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 const struct device_type disk_type;
43 static void disk_check_events(struct disk_events *ev,
44 unsigned int *clearing_ptr);
45 static void disk_alloc_events(struct gendisk *disk);
46 static void disk_add_events(struct gendisk *disk);
47 static void disk_del_events(struct gendisk *disk);
48 static void disk_release_events(struct gendisk *disk);
51 * Set disk capacity and notify if the size is not currently
52 * zero and will not be set to zero
54 void set_capacity_revalidate_and_notify(struct gendisk *disk, sector_t size,
57 sector_t capacity = get_capacity(disk);
59 set_capacity(disk, size);
62 revalidate_disk(disk);
64 if (capacity != size && capacity != 0 && size != 0) {
65 char *envp[] = { "RESIZE=1", NULL };
67 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
71 EXPORT_SYMBOL_GPL(set_capacity_revalidate_and_notify);
74 * Format the device name of the indicated disk into the supplied buffer and
75 * return a pointer to that same buffer for convenience.
77 char *disk_name(struct gendisk *hd, int partno, char *buf)
80 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
81 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
82 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
84 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
89 const char *bdevname(struct block_device *bdev, char *buf)
91 return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
93 EXPORT_SYMBOL(bdevname);
95 static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
99 memset(stat, 0, sizeof(struct disk_stats));
100 for_each_possible_cpu(cpu) {
101 struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
104 for (group = 0; group < NR_STAT_GROUPS; group++) {
105 stat->nsecs[group] += ptr->nsecs[group];
106 stat->sectors[group] += ptr->sectors[group];
107 stat->ios[group] += ptr->ios[group];
108 stat->merges[group] += ptr->merges[group];
111 stat->io_ticks += ptr->io_ticks;
115 static unsigned int part_in_flight(struct request_queue *q,
116 struct hd_struct *part)
118 unsigned int inflight = 0;
121 for_each_possible_cpu(cpu) {
122 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
123 part_stat_local_read_cpu(part, in_flight[1], cpu);
125 if ((int)inflight < 0)
131 static void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
132 unsigned int inflight[2])
138 for_each_possible_cpu(cpu) {
139 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
140 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
142 if ((int)inflight[0] < 0)
144 if ((int)inflight[1] < 0)
148 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
150 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
152 if (unlikely(partno < 0 || partno >= ptbl->len))
154 return rcu_dereference(ptbl->part[partno]);
158 * disk_get_part - get partition
159 * @disk: disk to look partition from
160 * @partno: partition number
162 * Look for partition @partno from @disk. If found, increment
163 * reference count and return it.
169 * Pointer to the found partition on success, NULL if not found.
171 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
173 struct hd_struct *part;
176 part = __disk_get_part(disk, partno);
178 get_device(part_to_dev(part));
185 * disk_part_iter_init - initialize partition iterator
186 * @piter: iterator to initialize
187 * @disk: disk to iterate over
188 * @flags: DISK_PITER_* flags
190 * Initialize @piter so that it iterates over partitions of @disk.
195 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
198 struct disk_part_tbl *ptbl;
201 ptbl = rcu_dereference(disk->part_tbl);
206 if (flags & DISK_PITER_REVERSE)
207 piter->idx = ptbl->len - 1;
208 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
213 piter->flags = flags;
217 EXPORT_SYMBOL_GPL(disk_part_iter_init);
220 * disk_part_iter_next - proceed iterator to the next partition and return it
221 * @piter: iterator of interest
223 * Proceed @piter to the next partition and return it.
228 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
230 struct disk_part_tbl *ptbl;
233 /* put the last partition */
234 disk_put_part(piter->part);
239 ptbl = rcu_dereference(piter->disk->part_tbl);
241 /* determine iteration parameters */
242 if (piter->flags & DISK_PITER_REVERSE) {
244 if (piter->flags & (DISK_PITER_INCL_PART0 |
245 DISK_PITER_INCL_EMPTY_PART0))
254 /* iterate to the next partition */
255 for (; piter->idx != end; piter->idx += inc) {
256 struct hd_struct *part;
258 part = rcu_dereference(ptbl->part[piter->idx]);
261 if (!part_nr_sects_read(part) &&
262 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
263 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
267 get_device(part_to_dev(part));
277 EXPORT_SYMBOL_GPL(disk_part_iter_next);
280 * disk_part_iter_exit - finish up partition iteration
281 * @piter: iter of interest
283 * Called when iteration is over. Cleans up @piter.
288 void disk_part_iter_exit(struct disk_part_iter *piter)
290 disk_put_part(piter->part);
293 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
295 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
297 return part->start_sect <= sector &&
298 sector < part->start_sect + part_nr_sects_read(part);
302 * disk_map_sector_rcu - map sector to partition
303 * @disk: gendisk of interest
304 * @sector: sector to map
306 * Find out which partition @sector maps to on @disk. This is
307 * primarily used for stats accounting.
310 * RCU read locked. The returned partition pointer is always valid
311 * because its refcount is grabbed except for part0, which lifetime
312 * is same with the disk.
315 * Found partition on success, part0 is returned if no partition matches
316 * or the matched partition is being deleted.
318 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
320 struct disk_part_tbl *ptbl;
321 struct hd_struct *part;
325 ptbl = rcu_dereference(disk->part_tbl);
327 part = rcu_dereference(ptbl->last_lookup);
328 if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
331 for (i = 1; i < ptbl->len; i++) {
332 part = rcu_dereference(ptbl->part[i]);
334 if (part && sector_in_part(part, sector)) {
336 * only live partition can be cached for lookup,
337 * so use-after-free on cached & deleting partition
340 if (!hd_struct_try_get(part))
342 rcu_assign_pointer(ptbl->last_lookup, part);
354 * disk_has_partitions
355 * @disk: gendisk of interest
357 * Walk through the partition table and check if valid partition exists.
363 * True if the gendisk has at least one valid non-zero size partition.
