ptbl =
1502 rcu_dereference_protected(disk->part_tbl, 1);
1503 struct disk_part_tbl *new_ptbl;
1504 int len = old_ptbl ? old_ptbl->len : 0;
1508 * check for int overflow, since we can get here from blkpg_ioctl()
1509 * with a user passed 'partno'.
1511 target = partno + 1;
1515 /* disk_max_parts() is zero during initialization, ignore if so */
1516 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1522 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1527 new_ptbl->len = target;
1529 for (i = 0; i < len; i++)
1530 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1532 disk_replace_part_tbl(disk, new_ptbl);
1537 * disk_release - releases all allocated resources of the gendisk
1538 * @dev: the device representing this disk
1540 * This function releases all allocated resources of the gendisk.
1542 * The struct gendisk refcount is incremented with get_gendisk() or
1543 * get_disk_and_module(), and its refcount is decremented with
1544 * put_disk_and_module() or put_disk(). Once the refcount reaches 0 this
1545 * function is called.
1547 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1548 * assigned. Since the request_queue sits on top of the gendisk for these
1549 * drivers we also call blk_put_queue() for them, and we expect the
1550 * request_queue refcount to reach 0 at this point, and so the request_queue
1551 * will also be freed prior to the disk.
1553 * Context: can sleep
1555 static void disk_release(struct device *dev)
1557 struct gendisk *disk = dev_to_disk(dev);
1561 blk_free_devt(dev->devt);
1562 disk_release_events(disk);
1563 kfree(disk->random);
1564 disk_replace_part_tbl(disk, NULL);
1565 hd_free_part(&disk->part0);
1567 blk_put_queue(disk->queue);
1570 struct class block_class = {
1574 static char *block_devnode(struct device *dev, umode_t *mode,
1575 kuid_t *uid, kgid_t *gid)
1577 struct gendisk *disk = dev_to_disk(dev);
1579 if (disk->fops->devnode)
1580 return disk->fops->devnode(disk, mode);
1584 const struct device_type disk_type = {
1586 .groups = disk_attr_groups,
1587 .release = disk_release,
1588 .devnode = block_devnode,
1591 #ifdef CONFIG_PROC_FS
1593 * aggregate disk stat collector. Uses the same stats that the sysfs
1594 * entries do, above, but makes them available through one seq_file.
1596 * The output looks suspiciously like /proc/partitions with a bunch of
1599 static int diskstats_show(struct seq_file *seqf, void *v)
1601 struct gendisk *gp = v;
1602 struct disk_part_iter piter;
1603 struct hd_struct *hd;
1604 char buf[BDEVNAME_SIZE];
1605 unsigned int inflight;
1606 struct disk_stats stat;
1609 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1610 seq_puts(seqf, "major minor name"
1611 " rio rmerge rsect ruse wio wmerge "
1612 "wsect wuse running use aveq"
1616 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1617 while ((hd = disk_part_iter_next(&piter))) {
1618 part_stat_read_all(hd, &stat);
1619 if (queue_is_mq(gp->queue))
1620 inflight = blk_mq_in_flight(gp->queue, hd);
1622 inflight = part_in_flight(hd);
1624 seq_printf(seqf, "%4d %7d %s "
1631 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1632 disk_name(gp, hd->partno, buf),
1633 stat.ios[STAT_READ],
1634 stat.merges[STAT_READ],
1635 stat.sectors[STAT_READ],
1636 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1638 stat.ios[STAT_WRITE],
1639 stat.merges[STAT_WRITE],
1640 stat.sectors[STAT_WRITE],
1641 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1644 jiffies_to_msecs(stat.io_ticks),
1645 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1646 stat.nsecs[STAT_WRITE] +
1647 stat.nsecs[STAT_DISCARD] +
1648 stat.nsecs[STAT_FLUSH],
1650 stat.ios[STAT_DISCARD],
1651 stat.merges[STAT_DISCARD],
1652 stat.sectors[STAT_DISCARD],
1653 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1655 stat.ios[STAT_FLUSH],
1656 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1660 disk_part_iter_exit(&piter);
1665 static const struct seq_operations diskstats_op = {
1666 .start = disk_seqf_start,
1667 .next = disk_seqf_next,
1668 .stop = disk_seqf_stop,
1669 .show = diskstats_show
1672 static int __init proc_genhd_init(void)
1674 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1675 proc_create_seq("partitions", 0, NULL, &partitions_op);
1678 module_init(proc_genhd_init);
1679 #endif /* CONFIG_PROC_FS */
1681 dev_t blk_lookup_devt(const char *name, int partno)
1683 dev_t devt = MKDEV(0, 0);
1684 struct class_dev_iter iter;
1687 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1688 while ((dev = class_dev_iter_next(&iter))) {
1689 struct gendisk *disk = dev_to_disk(dev);
1690 struct hd_struct *part;
1692 if (strcmp(dev_name(dev), name))
1695 if (partno < disk->minors) {
1696 /* We need to return the right devno, even
1697 * if the partition doesn't exist yet.
