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);
96 static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
100 memset(stat, 0, sizeof(struct disk_stats));
101 for_each_possible_cpu(cpu) {
102 struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
105 for (group = 0; group < NR_STAT_GROUPS; group++) {
106 stat->nsecs[group] += ptr->nsecs[group];
107 stat->sectors[group] += ptr->sectors[group];
108 stat->ios[group] += ptr->ios[group];
109 stat->merges[group] += ptr->merges[group];
112 stat->io_ticks += ptr->io_ticks;
115 #else /* CONFIG_SMP */
116 static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
118 memcpy(stat, &part->dkstats, sizeof(struct disk_stats));
120 #endif /* CONFIG_SMP */
122 void part_inc_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
127 part_stat_local_inc(part, in_flight[rw]);
129 part_stat_local_inc(&part_to_disk(part)->part0, in_flight[rw]);
132 void part_dec_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
137 part_stat_local_dec(part, in_flight[rw]);
139 part_stat_local_dec(&part_to_disk(part)->part0, in_flight[rw]);
142 static unsigned int part_in_flight(struct request_queue *q,
143 struct hd_struct *part)
146 unsigned int inflight;
148 if (queue_is_mq(q)) {
149 return blk_mq_in_flight(q, part);
153 for_each_possible_cpu(cpu) {
154 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
155 part_stat_local_read_cpu(part, in_flight[1], cpu);
157 if ((int)inflight < 0)
163 static void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
164 unsigned int inflight[2])
168 if (queue_is_mq(q)) {
169 blk_mq_in_flight_rw(q, part, inflight);
175 for_each_possible_cpu(cpu) {
176 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
177 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
179 if ((int)inflight[0] < 0)
181 if ((int)inflight[1] < 0)
185 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
187 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
189 if (unlikely(partno < 0 || partno >= ptbl->len))
191 return rcu_dereference(ptbl->part[partno]);
195 * disk_get_part - get partition
196 * @disk: disk to look partition from
197 * @partno: partition number
199 * Look for partition @partno from @disk. If found, increment
200 * reference count and return it.
206 * Pointer to the found partition on success, NULL if not found.
208 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
210 struct hd_struct *part;
213 part = __disk_get_part(disk, partno);
215 get_device(part_to_dev(part));
222 * disk_part_iter_init - initialize partition iterator
223 * @piter: iterator to initialize
224 * @disk: disk to iterate over
225 * @flags: DISK_PITER_* flags
227 * Initialize @piter so that it iterates over partitions of @disk.
232 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
235 struct disk_part_tbl *ptbl;
238 ptbl = rcu_dereference(disk->part_tbl);
243 if (flags & DISK_PITER_REVERSE)
244 piter->idx = ptbl->len - 1;
245 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
250 piter->flags = flags;
254 EXPORT_SYMBOL_GPL(disk_part_iter_init);
257 * disk_part_iter_next - proceed iterator to the next partition and return it
258 * @piter: iterator of interest
260 * Proceed @piter to the next partition and return it.
265 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
267 struct disk_part_tbl *ptbl;
270 /* put the last partition */
271 disk_put_part(piter->part);
276 ptbl = rcu_dereference(piter->disk->part_tbl);
278 /* determine iteration parameters */
279 if (piter->flags & DISK_PITER_REVERSE) {
281 if (piter->flags & (DISK_PITER_INCL_PART0 |
282 DISK_PITER_INCL_EMPTY_PART0))
291 /* iterate to the next partition */
292 for (; piter->idx != end; piter->idx += inc) {
293 struct hd_struct *part;
295 part = rcu_dereference(ptbl->part[piter->idx]);
298 if (!part_nr_sects_read(part) &&
299 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
300 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
304 get_device(part_to_dev(part));
314 EXPORT_SYMBOL_GPL(disk_part_iter_next);
317 * disk_part_iter_exit - finish up partition iteration
318 * @piter: iter of interest
320 * Called when iteration is over. Cleans up @piter.
325 void disk_part_iter_exit(struct disk_part_iter *piter)
327 disk_put_part(piter->part);
330 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
332 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
334 return part->start_sect <= sector &&
335 sector < part->start_sect + part_nr_sects_read(part);
339 * disk_map_sector_rcu - map sector to partition
340 * @disk: gendisk of interest
341 * @sector: sector to map
343 * Find out which partition @sector maps to on @disk. This is
344 * primarily used for stats accounting.
347 * RCU read locked. The returned partition pointer is always valid
348 * because its refcount is grabbed.
351 * Found partition on success, part0 is returned if no partition matches
352 * or the matched partition is being deleted.
354 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
356 struct disk_part_tbl *ptbl;
357 struct hd_struct *part;
360 ptbl = rcu_dereference(disk->part_tbl);
362 part = rcu_dereference(ptbl->last_lookup);
363 if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
366 for (i = 1; i < ptbl->len; i++) {
367 part = rcu_dereference(ptbl->part[i]);
369 if (part && sector_in_part(part, sector)) {
371 * only live partition can be cached for lookup,
372 * so use-after-free on cached & deleting partition
375 if (!hd_struct_try_get(part))
377 rcu_assign_pointer(ptbl->last_lookup, part);
381 hd_struct_get(&disk->part0);
386 * disk_has_partitions
387 * @disk: gendisk of interest
389 * Walk through the partition table and check if valid partition exists.
395 * True if the gendisk has at least one valid non-zero size partition.
398 bool disk_has_partitions(struct gendisk *disk)
400 struct disk_part_tbl *ptbl;
405 ptbl = rcu_dereference(disk->part_tbl);
407 /* Iterate partitions skipping the whole device at index 0 */
408 for (i = 1; i < ptbl->len; i++) {
409 if (rcu_dereference(ptbl->part[i])) {
419 EXPORT_SYMBOL_GPL(disk_has_partitions);
422 * Can be deleted altogether. Later.