366 bool disk_has_partitions(struct gendisk *disk)
368 struct disk_part_tbl *ptbl;
373 ptbl = rcu_dereference(disk->part_tbl);
375 /* Iterate partitions skipping the whole device at index 0 */
376 for (i = 1; i < ptbl->len; i++) {
377 if (rcu_dereference(ptbl->part[i])) {
387 EXPORT_SYMBOL_GPL(disk_has_partitions);
390 * Can be deleted altogether. Later.
393 #define BLKDEV_MAJOR_HASH_SIZE 255
394 static struct blk_major_name {
395 struct blk_major_name *next;
398 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
400 /* index in the above - for now: assume no multimajor ranges */
401 static inline int major_to_index(unsigned major)
403 return major % BLKDEV_MAJOR_HASH_SIZE;
406 #ifdef CONFIG_PROC_FS
407 void blkdev_show(struct seq_file *seqf, off_t offset)
409 struct blk_major_name *dp;
411 mutex_lock(&block_class_lock);
412 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
413 if (dp->major == offset)
414 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
415 mutex_unlock(&block_class_lock);
417 #endif /* CONFIG_PROC_FS */
420 * register_blkdev - register a new block device
422 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
423 * @major = 0, try to allocate any unused major number.
424 * @name: the name of the new block device as a zero terminated string
426 * The @name must be unique within the system.
428 * The return value depends on the @major input parameter:
430 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
431 * then the function returns zero on success, or a negative error code
432 * - if any unused major number was requested with @major = 0 parameter
433 * then the return value is the allocated major number in range
434 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
436 * See Documentation/admin-guide/devices.txt for the list of allocated
439 int register_blkdev(unsigned int major, const char *name)
441 struct blk_major_name **n, *p;
444 mutex_lock(&block_class_lock);
448 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
449 if (major_names[index] == NULL)
454 printk("%s: failed to get major for %s\n",
463 if (major >= BLKDEV_MAJOR_MAX) {
464 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
465 __func__, major, BLKDEV_MAJOR_MAX-1, name);
471 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
478 strlcpy(p->name, name, sizeof(p->name));
480 index = major_to_index(major);
482 for (n = &major_names[index]; *n; n = &(*n)->next) {
483 if ((*n)->major == major)
492 printk("register_blkdev: cannot get major %u for %s\n",
497 mutex_unlock(&block_class_lock);
501 EXPORT_SYMBOL(register_blkdev);
503 void unregister_blkdev(unsigned int major, const char *name)
505 struct blk_major_name **n;
506 struct blk_major_name *p = NULL;
507 int index = major_to_index(major);
509 mutex_lock(&block_class_lock);
510 for (n = &major_names[index]; *n; n = &(*n)->next)
511 if ((*n)->major == major)
513 if (!*n || strcmp((*n)->name, name)) {
519 mutex_unlock(&block_class_lock);
523 EXPORT_SYMBOL(unregister_blkdev);
525 static struct kobj_map *bdev_map;
528 * blk_mangle_minor - scatter minor numbers apart
529 * @minor: minor number to mangle
531 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
532 * is enabled. Mangling twice gives the original value.
540 static int blk_mangle_minor(int minor)
542 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
545 for (i = 0; i < MINORBITS / 2; i++) {
546 int low = minor & (1 << i);
547 int high = minor & (1 << (MINORBITS - 1 - i));
548 int distance = MINORBITS - 1 - 2 * i;
550 minor ^= low | high; /* clear both bits */
551 low <<= distance; /* swap the positions */
553 minor |= low | high; /* and set */
560 * blk_alloc_devt - allocate a dev_t for a partition
561 * @part: partition to allocate dev_t for
562 * @devt: out parameter for resulting dev_t
564 * Allocate a dev_t for block device.
567 * 0 on success, allocated dev_t is returned in *@devt. -errno on
573 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
575 struct gendisk *disk = part_to_disk(part);
578 /* in consecutive minor range? */
579 if (part->partno < disk->minors) {
580 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
584 /* allocate ext devt */
585 idr_preload(GFP_KERNEL);
587 spin_lock_bh(&ext_devt_lock);
588 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
589 spin_unlock_bh(&ext_devt_lock);
593 return idx == -ENOSPC ? -EBUSY : idx;
595 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
600 * blk_free_devt - free a dev_t
601 * @devt: dev_t to free
603 * Free @devt which was allocated using blk_alloc_devt().
608 void blk_free_devt(dev_t devt)
610 if (devt == MKDEV(0, 0))
613 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
614 spin_lock_bh(&ext_devt_lock);
615 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
616 spin_unlock_bh(&ext_devt_lock);
621 * We invalidate devt by assigning NULL pointer for devt in idr.
623 void blk_invalidate_devt(dev_t devt)
625 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
626 spin_lock_bh(&ext_devt_lock);
627 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
628 spin_unlock_bh(&ext_devt_lock);
632 static char *bdevt_str(dev_t devt, char *buf)
634 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
635 char tbuf[BDEVT_SIZE];
636 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
637 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
639 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
645 * Register device numbers dev..(dev+range-1)
646 * range must be nonzero
647 * The hash chain is sorted on range, so that subranges can override.