1699 devt = MKDEV(MAJOR(dev->devt),
1700 MINOR(dev->devt) + partno);
1703 part = disk_get_part(disk, partno);
1705 devt = part_devt(part);
1706 disk_put_part(part);
1709 disk_put_part(part);
1711 class_dev_iter_exit(&iter);
1715 struct gendisk *__alloc_disk_node(int minors, int node_id)
1717 struct gendisk *disk;
1718 struct disk_part_tbl *ptbl;
1720 if (minors > DISK_MAX_PARTS) {
1722 "block: can't allocate more than %d partitions\n",
1724 minors = DISK_MAX_PARTS;
1727 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1731 disk->part0.dkstats = alloc_percpu(struct disk_stats);
1732 if (!disk->part0.dkstats)
1735 init_rwsem(&disk->lookup_sem);
1736 disk->node_id = node_id;
1737 if (disk_expand_part_tbl(disk, 0)) {
1738 free_percpu(disk->part0.dkstats);
1742 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1743 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1746 * set_capacity() and get_capacity() currently don't use
1747 * seqcounter to read/update the part0->nr_sects. Still init
1748 * the counter as we can read the sectors in IO submission
1749 * patch using seqence counters.
1751 * TODO: Ideally set_capacity() and get_capacity() should be
1752 * converted to make use of bd_mutex and sequence counters.
1754 hd_sects_seq_init(&disk->part0);
1755 if (hd_ref_init(&disk->part0))
1756 goto out_free_part0;
1758 disk->minors = minors;
1759 rand_initialize_disk(disk);
1760 disk_to_dev(disk)->class = &block_class;
1761 disk_to_dev(disk)->type = &disk_type;
1762 device_initialize(disk_to_dev(disk));
1766 hd_free_part(&disk->part0);
1771 EXPORT_SYMBOL(__alloc_disk_node);
1774 * get_disk_and_module - increments the gendisk and gendisk fops module refcount
1775 * @disk: the struct gendisk to increment the refcount for
1777 * This increments the refcount for the struct gendisk, and the gendisk's
1778 * fops module owner.
1780 * Context: Any context.
1782 struct kobject *get_disk_and_module(struct gendisk *disk)
1784 struct module *owner;
1785 struct kobject *kobj;
1789 owner = disk->fops->owner;
1790 if (owner && !try_module_get(owner))
1792 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1800 EXPORT_SYMBOL(get_disk_and_module);
1803 * put_disk - decrements the gendisk refcount
1804 * @disk: the struct gendisk to decrement the refcount for
1806 * This decrements the refcount for the struct gendisk. When this reaches 0
1807 * we'll have disk_release() called.