425 #define BLKDEV_MAJOR_HASH_SIZE 255
426 static struct blk_major_name {
427 struct blk_major_name *next;
430 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
432 /* index in the above - for now: assume no multimajor ranges */
433 static inline int major_to_index(unsigned major)
435 return major % BLKDEV_MAJOR_HASH_SIZE;
438 #ifdef CONFIG_PROC_FS
439 void blkdev_show(struct seq_file *seqf, off_t offset)
441 struct blk_major_name *dp;
443 mutex_lock(&block_class_lock);
444 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
445 if (dp->major == offset)
446 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
447 mutex_unlock(&block_class_lock);
449 #endif /* CONFIG_PROC_FS */
452 * register_blkdev - register a new block device
454 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
455 * @major = 0, try to allocate any unused major number.
456 * @name: the name of the new block device as a zero terminated string
458 * The @name must be unique within the system.
460 * The return value depends on the @major input parameter:
462 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
463 * then the function returns zero on success, or a negative error code
464 * - if any unused major number was requested with @major = 0 parameter
465 * then the return value is the allocated major number in range
466 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
468 * See Documentation/admin-guide/devices.txt for the list of allocated
471 int register_blkdev(unsigned int major, const char *name)
473 struct blk_major_name **n, *p;
476 mutex_lock(&block_class_lock);
480 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
481 if (major_names[index] == NULL)
486 printk("%s: failed to get major for %s\n",
495 if (major >= BLKDEV_MAJOR_MAX) {
496 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
497 __func__, major, BLKDEV_MAJOR_MAX-1, name);
503 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
510 strlcpy(p->name, name, sizeof(p->name));
512 index = major_to_index(major);
514 for (n = &major_names[index]; *n; n = &(*n)->next) {
515 if ((*n)->major == major)
524 printk("register_blkdev: cannot get major %u for %s\n",
529 mutex_unlock(&block_class_lock);
533 EXPORT_SYMBOL(register_blkdev);
535 void unregister_blkdev(unsigned int major, const char *name)
537 struct blk_major_name **n;
538 struct blk_major_name *p = NULL;
539 int index = major_to_index(major);
541 mutex_lock(&block_class_lock);
542 for (n = &major_names[index]; *n; n = &(*n)->next)
543 if ((*n)->major == major)
545 if (!*n || strcmp((*n)->name, name)) {
551 mutex_unlock(&block_class_lock);
555 EXPORT_SYMBOL(unregister_blkdev);
557 static struct kobj_map *bdev_map;
560 * blk_mangle_minor - scatter minor numbers apart
561 * @minor: minor number to mangle
563 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
564 * is enabled. Mangling twice gives the original value.
572 static int blk_mangle_minor(int minor)
574 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
577 for (i = 0; i < MINORBITS / 2; i++) {
578 int low = minor & (1 << i);
579 int high = minor & (1 << (MINORBITS - 1 - i));
580 int distance = MINORBITS - 1 - 2 * i;
582 minor ^= low | high; /* clear both bits */
583 low <<= distance; /* swap the positions */
585 minor |= low | high; /* and set */
592 * blk_alloc_devt - allocate a dev_t for a partition
593 * @part: partition to allocate dev_t for
594 * @devt: out parameter for resulting dev_t
596 * Allocate a dev_t for block device.
599 * 0 on success, allocated dev_t is returned in *@devt. -errno on
605 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
607 struct gendisk *disk = part_to_disk(part);
610 /* in consecutive minor range? */
611 if (part->partno < disk->minors) {
612 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
616 /* allocate ext devt */
617 idr_preload(GFP_KERNEL);
619 spin_lock_bh(&ext_devt_lock);
620 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
621 spin_unlock_bh(&ext_devt_lock);
625 return idx == -ENOSPC ? -EBUSY : idx;
627 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
632 * blk_free_devt - free a dev_t
633 * @devt: dev_t to free
635 * Free @devt which was allocated using blk_alloc_devt().
640 void blk_free_devt(dev_t devt)
642 if (devt == MKDEV(0, 0))
645 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
646 spin_lock_bh(&ext_devt_lock);
647 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
648 spin_unlock_bh(&ext_devt_lock);
653 * We invalidate devt by assigning NULL pointer for devt in idr.
655 void blk_invalidate_devt(dev_t devt)
657 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
658 spin_lock_bh(&ext_devt_lock);
659 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
660 spin_unlock_bh(&ext_devt_lock);
664 static char *bdevt_str(dev_t devt, char *buf)
666 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
667 char tbuf[BDEVT_SIZE];
668 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
669 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
671 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
677 * Register device numbers dev..(dev+range-1)
678 * range must be nonzero
679 * The hash chain is sorted on range, so that subranges can override.