649 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
650 struct kobject *(*probe)(dev_t, int *, void *),
651 int (*lock)(dev_t, void *), void *data)
653 kobj_map(bdev_map, devt, range, module, probe, lock, data);
656 EXPORT_SYMBOL(blk_register_region);
658 void blk_unregister_region(dev_t devt, unsigned long range)
660 kobj_unmap(bdev_map, devt, range);
663 EXPORT_SYMBOL(blk_unregister_region);
665 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
667 struct gendisk *p = data;
669 return &disk_to_dev(p)->kobj;
672 static int exact_lock(dev_t devt, void *data)
674 struct gendisk *p = data;
676 if (!get_disk_and_module(p))
681 static void register_disk(struct device *parent, struct gendisk *disk,
682 const struct attribute_group **groups)
684 struct device *ddev = disk_to_dev(disk);
685 struct block_device *bdev;
686 struct disk_part_iter piter;
687 struct hd_struct *part;
690 ddev->parent = parent;
692 dev_set_name(ddev, "%s", disk->disk_name);
694 /* delay uevents, until we scanned partition table */
695 dev_set_uevent_suppress(ddev, 1);
698 WARN_ON(ddev->groups);
699 ddev->groups = groups;
701 if (device_add(ddev))
703 if (!sysfs_deprecated) {
704 err = sysfs_create_link(block_depr, &ddev->kobj,
705 kobject_name(&ddev->kobj));
713 * avoid probable deadlock caused by allocating memory with
714 * GFP_KERNEL in runtime_resume callback of its all ancestor
717 pm_runtime_set_memalloc_noio(ddev, true);
719 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
720 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
722 if (disk->flags & GENHD_FL_HIDDEN) {
723 dev_set_uevent_suppress(ddev, 0);
727 /* No minors to use for partitions */
728 if (!disk_part_scan_enabled(disk))
731 /* No such device (e.g., media were just removed) */
732 if (!get_capacity(disk))
735 bdev = bdget_disk(disk, 0);
739 bdev->bd_invalidated = 1;
740 err = blkdev_get(bdev, FMODE_READ, NULL);
743 blkdev_put(bdev, FMODE_READ);
746 /* announce disk after possible partitions are created */
747 dev_set_uevent_suppress(ddev, 0);
748 kobject_uevent(&ddev->kobj, KOBJ_ADD);
750 /* announce possible partitions */
751 disk_part_iter_init(&piter, disk, 0);
752 while ((part = disk_part_iter_next(&piter)))
753 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
754 disk_part_iter_exit(&piter);
756 if (disk->queue->backing_dev_info->dev) {
757 err = sysfs_create_link(&ddev->kobj,
758 &disk->queue->backing_dev_info->dev->kobj,
765 * __device_add_disk - add disk information to kernel list
766 * @parent: parent device for the disk
767 * @disk: per-device partitioning information
768 * @groups: Additional per-device sysfs groups
769 * @register_queue: register the queue if set to true
771 * This function registers the partitioning information in @disk
774 * FIXME: error handling
776 static void __device_add_disk(struct device *parent, struct gendisk *disk,
777 const struct attribute_group **groups,
784 * The disk queue should now be all set with enough information about
785 * the device for the elevator code to pick an adequate default
786 * elevator if one is needed, that is, for devices requesting queue
790 elevator_init_mq(disk->queue);
792 /* minors == 0 indicates to use ext devt from part0 and should
793 * be accompanied with EXT_DEVT flag. Make sure all
794 * parameters make sense.
796 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
797 WARN_ON(!disk->minors &&
798 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
800 disk->flags |= GENHD_FL_UP;
802 retval = blk_alloc_devt(&disk->part0, &devt);
807 disk->major = MAJOR(devt);
808 disk->first_minor = MINOR(devt);
810 disk_alloc_events(disk);
812 if (disk->flags & GENHD_FL_HIDDEN) {
814 * Don't let hidden disks show up in /proc/partitions,
815 * and don't bother scanning for partitions either.
817 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
818 disk->flags |= GENHD_FL_NO_PART_SCAN;
820 struct backing_dev_info *bdi = disk->queue->backing_dev_info;
821 struct device *dev = disk_to_dev(disk);
824 /* Register BDI before referencing it from bdev */
826 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
828 bdi_set_owner(bdi, dev);
829 blk_register_region(disk_devt(disk), disk->minors, NULL,
830 exact_match, exact_lock, disk);
832 register_disk(parent, disk, groups);
834 blk_register_queue(disk);
837 * Take an extra ref on queue which will be put on disk_release()
838 * so that it sticks around as long as @disk is there.
840 WARN_ON_ONCE(!blk_get_queue(disk->queue));
842 disk_add_events(disk);
843 blk_integrity_add(disk);
846 void device_add_disk(struct device *parent, struct gendisk *disk,
847 const struct attribute_group **groups)
850 __device_add_disk(parent, disk, groups, true);
852 EXPORT_SYMBOL(device_add_disk);
854 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
856 __device_add_disk(parent, disk, NULL, false);
858 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
860 static void invalidate_partition(struct gendisk *disk, int partno)
862 struct block_device *bdev;
864 bdev = bdget_disk(disk, partno);
869 __invalidate_device(bdev, true);
872 * Unhash the bdev inode for this device so that it gets evicted as soon
873 * as last inode reference is dropped.
875 remove_inode_hash(bdev->bd_inode);
879 void del_gendisk(struct gendisk *disk)
881 struct disk_part_iter piter;
882 struct hd_struct *part;
884 blk_integrity_del(disk);
885 disk_del_events(disk);
888 * Block lookups of the disk until all bdevs are unhashed and the
889 * disk is marked as dead (GENHD_FL_UP cleared).
891 down_write(&disk->lookup_sem);
892 /* invalidate stuff */
893 disk_part_iter_init(&piter, disk,
894 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
895 while ((part = disk_part_iter_next(&piter))) {
896 invalidate_partition(disk, part->partno);
897 delete_partition(disk, part);
899 disk_part_iter_exit(&piter);
901 invalidate_partition(disk, 0);
902 set_capacity(disk, 0);
903 disk->flags &= ~GENHD_FL_UP;
904 up_write(&disk->lookup_sem);
906 if (!(disk->flags & GENHD_FL_HIDDEN))
907 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
910 * Unregister bdi before releasing device numbers (as they can
911 * get reused and we'd get clashes in sysfs).
913 if (!(disk->flags & GENHD_FL_HIDDEN))
914 bdi_unregister(disk->queue->backing_dev_info);
915 blk_unregister_queue(disk);
920 if (!(disk->flags & GENHD_FL_HIDDEN))
921 blk_unregister_region(disk_devt(disk), disk->minors);
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);
968 * get_gendisk - get partitioning information for a given device
969 * @devt: device to get partitioning information for
970 * @partno: returned partition index
972 * This function gets the structure containing partitioning
973 * information for the given device @devt.