1809 * Context: Any context, but the last reference must not be dropped from
1812 void put_disk(struct gendisk *disk)
1815 kobject_put(&disk_to_dev(disk)->kobj);
1817 EXPORT_SYMBOL(put_disk);
1820 * put_disk_and_module - decrements the module and gendisk refcount
1821 * @disk: the struct gendisk to decrement the refcount for
1823 * This is a counterpart of get_disk_and_module() and thus also of
1826 * Context: Any context, but the last reference must not be dropped from
1829 void put_disk_and_module(struct gendisk *disk)
1832 struct module *owner = disk->fops->owner;
1838 EXPORT_SYMBOL(put_disk_and_module);
1840 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1842 char event[] = "DISK_RO=1";
1843 char *envp[] = { event, NULL };
1847 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1850 void set_device_ro(struct block_device *bdev, int flag)
1852 bdev->bd_part->policy = flag;
1855 EXPORT_SYMBOL(set_device_ro);
1857 void set_disk_ro(struct gendisk *disk, int flag)
1859 struct disk_part_iter piter;
1860 struct hd_struct *part;
1862 if (disk->part0.policy != flag) {
1863 set_disk_ro_uevent(disk, flag);
1864 disk->part0.policy = flag;
1867 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1868 while ((part = disk_part_iter_next(&piter)))
1869 part->policy = flag;
1870 disk_part_iter_exit(&piter);
1873 EXPORT_SYMBOL(set_disk_ro);
1875 int bdev_read_only(struct block_device *bdev)
1879 return bdev->bd_part->policy;
1882 EXPORT_SYMBOL(bdev_read_only);
1885 * Disk events - monitor disk events like media change and eject request.
1887 struct disk_events {
1888 struct list_head node; /* all disk_event's */
1889 struct gendisk *disk; /* the associated disk */
1892 struct mutex block_mutex; /* protects blocking */
1893 int block; /* event blocking depth */
1894 unsigned int pending; /* events already sent out */
1895 unsigned int clearing; /* events being cleared */
1897 long poll_msecs; /* interval, -1 for default */
1898 struct delayed_work dwork;
1901 static const char *disk_events_strs[] = {
1902 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1903 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1906 static char *disk_uevents[] = {
1907 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1908 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1911 /* list of all disk_events */
1912 static DEFINE_MUTEX(disk_events_mutex);
1913 static LIST_HEAD(disk_events);
1915 /* disable in-kernel polling by default */
1916 static unsigned long disk_events_dfl_poll_msecs;
1918 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1920 struct disk_events *ev = disk->ev;
1921 long intv_msecs = 0;
1924 * If device-specific poll interval is set, always use it. If
1925 * the default is being used, poll if the POLL flag is set.
1927 if (ev->poll_msecs >= 0)
1928 intv_msecs = ev->poll_msecs;
1929 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1930 intv_msecs = disk_events_dfl_poll_msecs;
1932 return msecs_to_jiffies(intv_msecs);
1936 * disk_block_events - block and flush disk event checking
1937 * @disk: disk to block events for
1939 * On return from this function, it is guaranteed that event checking
1940 * isn't in progress and won't happen until unblocked by
1941 * disk_unblock_events(). Events blocking is counted and the actual
1942 * unblocking happens after the matching number of unblocks are done.
1944 * Note that this intentionally does not block event checking from
1945 * disk_clear_events().
1950 void disk_block_events(struct gendisk *disk)
1952 struct disk_events *ev = disk->ev;
1953 unsigned long flags;
1960 * Outer mutex ensures that the first blocker completes canceling
1961 * the event work before further blockers are allowed to finish.
1963 mutex_lock(&ev->block_mutex);
1965 spin_lock_irqsave(&ev->lock, flags);
1966 cancel = !ev->block++;
1967 spin_unlock_irqrestore(&ev->lock, flags);
1970 cancel_delayed_work_sync(&disk->ev->dwork);
1972 mutex_unlock(&ev->block_mutex);
1975 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1977 struct disk_events *ev = disk->ev;
1979 unsigned long flags;
1981 spin_lock_irqsave(&ev->lock, flags);
1983 if (WARN_ON_ONCE(ev->block <= 0))
1989 intv = disk_events_poll_jiffies(disk);
1991 queue_delayed_work(system_freezable_power_efficient_wq,
1994 queue_delayed_work(system_freezable_power_efficient_wq,
1997 spin_unlock_irqrestore(&ev->lock, flags);
2001 * disk_unblock_events - unblock disk event checking
2002 * @disk: disk to unblock events for
2004 * Undo disk_block_events(). When the block count reaches zero, it
2005 * starts events polling if configured.
2008 * Don't care. Safe to call from irq context.
2010 void disk_unblock_events(struct gendisk *disk)
2013 __disk_unblock_events(disk, false);
2017 * disk_flush_events - schedule immediate event checking and flushing
2018 * @disk: disk to check and flush events for
2019 * @mask: events to flush
2021 * Schedule immediate event checking on @disk if not blocked. Events in
2022 * @mask are scheduled to be cleared from the driver. Note that this
2023 * doesn't clear the events from @disk->ev.