681 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
682 struct kobject *(*probe)(dev_t, int *, void *),
683 int (*lock)(dev_t, void *), void *data)
685 kobj_map(bdev_map, devt, range, module, probe, lock, data);
688 EXPORT_SYMBOL(blk_register_region);
690 void blk_unregister_region(dev_t devt, unsigned long range)
692 kobj_unmap(bdev_map, devt, range);
695 EXPORT_SYMBOL(blk_unregister_region);
697 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
699 struct gendisk *p = data;
701 return &disk_to_dev(p)->kobj;
704 static int exact_lock(dev_t devt, void *data)
706 struct gendisk *p = data;
708 if (!get_disk_and_module(p))
713 static void register_disk(struct device *parent, struct gendisk *disk,
714 const struct attribute_group **groups)
716 struct device *ddev = disk_to_dev(disk);
717 struct block_device *bdev;
718 struct disk_part_iter piter;
719 struct hd_struct *part;
722 ddev->parent = parent;
724 dev_set_name(ddev, "%s", disk->disk_name);
726 /* delay uevents, until we scanned partition table */
727 dev_set_uevent_suppress(ddev, 1);
730 WARN_ON(ddev->groups);
731 ddev->groups = groups;
733 if (device_add(ddev))
735 if (!sysfs_deprecated) {
736 err = sysfs_create_link(block_depr, &ddev->kobj,
737 kobject_name(&ddev->kobj));
745 * avoid probable deadlock caused by allocating memory with
746 * GFP_KERNEL in runtime_resume callback of its all ancestor
749 pm_runtime_set_memalloc_noio(ddev, true);
751 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
752 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
754 if (disk->flags & GENHD_FL_HIDDEN) {
755 dev_set_uevent_suppress(ddev, 0);
759 /* No minors to use for partitions */
760 if (!disk_part_scan_enabled(disk))
763 /* No such device (e.g., media were just removed) */
764 if (!get_capacity(disk))
767 bdev = bdget_disk(disk, 0);
771 bdev->bd_invalidated = 1;
772 err = blkdev_get(bdev, FMODE_READ, NULL);
775 blkdev_put(bdev, FMODE_READ);
778 /* announce disk after possible partitions are created */
779 dev_set_uevent_suppress(ddev, 0);
780 kobject_uevent(&ddev->kobj, KOBJ_ADD);
782 /* announce possible partitions */
783 disk_part_iter_init(&piter, disk, 0);
784 while ((part = disk_part_iter_next(&piter)))
785 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
786 disk_part_iter_exit(&piter);
788 if (disk->queue->backing_dev_info->dev) {
789 err = sysfs_create_link(&ddev->kobj,
790 &disk->queue->backing_dev_info->dev->kobj,
797 * __device_add_disk - add disk information to kernel list
798 * @parent: parent device for the disk
799 * @disk: per-device partitioning information
800 * @groups: Additional per-device sysfs groups
801 * @register_queue: register the queue if set to true
803 * This function registers the partitioning information in @disk
806 * FIXME: error handling
808 static void __device_add_disk(struct device *parent, struct gendisk *disk,
809 const struct attribute_group **groups,
816 * The disk queue should now be all set with enough information about
817 * the device for the elevator code to pick an adequate default
818 * elevator if one is needed, that is, for devices requesting queue
822 elevator_init_mq(disk->queue);
824 /* minors == 0 indicates to use ext devt from part0 and should
825 * be accompanied with EXT_DEVT flag. Make sure all
826 * parameters make sense.
828 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
829 WARN_ON(!disk->minors &&
830 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
832 disk->flags |= GENHD_FL_UP;
834 retval = blk_alloc_devt(&disk->part0, &devt);
839 disk->major = MAJOR(devt);
840 disk->first_minor = MINOR(devt);
842 disk_alloc_events(disk);
844 if (disk->flags & GENHD_FL_HIDDEN) {
846 * Don't let hidden disks show up in /proc/partitions,
847 * and don't bother scanning for partitions either.
849 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
850 disk->flags |= GENHD_FL_NO_PART_SCAN;
852 struct backing_dev_info *bdi = disk->queue->backing_dev_info;
853 struct device *dev = disk_to_dev(disk);
856 /* Register BDI before referencing it from bdev */
858 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
860 bdi_set_owner(bdi, dev);
861 blk_register_region(disk_devt(disk), disk->minors, NULL,
862 exact_match, exact_lock, disk);
864 register_disk(parent, disk, groups);
866 blk_register_queue(disk);
869 * Take an extra ref on queue which will be put on disk_release()
870 * so that it sticks around as long as @disk is there.
872 WARN_ON_ONCE(!blk_get_queue(disk->queue));
874 disk_add_events(disk);
875 blk_integrity_add(disk);
878 void device_add_disk(struct device *parent, struct gendisk *disk,
879 const struct attribute_group **groups)
882 __device_add_disk(parent, disk, groups, true);
884 EXPORT_SYMBOL(device_add_disk);
886 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
888 __device_add_disk(parent, disk, NULL, false);
890 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
892 static void invalidate_partition(struct gendisk *disk, int partno)
894 struct block_device *bdev;
896 bdev = bdget_disk(disk, partno);
901 __invalidate_device(bdev, true);
904 * Unhash the bdev inode for this device so that it gets evicted as soon
905 * as last inode reference is dropped.
907 remove_inode_hash(bdev->bd_inode);
911 void del_gendisk(struct gendisk *disk)
913 struct disk_part_iter piter;
914 struct hd_struct *part;
916 blk_integrity_del(disk);
917 disk_del_events(disk);
920 * Block lookups of the disk until all bdevs are unhashed and the
921 * disk is marked as dead (GENHD_FL_UP cleared).
923 down_write(&disk->lookup_sem);
924 /* invalidate stuff */
925 disk_part_iter_init(&piter, disk,
926 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
927 while ((part = disk_part_iter_next(&piter))) {
928 invalidate_partition(disk, part->partno);
929 delete_partition(disk, part);
931 disk_part_iter_exit(&piter);
933 invalidate_partition(disk, 0);
934 set_capacity(disk, 0);
935 disk->flags &= ~GENHD_FL_UP;
936 up_write(&disk->lookup_sem);
938 if (!(disk->flags & GENHD_FL_HIDDEN))
939 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
942 * Unregister bdi before releasing device numbers (as they can
943 * get reused and we'd get clashes in sysfs).
945 if (!(disk->flags & GENHD_FL_HIDDEN))
946 bdi_unregister(disk->queue->backing_dev_info);
947 blk_unregister_queue(disk);
952 if (!(disk->flags & GENHD_FL_HIDDEN))
953 blk_unregister_region(disk_devt(disk), disk->minors);
955 * Remove gendisk pointer from idr so that it cannot be looked up
956 * while RCU period before freeing gendisk is running to prevent
957 * use-after-free issues. Note that the device number stays
958 * "in-use" until we really free the gendisk.