975 struct gendisk *get_gendisk(dev_t devt, int *partno)
977 struct gendisk *disk = NULL;
979 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
980 struct kobject *kobj;
982 kobj = kobj_lookup(bdev_map, devt, partno);
984 disk = dev_to_disk(kobj_to_dev(kobj));
986 struct hd_struct *part;
988 spin_lock_bh(&ext_devt_lock);
989 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
990 if (part && get_disk_and_module(part_to_disk(part))) {
991 *partno = part->partno;
992 disk = part_to_disk(part);
994 spin_unlock_bh(&ext_devt_lock);
1001 * Synchronize with del_gendisk() to not return disk that is being
1004 down_read(&disk->lookup_sem);
1005 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1006 !(disk->flags & GENHD_FL_UP))) {
1007 up_read(&disk->lookup_sem);
1008 put_disk_and_module(disk);
1011 up_read(&disk->lookup_sem);
1017 * bdget_disk - do bdget() by gendisk and partition number
1018 * @disk: gendisk of interest
1019 * @partno: partition number
1021 * Find partition @partno from @disk, do bdget() on it.
1027 * Resulting block_device on success, NULL on failure.
1029 struct block_device *bdget_disk(struct gendisk *disk, int partno)
1031 struct hd_struct *part;
1032 struct block_device *bdev = NULL;
1034 part = disk_get_part(disk, partno);
1036 bdev = bdget(part_devt(part));
1037 disk_put_part(part);
1041 EXPORT_SYMBOL(bdget_disk);
1044 * print a full list of all partitions - intended for places where the root
1045 * filesystem can't be mounted and thus to give the victim some idea of what
1048 void __init printk_all_partitions(void)
1050 struct class_dev_iter iter;
1053 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1054 while ((dev = class_dev_iter_next(&iter))) {
1055 struct gendisk *disk = dev_to_disk(dev);
1056 struct disk_part_iter piter;
1057 struct hd_struct *part;
1058 char name_buf[BDEVNAME_SIZE];
1059 char devt_buf[BDEVT_SIZE];
1062 * Don't show empty devices or things that have been
1065 if (get_capacity(disk) == 0 ||
1066 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1070 * Note, unlike /proc/partitions, I am showing the
1071 * numbers in hex - the same format as the root=
1074 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1075 while ((part = disk_part_iter_next(&piter))) {
1076 bool is_part0 = part == &disk->part0;
1078 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1079 bdevt_str(part_devt(part), devt_buf),
1080 (unsigned long long)part_nr_sects_read(part) >> 1
1081 , disk_name(disk, part->partno, name_buf),
1082 part->info ? part->info->uuid : "");
1084 if (dev->parent && dev->parent->driver)
1085 printk(" driver: %s\n",
1086 dev->parent->driver->name);
1088 printk(" (driver?)\n");
1092 disk_part_iter_exit(&piter);
1094 class_dev_iter_exit(&iter);
1097 #ifdef CONFIG_PROC_FS
1099 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1102 struct class_dev_iter *iter;
1105 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1107 return ERR_PTR(-ENOMEM);
1109 seqf->private = iter;
1110 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1112 dev = class_dev_iter_next(iter);
1117 return dev_to_disk(dev);
1120 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1125 dev = class_dev_iter_next(seqf->private);
1127 return dev_to_disk(dev);
1132 static void disk_seqf_stop(struct seq_file *seqf, void *v)
1134 struct class_dev_iter *iter = seqf->private;
1136 /* stop is called even after start failed :-( */
1138 class_dev_iter_exit(iter);
1140 seqf->private = NULL;
1144 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1148 p = disk_seqf_start(seqf, pos);
1149 if (!IS_ERR_OR_NULL(p) && !*pos)
1150 seq_puts(seqf, "major minor #blocks name\n\n");
1154 static int show_partition(struct seq_file *seqf, void *v)
1156 struct gendisk *sgp = v;
1157 struct disk_part_iter piter;
1158 struct hd_struct *part;
1159 char buf[BDEVNAME_SIZE];
1161 /* Don't show non-partitionable removeable devices or empty devices */
1162 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1163 (sgp->flags & GENHD_FL_REMOVABLE)))
1165 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1168 /* show the full disk and all non-0 size partitions of it */
1169 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1170 while ((part = disk_part_iter_next(&piter)))
1171 seq_printf(seqf, "%4d %7d %10llu %s\n",
1172 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1173 (unsigned long long)part_nr_sects_read(part) >> 1,
1174 disk_name(sgp, part->partno, buf));
1175 disk_part_iter_exit(&piter);
1180 static const struct seq_operations partitions_op = {
1181 .start = show_partition_start,
1182 .next = disk_seqf_next,
1183 .stop = disk_seqf_stop,
1184 .show = show_partition
1189 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1191 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1192 /* Make old-style 2.4 aliases work */
1193 request_module("block-major-%d", MAJOR(devt));
1197 static int __init genhd_device_init(void)
1201 block_class.dev_kobj = sysfs_dev_block_kobj;
1202 error = class_register(&block_class);
1203 if (unlikely(error))
1205 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1208 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1210 /* create top-level block dir */
1211 if (!sysfs_deprecated)
1212 block_depr = kobject_create_and_add("block", NULL);
1216 subsys_initcall(genhd_device_init);
1218 static ssize_t disk_range_show(struct device *dev,
1219 struct device_attribute *attr, char *buf)
1221 struct gendisk *disk = dev_to_disk(dev);
1223 return sprintf(buf, "%d\n", disk->minors);
1226 static ssize_t disk_ext_range_show(struct device *dev,
1227 struct device_attribute *attr, char *buf)
1229 struct gendisk *disk = dev_to_disk(dev);
1231 return sprintf(buf, "%d\n", disk_max_parts(disk));
1234 static ssize_t disk_removable_show(struct device *dev,
1235 struct device_attribute *attr, char *buf)
1237 struct gendisk *disk = dev_to_disk(dev);
1239 return sprintf(buf, "%d\n",
1240 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1243 static ssize_t disk_hidden_show(struct device *dev,
1244 struct device_attribute *attr, char *buf)
1246 struct gendisk *disk = dev_to_disk(dev);
1248 return sprintf(buf, "%d\n",
1249 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1252 static ssize_t disk_ro_show(struct device *dev,
1253 struct device_attribute *attr, char *buf)
