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);
60 revalidate_disk(disk);
62 if (capacity != size && capacity != 0 && size != 0) {
63 char *envp[] = { "RESIZE=1", NULL };
65 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
69 EXPORT_SYMBOL_GPL(set_capacity_revalidate_and_notify);
72 * Format the device name of the indicated disk into the supplied buffer and
73 * return a pointer to that same buffer for convenience.
75 char *disk_name(struct gendisk *hd, int partno, char *buf)
78 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
79 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
80 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
82 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
87 const char *bdevname(struct block_device *bdev, char *buf)
89 return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
91 EXPORT_SYMBOL(bdevname);
93 static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
97 memset(stat, 0, sizeof(struct disk_stats));
98 for_each_possible_cpu(cpu) {
99 struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
102 for (group = 0; group < NR_STAT_GROUPS; group++) {
103 stat->nsecs[group] += ptr->nsecs[group];
104 stat->sectors[group] += ptr->sectors[group];
105 stat->ios[group] += ptr->ios[group];
106 stat->merges[group] += ptr->merges[group];
109 stat->io_ticks += ptr->io_ticks;
113 static unsigned int part_in_flight(struct request_queue *q,
114 struct hd_struct *part)
116 unsigned int inflight = 0;
119 for_each_possible_cpu(cpu) {
120 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
121 part_stat_local_read_cpu(part, in_flight[1], cpu);
123 if ((int)inflight < 0)
129 static void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
130 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 if (!part_nr_sects_read(part) &&
260 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
261 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
265 get_device(part_to_dev(part));
275 EXPORT_SYMBOL_GPL(disk_part_iter_next);
278 * disk_part_iter_exit - finish up partition iteration
279 * @piter: iter of interest
281 * Called when iteration is over. Cleans up @piter.
286 void disk_part_iter_exit(struct disk_part_iter *piter)
288 disk_put_part(piter->part);
291 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
293 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
295 return part->start_sect <= sector &&
296 sector < part->start_sect + part_nr_sects_read(part);
300 * disk_map_sector_rcu - map sector to partition
301 * @disk: gendisk of interest
302 * @sector: sector to map
304 * Find out which partition @sector maps to on @disk. This is
305 * primarily used for stats accounting.
308 * RCU read locked. The returned partition pointer is always valid
309 * because its refcount is grabbed except for part0, which lifetime
310 * is same with the disk.
313 * Found partition on success, part0 is returned if no partition matches
314 * or the matched partition is being deleted.
316 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
318 struct disk_part_tbl *ptbl;
319 struct hd_struct *part;
323 ptbl = rcu_dereference(disk->part_tbl);
325 part = rcu_dereference(ptbl->last_lookup);
326 if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
329 for (i = 1; i < ptbl->len; i++) {
330 part = rcu_dereference(ptbl->part[i]);
332 if (part && sector_in_part(part, sector)) {
334 * only live partition can be cached for lookup,
335 * so use-after-free on cached & deleting partition
338 if (!hd_struct_try_get(part))
340 rcu_assign_pointer(ptbl->last_lookup, part);
352 * disk_has_partitions
353 * @disk: gendisk of interest
355 * Walk through the partition table and check if valid partition exists.
361 * True if the gendisk has at least one valid non-zero size partition.
364 bool disk_has_partitions(struct gendisk *disk)
366 struct disk_part_tbl *ptbl;
371 ptbl = rcu_dereference(disk->part_tbl);
373 /* Iterate partitions skipping the whole device at index 0 */
374 for (i = 1; i < ptbl->len; i++) {
375 if (rcu_dereference(ptbl->part[i])) {
385 EXPORT_SYMBOL_GPL(disk_has_partitions);
388 * Can be deleted altogether. Later.
391 #define BLKDEV_MAJOR_HASH_SIZE 255
392 static struct blk_major_name {
393 struct blk_major_name *next;
396 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
398 /* index in the above - for now: assume no multimajor ranges */
399 static inline int major_to_index(unsigned major)
401 return major % BLKDEV_MAJOR_HASH_SIZE;
404 #ifdef CONFIG_PROC_FS
405 void blkdev_show(struct seq_file *seqf, off_t offset)
407 struct blk_major_name *dp;
409 mutex_lock(&block_class_lock);
410 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
411 if (dp->major == offset)
412 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
413 mutex_unlock(&block_class_lock);
415 #endif /* CONFIG_PROC_FS */
418 * register_blkdev - register a new block device
420 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
421 * @major = 0, try to allocate any unused major number.
422 * @name: the name of the new block device as a zero terminated string
424 * The @name must be unique within the system.
426 * The return value depends on the @major input parameter:
428 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
429 * then the function returns zero on success, or a negative error code
430 * - if any unused major number was requested with @major = 0 parameter
431 * then the return value is the allocated major number in range
432 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
434 * See Documentation/admin-guide/devices.txt for the list of allocated
437 int register_blkdev(unsigned int major, const char *name)
439 struct blk_major_name **n, *p;
442 mutex_lock(&block_class_lock);
446 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
447 if (major_names[index] == NULL)
452 printk("%s: failed to get major for %s\n",
461 if (major >= BLKDEV_MAJOR_MAX) {
462 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
463 __func__, major, BLKDEV_MAJOR_MAX-1, name);
469 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
476 strlcpy(p->name, name, sizeof(p->name));
478 index = major_to_index(major);
480 for (n = &major_names[index]; *n; n = &(*n)->next) {
481 if ((*n)->major == major)
490 printk("register_blkdev: cannot get major %u for %s\n",
495 mutex_unlock(&block_class_lock);
499 EXPORT_SYMBOL(register_blkdev);
501 void unregister_blkdev(unsigned int major, const char *name)
503 struct blk_major_name **n;
504 struct blk_major_name *p = NULL;
505 int index = major_to_index(major);
507 mutex_lock(&block_class_lock);
508 for (n = &major_names[index]; *n; n = &(*n)->next)
509 if ((*n)->major == major)
511 if (!*n || strcmp((*n)->name, name)) {
517 mutex_unlock(&block_class_lock);
521 EXPORT_SYMBOL(unregister_blkdev);
523 static struct kobj_map *bdev_map;
526 * blk_mangle_minor - scatter minor numbers apart
527 * @minor: minor number to mangle
529 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
530 * is enabled. Mangling twice gives the original value.