2026 * If @mask is non-zero must be called with bdev->bd_mutex held.
2028 void disk_flush_events(struct gendisk *disk, unsigned int mask)
2030 struct disk_events *ev = disk->ev;
2035 spin_lock_irq(&ev->lock);
2036 ev->clearing |= mask;
2038 mod_delayed_work(system_freezable_power_efficient_wq,
2040 spin_unlock_irq(&ev->lock);
2044 * disk_clear_events - synchronously check, clear and return pending events
2045 * @disk: disk to fetch and clear events from
2046 * @mask: mask of events to be fetched and cleared
2048 * Disk events are synchronously checked and pending events in @mask
2049 * are cleared and returned. This ignores the block count.
2054 static unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
2056 struct disk_events *ev = disk->ev;
2057 unsigned int pending;
2058 unsigned int clearing = mask;
2063 disk_block_events(disk);
2066 * store the union of mask and ev->clearing on the stack so that the
2067 * race with disk_flush_events does not cause ambiguity (ev->clearing
2068 * can still be modified even if events are blocked).
2070 spin_lock_irq(&ev->lock);
2071 clearing |= ev->clearing;
2073 spin_unlock_irq(&ev->lock);
2075 disk_check_events(ev, &clearing);
2077 * if ev->clearing is not 0, the disk_flush_events got called in the
2078 * middle of this function, so we want to run the workfn without delay.
2080 __disk_unblock_events(disk, ev->clearing ? true : false);
2082 /* then, fetch and clear pending events */
2083 spin_lock_irq(&ev->lock);
2084 pending = ev->pending & mask;
2085 ev->pending &= ~mask;
2086 spin_unlock_irq(&ev->lock);
2087 WARN_ON_ONCE(clearing & mask);
2093 * bdev_check_media_change - check if a removable media has been changed
2094 * @bdev: block device to check
2096 * Check whether a removable media has been changed, and attempt to free all
2097 * dentries and inodes and invalidates all block device page cache entries in
2100 * Returns %true if the block device changed, or %false if not.
2102 bool bdev_check_media_change(struct block_device *bdev)
2104 unsigned int events;
2106 events = disk_clear_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE |
2107 DISK_EVENT_EJECT_REQUEST);
2108 if (!(events & DISK_EVENT_MEDIA_CHANGE))
2111 if (__invalidate_device(bdev, true))
2112 pr_warn("VFS: busy inodes on changed media %s\n",
2113 bdev->bd_disk->disk_name);
2114 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
2117 EXPORT_SYMBOL(bdev_check_media_change);
2120 * Separate this part out so that a different pointer for clearing_ptr can be
2121 * passed in for disk_clear_events.
2123 static void disk_events_workfn(struct work_struct *work)
2125 struct delayed_work *dwork = to_delayed_work(work);
2126 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2128 disk_check_events(ev, &ev->clearing);
2131 static void disk_check_events(struct disk_events *ev,
2132 unsigned int *clearing_ptr)
2134 struct gendisk *disk = ev->disk;
2135 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2136 unsigned int clearing = *clearing_ptr;
2137 unsigned int events;
2139 int nr_events = 0, i;
2142 events = disk->fops->check_events(disk, clearing);
2144 /* accumulate pending events and schedule next poll if necessary */
2145 spin_lock_irq(&ev->lock);
2147 events &= ~ev->pending;
2148 ev->pending |= events;
2149 *clearing_ptr &= ~clearing;
2151 intv = disk_events_poll_jiffies(disk);
2152 if (!ev->block && intv)
2153 queue_delayed_work(system_freezable_power_efficient_wq,
2156 spin_unlock_irq(&ev->lock);
2159 * Tell userland about new events. Only the events listed in
2160 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2161 * is set. Otherwise, events are processed internally but never
2162 * get reported to userland.
2164 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2165 if ((events & disk->events & (1 << i)) &&
2166 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2167 envp[nr_events++] = disk_uevents[i];
2170 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2174 * A disk events enabled device has the following sysfs nodes under
2175 * its /sys/block/X/ directory.