1255 struct gendisk *disk = dev_to_disk(dev);
1257 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1260 ssize_t part_size_show(struct device *dev,
1261 struct device_attribute *attr, char *buf)
1263 struct hd_struct *p = dev_to_part(dev);
1265 return sprintf(buf, "%llu\n",
1266 (unsigned long long)part_nr_sects_read(p));
1269 ssize_t part_stat_show(struct device *dev,
1270 struct device_attribute *attr, char *buf)
1272 struct hd_struct *p = dev_to_part(dev);
1273 struct request_queue *q = part_to_disk(p)->queue;
1274 struct disk_stats stat;
1275 unsigned int inflight;
1277 part_stat_read_all(p, &stat);
1279 inflight = blk_mq_in_flight(q, p);
1281 inflight = part_in_flight(q, p);
1284 "%8lu %8lu %8llu %8u "
1285 "%8lu %8lu %8llu %8u "
1287 "%8lu %8lu %8llu %8u "
1290 stat.ios[STAT_READ],
1291 stat.merges[STAT_READ],
1292 (unsigned long long)stat.sectors[STAT_READ],
1293 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1294 stat.ios[STAT_WRITE],
1295 stat.merges[STAT_WRITE],
1296 (unsigned long long)stat.sectors[STAT_WRITE],
1297 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1299 jiffies_to_msecs(stat.io_ticks),
1300 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1301 stat.nsecs[STAT_WRITE] +
1302 stat.nsecs[STAT_DISCARD] +
1303 stat.nsecs[STAT_FLUSH],
1305 stat.ios[STAT_DISCARD],
1306 stat.merges[STAT_DISCARD],
1307 (unsigned long long)stat.sectors[STAT_DISCARD],
1308 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1309 stat.ios[STAT_FLUSH],
1310 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1313 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1316 struct hd_struct *p = dev_to_part(dev);
1317 struct request_queue *q = part_to_disk(p)->queue;
1318 unsigned int inflight[2];
1321 blk_mq_in_flight_rw(q, p, inflight);
1323 part_in_flight_rw(q, p, inflight);
1325 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1328 static ssize_t disk_capability_show(struct device *dev,
1329 struct device_attribute *attr, char *buf)
1331 struct gendisk *disk = dev_to_disk(dev);
1333 return sprintf(buf, "%x\n", disk->flags);
1336 static ssize_t disk_alignment_offset_show(struct device *dev,
1337 struct device_attribute *attr,
1340 struct gendisk *disk = dev_to_disk(dev);
1342 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1345 static ssize_t disk_discard_alignment_show(struct device *dev,
1346 struct device_attribute *attr,
1349 struct gendisk *disk = dev_to_disk(dev);
1351 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1354 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1355 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1356 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1357 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1358 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1359 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1360 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1361 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1362 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1363 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1364 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1365 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1367 #ifdef CONFIG_FAIL_MAKE_REQUEST
1368 ssize_t part_fail_show(struct device *dev,
1369 struct device_attribute *attr, char *buf)
1371 struct hd_struct *p = dev_to_part(dev);
1373 return sprintf(buf, "%d\n", p->make_it_fail);
1376 ssize_t part_fail_store(struct device *dev,
1377 struct device_attribute *attr,
1378 const char *buf, size_t count)
1380 struct hd_struct *p = dev_to_part(dev);
1383 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1384 p->make_it_fail = (i == 0) ? 0 : 1;
1389 static struct device_attribute dev_attr_fail =
1390 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1391 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1393 #ifdef CONFIG_FAIL_IO_TIMEOUT
1394 static struct device_attribute dev_attr_fail_timeout =
1395 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1398 static struct attribute *disk_attrs[] = {
1399 &dev_attr_range.attr,
1400 &dev_attr_ext_range.attr,
1401 &dev_attr_removable.attr,
1402 &dev_attr_hidden.attr,
1404 &dev_attr_size.attr,
1405 &dev_attr_alignment_offset.attr,
1406 &dev_attr_discard_alignment.attr,
1407 &dev_attr_capability.attr,
1408 &dev_attr_stat.attr,
1409 &dev_attr_inflight.attr,
1410 &dev_attr_badblocks.attr,
1411 #ifdef CONFIG_FAIL_MAKE_REQUEST
1412 &dev_attr_fail.attr,
1414 #ifdef CONFIG_FAIL_IO_TIMEOUT
1415 &dev_attr_fail_timeout.attr,
1420 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1422 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1423 struct gendisk *disk = dev_to_disk(dev);
1425 if (a == &dev_attr_badblocks.attr && !disk->bb)
1430 static struct attribute_group disk_attr_group = {
1431 .attrs = disk_attrs,
1432 .is_visible = disk_visible,
1435 static const struct attribute_group *disk_attr_groups[] = {
1441 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1442 * @disk: disk to replace part_tbl for
1443 * @new_ptbl: new part_tbl to install
1445 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1446 * original ptbl is freed using RCU callback.
1449 * Matching bd_mutex locked or the caller is the only user of @disk.
1451 static void disk_replace_part_tbl(struct gendisk *disk,
1452 struct disk_part_tbl *new_ptbl)
1454 struct disk_part_tbl *old_ptbl =
1455 rcu_dereference_protected(disk->part_tbl, 1);
1457 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1460 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1461 kfree_rcu(old_ptbl, rcu_head);
1466 * disk_expand_part_tbl - expand disk->part_tbl
1467 * @disk: disk to expand part_tbl for
1468 * @partno: expand such that this partno can fit in
1470 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1471 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1474 * Matching bd_mutex locked or the caller is the only user of @disk.
1478 * 0 on success, -errno on failure.
1480 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1482 struct disk_part_tbl *old_ptbl =
1483 rcu_dereference_protected(disk->part_tbl, 1);
1484 struct disk_part_tbl *new_ptbl;
1485 int len = old_ptbl ? old_ptbl->len : 0;
1489 * check for int overflow, since we can get here from blkpg_ioctl()
1490 * with a user passed 'partno'.