538 static int blk_mangle_minor(int minor)
540 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
543 for (i = 0; i < MINORBITS / 2; i++) {
544 int low = minor & (1 << i);
545 int high = minor & (1 << (MINORBITS - 1 - i));
546 int distance = MINORBITS - 1 - 2 * i;
548 minor ^= low | high; /* clear both bits */
549 low <<= distance; /* swap the positions */
551 minor |= low | high; /* and set */
558 * blk_alloc_devt - allocate a dev_t for a partition
559 * @part: partition to allocate dev_t for
560 * @devt: out parameter for resulting dev_t
562 * Allocate a dev_t for block device.
565 * 0 on success, allocated dev_t is returned in *@devt. -errno on
571 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
573 struct gendisk *disk = part_to_disk(part);
576 /* in consecutive minor range? */
577 if (part->partno < disk->minors) {
578 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
582 /* allocate ext devt */
583 idr_preload(GFP_KERNEL);
585 spin_lock_bh(&ext_devt_lock);
586 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
587 spin_unlock_bh(&ext_devt_lock);
591 return idx == -ENOSPC ? -EBUSY : idx;
593 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
598 * blk_free_devt - free a dev_t
599 * @devt: dev_t to free
601 * Free @devt which was allocated using blk_alloc_devt().
606 void blk_free_devt(dev_t devt)
608 if (devt == MKDEV(0, 0))
611 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
612 spin_lock_bh(&ext_devt_lock);
613 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
614 spin_unlock_bh(&ext_devt_lock);
619 * We invalidate devt by assigning NULL pointer for devt in idr.
621 void blk_invalidate_devt(dev_t devt)
623 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
624 spin_lock_bh(&ext_devt_lock);
625 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
626 spin_unlock_bh(&ext_devt_lock);
630 static char *bdevt_str(dev_t devt, char *buf)
632 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
633 char tbuf[BDEVT_SIZE];
634 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
635 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
637 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
643 * Register device numbers dev..(dev+range-1)
644 * range must be nonzero
645 * The hash chain is sorted on range, so that subranges can override.
647 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
648 struct kobject *(*probe)(dev_t, int *, void *),
649 int (*lock)(dev_t, void *), void *data)
651 kobj_map(bdev_map, devt, range, module, probe, lock, data);
654 EXPORT_SYMBOL(blk_register_region);
656 void blk_unregister_region(dev_t devt, unsigned long range)
658 kobj_unmap(bdev_map, devt, range);
661 EXPORT_SYMBOL(blk_unregister_region);
663 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
665 struct gendisk *p = data;
667 return &disk_to_dev(p)->kobj;
670 static int exact_lock(dev_t devt, void *data)
672 struct gendisk *p = data;
674 if (!get_disk_and_module(p))
679 static void register_disk(struct device *parent, struct gendisk *disk,
680 const struct attribute_group **groups)
682 struct device *ddev = disk_to_dev(disk);
683 struct block_device *bdev;
684 struct disk_part_iter piter;
685 struct hd_struct *part;
688 ddev->parent = parent;
690 dev_set_name(ddev, "%s", disk->disk_name);
692 /* delay uevents, until we scanned partition table */
693 dev_set_uevent_suppress(ddev, 1);
696 WARN_ON(ddev->groups);
697 ddev->groups = groups;
699 if (device_add(ddev))
701 if (!sysfs_deprecated) {
702 err = sysfs_create_link(block_depr, &ddev->kobj,
703 kobject_name(&ddev->kobj));
711 * avoid probable deadlock caused by allocating memory with
712 * GFP_KERNEL in runtime_resume callback of its all ancestor
715 pm_runtime_set_memalloc_noio(ddev, true);
717 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
718 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
720 if (disk->flags & GENHD_FL_HIDDEN) {
721 dev_set_uevent_suppress(ddev, 0);
725 /* No minors to use for partitions */
726 if (!disk_part_scan_enabled(disk))
729 /* No such device (e.g., media were just removed) */
730 if (!get_capacity(disk))
733 bdev = bdget_disk(disk, 0);
737 bdev->bd_invalidated = 1;
738 err = blkdev_get(bdev, FMODE_READ, NULL);
741 blkdev_put(bdev, FMODE_READ);
744 /* announce disk after possible partitions are created */
745 dev_set_uevent_suppress(ddev, 0);
746 kobject_uevent(&ddev->kobj, KOBJ_ADD);
748 /* announce possible partitions */
749 disk_part_iter_init(&piter, disk, 0);
750 while ((part = disk_part_iter_next(&piter)))
751 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
752 disk_part_iter_exit(&piter);
754 if (disk->queue->backing_dev_info->dev) {
755 err = sysfs_create_link(&ddev->kobj,
756 &disk->queue->backing_dev_info->dev->kobj,
763 * __device_add_disk - add disk information to kernel list
764 * @parent: parent device for the disk
765 * @disk: per-device partitioning information
766 * @groups: Additional per-device sysfs groups
767 * @register_queue: register the queue if set to true
769 * This function registers the partitioning information in @disk
772 * FIXME: error handling
774 static void __device_add_disk(struct device *parent, struct gendisk *disk,
775 const struct attribute_group **groups,
782 * The disk queue should now be all set with enough information about
783 * the device for the elevator code to pick an adequate default
784 * elevator if one is needed, that is, for devices requesting queue
788 elevator_init_mq(disk->queue);
790 /* minors == 0 indicates to use ext devt from part0 and should
791 * be accompanied with EXT_DEVT flag. Make sure all
792 * parameters make sense.
794 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
795 WARN_ON(!disk->minors &&
796 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
798 disk->flags |= GENHD_FL_UP;
800 retval = blk_alloc_devt(&disk->part0, &devt);
805 disk->major = MAJOR(devt);
806 disk->first_minor = MINOR(devt);
808 disk_alloc_events(disk);
810 if (disk->flags & GENHD_FL_HIDDEN) {
812 * Don't let hidden disks show up in /proc/partitions,
813 * and don't bother scanning for partitions either.