2177 * events : list of all supported events
2178 * events_async : list of events which can be detected w/o polling
2179 * (always empty, only for backwards compatibility)
2180 * events_poll_msecs : polling interval, 0: disable, -1: system default
2182 static ssize_t __disk_events_show(unsigned int events, char *buf)
2184 const char *delim = "";
2188 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2189 if (events & (1 << i)) {
2190 pos += sprintf(buf + pos, "%s%s",
2191 delim, disk_events_strs[i]);
2195 pos += sprintf(buf + pos, "\n");
2199 static ssize_t disk_events_show(struct device *dev,
2200 struct device_attribute *attr, char *buf)
2202 struct gendisk *disk = dev_to_disk(dev);
2204 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2207 return __disk_events_show(disk->events, buf);
2210 static ssize_t disk_events_async_show(struct device *dev,
2211 struct device_attribute *attr, char *buf)
2216 static ssize_t disk_events_poll_msecs_show(struct device *dev,
2217 struct device_attribute *attr,
2220 struct gendisk *disk = dev_to_disk(dev);
2223 return sprintf(buf, "-1\n");
2225 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2228 static ssize_t disk_events_poll_msecs_store(struct device *dev,
2229 struct device_attribute *attr,
2230 const char *buf, size_t count)
2232 struct gendisk *disk = dev_to_disk(dev);
2235 if (!count || !sscanf(buf, "%ld", &intv))
2238 if (intv < 0 && intv != -1)
2244 disk_block_events(disk);
2245 disk->ev->poll_msecs = intv;
2246 __disk_unblock_events(disk, true);
2251 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2252 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2253 static const DEVICE_ATTR(events_poll_msecs, 0644,
2254 disk_events_poll_msecs_show,
2255 disk_events_poll_msecs_store);
2257 static const struct attribute *disk_events_attrs[] = {
2258 &dev_attr_events.attr,
2259 &dev_attr_events_async.attr,
2260 &dev_attr_events_poll_msecs.attr,
2265 * The default polling interval can be specified by the kernel
2266 * parameter block.events_dfl_poll_msecs which defaults to 0
2267 * (disable). This can also be modified runtime by writing to
2268 * /sys/module/block/parameters/events_dfl_poll_msecs.
2270 static int disk_events_set_dfl_poll_msecs(const char *val,
2271 const struct kernel_param *kp)
2273 struct disk_events *ev;
2276 ret = param_set_ulong(val, kp);
2280 mutex_lock(&disk_events_mutex);
2282 list_for_each_entry(ev, &disk_events, node)
2283 disk_flush_events(ev->disk, 0);
2285 mutex_unlock(&disk_events_mutex);
2290 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2291 .set = disk_events_set_dfl_poll_msecs,
2292 .get = param_get_ulong,
2295 #undef MODULE_PARAM_PREFIX
2296 #define MODULE_PARAM_PREFIX "block."
2298 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2299 &disk_events_dfl_poll_msecs, 0644);
2302 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2304 static void disk_alloc_events(struct gendisk *disk)
2306 struct disk_events *ev;
2308 if (!disk->fops->check_events || !disk->events)
2311 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2313 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2317 INIT_LIST_HEAD(&ev->node);
2319 spin_lock_init(&ev->lock);
2320 mutex_init(&ev->block_mutex);
2322 ev->poll_msecs = -1;
2323 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2328 static void disk_add_events(struct gendisk *disk)
2330 /* FIXME: error handling */
2331 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2332 pr_warn("%s: failed to create sysfs files for events\n",
2338 mutex_lock(&disk_events_mutex);
2339 list_add_tail(&disk->ev->node, &disk_events);
2340 mutex_unlock(&disk_events_mutex);
2343 * Block count is initialized to 1 and the following initial
2344 * unblock kicks it into action.
2346 __disk_unblock_events(disk, true);
2349 static void disk_del_events(struct gendisk *disk)
2352 disk_block_events(disk);
2354 mutex_lock(&disk_events_mutex);
2355 list_del_init(&disk->ev->node);
2356 mutex_unlock(&disk_events_mutex);
2359 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2362 static void disk_release_events(struct gendisk *disk)
2364 /* the block count should be 1 from disk_del_events() */
2365 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);