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/mutex.h>
21 #include <linux/idr.h>
22 #include <linux/log2.h>
23 #include <linux/pm_runtime.h>
24 #include <linux/badblocks.h>
28 static struct kobject *block_depr;
30 static DEFINE_XARRAY(bdev_map);
31 static DEFINE_MUTEX(bdev_map_lock);
33 /* for extended dynamic devt allocation, currently only one major is used */
34 #define NR_EXT_DEVT (1 << MINORBITS)
36 /* For extended devt allocation. ext_devt_lock prevents look up
37 * results from going away underneath its user.
39 static DEFINE_SPINLOCK(ext_devt_lock);
40 static DEFINE_IDR(ext_devt_idr);
42 static void disk_check_events(struct disk_events *ev,
43 unsigned int *clearing_ptr);
44 static void disk_alloc_events(struct gendisk *disk);
45 static void disk_add_events(struct gendisk *disk);
46 static void disk_del_events(struct gendisk *disk);
47 static void disk_release_events(struct gendisk *disk);
50 * Set disk capacity and notify if the size is not currently zero and will not
51 * be set to zero. Returns true if a uevent was sent, otherwise false.
53 bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
55 sector_t capacity = get_capacity(disk);
57 set_capacity(disk, size);
58 revalidate_disk_size(disk, true);
60 if (capacity != size && capacity != 0 && size != 0) {
61 char *envp[] = { "RESIZE=1", NULL };
63 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
69 EXPORT_SYMBOL_GPL(set_capacity_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_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 hd_struct *part)
115 unsigned int inflight = 0;
118 for_each_possible_cpu(cpu) {
119 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
120 part_stat_local_read_cpu(part, in_flight[1], cpu);
122 if ((int)inflight < 0)
128 static void part_in_flight_rw(struct hd_struct *part, unsigned int inflight[2])
134 for_each_possible_cpu(cpu) {
135 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
136 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
138 if ((int)inflight[0] < 0)
140 if ((int)inflight[1] < 0)
144 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
146 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
148 if (unlikely(partno < 0 || partno >= ptbl->len))
150 return rcu_dereference(ptbl->part[partno]);
154 * disk_get_part - get partition
155 * @disk: disk to look partition from
156 * @partno: partition number
158 * Look for partition @partno from @disk. If found, increment
159 * reference count and return it.
165 * Pointer to the found partition on success, NULL if not found.
167 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
169 struct hd_struct *part;
172 part = __disk_get_part(disk, partno);
174 get_device(part_to_dev(part));
181 * disk_part_iter_init - initialize partition iterator
182 * @piter: iterator to initialize
183 * @disk: disk to iterate over
184 * @flags: DISK_PITER_* flags
186 * Initialize @piter so that it iterates over partitions of @disk.
191 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
194 struct disk_part_tbl *ptbl;
197 ptbl = rcu_dereference(disk->part_tbl);
202 if (flags & DISK_PITER_REVERSE)
203 piter->idx = ptbl->len - 1;
204 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
209 piter->flags = flags;
213 EXPORT_SYMBOL_GPL(disk_part_iter_init);
216 * disk_part_iter_next - proceed iterator to the next partition and return it
217 * @piter: iterator of interest
219 * Proceed @piter to the next partition and return it.
224 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
226 struct disk_part_tbl *ptbl;
229 /* put the last partition */
230 disk_part_iter_exit(piter);
234 ptbl = rcu_dereference(piter->disk->part_tbl);
236 /* determine iteration parameters */
237 if (piter->flags & DISK_PITER_REVERSE) {
239 if (piter->flags & (DISK_PITER_INCL_PART0 |
240 DISK_PITER_INCL_EMPTY_PART0))
249 /* iterate to the next partition */
250 for (; piter->idx != end; piter->idx += inc) {
251 struct hd_struct *part;
253 part = rcu_dereference(ptbl->part[piter->idx]);
256 if (!part_nr_sects_read(part) &&
257 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
258 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
262 get_device(part_to_dev(part));
272 EXPORT_SYMBOL_GPL(disk_part_iter_next);
275 * disk_part_iter_exit - finish up partition iteration
276 * @piter: iter of interest
278 * Called when iteration is over. Cleans up @piter.
283 void disk_part_iter_exit(struct disk_part_iter *piter)
285 disk_put_part(piter->part);
288 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
290 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
292 return part->start_sect <= sector &&
293 sector < part->start_sect + part_nr_sects_read(part);
297 * disk_map_sector_rcu - map sector to partition
298 * @disk: gendisk of interest
299 * @sector: sector to map
301 * Find out which partition @sector maps to on @disk. This is
302 * primarily used for stats accounting.
305 * RCU read locked. The returned partition pointer is always valid
306 * because its refcount is grabbed except for part0, which lifetime
307 * is same with the disk.
310 * Found partition on success, part0 is returned if no partition matches
311 * or the matched partition is being deleted.
313 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
315 struct disk_part_tbl *ptbl;
316 struct hd_struct *part;
320 ptbl = rcu_dereference(disk->part_tbl);
322 part = rcu_dereference(ptbl->last_lookup);
323 if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
326 for (i = 1; i < ptbl->len; i++) {
327 part = rcu_dereference(ptbl->part[i]);
329 if (part && sector_in_part(part, sector)) {
331 * only live partition can be cached for lookup,
332 * so use-after-free on cached & deleting partition
335 if (!hd_struct_try_get(part))
337 rcu_assign_pointer(ptbl->last_lookup, part);
349 * disk_has_partitions
350 * @disk: gendisk of interest
352 * Walk through the partition table and check if valid partition exists.
358 * True if the gendisk has at least one valid non-zero size partition.
361 bool disk_has_partitions(struct gendisk *disk)
363 struct disk_part_tbl *ptbl;
368 ptbl = rcu_dereference(disk->part_tbl);
370 /* Iterate partitions skipping the whole device at index 0 */
371 for (i = 1; i < ptbl->len; i++) {
372 if (rcu_dereference(ptbl->part[i])) {
382 EXPORT_SYMBOL_GPL(disk_has_partitions);
385 * Can be deleted altogether. Later.