815 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
816 disk->flags |= GENHD_FL_NO_PART_SCAN;
818 struct backing_dev_info *bdi = disk->queue->backing_dev_info;
819 struct device *dev = disk_to_dev(disk);
822 /* Register BDI before referencing it from bdev */
824 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
826 bdi_set_owner(bdi, dev);
827 blk_register_region(disk_devt(disk), disk->minors, NULL,
828 exact_match, exact_lock, disk);
830 register_disk(parent, disk, groups);
832 blk_register_queue(disk);
835 * Take an extra ref on queue which will be put on disk_release()
836 * so that it sticks around as long as @disk is there.
838 WARN_ON_ONCE(!blk_get_queue(disk->queue));
840 disk_add_events(disk);
841 blk_integrity_add(disk);
844 void device_add_disk(struct device *parent, struct gendisk *disk,
845 const struct attribute_group **groups)
848 __device_add_disk(parent, disk, groups, true);
850 EXPORT_SYMBOL(device_add_disk);
852 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
854 __device_add_disk(parent, disk, NULL, false);
856 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
858 static void invalidate_partition(struct gendisk *disk, int partno)
860 struct block_device *bdev;
862 bdev = bdget_disk(disk, partno);
867 __invalidate_device(bdev, true);
870 * Unhash the bdev inode for this device so that it gets evicted as soon
871 * as last inode reference is dropped.
873 remove_inode_hash(bdev->bd_inode);
878 * del_gendisk - remove the gendisk
879 * @disk: the struct gendisk to remove
881 * Removes the gendisk and all its associated resources. This deletes the
882 * partitions associated with the gendisk, and unregisters the associated
885 * This is the counter to the respective __device_add_disk() call.
887 * The final removal of the struct gendisk happens when its refcount reaches 0
888 * with put_disk(), which should be called after del_gendisk(), if
889 * __device_add_disk() was used.
891 * Drivers exist which depend on the release of the gendisk to be synchronous,
892 * it should not be deferred.
896 void del_gendisk(struct gendisk *disk)
898 struct disk_part_iter piter;
899 struct hd_struct *part;
903 blk_integrity_del(disk);
904 disk_del_events(disk);
907 * Block lookups of the disk until all bdevs are unhashed and the
908 * disk is marked as dead (GENHD_FL_UP cleared).
910 down_write(&disk->lookup_sem);
911 /* invalidate stuff */
912 disk_part_iter_init(&piter, disk,
913 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
914 while ((part = disk_part_iter_next(&piter))) {
915 invalidate_partition(disk, part->partno);
916 delete_partition(disk, part);
918 disk_part_iter_exit(&piter);
920 invalidate_partition(disk, 0);
921 set_capacity(disk, 0);
922 disk->flags &= ~GENHD_FL_UP;
923 up_write(&disk->lookup_sem);
925 if (!(disk->flags & GENHD_FL_HIDDEN))
926 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
929 * Unregister bdi before releasing device numbers (as they can
930 * get reused and we'd get clashes in sysfs).
932 if (!(disk->flags & GENHD_FL_HIDDEN))
933 bdi_unregister(disk->queue->backing_dev_info);
934 blk_unregister_queue(disk);
939 if (!(disk->flags & GENHD_FL_HIDDEN))
940 blk_unregister_region(disk_devt(disk), disk->minors);
942 * Remove gendisk pointer from idr so that it cannot be looked up
943 * while RCU period before freeing gendisk is running to prevent
944 * use-after-free issues. Note that the device number stays
945 * "in-use" until we really free the gendisk.
947 blk_invalidate_devt(disk_devt(disk));
949 kobject_put(disk->part0.holder_dir);
950 kobject_put(disk->slave_dir);
952 part_stat_set_all(&disk->part0, 0);
953 disk->part0.stamp = 0;
954 if (!sysfs_deprecated)
955 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
956 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
957 device_del(disk_to_dev(disk));
959 EXPORT_SYMBOL(del_gendisk);
961 /* sysfs access to bad-blocks list. */
962 static ssize_t disk_badblocks_show(struct device *dev,
963 struct device_attribute *attr,
966 struct gendisk *disk = dev_to_disk(dev);
969 return sprintf(page, "\n");
971 return badblocks_show(disk->bb, page, 0);
974 static ssize_t disk_badblocks_store(struct device *dev,
975 struct device_attribute *attr,
976 const char *page, size_t len)
978 struct gendisk *disk = dev_to_disk(dev);
983 return badblocks_store(disk->bb, page, len, 0);
987 * get_gendisk - get partitioning information for a given device
988 * @devt: device to get partitioning information for
989 * @partno: returned partition index
991 * This function gets the structure containing partitioning
992 * information for the given device @devt.
996 struct gendisk *get_gendisk(dev_t devt, int *partno)
998 struct gendisk *disk = NULL;
1002 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
1003 struct kobject *kobj;
1005 kobj = kobj_lookup(bdev_map, devt, partno);
1007 disk = dev_to_disk(kobj_to_dev(kobj));
1009 struct hd_struct *part;
1011 spin_lock_bh(&ext_devt_lock);
1012 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
1013 if (part && get_disk_and_module(part_to_disk(part))) {
1014 *partno = part->partno;
1015 disk = part_to_disk(part);
1017 spin_unlock_bh(&ext_devt_lock);
1024 * Synchronize with del_gendisk() to not return disk that is being
1027 down_read(&disk->lookup_sem);
1028 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1029 !(disk->flags & GENHD_FL_UP))) {
1030 up_read(&disk->lookup_sem);
1031 put_disk_and_module(disk);
1034 up_read(&disk->lookup_sem);
1040 * bdget_disk - do bdget() by gendisk and partition number
1041 * @disk: gendisk of interest
1042 * @partno: partition number
1044 * Find partition @partno from @disk, do bdget() on it.
1050 * Resulting block_device on success, NULL on failure.