960 blk_invalidate_devt(disk_devt(disk));
962 kobject_put(disk->part0.holder_dir);
963 kobject_put(disk->slave_dir);
965 part_stat_set_all(&disk->part0, 0);
966 disk->part0.stamp = 0;
967 if (!sysfs_deprecated)
968 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
969 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
970 device_del(disk_to_dev(disk));
972 EXPORT_SYMBOL(del_gendisk);
974 /* sysfs access to bad-blocks list. */
975 static ssize_t disk_badblocks_show(struct device *dev,
976 struct device_attribute *attr,
979 struct gendisk *disk = dev_to_disk(dev);
982 return sprintf(page, "\n");
984 return badblocks_show(disk->bb, page, 0);
987 static ssize_t disk_badblocks_store(struct device *dev,
988 struct device_attribute *attr,
989 const char *page, size_t len)
991 struct gendisk *disk = dev_to_disk(dev);
996 return badblocks_store(disk->bb, page, len, 0);
1000 * get_gendisk - get partitioning information for a given device
1001 * @devt: device to get partitioning information for
1002 * @partno: returned partition index
1004 * This function gets the structure containing partitioning
1005 * information for the given device @devt.
1007 struct gendisk *get_gendisk(dev_t devt, int *partno)
1009 struct gendisk *disk = NULL;
1011 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
1012 struct kobject *kobj;
1014 kobj = kobj_lookup(bdev_map, devt, partno);
1016 disk = dev_to_disk(kobj_to_dev(kobj));
1018 struct hd_struct *part;
1020 spin_lock_bh(&ext_devt_lock);
1021 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
1022 if (part && get_disk_and_module(part_to_disk(part))) {
1023 *partno = part->partno;
1024 disk = part_to_disk(part);
1026 spin_unlock_bh(&ext_devt_lock);
1033 * Synchronize with del_gendisk() to not return disk that is being
1036 down_read(&disk->lookup_sem);
1037 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1038 !(disk->flags & GENHD_FL_UP))) {
1039 up_read(&disk->lookup_sem);
1040 put_disk_and_module(disk);
1043 up_read(&disk->lookup_sem);
1049 * bdget_disk - do bdget() by gendisk and partition number
1050 * @disk: gendisk of interest
1051 * @partno: partition number
1053 * Find partition @partno from @disk, do bdget() on it.
1059 * Resulting block_device on success, NULL on failure.
1061 struct block_device *bdget_disk(struct gendisk *disk, int partno)
1063 struct hd_struct *part;
1064 struct block_device *bdev = NULL;
1066 part = disk_get_part(disk, partno);
1068 bdev = bdget(part_devt(part));
1069 disk_put_part(part);
1073 EXPORT_SYMBOL(bdget_disk);
1076 * print a full list of all partitions - intended for places where the root
1077 * filesystem can't be mounted and thus to give the victim some idea of what
1080 void __init printk_all_partitions(void)
1082 struct class_dev_iter iter;
1085 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1086 while ((dev = class_dev_iter_next(&iter))) {
1087 struct gendisk *disk = dev_to_disk(dev);
1088 struct disk_part_iter piter;
1089 struct hd_struct *part;
1090 char name_buf[BDEVNAME_SIZE];
1091 char devt_buf[BDEVT_SIZE];
1094 * Don't show empty devices or things that have been
1097 if (get_capacity(disk) == 0 ||
1098 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1102 * Note, unlike /proc/partitions, I am showing the
1103 * numbers in hex - the same format as the root=
1106 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1107 while ((part = disk_part_iter_next(&piter))) {
1108 bool is_part0 = part == &disk->part0;
1110 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1111 bdevt_str(part_devt(part), devt_buf),
1112 (unsigned long long)part_nr_sects_read(part) >> 1
1113 , disk_name(disk, part->partno, name_buf),
1114 part->info ? part->info->uuid : "");
1116 if (dev->parent && dev->parent->driver)
1117 printk(" driver: %s\n",
1118 dev->parent->driver->name);
1120 printk(" (driver?)\n");
1124 disk_part_iter_exit(&piter);
1126 class_dev_iter_exit(&iter);
1129 #ifdef CONFIG_PROC_FS
1131 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1134 struct class_dev_iter *iter;
1137 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1139 return ERR_PTR(-ENOMEM);
1141 seqf->private = iter;
1142 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1144 dev = class_dev_iter_next(iter);
1149 return dev_to_disk(dev);
1152 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1157 dev = class_dev_iter_next(seqf->private);
1159 return dev_to_disk(dev);
1164 static void disk_seqf_stop(struct seq_file *seqf, void *v)
1166 struct class_dev_iter *iter = seqf->private;
1168 /* stop is called even after start failed :-( */
1170 class_dev_iter_exit(iter);
1172 seqf->private = NULL;
1176 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1180 p = disk_seqf_start(seqf, pos);
1181 if (!IS_ERR_OR_NULL(p) && !*pos)
1182 seq_puts(seqf, "major minor #blocks name\n\n");
1186 static int show_partition(struct seq_file *seqf, void *v)
1188 struct gendisk *sgp = v;
1189 struct disk_part_iter piter;
1190 struct hd_struct *part;
1191 char buf[BDEVNAME_SIZE];
1193 /* Don't show non-partitionable removeable devices or empty devices */
1194 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1195 (sgp->flags & GENHD_FL_REMOVABLE)))
1197 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1200 /* show the full disk and all non-0 size partitions of it */
1201 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1202 while ((part = disk_part_iter_next(&piter)))
1203 seq_printf(seqf, "%4d %7d %10llu %s\n",
1204 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1205 (unsigned long long)part_nr_sects_read(part) >> 1,
1206 disk_name(sgp, part->partno, buf));
1207 disk_part_iter_exit(&piter);
1212 static const struct seq_operations partitions_op = {
1213 .start = show_partition_start,
1214 .next = disk_seqf_next,
1215 .stop = disk_seqf_stop,
1216 .show = show_partition
1221 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1223 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1224 /* Make old-style 2.4 aliases work */
1225 request_module("block-major-%d", MAJOR(devt));
1229 static int __init genhd_device_init(void)
1233 block_class.dev_kobj = sysfs_dev_block_kobj;
1234 error = class_register(&block_class);
1235 if (unlikely(error))