388 #define BLKDEV_MAJOR_HASH_SIZE 255
389 static struct blk_major_name {
390 struct blk_major_name *next;
393 void (*probe)(dev_t devt);
394 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
395 static DEFINE_MUTEX(major_names_lock);
397 /* index in the above - for now: assume no multimajor ranges */
398 static inline int major_to_index(unsigned major)
400 return major % BLKDEV_MAJOR_HASH_SIZE;
403 #ifdef CONFIG_PROC_FS
404 void blkdev_show(struct seq_file *seqf, off_t offset)
406 struct blk_major_name *dp;
408 mutex_lock(&major_names_lock);
409 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
410 if (dp->major == offset)
411 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
412 mutex_unlock(&major_names_lock);
414 #endif /* CONFIG_PROC_FS */
417 * __register_blkdev - register a new block device
419 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
420 * @major = 0, try to allocate any unused major number.
421 * @name: the name of the new block device as a zero terminated string
422 * @probe: allback that is called on access to any minor number of @major
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 * Use register_blkdev instead for any new code.
439 int __register_blkdev(unsigned int major, const char *name,
440 void (*probe)(dev_t devt))
442 struct blk_major_name **n, *p;
445 mutex_lock(&major_names_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);
480 strlcpy(p->name, name, sizeof(p->name));
482 index = major_to_index(major);
484 for (n = &major_names[index]; *n; n = &(*n)->next) {
485 if ((*n)->major == major)
494 printk("register_blkdev: cannot get major %u for %s\n",
499 mutex_unlock(&major_names_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(&major_names_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(&major_names_lock);
524 EXPORT_SYMBOL(unregister_blkdev);
527 * blk_mangle_minor - scatter minor numbers apart
528 * @minor: minor number to mangle
530 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
531 * is enabled. Mangling twice gives the original value.
539 static int blk_mangle_minor(int minor)
541 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
544 for (i = 0; i < MINORBITS / 2; i++) {
545 int low = minor & (1 << i);
546 int high = minor & (1 << (MINORBITS - 1 - i));
547 int distance = MINORBITS - 1 - 2 * i;
549 minor ^= low | high; /* clear both bits */
550 low <<= distance; /* swap the positions */
552 minor |= low | high; /* and set */
559 * blk_alloc_devt - allocate a dev_t for a partition
560 * @part: partition to allocate dev_t for
561 * @devt: out parameter for resulting dev_t
563 * Allocate a dev_t for block device.
566 * 0 on success, allocated dev_t is returned in *@devt. -errno on
572 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
574 struct gendisk *disk = part_to_disk(part);
577 /* in consecutive minor range? */
578 if (part->partno < disk->minors) {
579 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
583 /* allocate ext devt */
584 idr_preload(GFP_KERNEL);
586 spin_lock_bh(&ext_devt_lock);
587 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
588 spin_unlock_bh(&ext_devt_lock);
592 return idx == -ENOSPC ? -EBUSY : idx;
594 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
599 * blk_free_devt - free a dev_t
600 * @devt: dev_t to free
602 * Free @devt which was allocated using blk_alloc_devt().
607 void blk_free_devt(dev_t devt)
609 if (devt == MKDEV(0, 0))
612 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
613 spin_lock_bh(&ext_devt_lock);
614 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
615 spin_unlock_bh(&ext_devt_lock);
620 * We invalidate devt by assigning NULL pointer for devt in idr.
622 void blk_invalidate_devt(dev_t devt)
624 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
625 spin_lock_bh(&ext_devt_lock);
626 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
627 spin_unlock_bh(&ext_devt_lock);
631 static char *bdevt_str(dev_t devt, char *buf)
633 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
634 char tbuf[BDEVT_SIZE];
635 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
636 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
638 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
643 static void blk_register_region(struct gendisk *disk)
647 mutex_lock(&bdev_map_lock);
648 for (i = 0; i < disk->minors; i++) {
649 if (xa_insert(&bdev_map, disk_devt(disk) + i, disk, GFP_KERNEL))
652 mutex_unlock(&bdev_map_lock);
655 static void blk_unregister_region(struct gendisk *disk)
659 mutex_lock(&bdev_map_lock);
660 for (i = 0; i < disk->minors; i++)
661 xa_erase(&bdev_map, disk_devt(disk) + i);
662 mutex_unlock(&bdev_map_lock);
665 static void disk_scan_partitions(struct gendisk *disk)
667 struct block_device *bdev;
669 if (!get_capacity(disk) || !disk_part_scan_enabled(disk))
672 set_bit(GD_NEED_PART_SCAN, &disk->state);
673 bdev = blkdev_get_by_dev(disk_devt(disk), FMODE_READ, NULL);
675 blkdev_put(bdev, FMODE_READ);
678 static void register_disk(struct device *parent, struct gendisk *disk,
679 const struct attribute_group **groups)
681 struct device *ddev = disk_to_dev(disk);
682 struct disk_part_iter piter;
683 struct hd_struct *part;
686 ddev->parent = parent;
688 dev_set_name(ddev, "%s", disk->disk_name);
690 /* delay uevents, until we scanned partition table */
691 dev_set_uevent_suppress(ddev, 1);
694 WARN_ON(ddev->groups);
695 ddev->groups = groups;
697 if (device_add(ddev))
699 if (!sysfs_deprecated) {
700 err = sysfs_create_link(block_depr, &ddev->kobj,
701 kobject_name(&ddev->kobj));
709 * avoid probable deadlock caused by allocating memory with
710 * GFP_KERNEL in runtime_resume callback of its all ancestor
713 pm_runtime_set_memalloc_noio(ddev, true);
715 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
716 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
718 if (disk->flags & GENHD_FL_HIDDEN) {
719 dev_set_uevent_suppress(ddev, 0);
723 disk_scan_partitions(disk);
725 /* announce disk after possible partitions are created */
726 dev_set_uevent_suppress(ddev, 0);
727 kobject_uevent(&ddev->kobj, KOBJ_ADD);
729 /* announce possible partitions */
730 disk_part_iter_init(&piter, disk, 0);
731 while ((part = disk_part_iter_next(&piter)))
732 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
733 disk_part_iter_exit(&piter);
735 if (disk->queue->backing_dev_info->dev) {
736 err = sysfs_create_link(&ddev->kobj,
737 &disk->queue->backing_dev_info->dev->kobj,
744 * __device_add_disk - add disk information to kernel list
745 * @parent: parent device for the disk
746 * @disk: per-device partitioning information
747 * @groups: Additional per-device sysfs groups
748 * @register_queue: register the queue if set to true
750 * This function registers the partitioning information in @disk
753 * FIXME: error handling
755 static void __device_add_disk(struct device *parent, struct gendisk *disk,
756 const struct attribute_group **groups,
763 * The disk queue should now be all set with enough information about
764 * the device for the elevator code to pick an adequate default
765 * elevator if one is needed, that is, for devices requesting queue
769 elevator_init_mq(disk->queue);
771 /* minors == 0 indicates to use ext devt from part0 and should
772 * be accompanied with EXT_DEVT flag. Make sure all
773 * parameters make sense.