1052 struct block_device *bdget_disk(struct gendisk *disk, int partno)
1054 struct hd_struct *part;
1055 struct block_device *bdev = NULL;
1057 part = disk_get_part(disk, partno);
1059 bdev = bdget(part_devt(part));
1060 disk_put_part(part);
1064 EXPORT_SYMBOL(bdget_disk);
1067 * print a full list of all partitions - intended for places where the root
1068 * filesystem can't be mounted and thus to give the victim some idea of what
1071 void __init printk_all_partitions(void)
1073 struct class_dev_iter iter;
1076 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1077 while ((dev = class_dev_iter_next(&iter))) {
1078 struct gendisk *disk = dev_to_disk(dev);
1079 struct disk_part_iter piter;
1080 struct hd_struct *part;
1081 char name_buf[BDEVNAME_SIZE];
1082 char devt_buf[BDEVT_SIZE];
1085 * Don't show empty devices or things that have been
1088 if (get_capacity(disk) == 0 ||
1089 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1093 * Note, unlike /proc/partitions, I am showing the
1094 * numbers in hex - the same format as the root=
1097 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1098 while ((part = disk_part_iter_next(&piter))) {
1099 bool is_part0 = part == &disk->part0;
1101 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1102 bdevt_str(part_devt(part), devt_buf),
1103 (unsigned long long)part_nr_sects_read(part) >> 1
1104 , disk_name(disk, part->partno, name_buf),
1105 part->info ? part->info->uuid : "");
1107 if (dev->parent && dev->parent->driver)
1108 printk(" driver: %s\n",
1109 dev->parent->driver->name);
1111 printk(" (driver?)\n");
1115 disk_part_iter_exit(&piter);
1117 class_dev_iter_exit(&iter);
1120 #ifdef CONFIG_PROC_FS
1122 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1125 struct class_dev_iter *iter;
1128 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1130 return ERR_PTR(-ENOMEM);
1132 seqf->private = iter;
1133 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1135 dev = class_dev_iter_next(iter);
1140 return dev_to_disk(dev);
1143 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1148 dev = class_dev_iter_next(seqf->private);
1150 return dev_to_disk(dev);
1155 static void disk_seqf_stop(struct seq_file *seqf, void *v)
1157 struct class_dev_iter *iter = seqf->private;
1159 /* stop is called even after start failed :-( */
1161 class_dev_iter_exit(iter);
1163 seqf->private = NULL;
1167 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1171 p = disk_seqf_start(seqf, pos);
1172 if (!IS_ERR_OR_NULL(p) && !*pos)
1173 seq_puts(seqf, "major minor #blocks name\n\n");
1177 static int show_partition(struct seq_file *seqf, void *v)
1179 struct gendisk *sgp = v;
1180 struct disk_part_iter piter;
1181 struct hd_struct *part;
1182 char buf[BDEVNAME_SIZE];
1184 /* Don't show non-partitionable removeable devices or empty devices */
1185 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1186 (sgp->flags & GENHD_FL_REMOVABLE)))
1188 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1191 /* show the full disk and all non-0 size partitions of it */
1192 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1193 while ((part = disk_part_iter_next(&piter)))
1194 seq_printf(seqf, "%4d %7d %10llu %s\n",
1195 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1196 (unsigned long long)part_nr_sects_read(part) >> 1,
1197 disk_name(sgp, part->partno, buf));
1198 disk_part_iter_exit(&piter);
1203 static const struct seq_operations partitions_op = {
1204 .start = show_partition_start,
1205 .next = disk_seqf_next,
1206 .stop = disk_seqf_stop,
1207 .show = show_partition
1212 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1214 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1215 /* Make old-style 2.4 aliases work */
1216 request_module("block-major-%d", MAJOR(devt));
1220 static int __init genhd_device_init(void)
1224 block_class.dev_kobj = sysfs_dev_block_kobj;
1225 error = class_register(&block_class);
1226 if (unlikely(error))
1228 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1231 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1233 /* create top-level block dir */
1234 if (!sysfs_deprecated)
1235 block_depr = kobject_create_and_add("block", NULL);
1239 subsys_initcall(genhd_device_init);
1241 static ssize_t disk_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->minors);
1249 static ssize_t disk_ext_range_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", disk_max_parts(disk));
1257 static ssize_t disk_removable_show(struct device *dev,
1258 struct device_attribute *attr, char *buf)
1260 struct gendisk *disk = dev_to_disk(dev);
1262 return sprintf(buf, "%d\n",
1263 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1266 static ssize_t disk_hidden_show(struct device *dev,
1267 struct device_attribute *attr, char *buf)
1269 struct gendisk *disk = dev_to_disk(dev);
1271 return sprintf(buf, "%d\n",
1272 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1275 static ssize_t disk_ro_show(struct device *dev,
1276 struct device_attribute *attr, char *buf)
1278 struct gendisk *disk = dev_to_disk(dev);
1280 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1283 ssize_t part_size_show(struct device *dev,
1284 struct device_attribute *attr, char *buf)
1286 struct hd_struct *p = dev_to_part(dev);
1288 return sprintf(buf, "%llu\n",
1289 (unsigned long long)part_nr_sects_read(p));
1292 ssize_t part_stat_show(struct device *dev,
1293 struct device_attribute *attr, char *buf)
1295 struct hd_struct *p = dev_to_part(dev);
1296 struct request_queue *q = part_to_disk(p)->queue;
1297 struct disk_stats stat;
1298 unsigned int inflight;
1300 part_stat_read_all(p, &stat);
1302 inflight = blk_mq_in_flight(q, p);
1304 inflight = part_in_flight(q, p);
1307 "%8lu %8lu %8llu %8u "
1308 "%8lu %8lu %8llu %8u "
1310 "%8lu %8lu %8llu %8u "
1313 stat.ios[STAT_READ],
1314 stat.merges[STAT_READ],
1315 (unsigned long long)stat.sectors[STAT_READ],
1316 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1317 stat.ios[STAT_WRITE],
1318 stat.merges[STAT_WRITE],
1319 (unsigned long long)stat.sectors[STAT_WRITE],
1320 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1322 jiffies_to_msecs(stat.io_ticks),
1323 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1324 stat.nsecs[STAT_WRITE] +
1325 stat.nsecs[STAT_DISCARD] +
1326 stat.nsecs[STAT_FLUSH],
1328 stat.ios[STAT_DISCARD],
1329 stat.merges[STAT_DISCARD],
1330 (unsigned long long)stat.sectors[STAT_DISCARD],
1331 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1332 stat.ios[STAT_FLUSH],
1333 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1336 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1339 struct hd_struct *p = dev_to_part(dev);