1237 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1240 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1242 /* create top-level block dir */
1243 if (!sysfs_deprecated)
1244 block_depr = kobject_create_and_add("block", NULL);
1248 subsys_initcall(genhd_device_init);
1250 static ssize_t disk_range_show(struct device *dev,
1251 struct device_attribute *attr, char *buf)
1253 struct gendisk *disk = dev_to_disk(dev);
1255 return sprintf(buf, "%d\n", disk->minors);
1258 static ssize_t disk_ext_range_show(struct device *dev,
1259 struct device_attribute *attr, char *buf)
1261 struct gendisk *disk = dev_to_disk(dev);
1263 return sprintf(buf, "%d\n", disk_max_parts(disk));
1266 static ssize_t disk_removable_show(struct device *dev,
1267 struct device_attribute *attr, char *buf)
1269 struct gendisk *disk = dev_to_disk(dev);
1271 return sprintf(buf, "%d\n",
1272 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1275 static ssize_t disk_hidden_show(struct device *dev,
1276 struct device_attribute *attr, char *buf)
1278 struct gendisk *disk = dev_to_disk(dev);
1280 return sprintf(buf, "%d\n",
1281 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1284 static ssize_t disk_ro_show(struct device *dev,
1285 struct device_attribute *attr, char *buf)
1287 struct gendisk *disk = dev_to_disk(dev);
1289 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1292 ssize_t part_size_show(struct device *dev,
1293 struct device_attribute *attr, char *buf)
1295 struct hd_struct *p = dev_to_part(dev);
1297 return sprintf(buf, "%llu\n",
1298 (unsigned long long)part_nr_sects_read(p));
1301 ssize_t part_stat_show(struct device *dev,
1302 struct device_attribute *attr, char *buf)
1304 struct hd_struct *p = dev_to_part(dev);
1305 struct request_queue *q = part_to_disk(p)->queue;
1306 struct disk_stats stat;
1307 unsigned int inflight;
1309 part_stat_read_all(p, &stat);
1310 inflight = part_in_flight(q, p);
1313 "%8lu %8lu %8llu %8u "
1314 "%8lu %8lu %8llu %8u "
1316 "%8lu %8lu %8llu %8u "
1319 stat.ios[STAT_READ],
1320 stat.merges[STAT_READ],
1321 (unsigned long long)stat.sectors[STAT_READ],
1322 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1323 stat.ios[STAT_WRITE],
1324 stat.merges[STAT_WRITE],
1325 (unsigned long long)stat.sectors[STAT_WRITE],
1326 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1328 jiffies_to_msecs(stat.io_ticks),
1329 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1330 stat.nsecs[STAT_WRITE] +
1331 stat.nsecs[STAT_DISCARD] +
1332 stat.nsecs[STAT_FLUSH],
1334 stat.ios[STAT_DISCARD],
1335 stat.merges[STAT_DISCARD],
1336 (unsigned long long)stat.sectors[STAT_DISCARD],
1337 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1338 stat.ios[STAT_FLUSH],
1339 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1342 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1345 struct hd_struct *p = dev_to_part(dev);
1346 struct request_queue *q = part_to_disk(p)->queue;
1347 unsigned int inflight[2];
1349 part_in_flight_rw(q, p, inflight);
1350 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1353 static ssize_t disk_capability_show(struct device *dev,
1354 struct device_attribute *attr, char *buf)
1356 struct gendisk *disk = dev_to_disk(dev);
1358 return sprintf(buf, "%x\n", disk->flags);
1361 static ssize_t disk_alignment_offset_show(struct device *dev,
1362 struct device_attribute *attr,
1365 struct gendisk *disk = dev_to_disk(dev);
1367 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1370 static ssize_t disk_discard_alignment_show(struct device *dev,
1371 struct device_attribute *attr,
1374 struct gendisk *disk = dev_to_disk(dev);
1376 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1379 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1380 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1381 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1382 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1383 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1384 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1385 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1386 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1387 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1388 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1389 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1390 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1392 #ifdef CONFIG_FAIL_MAKE_REQUEST
1393 ssize_t part_fail_show(struct device *dev,
1394 struct device_attribute *attr, char *buf)
1396 struct hd_struct *p = dev_to_part(dev);
1398 return sprintf(buf, "%d\n", p->make_it_fail);
1401 ssize_t part_fail_store(struct device *dev,
1402 struct device_attribute *attr,
1403 const char *buf, size_t count)
1405 struct hd_struct *p = dev_to_part(dev);
1408 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1409 p->make_it_fail = (i == 0) ? 0 : 1;
1414 static struct device_attribute dev_attr_fail =
1415 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1416 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1418 #ifdef CONFIG_FAIL_IO_TIMEOUT
1419 static struct device_attribute dev_attr_fail_timeout =
1420 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1423 static struct attribute *disk_attrs[] = {
1424 &dev_attr_range.attr,
1425 &dev_attr_ext_range.attr,
1426 &dev_attr_removable.attr,
1427 &dev_attr_hidden.attr,
1429 &dev_attr_size.attr,
1430 &dev_attr_alignment_offset.attr,
1431 &dev_attr_discard_alignment.attr,
1432 &dev_attr_capability.attr,
1433 &dev_attr_stat.attr,
1434 &dev_attr_inflight.attr,
1435 &dev_attr_badblocks.attr,
1436 #ifdef CONFIG_FAIL_MAKE_REQUEST
1437 &dev_attr_fail.attr,
1439 #ifdef CONFIG_FAIL_IO_TIMEOUT
1440 &dev_attr_fail_timeout.attr,
1445 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1447 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1448 struct gendisk *disk = dev_to_disk(dev);
1450 if (a == &dev_attr_badblocks.attr && !disk->bb)
1455 static struct attribute_group disk_attr_group = {
1456 .attrs = disk_attrs,
1457 .is_visible = disk_visible,
1460 static const struct attribute_group *disk_attr_groups[] = {
1466 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1467 * @disk: disk to replace part_tbl for
1468 * @new_ptbl: new part_tbl to install
1470 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1471 * original ptbl is freed using RCU callback.