775 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
776 WARN_ON(!disk->minors &&
777 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
779 disk->flags |= GENHD_FL_UP;
781 retval = blk_alloc_devt(&disk->part0, &devt);
786 disk->major = MAJOR(devt);
787 disk->first_minor = MINOR(devt);
789 disk_alloc_events(disk);
791 if (disk->flags & GENHD_FL_HIDDEN) {
793 * Don't let hidden disks show up in /proc/partitions,
794 * and don't bother scanning for partitions either.
796 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
797 disk->flags |= GENHD_FL_NO_PART_SCAN;
799 struct backing_dev_info *bdi = disk->queue->backing_dev_info;
800 struct device *dev = disk_to_dev(disk);
803 /* Register BDI before referencing it from bdev */
805 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
807 bdi_set_owner(bdi, dev);
808 blk_register_region(disk);
810 register_disk(parent, disk, groups);
812 blk_register_queue(disk);
815 * Take an extra ref on queue which will be put on disk_release()
816 * so that it sticks around as long as @disk is there.
818 WARN_ON_ONCE(!blk_get_queue(disk->queue));
820 disk_add_events(disk);
821 blk_integrity_add(disk);
824 void device_add_disk(struct device *parent, struct gendisk *disk,
825 const struct attribute_group **groups)
828 __device_add_disk(parent, disk, groups, true);
830 EXPORT_SYMBOL(device_add_disk);
832 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
834 __device_add_disk(parent, disk, NULL, false);
836 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
838 static void invalidate_partition(struct gendisk *disk, int partno)
840 struct block_device *bdev;
842 bdev = bdget_disk(disk, partno);
847 __invalidate_device(bdev, true);
850 * Unhash the bdev inode for this device so that it gets evicted as soon
851 * as last inode reference is dropped.
853 remove_inode_hash(bdev->bd_inode);
858 * del_gendisk - remove the gendisk
859 * @disk: the struct gendisk to remove
861 * Removes the gendisk and all its associated resources. This deletes the
862 * partitions associated with the gendisk, and unregisters the associated
865 * This is the counter to the respective __device_add_disk() call.
867 * The final removal of the struct gendisk happens when its refcount reaches 0
868 * with put_disk(), which should be called after del_gendisk(), if
869 * __device_add_disk() was used.
871 * Drivers exist which depend on the release of the gendisk to be synchronous,
872 * it should not be deferred.
876 void del_gendisk(struct gendisk *disk)
878 struct disk_part_iter piter;
879 struct hd_struct *part;
883 if (WARN_ON_ONCE(!disk->queue))
886 blk_integrity_del(disk);
887 disk_del_events(disk);
890 * Block lookups of the disk until all bdevs are unhashed and the
891 * disk is marked as dead (GENHD_FL_UP cleared).
893 down_write(&disk->lookup_sem);
894 /* invalidate stuff */
895 disk_part_iter_init(&piter, disk,
896 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
897 while ((part = disk_part_iter_next(&piter))) {
898 invalidate_partition(disk, part->partno);
899 delete_partition(part);
901 disk_part_iter_exit(&piter);
903 invalidate_partition(disk, 0);
904 set_capacity(disk, 0);
905 disk->flags &= ~GENHD_FL_UP;
906 up_write(&disk->lookup_sem);
908 if (!(disk->flags & GENHD_FL_HIDDEN)) {
909 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
912 * Unregister bdi before releasing device numbers (as they can
913 * get reused and we'd get clashes in sysfs).
915 bdi_unregister(disk->queue->backing_dev_info);
918 blk_unregister_queue(disk);
920 if (!(disk->flags & GENHD_FL_HIDDEN))
921 blk_unregister_region(disk);
923 * Remove gendisk pointer from idr so that it cannot be looked up
924 * while RCU period before freeing gendisk is running to prevent
925 * use-after-free issues. Note that the device number stays
926 * "in-use" until we really free the gendisk.
928 blk_invalidate_devt(disk_devt(disk));
930 kobject_put(disk->part0.holder_dir);
931 kobject_put(disk->slave_dir);
933 part_stat_set_all(&disk->part0, 0);
934 disk->part0.stamp = 0;
935 if (!sysfs_deprecated)
936 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
937 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
938 device_del(disk_to_dev(disk));
940 EXPORT_SYMBOL(del_gendisk);
942 /* sysfs access to bad-blocks list. */
943 static ssize_t disk_badblocks_show(struct device *dev,
944 struct device_attribute *attr,
947 struct gendisk *disk = dev_to_disk(dev);
950 return sprintf(page, "\n");
952 return badblocks_show(disk->bb, page, 0);
955 static ssize_t disk_badblocks_store(struct device *dev,
956 struct device_attribute *attr,
957 const char *page, size_t len)
959 struct gendisk *disk = dev_to_disk(dev);
964 return badblocks_store(disk->bb, page, len, 0);
967 static void request_gendisk_module(dev_t devt)
969 unsigned int major = MAJOR(devt);
970 struct blk_major_name **n;
972 mutex_lock(&major_names_lock);
973 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
974 if ((*n)->major == major && (*n)->probe) {
976 mutex_unlock(&major_names_lock);
980 mutex_unlock(&major_names_lock);
982 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
983 /* Make old-style 2.4 aliases work */
984 request_module("block-major-%d", MAJOR(devt));
987 static bool get_disk_and_module(struct gendisk *disk)
989 struct module *owner;
993 owner = disk->fops->owner;
994 if (owner && !try_module_get(owner))
996 if (!kobject_get_unless_zero(&disk_to_dev(disk)->kobj)) {
1005 * get_gendisk - get partitioning information for a given device
1006 * @devt: device to get partitioning information for
1007 * @partno: returned partition index
1009 * This function gets the structure containing partitioning
1010 * information for the given device @devt.