1340 struct request_queue *q = part_to_disk(p)->queue;
1341 unsigned int inflight[2];
1344 blk_mq_in_flight_rw(q, p, inflight);
1346 part_in_flight_rw(q, p, inflight);
1348 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1351 static ssize_t disk_capability_show(struct device *dev,
1352 struct device_attribute *attr, char *buf)
1354 struct gendisk *disk = dev_to_disk(dev);
1356 return sprintf(buf, "%x\n", disk->flags);
1359 static ssize_t disk_alignment_offset_show(struct device *dev,
1360 struct device_attribute *attr,
1363 struct gendisk *disk = dev_to_disk(dev);
1365 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1368 static ssize_t disk_discard_alignment_show(struct device *dev,
1369 struct device_attribute *attr,
1372 struct gendisk *disk = dev_to_disk(dev);
1374 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1377 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1378 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1379 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1380 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1381 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1382 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1383 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1384 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1385 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1386 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1387 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1388 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1390 #ifdef CONFIG_FAIL_MAKE_REQUEST
1391 ssize_t part_fail_show(struct device *dev,
1392 struct device_attribute *attr, char *buf)
1394 struct hd_struct *p = dev_to_part(dev);
1396 return sprintf(buf, "%d\n", p->make_it_fail);
1399 ssize_t part_fail_store(struct device *dev,
1400 struct device_attribute *attr,
1401 const char *buf, size_t count)
1403 struct hd_struct *p = dev_to_part(dev);
1406 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1407 p->make_it_fail = (i == 0) ? 0 : 1;
1412 static struct device_attribute dev_attr_fail =
1413 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1414 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1416 #ifdef CONFIG_FAIL_IO_TIMEOUT
1417 static struct device_attribute dev_attr_fail_timeout =
1418 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1421 static struct attribute *disk_attrs[] = {
1422 &dev_attr_range.attr,
1423 &dev_attr_ext_range.attr,
1424 &dev_attr_removable.attr,
1425 &dev_attr_hidden.attr,
1427 &dev_attr_size.attr,
1428 &dev_attr_alignment_offset.attr,
1429 &dev_attr_discard_alignment.attr,
1430 &dev_attr_capability.attr,
1431 &dev_attr_stat.attr,
1432 &dev_attr_inflight.attr,
1433 &dev_attr_badblocks.attr,
1434 #ifdef CONFIG_FAIL_MAKE_REQUEST
1435 &dev_attr_fail.attr,
1437 #ifdef CONFIG_FAIL_IO_TIMEOUT
1438 &dev_attr_fail_timeout.attr,
1443 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1445 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1446 struct gendisk *disk = dev_to_disk(dev);
1448 if (a == &dev_attr_badblocks.attr && !disk->bb)
1453 static struct attribute_group disk_attr_group = {
1454 .attrs = disk_attrs,
1455 .is_visible = disk_visible,
1458 static const struct attribute_group *disk_attr_groups[] = {
1464 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1465 * @disk: disk to replace part_tbl for
1466 * @new_ptbl: new part_tbl to install
1468 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1469 * original ptbl is freed using RCU callback.
1472 * Matching bd_mutex locked or the caller is the only user of @disk.
1474 static void disk_replace_part_tbl(struct gendisk *disk,
1475 struct disk_part_tbl *new_ptbl)
1477 struct disk_part_tbl *old_ptbl =
1478 rcu_dereference_protected(disk->part_tbl, 1);
1480 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1483 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1484 kfree_rcu(old_ptbl, rcu_head);
1489 * disk_expand_part_tbl - expand disk->part_tbl
1490 * @disk: disk to expand part_tbl for
1491 * @partno: expand such that this partno can fit in
1493 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1494 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1497 * Matching bd_mutex locked or the caller is the only user of @disk.
1501 * 0 on success, -errno on failure.
1503 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1505 struct disk_part_tbl *old_ptbl =
1506 rcu_dereference_protected(disk->part_tbl, 1);
1507 struct disk_part_tbl *new_ptbl;
1508 int len = old_ptbl ? old_ptbl->len : 0;
1512 * check for int overflow, since we can get here from blkpg_ioctl()
1513 * with a user passed 'partno'.
1515 target = partno + 1;
1519 /* disk_max_parts() is zero during initialization, ignore if so */
1520 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1526 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1531 new_ptbl->len = target;
1533 for (i = 0; i < len; i++)
1534 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1536 disk_replace_part_tbl(disk, new_ptbl);
1541 * disk_release - releases all allocated resources of the gendisk
1542 * @dev: the device representing this disk
1544 * This function releases all allocated resources of the gendisk.
1546 * The struct gendisk refcount is incremented with get_gendisk() or
1547 * get_disk_and_module(), and its refcount is decremented with
1548 * put_disk_and_module() or put_disk(). Once the refcount reaches 0 this
1549 * function is called.
1551 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1552 * assigned. Since the request_queue sits on top of the gendisk for these
1553 * drivers we also call blk_put_queue() for them, and we expect the
1554 * request_queue refcount to reach 0 at this point, and so the request_queue
1555 * will also be freed prior to the disk.
1557 * Context: can sleep
1559 static void disk_release(struct device *dev)
1561 struct gendisk *disk = dev_to_disk(dev);
1565 blk_free_devt(dev->devt);
1566 disk_release_events(disk);
1567 kfree(disk->random);
1568 disk_replace_part_tbl(disk, NULL);
1569 hd_free_part(&disk->part0);
1571 blk_put_queue(disk->queue);
1574 struct class block_class = {
1578 static char *block_devnode(struct device *dev, umode_t *mode,
1579 kuid_t *uid, kgid_t *gid)
1581 struct gendisk *disk = dev_to_disk(dev);
1583 if (disk->fops->devnode)
1584 return disk->fops->devnode(disk, mode);
1588 const struct device_type disk_type = {
1590 .groups = disk_attr_groups,
1591 .release = disk_release,
1592 .devnode = block_devnode,
1595 #ifdef CONFIG_PROC_FS
1597 * aggregate disk stat collector. Uses the same stats that the sysfs
1598 * entries do, above, but makes them available through one seq_file.