1492 target = partno + 1;
1496 /* disk_max_parts() is zero during initialization, ignore if so */
1497 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1503 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1508 new_ptbl->len = target;
1510 for (i = 0; i < len; i++)
1511 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1513 disk_replace_part_tbl(disk, new_ptbl);
1517 static void disk_release(struct device *dev)
1519 struct gendisk *disk = dev_to_disk(dev);
1521 blk_free_devt(dev->devt);
1522 disk_release_events(disk);
1523 kfree(disk->random);
1524 disk_replace_part_tbl(disk, NULL);
1525 hd_free_part(&disk->part0);
1527 blk_put_queue(disk->queue);
1530 struct class block_class = {
1534 static char *block_devnode(struct device *dev, umode_t *mode,
1535 kuid_t *uid, kgid_t *gid)
1537 struct gendisk *disk = dev_to_disk(dev);
1539 if (disk->fops->devnode)
1540 return disk->fops->devnode(disk, mode);
1544 static const struct device_type disk_type = {
1546 .groups = disk_attr_groups,
1547 .release = disk_release,
1548 .devnode = block_devnode,
1551 #ifdef CONFIG_PROC_FS
1553 * aggregate disk stat collector. Uses the same stats that the sysfs
1554 * entries do, above, but makes them available through one seq_file.
1556 * The output looks suspiciously like /proc/partitions with a bunch of
1559 static int diskstats_show(struct seq_file *seqf, void *v)
1561 struct gendisk *gp = v;
1562 struct disk_part_iter piter;
1563 struct hd_struct *hd;
1564 char buf[BDEVNAME_SIZE];
1565 unsigned int inflight;
1566 struct disk_stats stat;
1569 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1570 seq_puts(seqf, "major minor name"
1571 " rio rmerge rsect ruse wio wmerge "
1572 "wsect wuse running use aveq"
1576 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1577 while ((hd = disk_part_iter_next(&piter))) {
1578 part_stat_read_all(hd, &stat);
1579 if (queue_is_mq(gp->queue))
1580 inflight = blk_mq_in_flight(gp->queue, hd);
1582 inflight = part_in_flight(gp->queue, hd);
1584 seq_printf(seqf, "%4d %7d %s "
1591 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1592 disk_name(gp, hd->partno, buf),
1593 stat.ios[STAT_READ],
1594 stat.merges[STAT_READ],
1595 stat.sectors[STAT_READ],
1596 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1598 stat.ios[STAT_WRITE],
1599 stat.merges[STAT_WRITE],
1600 stat.sectors[STAT_WRITE],
1601 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1604 jiffies_to_msecs(stat.io_ticks),
1605 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1606 stat.nsecs[STAT_WRITE] +
1607 stat.nsecs[STAT_DISCARD] +
1608 stat.nsecs[STAT_FLUSH],
1610 stat.ios[STAT_DISCARD],
1611 stat.merges[STAT_DISCARD],
1612 stat.sectors[STAT_DISCARD],
1613 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1615 stat.ios[STAT_FLUSH],
1616 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1620 disk_part_iter_exit(&piter);
1625 static const struct seq_operations diskstats_op = {
1626 .start = disk_seqf_start,
1627 .next = disk_seqf_next,
1628 .stop = disk_seqf_stop,
1629 .show = diskstats_show
1632 static int __init proc_genhd_init(void)
1634 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1635 proc_create_seq("partitions", 0, NULL, &partitions_op);
1638 module_init(proc_genhd_init);
1639 #endif /* CONFIG_PROC_FS */
1641 dev_t blk_lookup_devt(const char *name, int partno)
1643 dev_t devt = MKDEV(0, 0);
1644 struct class_dev_iter iter;
1647 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1648 while ((dev = class_dev_iter_next(&iter))) {
1649 struct gendisk *disk = dev_to_disk(dev);
1650 struct hd_struct *part;
1652 if (strcmp(dev_name(dev), name))
1655 if (partno < disk->minors) {
1656 /* We need to return the right devno, even
1657 * if the partition doesn't exist yet.
1659 devt = MKDEV(MAJOR(dev->devt),
1660 MINOR(dev->devt) + partno);
1663 part = disk_get_part(disk, partno);
1665 devt = part_devt(part);
1666 disk_put_part(part);
1669 disk_put_part(part);
1671 class_dev_iter_exit(&iter);
1675 struct gendisk *__alloc_disk_node(int minors, int node_id)
1677 struct gendisk *disk;
1678 struct disk_part_tbl *ptbl;
1680 if (minors > DISK_MAX_PARTS) {
1682 "block: can't allocate more than %d partitions\n",
1684 minors = DISK_MAX_PARTS;
1687 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1689 disk->part0.dkstats = alloc_percpu(struct disk_stats);
1690 if (!disk->part0.dkstats) {
1694 init_rwsem(&disk->lookup_sem);
1695 disk->node_id = node_id;
1696 if (disk_expand_part_tbl(disk, 0)) {
1697 free_percpu(disk->part0.dkstats);
1701 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1702 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1705 * set_capacity() and get_capacity() currently don't use
1706 * seqcounter to read/update the part0->nr_sects. Still init
1707 * the counter as we can read the sectors in IO submission
1708 * patch using seqence counters.
1710 * TODO: Ideally set_capacity() and get_capacity() should be
1711 * converted to make use of bd_mutex and sequence counters.