1012 * Context: can sleep
1014 struct gendisk *get_gendisk(dev_t devt, int *partno)
1016 struct gendisk *disk = NULL;
1020 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
1021 mutex_lock(&bdev_map_lock);
1022 disk = xa_load(&bdev_map, devt);
1024 mutex_unlock(&bdev_map_lock);
1025 request_gendisk_module(devt);
1026 mutex_lock(&bdev_map_lock);
1027 disk = xa_load(&bdev_map, devt);
1029 if (disk && !get_disk_and_module(disk))
1032 *partno = devt - disk_devt(disk);
1033 mutex_unlock(&bdev_map_lock);
1035 struct hd_struct *part;
1037 spin_lock_bh(&ext_devt_lock);
1038 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
1039 if (part && get_disk_and_module(part_to_disk(part))) {
1040 *partno = part->partno;
1041 disk = part_to_disk(part);
1043 spin_unlock_bh(&ext_devt_lock);
1050 * Synchronize with del_gendisk() to not return disk that is being
1053 down_read(&disk->lookup_sem);
1054 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1055 !(disk->flags & GENHD_FL_UP))) {
1056 up_read(&disk->lookup_sem);
1057 put_disk_and_module(disk);
1060 up_read(&disk->lookup_sem);
1066 * bdget_disk - do bdget() by gendisk and partition number
1067 * @disk: gendisk of interest
1068 * @partno: partition number
1070 * Find partition @partno from @disk, do bdget() on it.
1076 * Resulting block_device on success, NULL on failure.
1078 struct block_device *bdget_disk(struct gendisk *disk, int partno)
1080 struct hd_struct *part;
1081 struct block_device *bdev = NULL;
1083 part = disk_get_part(disk, partno);
1085 bdev = bdget_part(part);
1086 disk_put_part(part);
1090 EXPORT_SYMBOL(bdget_disk);
1093 * print a full list of all partitions - intended for places where the root
1094 * filesystem can't be mounted and thus to give the victim some idea of what
1097 void __init printk_all_partitions(void)
1099 struct class_dev_iter iter;
1102 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1103 while ((dev = class_dev_iter_next(&iter))) {
1104 struct gendisk *disk = dev_to_disk(dev);
1105 struct disk_part_iter piter;
1106 struct hd_struct *part;
1107 char name_buf[BDEVNAME_SIZE];
1108 char devt_buf[BDEVT_SIZE];
1111 * Don't show empty devices or things that have been
1114 if (get_capacity(disk) == 0 ||
1115 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1119 * Note, unlike /proc/partitions, I am showing the
1120 * numbers in hex - the same format as the root=
1123 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1124 while ((part = disk_part_iter_next(&piter))) {
1125 bool is_part0 = part == &disk->part0;
1127 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1128 bdevt_str(part_devt(part), devt_buf),
1129 (unsigned long long)part_nr_sects_read(part) >> 1
1130 , disk_name(disk, part->partno, name_buf),
1131 part->info ? part->info->uuid : "");
1133 if (dev->parent && dev->parent->driver)
1134 printk(" driver: %s\n",
1135 dev->parent->driver->name);
1137 printk(" (driver?)\n");
1141 disk_part_iter_exit(&piter);
1143 class_dev_iter_exit(&iter);
1146 #ifdef CONFIG_PROC_FS
1148 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1151 struct class_dev_iter *iter;
1154 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1156 return ERR_PTR(-ENOMEM);
1158 seqf->private = iter;
1159 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1161 dev = class_dev_iter_next(iter);
1166 return dev_to_disk(dev);
1169 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1174 dev = class_dev_iter_next(seqf->private);
1176 return dev_to_disk(dev);
1181 static void disk_seqf_stop(struct seq_file *seqf, void *v)
1183 struct class_dev_iter *iter = seqf->private;
1185 /* stop is called even after start failed :-( */
1187 class_dev_iter_exit(iter);
1189 seqf->private = NULL;
1193 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1197 p = disk_seqf_start(seqf, pos);
1198 if (!IS_ERR_OR_NULL(p) && !*pos)
1199 seq_puts(seqf, "major minor #blocks name\n\n");
1203 static int show_partition(struct seq_file *seqf, void *v)
1205 struct gendisk *sgp = v;
1206 struct disk_part_iter piter;
1207 struct hd_struct *part;
1208 char buf[BDEVNAME_SIZE];
1210 /* Don't show non-partitionable removeable devices or empty devices */
1211 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1212 (sgp->flags & GENHD_FL_REMOVABLE)))
1214 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1217 /* show the full disk and all non-0 size partitions of it */
1218 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1219 while ((part = disk_part_iter_next(&piter)))
1220 seq_printf(seqf, "%4d %7d %10llu %s\n",
1221 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1222 (unsigned long long)part_nr_sects_read(part) >> 1,
1223 disk_name(sgp, part->partno, buf));
1224 disk_part_iter_exit(&piter);
1229 static const struct seq_operations partitions_op = {
1230 .start = show_partition_start,
1231 .next = disk_seqf_next,
1232 .stop = disk_seqf_stop,
1233 .show = show_partition
1237 static int __init genhd_device_init(void)
1241 block_class.dev_kobj = sysfs_dev_block_kobj;
1242 error = class_register(&block_class);
1243 if (unlikely(error))
1247 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1249 /* create top-level block dir */
1250 if (!sysfs_deprecated)
1251 block_depr = kobject_create_and_add("block", NULL);
1255 subsys_initcall(genhd_device_init);
1257 static ssize_t disk_range_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", disk->minors);
1265 static ssize_t disk_ext_range_show(struct device *dev,
1266 struct device_attribute *attr, char *buf)
1268 struct gendisk *disk = dev_to_disk(dev);
1270 return sprintf(buf, "%d\n", disk_max_parts(disk));
1273 static ssize_t disk_removable_show(struct device *dev,
1274 struct device_attribute *attr, char *buf)
1276 struct gendisk *disk = dev_to_disk(dev);
1278 return sprintf(buf, "%d\n",
1279 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1282 static ssize_t disk_hidden_show(struct device *dev,
1283 struct device_attribute *attr, char *buf)
1285 struct gendisk *disk = dev_to_disk(dev);
1287 return sprintf(buf, "%d\n",
1288 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1291 static ssize_t disk_ro_show(struct device *dev,
1292 struct device_attribute *attr, char *buf)
1294 struct gendisk *disk = dev_to_disk(dev);
1296 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1299 ssize_t part_size_show(struct device *dev,
1300 struct device_attribute *attr, char *buf)
1302 struct hd_struct *p = dev_to_part(dev);
1304 return sprintf(buf, "%llu\n",
1305 (unsigned long long)part_nr_sects_read(p));
1308 ssize_t part_stat_show(struct device *dev,
1309 struct device_attribute *attr, char *buf)
1311 struct hd_struct *p = dev_to_part(dev);