1474 * Matching bd_mutex locked or the caller is the only user of @disk.
1476 static void disk_replace_part_tbl(struct gendisk *disk,
1477 struct disk_part_tbl *new_ptbl)
1479 struct disk_part_tbl *old_ptbl =
1480 rcu_dereference_protected(disk->part_tbl, 1);
1482 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1485 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1486 kfree_rcu(old_ptbl, rcu_head);
1491 * disk_expand_part_tbl - expand disk->part_tbl
1492 * @disk: disk to expand part_tbl for
1493 * @partno: expand such that this partno can fit in
1495 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1496 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1499 * Matching bd_mutex locked or the caller is the only user of @disk.
1503 * 0 on success, -errno on failure.
1505 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1507 struct disk_part_tbl *old_ptbl =
1508 rcu_dereference_protected(disk->part_tbl, 1);
1509 struct disk_part_tbl *new_ptbl;
1510 int len = old_ptbl ? old_ptbl->len : 0;
1514 * check for int overflow, since we can get here from blkpg_ioctl()
1515 * with a user passed 'partno'.
1517 target = partno + 1;
1521 /* disk_max_parts() is zero during initialization, ignore if so */
1522 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1528 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1533 new_ptbl->len = target;
1535 for (i = 0; i < len; i++)
1536 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1538 disk_replace_part_tbl(disk, new_ptbl);
1542 static void disk_release(struct device *dev)
1544 struct gendisk *disk = dev_to_disk(dev);
1546 blk_free_devt(dev->devt);
1547 disk_release_events(disk);
1548 kfree(disk->random);
1549 disk_replace_part_tbl(disk, NULL);
1550 hd_free_part(&disk->part0);
1552 blk_put_queue(disk->queue);
1555 struct class block_class = {
1559 static char *block_devnode(struct device *dev, umode_t *mode,
1560 kuid_t *uid, kgid_t *gid)
1562 struct gendisk *disk = dev_to_disk(dev);
1564 if (disk->fops->devnode)
1565 return disk->fops->devnode(disk, mode);
1569 static const struct device_type disk_type = {
1571 .groups = disk_attr_groups,
1572 .release = disk_release,
1573 .devnode = block_devnode,
1576 #ifdef CONFIG_PROC_FS
1578 * aggregate disk stat collector. Uses the same stats that the sysfs
1579 * entries do, above, but makes them available through one seq_file.
1581 * The output looks suspiciously like /proc/partitions with a bunch of
1584 static int diskstats_show(struct seq_file *seqf, void *v)
1586 struct gendisk *gp = v;
1587 struct disk_part_iter piter;
1588 struct hd_struct *hd;
1589 char buf[BDEVNAME_SIZE];
1590 unsigned int inflight;
1591 struct disk_stats stat;
1594 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1595 seq_puts(seqf, "major minor name"
1596 " rio rmerge rsect ruse wio wmerge "
1597 "wsect wuse running use aveq"
1601 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1602 while ((hd = disk_part_iter_next(&piter))) {
1603 part_stat_read_all(hd, &stat);
1604 inflight = part_in_flight(gp->queue, hd);
1606 seq_printf(seqf, "%4d %7d %s "
1613 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1614 disk_name(gp, hd->partno, buf),
1615 stat.ios[STAT_READ],
1616 stat.merges[STAT_READ],
1617 stat.sectors[STAT_READ],
1618 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1620 stat.ios[STAT_WRITE],
1621 stat.merges[STAT_WRITE],
1622 stat.sectors[STAT_WRITE],
1623 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1626 jiffies_to_msecs(stat.io_ticks),
1627 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1628 stat.nsecs[STAT_WRITE] +
1629 stat.nsecs[STAT_DISCARD] +
1630 stat.nsecs[STAT_FLUSH],
1632 stat.ios[STAT_DISCARD],
1633 stat.merges[STAT_DISCARD],
1634 stat.sectors[STAT_DISCARD],
1635 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1637 stat.ios[STAT_FLUSH],
1638 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1642 disk_part_iter_exit(&piter);
1647 static const struct seq_operations diskstats_op = {
1648 .start = disk_seqf_start,
1649 .next = disk_seqf_next,
1650 .stop = disk_seqf_stop,
1651 .show = diskstats_show
1654 static int __init proc_genhd_init(void)
1656 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1657 proc_create_seq("partitions", 0, NULL, &partitions_op);
1660 module_init(proc_genhd_init);
1661 #endif /* CONFIG_PROC_FS */
1663 dev_t blk_lookup_devt(const char *name, int partno)
1665 dev_t devt = MKDEV(0, 0);
1666 struct class_dev_iter iter;
1669 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1670 while ((dev = class_dev_iter_next(&iter))) {
1671 struct gendisk *disk = dev_to_disk(dev);
1672 struct hd_struct *part;
1674 if (strcmp(dev_name(dev), name))
1677 if (partno < disk->minors) {
1678 /* We need to return the right devno, even
1679 * if the partition doesn't exist yet.