1600 * The output looks suspiciously like /proc/partitions with a bunch of
1603 static int diskstats_show(struct seq_file *seqf, void *v)
1605 struct gendisk *gp = v;
1606 struct disk_part_iter piter;
1607 struct hd_struct *hd;
1608 char buf[BDEVNAME_SIZE];
1609 unsigned int inflight;
1610 struct disk_stats stat;
1613 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1614 seq_puts(seqf, "major minor name"
1615 " rio rmerge rsect ruse wio wmerge "
1616 "wsect wuse running use aveq"
1620 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1621 while ((hd = disk_part_iter_next(&piter))) {
1622 part_stat_read_all(hd, &stat);
1623 if (queue_is_mq(gp->queue))
1624 inflight = blk_mq_in_flight(gp->queue, hd);
1626 inflight = part_in_flight(gp->queue, hd);
1628 seq_printf(seqf, "%4d %7d %s "
1635 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1636 disk_name(gp, hd->partno, buf),
1637 stat.ios[STAT_READ],
1638 stat.merges[STAT_READ],
1639 stat.sectors[STAT_READ],
1640 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1642 stat.ios[STAT_WRITE],
1643 stat.merges[STAT_WRITE],
1644 stat.sectors[STAT_WRITE],
1645 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1648 jiffies_to_msecs(stat.io_ticks),
1649 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1650 stat.nsecs[STAT_WRITE] +
1651 stat.nsecs[STAT_DISCARD] +
1652 stat.nsecs[STAT_FLUSH],
1654 stat.ios[STAT_DISCARD],
1655 stat.merges[STAT_DISCARD],
1656 stat.sectors[STAT_DISCARD],
1657 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1659 stat.ios[STAT_FLUSH],
1660 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1664 disk_part_iter_exit(&piter);
1669 static const struct seq_operations diskstats_op = {
1670 .start = disk_seqf_start,
1671 .next = disk_seqf_next,
1672 .stop = disk_seqf_stop,
1673 .show = diskstats_show
1676 static int __init proc_genhd_init(void)
1678 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1679 proc_create_seq("partitions", 0, NULL, &partitions_op);
1682 module_init(proc_genhd_init);
1683 #endif /* CONFIG_PROC_FS */
1685 dev_t blk_lookup_devt(const char *name, int partno)
1687 dev_t devt = MKDEV(0, 0);
1688 struct class_dev_iter iter;
1691 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1692 while ((dev = class_dev_iter_next(&iter))) {
1693 struct gendisk *disk = dev_to_disk(dev);
1694 struct hd_struct *part;
1696 if (strcmp(dev_name(dev), name))
1699 if (partno < disk->minors) {
1700 /* We need to return the right devno, even
1701 * if the partition doesn't exist yet.
1703 devt = MKDEV(MAJOR(dev->devt),
1704 MINOR(dev->devt) + partno);
1707 part = disk_get_part(disk, partno);
1709 devt = part_devt(part);
1710 disk_put_part(part);
1713 disk_put_part(part);
1715 class_dev_iter_exit(&iter);
1719 struct gendisk *__alloc_disk_node(int minors, int node_id)
1721 struct gendisk *disk;
1722 struct disk_part_tbl *ptbl;
1724 if (minors > DISK_MAX_PARTS) {
1726 "block: can't allocate more than %d partitions\n",
1728 minors = DISK_MAX_PARTS;
1731 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1733 disk->part0.dkstats = alloc_percpu(struct disk_stats);
1734 if (!disk->part0.dkstats) {
1738 init_rwsem(&disk->lookup_sem);
1739 disk->node_id = node_id;
1740 if (disk_expand_part_tbl(disk, 0)) {
1741 free_percpu(disk->part0.dkstats);
1745 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1746 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1749 * set_capacity() and get_capacity() currently don't use
1750 * seqcounter to read/update the part0->nr_sects. Still init
1751 * the counter as we can read the sectors in IO submission
1752 * patch using seqence counters.
1754 * TODO: Ideally set_capacity() and get_capacity() should be
1755 * converted to make use of bd_mutex and sequence counters.
1757 hd_sects_seq_init(&disk->part0);
1758 if (hd_ref_init(&disk->part0)) {
1759 hd_free_part(&disk->part0);
1764 disk->minors = minors;
1765 rand_initialize_disk(disk);
1766 disk_to_dev(disk)->class = &block_class;
1767 disk_to_dev(disk)->type = &disk_type;
1768 device_initialize(disk_to_dev(disk));
1772 EXPORT_SYMBOL(__alloc_disk_node);
1775 * get_disk_and_module - increments the gendisk and gendisk fops module refcount
1776 * @disk: the struct gendisk to increment the refcount for
1778 * This increments the refcount for the struct gendisk, and the gendisk's
1779 * fops module owner.
1781 * Context: Any context.
1783 struct kobject *get_disk_and_module(struct gendisk *disk)
1785 struct module *owner;
1786 struct kobject *kobj;
1790 owner = disk->fops->owner;
1791 if (owner && !try_module_get(owner))
1793 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1801 EXPORT_SYMBOL(get_disk_and_module);
1804 * put_disk - decrements the gendisk refcount
1805 * @disk: the struct gendisk to decrement the refcount for
1807 * This decrements the refcount for the struct gendisk. When this reaches 0
1808 * we'll have disk_release() called.