1312 struct request_queue *q = part_to_disk(p)->queue;
1313 struct disk_stats stat;
1314 unsigned int inflight;
1316 part_stat_read_all(p, &stat);
1318 inflight = blk_mq_in_flight(q, p);
1320 inflight = part_in_flight(p);
1323 "%8lu %8lu %8llu %8u "
1324 "%8lu %8lu %8llu %8u "
1326 "%8lu %8lu %8llu %8u "
1329 stat.ios[STAT_READ],
1330 stat.merges[STAT_READ],
1331 (unsigned long long)stat.sectors[STAT_READ],
1332 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1333 stat.ios[STAT_WRITE],
1334 stat.merges[STAT_WRITE],
1335 (unsigned long long)stat.sectors[STAT_WRITE],
1336 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1338 jiffies_to_msecs(stat.io_ticks),
1339 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1340 stat.nsecs[STAT_WRITE] +
1341 stat.nsecs[STAT_DISCARD] +
1342 stat.nsecs[STAT_FLUSH],
1344 stat.ios[STAT_DISCARD],
1345 stat.merges[STAT_DISCARD],
1346 (unsigned long long)stat.sectors[STAT_DISCARD],
1347 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1348 stat.ios[STAT_FLUSH],
1349 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1352 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1355 struct hd_struct *p = dev_to_part(dev);
1356 struct request_queue *q = part_to_disk(p)->queue;
1357 unsigned int inflight[2];
1360 blk_mq_in_flight_rw(q, p, inflight);
1362 part_in_flight_rw(p, inflight);
1364 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1367 static ssize_t disk_capability_show(struct device *dev,
1368 struct device_attribute *attr, char *buf)
1370 struct gendisk *disk = dev_to_disk(dev);
1372 return sprintf(buf, "%x\n", disk->flags);
1375 static ssize_t disk_alignment_offset_show(struct device *dev,
1376 struct device_attribute *attr,
1379 struct gendisk *disk = dev_to_disk(dev);
1381 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1384 static ssize_t disk_discard_alignment_show(struct device *dev,
1385 struct device_attribute *attr,
1388 struct gendisk *disk = dev_to_disk(dev);
1390 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1393 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1394 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1395 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1396 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1397 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1398 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1399 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1400 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1401 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1402 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1403 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1404 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1406 #ifdef CONFIG_FAIL_MAKE_REQUEST
1407 ssize_t part_fail_show(struct device *dev,
1408 struct device_attribute *attr, char *buf)
1410 struct hd_struct *p = dev_to_part(dev);
1412 return sprintf(buf, "%d\n", p->make_it_fail);
1415 ssize_t part_fail_store(struct device *dev,
1416 struct device_attribute *attr,
1417 const char *buf, size_t count)
1419 struct hd_struct *p = dev_to_part(dev);
1422 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1423 p->make_it_fail = (i == 0) ? 0 : 1;
1428 static struct device_attribute dev_attr_fail =
1429 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1430 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1432 #ifdef CONFIG_FAIL_IO_TIMEOUT
1433 static struct device_attribute dev_attr_fail_timeout =
1434 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1437 static struct attribute *disk_attrs[] = {
1438 &dev_attr_range.attr,
1439 &dev_attr_ext_range.attr,
1440 &dev_attr_removable.attr,
1441 &dev_attr_hidden.attr,
1443 &dev_attr_size.attr,
1444 &dev_attr_alignment_offset.attr,
1445 &dev_attr_discard_alignment.attr,
1446 &dev_attr_capability.attr,
1447 &dev_attr_stat.attr,
1448 &dev_attr_inflight.attr,
1449 &dev_attr_badblocks.attr,
1450 #ifdef CONFIG_FAIL_MAKE_REQUEST
1451 &dev_attr_fail.attr,
1453 #ifdef CONFIG_FAIL_IO_TIMEOUT
1454 &dev_attr_fail_timeout.attr,
1459 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1461 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1462 struct gendisk *disk = dev_to_disk(dev);
1464 if (a == &dev_attr_badblocks.attr && !disk->bb)
1469 static struct attribute_group disk_attr_group = {
1470 .attrs = disk_attrs,
1471 .is_visible = disk_visible,
1474 static const struct attribute_group *disk_attr_groups[] = {
1480 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1481 * @disk: disk to replace part_tbl for
1482 * @new_ptbl: new part_tbl to install
1484 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1485 * original ptbl is freed using RCU callback.
1488 * Matching bd_mutex locked or the caller is the only user of @disk.
1490 static void disk_replace_part_tbl(struct gendisk *disk,
1491 struct disk_part_tbl *new_ptbl)
1493 struct disk_part_tbl *old_ptbl =
1494 rcu_dereference_protected(disk->part_tbl, 1);
1496 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1499 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1500 kfree_rcu(old_ptbl, rcu_head);
1505 * disk_expand_part_tbl - expand disk->part_tbl
1506 * @disk: disk to expand part_tbl for
1507 * @partno: expand such that this partno can fit in
1509 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1510 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1513 * Matching bd_mutex locked or the caller is the only user of @disk.
1517 * 0 on success, -errno on failure.
1519 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1521 struct disk_part_tbl *old_ptbl =
1522 rcu_dereference_protected(disk->part_tbl, 1);
1523 struct disk_part_tbl *new_ptbl;
1524 int len = old_ptbl ? old_ptbl->len : 0;
1528 * check for int overflow, since we can get here from blkpg_ioctl()
1529 * with a user passed 'partno'.
1531 target = partno + 1;
1535 /* disk_max_parts() is zero during initialization, ignore if so */
1536 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1542 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1547 new_ptbl->len = target;
1549 for (i = 0; i < len; i++)
1550 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1552 disk_replace_part_tbl(disk, new_ptbl);
1557 * disk_release - releases all allocated resources of the gendisk
1558 * @dev: the device representing this disk
1560 * This function releases all allocated resources of the gendisk.
1562 * The struct gendisk refcount is incremented with get_gendisk() or
1563 * get_disk_and_module(), and its refcount is decremented with
1564 * put_disk_and_module() or put_disk(). Once the refcount reaches 0 this
1565 * function is called.