1713 hd_sects_seq_init(&disk->part0);
1714 if (hd_ref_init(&disk->part0)) {
1715 hd_free_part(&disk->part0);
1720 disk->minors = minors;
1721 rand_initialize_disk(disk);
1722 disk_to_dev(disk)->class = &block_class;
1723 disk_to_dev(disk)->type = &disk_type;
1724 device_initialize(disk_to_dev(disk));
1728 EXPORT_SYMBOL(__alloc_disk_node);
1730 struct kobject *get_disk_and_module(struct gendisk *disk)
1732 struct module *owner;
1733 struct kobject *kobj;
1737 owner = disk->fops->owner;
1738 if (owner && !try_module_get(owner))
1740 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1748 EXPORT_SYMBOL(get_disk_and_module);
1750 void put_disk(struct gendisk *disk)
1753 kobject_put(&disk_to_dev(disk)->kobj);
1755 EXPORT_SYMBOL(put_disk);
1758 * This is a counterpart of get_disk_and_module() and thus also of
1761 void put_disk_and_module(struct gendisk *disk)
1764 struct module *owner = disk->fops->owner;
1770 EXPORT_SYMBOL(put_disk_and_module);
1772 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1774 char event[] = "DISK_RO=1";
1775 char *envp[] = { event, NULL };
1779 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1782 void set_device_ro(struct block_device *bdev, int flag)
1784 bdev->bd_part->policy = flag;
1787 EXPORT_SYMBOL(set_device_ro);
1789 void set_disk_ro(struct gendisk *disk, int flag)
1791 struct disk_part_iter piter;
1792 struct hd_struct *part;
1794 if (disk->part0.policy != flag) {
1795 set_disk_ro_uevent(disk, flag);
1796 disk->part0.policy = flag;
1799 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1800 while ((part = disk_part_iter_next(&piter)))
1801 part->policy = flag;
1802 disk_part_iter_exit(&piter);
1805 EXPORT_SYMBOL(set_disk_ro);
1807 int bdev_read_only(struct block_device *bdev)
1811 return bdev->bd_part->policy;
1814 EXPORT_SYMBOL(bdev_read_only);
1817 * Disk events - monitor disk events like media change and eject request.
1819 struct disk_events {
1820 struct list_head node; /* all disk_event's */
1821 struct gendisk *disk; /* the associated disk */
1824 struct mutex block_mutex; /* protects blocking */
1825 int block; /* event blocking depth */
1826 unsigned int pending; /* events already sent out */
1827 unsigned int clearing; /* events being cleared */
1829 long poll_msecs; /* interval, -1 for default */
1830 struct delayed_work dwork;
1833 static const char *disk_events_strs[] = {
1834 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1835 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1838 static char *disk_uevents[] = {
1839 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1840 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1843 /* list of all disk_events */
1844 static DEFINE_MUTEX(disk_events_mutex);
1845 static LIST_HEAD(disk_events);
1847 /* disable in-kernel polling by default */
1848 static unsigned long disk_events_dfl_poll_msecs;
1850 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1852 struct disk_events *ev = disk->ev;
1853 long intv_msecs = 0;
1856 * If device-specific poll interval is set, always use it. If
1857 * the default is being used, poll if the POLL flag is set.
1859 if (ev->poll_msecs >= 0)
1860 intv_msecs = ev->poll_msecs;
1861 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1862 intv_msecs = disk_events_dfl_poll_msecs;
1864 return msecs_to_jiffies(intv_msecs);
1868 * disk_block_events - block and flush disk event checking
1869 * @disk: disk to block events for
1871 * On return from this function, it is guaranteed that event checking
1872 * isn't in progress and won't happen until unblocked by
1873 * disk_unblock_events(). Events blocking is counted and the actual
1874 * unblocking happens after the matching number of unblocks are done.
1876 * Note that this intentionally does not block event checking from
1877 * disk_clear_events().
1882 void disk_block_events(struct gendisk *disk)
1884 struct disk_events *ev = disk->ev;
1885 unsigned long flags;
1892 * Outer mutex ensures that the first blocker completes canceling
1893 * the event work before further blockers are allowed to finish.
1895 mutex_lock(&ev->block_mutex);
1897 spin_lock_irqsave(&ev->lock, flags);
1898 cancel = !ev->block++;
1899 spin_unlock_irqrestore(&ev->lock, flags);
1902 cancel_delayed_work_sync(&disk->ev->dwork);
1904 mutex_unlock(&ev->block_mutex);
1907 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1909 struct disk_events *ev = disk->ev;
1911 unsigned long flags;
1913 spin_lock_irqsave(&ev->lock, flags);
1915 if (WARN_ON_ONCE(ev->block <= 0))
1921 intv = disk_events_poll_jiffies(disk);
1923 queue_delayed_work(system_freezable_power_efficient_wq,
1926 queue_delayed_work(system_freezable_power_efficient_wq,
1929 spin_unlock_irqrestore(&ev->lock, flags);
1933 * disk_unblock_events - unblock disk event checking
1934 * @disk: disk to unblock events for
1936 * Undo disk_block_events(). When the block count reaches zero, it
1937 * starts events polling if configured.
1940 * Don't care. Safe to call from irq context.
1942 void disk_unblock_events(struct gendisk *disk)
1945 __disk_unblock_events(disk, false);
1949 * disk_flush_events - schedule immediate event checking and flushing
1950 * @disk: disk to check and flush events for
1951 * @mask: events to flush
1953 * Schedule immediate event checking on @disk if not blocked. Events in
1954 * @mask are scheduled to be cleared from the driver. Note that this
1955 * doesn't clear the events from @disk->ev.
1958 * If @mask is non-zero must be called with bdev->bd_mutex held.
1960 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1962 struct disk_events *ev = disk->ev;
1967 spin_lock_irq(&ev->lock);
1968 ev->clearing |= mask;
1970 mod_delayed_work(system_freezable_power_efficient_wq,
1972 spin_unlock_irq(&ev->lock);
1976 * disk_clear_events - synchronously check, clear and return pending events
1977 * @disk: disk to fetch and clear events from
1978 * @mask: mask of events to be fetched and cleared
1980 * Disk events are synchronously checked and pending events in @mask
1981 * are cleared and returned. This ignores the block count.
1986 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1988 const struct block_device_operations *bdops = disk->fops;
1989 struct disk_events *ev = disk->ev;
1990 unsigned int pending;
1991 unsigned int clearing = mask;
1994 /* for drivers still using the old ->media_changed method */
1995 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1996 bdops->media_changed && bdops->media_changed(disk))
1997 return DISK_EVENT_MEDIA_CHANGE;
2001 disk_block_events(disk);
2004 * store the union of mask and ev->clearing on the stack so that the
2005 * race with disk_flush_events does not cause ambiguity (ev->clearing
2006 * can still be modified even if events are blocked).
2008 spin_lock_irq(&ev->lock);
2009 clearing |= ev->clearing;
2011 spin_unlock_irq(&ev->lock);
2013 disk_check_events(ev, &clearing);
2015 * if ev->clearing is not 0, the disk_flush_events got called in the
2016 * middle of this function, so we want to run the workfn without delay.