1810 * Context: Any context, but the last reference must not be dropped from
1813 void put_disk(struct gendisk *disk)
1816 kobject_put(&disk_to_dev(disk)->kobj);
1818 EXPORT_SYMBOL(put_disk);
1821 * put_disk_and_module - decrements the module and gendisk refcount
1822 * @disk: the struct gendisk to decrement the refcount for
1824 * This is a counterpart of get_disk_and_module() and thus also of
1827 * Context: Any context, but the last reference must not be dropped from
1830 void put_disk_and_module(struct gendisk *disk)
1833 struct module *owner = disk->fops->owner;
1839 EXPORT_SYMBOL(put_disk_and_module);
1841 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1843 char event[] = "DISK_RO=1";
1844 char *envp[] = { event, NULL };
1848 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1851 void set_device_ro(struct block_device *bdev, int flag)
1853 bdev->bd_part->policy = flag;
1856 EXPORT_SYMBOL(set_device_ro);
1858 void set_disk_ro(struct gendisk *disk, int flag)
1860 struct disk_part_iter piter;
1861 struct hd_struct *part;
1863 if (disk->part0.policy != flag) {
1864 set_disk_ro_uevent(disk, flag);
1865 disk->part0.policy = flag;
1868 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1869 while ((part = disk_part_iter_next(&piter)))
1870 part->policy = flag;
1871 disk_part_iter_exit(&piter);
1874 EXPORT_SYMBOL(set_disk_ro);
1876 int bdev_read_only(struct block_device *bdev)
1880 return bdev->bd_part->policy;
1883 EXPORT_SYMBOL(bdev_read_only);
1886 * Disk events - monitor disk events like media change and eject request.
1888 struct disk_events {
1889 struct list_head node; /* all disk_event's */
1890 struct gendisk *disk; /* the associated disk */
1893 struct mutex block_mutex; /* protects blocking */
1894 int block; /* event blocking depth */
1895 unsigned int pending; /* events already sent out */
1896 unsigned int clearing; /* events being cleared */
1898 long poll_msecs; /* interval, -1 for default */
1899 struct delayed_work dwork;
1902 static const char *disk_events_strs[] = {
1903 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1904 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1907 static char *disk_uevents[] = {
1908 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1909 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1912 /* list of all disk_events */
1913 static DEFINE_MUTEX(disk_events_mutex);
1914 static LIST_HEAD(disk_events);
1916 /* disable in-kernel polling by default */
1917 static unsigned long disk_events_dfl_poll_msecs;
1919 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1921 struct disk_events *ev = disk->ev;
1922 long intv_msecs = 0;
1925 * If device-specific poll interval is set, always use it. If
1926 * the default is being used, poll if the POLL flag is set.
1928 if (ev->poll_msecs >= 0)
1929 intv_msecs = ev->poll_msecs;
1930 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1931 intv_msecs = disk_events_dfl_poll_msecs;
1933 return msecs_to_jiffies(intv_msecs);
1937 * disk_block_events - block and flush disk event checking
1938 * @disk: disk to block events for
1940 * On return from this function, it is guaranteed that event checking
1941 * isn't in progress and won't happen until unblocked by
1942 * disk_unblock_events(). Events blocking is counted and the actual
1943 * unblocking happens after the matching number of unblocks are done.
1945 * Note that this intentionally does not block event checking from
1946 * disk_clear_events().
1951 void disk_block_events(struct gendisk *disk)
1953 struct disk_events *ev = disk->ev;
1954 unsigned long flags;
1961 * Outer mutex ensures that the first blocker completes canceling
1962 * the event work before further blockers are allowed to finish.
1964 mutex_lock(&ev->block_mutex);
1966 spin_lock_irqsave(&ev->lock, flags);
1967 cancel = !ev->block++;
1968 spin_unlock_irqrestore(&ev->lock, flags);
1971 cancel_delayed_work_sync(&disk->ev->dwork);
1973 mutex_unlock(&ev->block_mutex);
1976 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1978 struct disk_events *ev = disk->ev;
1980 unsigned long flags;
1982 spin_lock_irqsave(&ev->lock, flags);
1984 if (WARN_ON_ONCE(ev->block <= 0))
1990 intv = disk_events_poll_jiffies(disk);
1992 queue_delayed_work(system_freezable_power_efficient_wq,
1995 queue_delayed_work(system_freezable_power_efficient_wq,
1998 spin_unlock_irqrestore(&ev->lock, flags);
2002 * disk_unblock_events - unblock disk event checking
2003 * @disk: disk to unblock events for
2005 * Undo disk_block_events(). When the block count reaches zero, it
2006 * starts events polling if configured.
2009 * Don't care. Safe to call from irq context.
2011 void disk_unblock_events(struct gendisk *disk)
2014 __disk_unblock_events(disk, false);
2018 * disk_flush_events - schedule immediate event checking and flushing
2019 * @disk: disk to check and flush events for
2020 * @mask: events to flush
2022 * Schedule immediate event checking on @disk if not blocked. Events in
2023 * @mask are scheduled to be cleared from the driver. Note that this
2024 * doesn't clear the events from @disk->ev.
2027 * If @mask is non-zero must be called with bdev->bd_mutex held.
2029 void disk_flush_events(struct gendisk *disk, unsigned int mask)
2031 struct disk_events *ev = disk->ev;
2036 spin_lock_irq(&ev->lock);
2037 ev->clearing |= mask;
2039 mod_delayed_work(system_freezable_power_efficient_wq,
2041 spin_unlock_irq(&ev->lock);
2045 * disk_clear_events - synchronously check, clear and return pending events
2046 * @disk: disk to fetch and clear events from
2047 * @mask: mask of events to be fetched and cleared
2049 * Disk events are synchronously checked and pending events in @mask
2050 * are cleared and returned. This ignores the block count.
2055 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
2057 struct disk_events *ev = disk->ev;
2058 unsigned int pending;
2059 unsigned int clearing = mask;
2064 disk_block_events(disk);
2067 * store the union of mask and ev->clearing on the stack so that the
2068 * race with disk_flush_events does not cause ambiguity (ev->clearing
2069 * can still be modified even if events are blocked).
2071 spin_lock_irq(&ev->lock);
2072 clearing |= ev->clearing;
2074 spin_unlock_irq(&ev->lock);
2076 disk_check_events(ev, &clearing);
2078 * if ev->clearing is not 0, the disk_flush_events got called in the
2079 * middle of this function, so we want to run the workfn without delay.