1567 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1568 * assigned. Since the request_queue sits on top of the gendisk for these
1569 * drivers we also call blk_put_queue() for them, and we expect the
1570 * request_queue refcount to reach 0 at this point, and so the request_queue
1571 * will also be freed prior to the disk.
1573 * Context: can sleep
1575 static void disk_release(struct device *dev)
1577 struct gendisk *disk = dev_to_disk(dev);
1581 blk_free_devt(dev->devt);
1582 disk_release_events(disk);
1583 kfree(disk->random);
1584 disk_replace_part_tbl(disk, NULL);
1585 hd_free_part(&disk->part0);
1587 blk_put_queue(disk->queue);
1590 struct class block_class = {
1594 static char *block_devnode(struct device *dev, umode_t *mode,
1595 kuid_t *uid, kgid_t *gid)
1597 struct gendisk *disk = dev_to_disk(dev);
1599 if (disk->fops->devnode)
1600 return disk->fops->devnode(disk, mode);
1604 const struct device_type disk_type = {
1606 .groups = disk_attr_groups,
1607 .release = disk_release,
1608 .devnode = block_devnode,
1611 #ifdef CONFIG_PROC_FS
1613 * aggregate disk stat collector. Uses the same stats that the sysfs
1614 * entries do, above, but makes them available through one seq_file.
1616 * The output looks suspiciously like /proc/partitions with a bunch of
1619 static int diskstats_show(struct seq_file *seqf, void *v)
1621 struct gendisk *gp = v;
1622 struct disk_part_iter piter;
1623 struct hd_struct *hd;
1624 char buf[BDEVNAME_SIZE];
1625 unsigned int inflight;
1626 struct disk_stats stat;
1629 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1630 seq_puts(seqf, "major minor name"
1631 " rio rmerge rsect ruse wio wmerge "
1632 "wsect wuse running use aveq"
1636 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1637 while ((hd = disk_part_iter_next(&piter))) {
1638 part_stat_read_all(hd, &stat);
1639 if (queue_is_mq(gp->queue))
1640 inflight = blk_mq_in_flight(gp->queue, hd);
1642 inflight = part_in_flight(hd);
1644 seq_printf(seqf, "%4d %7d %s "
1651 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1652 disk_name(gp, hd->partno, buf),
1653 stat.ios[STAT_READ],
1654 stat.merges[STAT_READ],
1655 stat.sectors[STAT_READ],
1656 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1658 stat.ios[STAT_WRITE],
1659 stat.merges[STAT_WRITE],
1660 stat.sectors[STAT_WRITE],
1661 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1664 jiffies_to_msecs(stat.io_ticks),
1665 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1666 stat.nsecs[STAT_WRITE] +
1667 stat.nsecs[STAT_DISCARD] +
1668 stat.nsecs[STAT_FLUSH],
1670 stat.ios[STAT_DISCARD],
1671 stat.merges[STAT_DISCARD],
1672 stat.sectors[STAT_DISCARD],
1673 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1675 stat.ios[STAT_FLUSH],
1676 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1680 disk_part_iter_exit(&piter);
1685 static const struct seq_operations diskstats_op = {
1686 .start = disk_seqf_start,
1687 .next = disk_seqf_next,
1688 .stop = disk_seqf_stop,
1689 .show = diskstats_show
1692 static int __init proc_genhd_init(void)
1694 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1695 proc_create_seq("partitions", 0, NULL, &partitions_op);
1698 module_init(proc_genhd_init);
1699 #endif /* CONFIG_PROC_FS */
1701 dev_t blk_lookup_devt(const char *name, int partno)
1703 dev_t devt = MKDEV(0, 0);
1704 struct class_dev_iter iter;
1707 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1708 while ((dev = class_dev_iter_next(&iter))) {
1709 struct gendisk *disk = dev_to_disk(dev);
1710 struct hd_struct *part;
1712 if (strcmp(dev_name(dev), name))
1715 if (partno < disk->minors) {
1716 /* We need to return the right devno, even
1717 * if the partition doesn't exist yet.
1719 devt = MKDEV(MAJOR(dev->devt),
1720 MINOR(dev->devt) + partno);
1723 part = disk_get_part(disk, partno);
1725 devt = part_devt(part);
1726 disk_put_part(part);
1729 disk_put_part(part);
1731 class_dev_iter_exit(&iter);
1735 struct gendisk *__alloc_disk_node(int minors, int node_id)
1737 struct gendisk *disk;
1738 struct disk_part_tbl *ptbl;
1740 if (minors > DISK_MAX_PARTS) {
1742 "block: can't allocate more than %d partitions\n",
1744 minors = DISK_MAX_PARTS;
1747 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1751 disk->part0.dkstats = alloc_percpu(struct disk_stats);
1752 if (!disk->part0.dkstats)
1755 init_rwsem(&disk->lookup_sem);
1756 disk->node_id = node_id;
1757 if (disk_expand_part_tbl(disk, 0)) {
1758 free_percpu(disk->part0.dkstats);
1762 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1763 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1766 * set_capacity() and get_capacity() currently don't use
1767 * seqcounter to read/update the part0->nr_sects. Still init
1768 * the counter as we can read the sectors in IO submission
1769 * patch using seqence counters.
1771 * TODO: Ideally set_capacity() and get_capacity() should be
1772 * converted to make use of bd_mutex and sequence counters.
1774 hd_sects_seq_init(&disk->part0);
1775 if (hd_ref_init(&disk->part0))
1776 goto out_free_part0;
1778 disk->minors = minors;
1779 rand_initialize_disk(disk);
1780 disk_to_dev(disk)->class = &block_class;
1781 disk_to_dev(disk)->type = &disk_type;
1782 device_initialize(disk_to_dev(disk));
1786 hd_free_part(&disk->part0);
1791 EXPORT_SYMBOL(__alloc_disk_node);
1794 * put_disk - decrements the gendisk refcount
1795 * @disk: the struct gendisk to decrement the refcount for
1797 * This decrements the refcount for the struct gendisk. When this reaches 0
1798 * we'll have disk_release() called.