2018 __disk_unblock_events(disk, ev->clearing ? true : false);
2020 /* then, fetch and clear pending events */
2021 spin_lock_irq(&ev->lock);
2022 pending = ev->pending & mask;
2023 ev->pending &= ~mask;
2024 spin_unlock_irq(&ev->lock);
2025 WARN_ON_ONCE(clearing & mask);
2031 * Separate this part out so that a different pointer for clearing_ptr can be
2032 * passed in for disk_clear_events.
2034 static void disk_events_workfn(struct work_struct *work)
2036 struct delayed_work *dwork = to_delayed_work(work);
2037 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2039 disk_check_events(ev, &ev->clearing);
2042 static void disk_check_events(struct disk_events *ev,
2043 unsigned int *clearing_ptr)
2045 struct gendisk *disk = ev->disk;
2046 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2047 unsigned int clearing = *clearing_ptr;
2048 unsigned int events;
2050 int nr_events = 0, i;
2053 events = disk->fops->check_events(disk, clearing);
2055 /* accumulate pending events and schedule next poll if necessary */
2056 spin_lock_irq(&ev->lock);
2058 events &= ~ev->pending;
2059 ev->pending |= events;
2060 *clearing_ptr &= ~clearing;
2062 intv = disk_events_poll_jiffies(disk);
2063 if (!ev->block && intv)
2064 queue_delayed_work(system_freezable_power_efficient_wq,
2067 spin_unlock_irq(&ev->lock);
2070 * Tell userland about new events. Only the events listed in
2071 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2072 * is set. Otherwise, events are processed internally but never
2073 * get reported to userland.
2075 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2076 if ((events & disk->events & (1 << i)) &&
2077 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2078 envp[nr_events++] = disk_uevents[i];
2081 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2085 * A disk events enabled device has the following sysfs nodes under
2086 * its /sys/block/X/ directory.
2088 * events : list of all supported events
2089 * events_async : list of events which can be detected w/o polling
2090 * (always empty, only for backwards compatibility)
2091 * events_poll_msecs : polling interval, 0: disable, -1: system default
2093 static ssize_t __disk_events_show(unsigned int events, char *buf)
2095 const char *delim = "";
2099 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2100 if (events & (1 << i)) {
2101 pos += sprintf(buf + pos, "%s%s",
2102 delim, disk_events_strs[i]);
2106 pos += sprintf(buf + pos, "\n");
2110 static ssize_t disk_events_show(struct device *dev,
2111 struct device_attribute *attr, char *buf)
2113 struct gendisk *disk = dev_to_disk(dev);
2115 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2118 return __disk_events_show(disk->events, buf);
2121 static ssize_t disk_events_async_show(struct device *dev,
2122 struct device_attribute *attr, char *buf)
2127 static ssize_t disk_events_poll_msecs_show(struct device *dev,
2128 struct device_attribute *attr,
2131 struct gendisk *disk = dev_to_disk(dev);
2134 return sprintf(buf, "-1\n");
2136 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2139 static ssize_t disk_events_poll_msecs_store(struct device *dev,
2140 struct device_attribute *attr,
2141 const char *buf, size_t count)
2143 struct gendisk *disk = dev_to_disk(dev);
2146 if (!count || !sscanf(buf, "%ld", &intv))
2149 if (intv < 0 && intv != -1)
2155 disk_block_events(disk);
2156 disk->ev->poll_msecs = intv;
2157 __disk_unblock_events(disk, true);
2162 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2163 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2164 static const DEVICE_ATTR(events_poll_msecs, 0644,
2165 disk_events_poll_msecs_show,
2166 disk_events_poll_msecs_store);
2168 static const struct attribute *disk_events_attrs[] = {
2169 &dev_attr_events.attr,
2170 &dev_attr_events_async.attr,
2171 &dev_attr_events_poll_msecs.attr,
2176 * The default polling interval can be specified by the kernel
2177 * parameter block.events_dfl_poll_msecs which defaults to 0
2178 * (disable). This can also be modified runtime by writing to
2179 * /sys/module/block/parameters/events_dfl_poll_msecs.
2181 static int disk_events_set_dfl_poll_msecs(const char *val,
2182 const struct kernel_param *kp)
2184 struct disk_events *ev;
2187 ret = param_set_ulong(val, kp);
2191 mutex_lock(&disk_events_mutex);
2193 list_for_each_entry(ev, &disk_events, node)
2194 disk_flush_events(ev->disk, 0);
2196 mutex_unlock(&disk_events_mutex);
2201 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2202 .set = disk_events_set_dfl_poll_msecs,
2203 .get = param_get_ulong,
2206 #undef MODULE_PARAM_PREFIX
2207 #define MODULE_PARAM_PREFIX "block."
2209 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2210 &disk_events_dfl_poll_msecs, 0644);
2213 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2215 static void disk_alloc_events(struct gendisk *disk)
2217 struct disk_events *ev;
2219 if (!disk->fops->check_events || !disk->events)
2222 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2224 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2228 INIT_LIST_HEAD(&ev->node);
2230 spin_lock_init(&ev->lock);
2231 mutex_init(&ev->block_mutex);
2233 ev->poll_msecs = -1;
2234 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2239 static void disk_add_events(struct gendisk *disk)
2241 /* FIXME: error handling */
2242 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2243 pr_warn("%s: failed to create sysfs files for events\n",
2249 mutex_lock(&disk_events_mutex);
2250 list_add_tail(&disk->ev->node, &disk_events);
2251 mutex_unlock(&disk_events_mutex);
2254 * Block count is initialized to 1 and the following initial
2255 * unblock kicks it into action.
2257 __disk_unblock_events(disk, true);
2260 static void disk_del_events(struct gendisk *disk)
2263 disk_block_events(disk);
2265 mutex_lock(&disk_events_mutex);
2266 list_del_init(&disk->ev->node);
2267 mutex_unlock(&disk_events_mutex);
2270 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2273 static void disk_release_events(struct gendisk *disk)
2275 /* the block count should be 1 from disk_del_events() */
2276 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);