2081 __disk_unblock_events(disk, ev->clearing ? true : false);
2083 /* then, fetch and clear pending events */
2084 spin_lock_irq(&ev->lock);
2085 pending = ev->pending & mask;
2086 ev->pending &= ~mask;
2087 spin_unlock_irq(&ev->lock);
2088 WARN_ON_ONCE(clearing & mask);
2094 * Separate this part out so that a different pointer for clearing_ptr can be
2095 * passed in for disk_clear_events.
2097 static void disk_events_workfn(struct work_struct *work)
2099 struct delayed_work *dwork = to_delayed_work(work);
2100 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2102 disk_check_events(ev, &ev->clearing);
2105 static void disk_check_events(struct disk_events *ev,
2106 unsigned int *clearing_ptr)
2108 struct gendisk *disk = ev->disk;
2109 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2110 unsigned int clearing = *clearing_ptr;
2111 unsigned int events;
2113 int nr_events = 0, i;
2116 events = disk->fops->check_events(disk, clearing);
2118 /* accumulate pending events and schedule next poll if necessary */
2119 spin_lock_irq(&ev->lock);
2121 events &= ~ev->pending;
2122 ev->pending |= events;
2123 *clearing_ptr &= ~clearing;
2125 intv = disk_events_poll_jiffies(disk);
2126 if (!ev->block && intv)
2127 queue_delayed_work(system_freezable_power_efficient_wq,
2130 spin_unlock_irq(&ev->lock);
2133 * Tell userland about new events. Only the events listed in
2134 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2135 * is set. Otherwise, events are processed internally but never
2136 * get reported to userland.
2138 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2139 if ((events & disk->events & (1 << i)) &&
2140 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2141 envp[nr_events++] = disk_uevents[i];
2144 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2148 * A disk events enabled device has the following sysfs nodes under
2149 * its /sys/block/X/ directory.
2151 * events : list of all supported events
2152 * events_async : list of events which can be detected w/o polling
2153 * (always empty, only for backwards compatibility)
2154 * events_poll_msecs : polling interval, 0: disable, -1: system default
2156 static ssize_t __disk_events_show(unsigned int events, char *buf)
2158 const char *delim = "";
2162 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2163 if (events & (1 << i)) {
2164 pos += sprintf(buf + pos, "%s%s",
2165 delim, disk_events_strs[i]);
2169 pos += sprintf(buf + pos, "\n");
2173 static ssize_t disk_events_show(struct device *dev,
2174 struct device_attribute *attr, char *buf)
2176 struct gendisk *disk = dev_to_disk(dev);
2178 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2181 return __disk_events_show(disk->events, buf);
2184 static ssize_t disk_events_async_show(struct device *dev,
2185 struct device_attribute *attr, char *buf)
2190 static ssize_t disk_events_poll_msecs_show(struct device *dev,
2191 struct device_attribute *attr,
2194 struct gendisk *disk = dev_to_disk(dev);
2197 return sprintf(buf, "-1\n");
2199 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2202 static ssize_t disk_events_poll_msecs_store(struct device *dev,
2203 struct device_attribute *attr,
2204 const char *buf, size_t count)
2206 struct gendisk *disk = dev_to_disk(dev);
2209 if (!count || !sscanf(buf, "%ld", &intv))
2212 if (intv < 0 && intv != -1)
2218 disk_block_events(disk);
2219 disk->ev->poll_msecs = intv;
2220 __disk_unblock_events(disk, true);
2225 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2226 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2227 static const DEVICE_ATTR(events_poll_msecs, 0644,
2228 disk_events_poll_msecs_show,
2229 disk_events_poll_msecs_store);
2231 static const struct attribute *disk_events_attrs[] = {
2232 &dev_attr_events.attr,
2233 &dev_attr_events_async.attr,
2234 &dev_attr_events_poll_msecs.attr,
2239 * The default polling interval can be specified by the kernel
2240 * parameter block.events_dfl_poll_msecs which defaults to 0
2241 * (disable). This can also be modified runtime by writing to
2242 * /sys/module/block/parameters/events_dfl_poll_msecs.
2244 static int disk_events_set_dfl_poll_msecs(const char *val,
2245 const struct kernel_param *kp)
2247 struct disk_events *ev;
2250 ret = param_set_ulong(val, kp);
2254 mutex_lock(&disk_events_mutex);
2256 list_for_each_entry(ev, &disk_events, node)
2257 disk_flush_events(ev->disk, 0);
2259 mutex_unlock(&disk_events_mutex);
2264 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2265 .set = disk_events_set_dfl_poll_msecs,
2266 .get = param_get_ulong,
2269 #undef MODULE_PARAM_PREFIX
2270 #define MODULE_PARAM_PREFIX "block."
2272 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2273 &disk_events_dfl_poll_msecs, 0644);
2276 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2278 static void disk_alloc_events(struct gendisk *disk)
2280 struct disk_events *ev;
2282 if (!disk->fops->check_events || !disk->events)
2285 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2287 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2291 INIT_LIST_HEAD(&ev->node);
2293 spin_lock_init(&ev->lock);
2294 mutex_init(&ev->block_mutex);
2296 ev->poll_msecs = -1;
2297 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2302 static void disk_add_events(struct gendisk *disk)
2304 /* FIXME: error handling */
2305 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2306 pr_warn("%s: failed to create sysfs files for events\n",
2312 mutex_lock(&disk_events_mutex);
2313 list_add_tail(&disk->ev->node, &disk_events);
2314 mutex_unlock(&disk_events_mutex);
2317 * Block count is initialized to 1 and the following initial
2318 * unblock kicks it into action.
2320 __disk_unblock_events(disk, true);
2323 static void disk_del_events(struct gendisk *disk)
2326 disk_block_events(disk);
2328 mutex_lock(&disk_events_mutex);
2329 list_del_init(&disk->ev->node);
2330 mutex_unlock(&disk_events_mutex);
2333 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2336 static void disk_release_events(struct gendisk *disk)
2338 /* the block count should be 1 from disk_del_events() */
2339 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);