1800 * Context: Any context, but the last reference must not be dropped from
1803 void put_disk(struct gendisk *disk)
1806 kobject_put(&disk_to_dev(disk)->kobj);
1808 EXPORT_SYMBOL(put_disk);
1811 * put_disk_and_module - decrements the module and gendisk refcount
1812 * @disk: the struct gendisk to decrement the refcount for
1814 * This is a counterpart of get_disk_and_module() and thus also of
1817 * Context: Any context, but the last reference must not be dropped from
1820 void put_disk_and_module(struct gendisk *disk)
1823 struct module *owner = disk->fops->owner;
1830 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1832 char event[] = "DISK_RO=1";
1833 char *envp[] = { event, NULL };
1837 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1840 void set_disk_ro(struct gendisk *disk, int flag)
1842 struct disk_part_iter piter;
1843 struct hd_struct *part;
1845 if (disk->part0.policy != flag) {
1846 set_disk_ro_uevent(disk, flag);
1847 disk->part0.policy = flag;
1850 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1851 while ((part = disk_part_iter_next(&piter)))
1852 part->policy = flag;
1853 disk_part_iter_exit(&piter);
1856 EXPORT_SYMBOL(set_disk_ro);
1858 int bdev_read_only(struct block_device *bdev)
1862 return bdev->bd_part->policy;
1865 EXPORT_SYMBOL(bdev_read_only);
1868 * Disk events - monitor disk events like media change and eject request.
1870 struct disk_events {
1871 struct list_head node; /* all disk_event's */
1872 struct gendisk *disk; /* the associated disk */
1875 struct mutex block_mutex; /* protects blocking */
1876 int block; /* event blocking depth */
1877 unsigned int pending; /* events already sent out */
1878 unsigned int clearing; /* events being cleared */
1880 long poll_msecs; /* interval, -1 for default */
1881 struct delayed_work dwork;
1884 static const char *disk_events_strs[] = {
1885 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1886 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1889 static char *disk_uevents[] = {
1890 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1891 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1894 /* list of all disk_events */
1895 static DEFINE_MUTEX(disk_events_mutex);
1896 static LIST_HEAD(disk_events);
1898 /* disable in-kernel polling by default */
1899 static unsigned long disk_events_dfl_poll_msecs;
1901 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1903 struct disk_events *ev = disk->ev;
1904 long intv_msecs = 0;
1907 * If device-specific poll interval is set, always use it. If
1908 * the default is being used, poll if the POLL flag is set.
1910 if (ev->poll_msecs >= 0)
1911 intv_msecs = ev->poll_msecs;
1912 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1913 intv_msecs = disk_events_dfl_poll_msecs;
1915 return msecs_to_jiffies(intv_msecs);
1919 * disk_block_events - block and flush disk event checking
1920 * @disk: disk to block events for
1922 * On return from this function, it is guaranteed that event checking
1923 * isn't in progress and won't happen until unblocked by
1924 * disk_unblock_events(). Events blocking is counted and the actual
1925 * unblocking happens after the matching number of unblocks are done.
1927 * Note that this intentionally does not block event checking from
1928 * disk_clear_events().
1933 void disk_block_events(struct gendisk *disk)
1935 struct disk_events *ev = disk->ev;
1936 unsigned long flags;
1943 * Outer mutex ensures that the first blocker completes canceling
1944 * the event work before further blockers are allowed to finish.
1946 mutex_lock(&ev->block_mutex);
1948 spin_lock_irqsave(&ev->lock, flags);
1949 cancel = !ev->block++;
1950 spin_unlock_irqrestore(&ev->lock, flags);
1953 cancel_delayed_work_sync(&disk->ev->dwork);
1955 mutex_unlock(&ev->block_mutex);
1958 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1960 struct disk_events *ev = disk->ev;
1962 unsigned long flags;
1964 spin_lock_irqsave(&ev->lock, flags);
1966 if (WARN_ON_ONCE(ev->block <= 0))
1972 intv = disk_events_poll_jiffies(disk);
1974 queue_delayed_work(system_freezable_power_efficient_wq,
1977 queue_delayed_work(system_freezable_power_efficient_wq,
1980 spin_unlock_irqrestore(&ev->lock, flags);
1984 * disk_unblock_events - unblock disk event checking
1985 * @disk: disk to unblock events for
1987 * Undo disk_block_events(). When the block count reaches zero, it
1988 * starts events polling if configured.
1991 * Don't care. Safe to call from irq context.
1993 void disk_unblock_events(struct gendisk *disk)
1996 __disk_unblock_events(disk, false);
2000 * disk_flush_events - schedule immediate event checking and flushing
2001 * @disk: disk to check and flush events for
2002 * @mask: events to flush
2004 * Schedule immediate event checking on @disk if not blocked. Events in
2005 * @mask are scheduled to be cleared from the driver. Note that this
2006 * doesn't clear the events from @disk->ev.
2009 * If @mask is non-zero must be called with bdev->bd_mutex held.
2011 void disk_flush_events(struct gendisk *disk, unsigned int mask)
2013 struct disk_events *ev = disk->ev;
2018 spin_lock_irq(&ev->lock);
2019 ev->clearing |= mask;
2021 mod_delayed_work(system_freezable_power_efficient_wq,
2023 spin_unlock_irq(&ev->lock);
2027 * disk_clear_events - synchronously check, clear and return pending events
2028 * @disk: disk to fetch and clear events from
2029 * @mask: mask of events to be fetched and cleared
2031 * Disk events are synchronously checked and pending events in @mask
2032 * are cleared and returned. This ignores the block count.
2037 static unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
2039 struct disk_events *ev = disk->ev;
2040 unsigned int pending;
2041 unsigned int clearing = mask;
2046 disk_block_events(disk);
2049 * store the union of mask and ev->clearing on the stack so that the
2050 * race with disk_flush_events does not cause ambiguity (ev->clearing
2051 * can still be modified even if events are blocked).
2053 spin_lock_irq(&ev->lock);
2054 clearing |= ev->clearing;
2056 spin_unlock_irq(&ev->lock);
2058 disk_check_events(ev, &clearing);
2060 * if ev->clearing is not 0, the disk_flush_events got called in the
2061 * middle of this function, so we want to run the workfn without delay.
2063 __disk_unblock_events(disk, ev->clearing ? true : false);
2065 /* then, fetch and clear pending events */
2066 spin_lock_irq(&ev->lock);
2067 pending = ev->pending & mask;
2068 ev->pending &= ~mask;
2069 spin_unlock_irq(&ev->lock);
2070 WARN_ON_ONCE(clearing & mask);
2076 * bdev_check_media_change - check if a removable media has been changed
2077 * @bdev: block device to check
2079 * Check whether a removable media has been changed, and attempt to free all
2080 * dentries and inodes and invalidates all block device page cache entries in
2083 * Returns %true if the block device changed, or %false if not.