1681 devt = MKDEV(MAJOR(dev->devt),
1682 MINOR(dev->devt) + partno);
1685 part = disk_get_part(disk, partno);
1687 devt = part_devt(part);
1688 disk_put_part(part);
1691 disk_put_part(part);
1693 class_dev_iter_exit(&iter);
1697 struct gendisk *__alloc_disk_node(int minors, int node_id)
1699 struct gendisk *disk;
1700 struct disk_part_tbl *ptbl;
1702 if (minors > DISK_MAX_PARTS) {
1704 "block: can't allocate more than %d partitions\n",
1706 minors = DISK_MAX_PARTS;
1709 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1711 if (!init_part_stats(&disk->part0)) {
1715 init_rwsem(&disk->lookup_sem);
1716 disk->node_id = node_id;
1717 if (disk_expand_part_tbl(disk, 0)) {
1718 free_part_stats(&disk->part0);
1722 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1723 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1726 * set_capacity() and get_capacity() currently don't use
1727 * seqcounter to read/update the part0->nr_sects. Still init
1728 * the counter as we can read the sectors in IO submission
1729 * patch using seqence counters.
1731 * TODO: Ideally set_capacity() and get_capacity() should be
1732 * converted to make use of bd_mutex and sequence counters.
1734 seqcount_init(&disk->part0.nr_sects_seq);
1735 if (hd_ref_init(&disk->part0)) {
1736 hd_free_part(&disk->part0);
1741 disk->minors = minors;
1742 rand_initialize_disk(disk);
1743 disk_to_dev(disk)->class = &block_class;
1744 disk_to_dev(disk)->type = &disk_type;
1745 device_initialize(disk_to_dev(disk));
1749 EXPORT_SYMBOL(__alloc_disk_node);
1751 struct kobject *get_disk_and_module(struct gendisk *disk)
1753 struct module *owner;
1754 struct kobject *kobj;
1758 owner = disk->fops->owner;
1759 if (owner && !try_module_get(owner))
1761 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1769 EXPORT_SYMBOL(get_disk_and_module);
1771 void put_disk(struct gendisk *disk)
1774 kobject_put(&disk_to_dev(disk)->kobj);
1776 EXPORT_SYMBOL(put_disk);
1779 * This is a counterpart of get_disk_and_module() and thus also of
1782 void put_disk_and_module(struct gendisk *disk)
1785 struct module *owner = disk->fops->owner;
1791 EXPORT_SYMBOL(put_disk_and_module);
1793 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1795 char event[] = "DISK_RO=1";
1796 char *envp[] = { event, NULL };
1800 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1803 void set_device_ro(struct block_device *bdev, int flag)
1805 bdev->bd_part->policy = flag;
1808 EXPORT_SYMBOL(set_device_ro);
1810 void set_disk_ro(struct gendisk *disk, int flag)
1812 struct disk_part_iter piter;
1813 struct hd_struct *part;
1815 if (disk->part0.policy != flag) {
1816 set_disk_ro_uevent(disk, flag);
1817 disk->part0.policy = flag;
1820 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1821 while ((part = disk_part_iter_next(&piter)))
1822 part->policy = flag;
1823 disk_part_iter_exit(&piter);
1826 EXPORT_SYMBOL(set_disk_ro);
1828 int bdev_read_only(struct block_device *bdev)
1832 return bdev->bd_part->policy;
1835 EXPORT_SYMBOL(bdev_read_only);
1838 * Disk events - monitor disk events like media change and eject request.
1840 struct disk_events {
1841 struct list_head node; /* all disk_event's */
1842 struct gendisk *disk; /* the associated disk */
1845 struct mutex block_mutex; /* protects blocking */
1846 int block; /* event blocking depth */
1847 unsigned int pending; /* events already sent out */
1848 unsigned int clearing; /* events being cleared */
1850 long poll_msecs; /* interval, -1 for default */
1851 struct delayed_work dwork;
1854 static const char *disk_events_strs[] = {
1855 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1856 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1859 static char *disk_uevents[] = {
1860 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1861 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1864 /* list of all disk_events */
1865 static DEFINE_MUTEX(disk_events_mutex);
1866 static LIST_HEAD(disk_events);
1868 /* disable in-kernel polling by default */
1869 static unsigned long disk_events_dfl_poll_msecs;
1871 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1873 struct disk_events *ev = disk->ev;
1874 long intv_msecs = 0;
1877 * If device-specific poll interval is set, always use it. If
1878 * the default is being used, poll if the POLL flag is set.
1880 if (ev->poll_msecs >= 0)
1881 intv_msecs = ev->poll_msecs;
1882 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1883 intv_msecs = disk_events_dfl_poll_msecs;
1885 return msecs_to_jiffies(intv_msecs);
1889 * disk_block_events - block and flush disk event checking
1890 * @disk: disk to block events for
1892 * On return from this function, it is guaranteed that event checking
1893 * isn't in progress and won't happen until unblocked by
1894 * disk_unblock_events(). Events blocking is counted and the actual
1895 * unblocking happens after the matching number of unblocks are done.
1897 * Note that this intentionally does not block event checking from
1898 * disk_clear_events().
1903 void disk_block_events(struct gendisk *disk)
1905 struct disk_events *ev = disk->ev;
1906 unsigned long flags;
1913 * Outer mutex ensures that the first blocker completes canceling
1914 * the event work before further blockers are allowed to finish.
1916 mutex_lock(&ev->block_mutex);
1918 spin_lock_irqsave(&ev->lock, flags);
1919 cancel = !ev->block++;
1920 spin_unlock_irqrestore(&ev->lock, flags);
1923 cancel_delayed_work_sync(&disk->ev->dwork);
1925 mutex_unlock(&ev->block_mutex);
1928 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1930 struct disk_events *ev = disk->ev;
1932 unsigned long flags;
1934 spin_lock_irqsave(&ev->lock, flags);
1936 if (WARN_ON_ONCE(ev->block <= 0))
1942 intv = disk_events_poll_jiffies(disk);
1944 queue_delayed_work(system_freezable_power_efficient_wq,
1947 queue_delayed_work(system_freezable_power_efficient_wq,
1950 spin_unlock_irqrestore(&ev->lock, flags);
1954 * disk_unblock_events - unblock disk event checking
1955 * @disk: disk to unblock events for
1957 * Undo disk_block_events(). When the block count reaches zero, it
1958 * starts events polling if configured.