2085 bool bdev_check_media_change(struct block_device *bdev)
2087 unsigned int events;
2089 events = disk_clear_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE |
2090 DISK_EVENT_EJECT_REQUEST);
2091 if (!(events & DISK_EVENT_MEDIA_CHANGE))
2094 if (__invalidate_device(bdev, true))
2095 pr_warn("VFS: busy inodes on changed media %s\n",
2096 bdev->bd_disk->disk_name);
2097 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
2100 EXPORT_SYMBOL(bdev_check_media_change);
2103 * Separate this part out so that a different pointer for clearing_ptr can be
2104 * passed in for disk_clear_events.
2106 static void disk_events_workfn(struct work_struct *work)
2108 struct delayed_work *dwork = to_delayed_work(work);
2109 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2111 disk_check_events(ev, &ev->clearing);
2114 static void disk_check_events(struct disk_events *ev,
2115 unsigned int *clearing_ptr)
2117 struct gendisk *disk = ev->disk;
2118 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2119 unsigned int clearing = *clearing_ptr;
2120 unsigned int events;
2122 int nr_events = 0, i;
2125 events = disk->fops->check_events(disk, clearing);
2127 /* accumulate pending events and schedule next poll if necessary */
2128 spin_lock_irq(&ev->lock);
2130 events &= ~ev->pending;
2131 ev->pending |= events;
2132 *clearing_ptr &= ~clearing;
2134 intv = disk_events_poll_jiffies(disk);
2135 if (!ev->block && intv)
2136 queue_delayed_work(system_freezable_power_efficient_wq,
2139 spin_unlock_irq(&ev->lock);
2142 * Tell userland about new events. Only the events listed in
2143 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2144 * is set. Otherwise, events are processed internally but never
2145 * get reported to userland.
2147 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2148 if ((events & disk->events & (1 << i)) &&
2149 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2150 envp[nr_events++] = disk_uevents[i];
2153 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2157 * A disk events enabled device has the following sysfs nodes under
2158 * its /sys/block/X/ directory.
2160 * events : list of all supported events
2161 * events_async : list of events which can be detected w/o polling
2162 * (always empty, only for backwards compatibility)
2163 * events_poll_msecs : polling interval, 0: disable, -1: system default
2165 static ssize_t __disk_events_show(unsigned int events, char *buf)
2167 const char *delim = "";
2171 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2172 if (events & (1 << i)) {
2173 pos += sprintf(buf + pos, "%s%s",
2174 delim, disk_events_strs[i]);
2178 pos += sprintf(buf + pos, "\n");
2182 static ssize_t disk_events_show(struct device *dev,
2183 struct device_attribute *attr, char *buf)
2185 struct gendisk *disk = dev_to_disk(dev);
2187 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2190 return __disk_events_show(disk->events, buf);
2193 static ssize_t disk_events_async_show(struct device *dev,
2194 struct device_attribute *attr, char *buf)
2199 static ssize_t disk_events_poll_msecs_show(struct device *dev,
2200 struct device_attribute *attr,
2203 struct gendisk *disk = dev_to_disk(dev);
2206 return sprintf(buf, "-1\n");
2208 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2211 static ssize_t disk_events_poll_msecs_store(struct device *dev,
2212 struct device_attribute *attr,
2213 const char *buf, size_t count)
2215 struct gendisk *disk = dev_to_disk(dev);
2218 if (!count || !sscanf(buf, "%ld", &intv))
2221 if (intv < 0 && intv != -1)
2227 disk_block_events(disk);
2228 disk->ev->poll_msecs = intv;
2229 __disk_unblock_events(disk, true);
2234 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2235 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2236 static const DEVICE_ATTR(events_poll_msecs, 0644,
2237 disk_events_poll_msecs_show,
2238 disk_events_poll_msecs_store);
2240 static const struct attribute *disk_events_attrs[] = {
2241 &dev_attr_events.attr,
2242 &dev_attr_events_async.attr,
2243 &dev_attr_events_poll_msecs.attr,
2248 * The default polling interval can be specified by the kernel
2249 * parameter block.events_dfl_poll_msecs which defaults to 0
2250 * (disable). This can also be modified runtime by writing to
2251 * /sys/module/block/parameters/events_dfl_poll_msecs.
2253 static int disk_events_set_dfl_poll_msecs(const char *val,
2254 const struct kernel_param *kp)
2256 struct disk_events *ev;
2259 ret = param_set_ulong(val, kp);
2263 mutex_lock(&disk_events_mutex);
2265 list_for_each_entry(ev, &disk_events, node)
2266 disk_flush_events(ev->disk, 0);
2268 mutex_unlock(&disk_events_mutex);
2273 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2274 .set = disk_events_set_dfl_poll_msecs,
2275 .get = param_get_ulong,
2278 #undef MODULE_PARAM_PREFIX
2279 #define MODULE_PARAM_PREFIX "block."
2281 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2282 &disk_events_dfl_poll_msecs, 0644);
2285 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2287 static void disk_alloc_events(struct gendisk *disk)
2289 struct disk_events *ev;
2291 if (!disk->fops->check_events || !disk->events)
2294 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2296 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2300 INIT_LIST_HEAD(&ev->node);
2302 spin_lock_init(&ev->lock);
2303 mutex_init(&ev->block_mutex);
2305 ev->poll_msecs = -1;
2306 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2311 static void disk_add_events(struct gendisk *disk)
2313 /* FIXME: error handling */
2314 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2315 pr_warn("%s: failed to create sysfs files for events\n",
2321 mutex_lock(&disk_events_mutex);
2322 list_add_tail(&disk->ev->node, &disk_events);
2323 mutex_unlock(&disk_events_mutex);
2326 * Block count is initialized to 1 and the following initial
2327 * unblock kicks it into action.
2329 __disk_unblock_events(disk, true);
2332 static void disk_del_events(struct gendisk *disk)
2335 disk_block_events(disk);
2337 mutex_lock(&disk_events_mutex);
2338 list_del_init(&disk->ev->node);
2339 mutex_unlock(&disk_events_mutex);
2342 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2345 static void disk_release_events(struct gendisk *disk)
2347 /* the block count should be 1 from disk_del_events() */
2348 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);