1961 * Don't care. Safe to call from irq context.
1963 void disk_unblock_events(struct gendisk *disk)
1966 __disk_unblock_events(disk, false);
1970 * disk_flush_events - schedule immediate event checking and flushing
1971 * @disk: disk to check and flush events for
1972 * @mask: events to flush
1974 * Schedule immediate event checking on @disk if not blocked. Events in
1975 * @mask are scheduled to be cleared from the driver. Note that this
1976 * doesn't clear the events from @disk->ev.
1979 * If @mask is non-zero must be called with bdev->bd_mutex held.
1981 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1983 struct disk_events *ev = disk->ev;
1988 spin_lock_irq(&ev->lock);
1989 ev->clearing |= mask;
1991 mod_delayed_work(system_freezable_power_efficient_wq,
1993 spin_unlock_irq(&ev->lock);
1997 * disk_clear_events - synchronously check, clear and return pending events
1998 * @disk: disk to fetch and clear events from
1999 * @mask: mask of events to be fetched and cleared
2001 * Disk events are synchronously checked and pending events in @mask
2002 * are cleared and returned. This ignores the block count.
2007 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
2009 const struct block_device_operations *bdops = disk->fops;
2010 struct disk_events *ev = disk->ev;
2011 unsigned int pending;
2012 unsigned int clearing = mask;
2015 /* for drivers still using the old ->media_changed method */
2016 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
2017 bdops->media_changed && bdops->media_changed(disk))
2018 return DISK_EVENT_MEDIA_CHANGE;
2022 disk_block_events(disk);
2025 * store the union of mask and ev->clearing on the stack so that the
2026 * race with disk_flush_events does not cause ambiguity (ev->clearing
2027 * can still be modified even if events are blocked).
2029 spin_lock_irq(&ev->lock);
2030 clearing |= ev->clearing;
2032 spin_unlock_irq(&ev->lock);
2034 disk_check_events(ev, &clearing);
2036 * if ev->clearing is not 0, the disk_flush_events got called in the
2037 * middle of this function, so we want to run the workfn without delay.
2039 __disk_unblock_events(disk, ev->clearing ? true : false);
2041 /* then, fetch and clear pending events */
2042 spin_lock_irq(&ev->lock);
2043 pending = ev->pending & mask;
2044 ev->pending &= ~mask;
2045 spin_unlock_irq(&ev->lock);
2046 WARN_ON_ONCE(clearing & mask);
2052 * Separate this part out so that a different pointer for clearing_ptr can be
2053 * passed in for disk_clear_events.
2055 static void disk_events_workfn(struct work_struct *work)
2057 struct delayed_work *dwork = to_delayed_work(work);
2058 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2060 disk_check_events(ev, &ev->clearing);
2063 static void disk_check_events(struct disk_events *ev,
2064 unsigned int *clearing_ptr)
2066 struct gendisk *disk = ev->disk;
2067 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2068 unsigned int clearing = *clearing_ptr;
2069 unsigned int events;
2071 int nr_events = 0, i;
2074 events = disk->fops->check_events(disk, clearing);
2076 /* accumulate pending events and schedule next poll if necessary */
2077 spin_lock_irq(&ev->lock);
2079 events &= ~ev->pending;
2080 ev->pending |= events;
2081 *clearing_ptr &= ~clearing;
2083 intv = disk_events_poll_jiffies(disk);
2084 if (!ev->block && intv)
2085 queue_delayed_work(system_freezable_power_efficient_wq,
2088 spin_unlock_irq(&ev->lock);
2091 * Tell userland about new events. Only the events listed in
2092 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2093 * is set. Otherwise, events are processed internally but never
2094 * get reported to userland.
2096 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2097 if ((events & disk->events & (1 << i)) &&
2098 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2099 envp[nr_events++] = disk_uevents[i];
2102 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2106 * A disk events enabled device has the following sysfs nodes under
2107 * its /sys/block/X/ directory.
2109 * events : list of all supported events
2110 * events_async : list of events which can be detected w/o polling
2111 * (always empty, only for backwards compatibility)
2112 * events_poll_msecs : polling interval, 0: disable, -1: system default
2114 static ssize_t __disk_events_show(unsigned int events, char *buf)
2116 const char *delim = "";
2120 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2121 if (events & (1 << i)) {
2122 pos += sprintf(buf + pos, "%s%s",
2123 delim, disk_events_strs[i]);
2127 pos += sprintf(buf + pos, "\n");
2131 static ssize_t disk_events_show(struct device *dev,
2132 struct device_attribute *attr, char *buf)
2134 struct gendisk *disk = dev_to_disk(dev);
2136 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2139 return __disk_events_show(disk->events, buf);
2142 static ssize_t disk_events_async_show(struct device *dev,
2143 struct device_attribute *attr, char *buf)
2148 static ssize_t disk_events_poll_msecs_show(struct device *dev,
2149 struct device_attribute *attr,
2152 struct gendisk *disk = dev_to_disk(dev);
2155 return sprintf(buf, "-1\n");
2157 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2160 static ssize_t disk_events_poll_msecs_store(struct device *dev,
2161 struct device_attribute *attr,
2162 const char *buf, size_t count)
2164 struct gendisk *disk = dev_to_disk(dev);
2167 if (!count || !sscanf(buf, "%ld", &intv))
2170 if (intv < 0 && intv != -1)
2176 disk_block_events(disk);
2177 disk->ev->poll_msecs = intv;
2178 __disk_unblock_events(disk, true);
2183 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2184 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2185 static const DEVICE_ATTR(events_poll_msecs, 0644,
2186 disk_events_poll_msecs_show,
2187 disk_events_poll_msecs_store);
2189 static const struct attribute *disk_events_attrs[] = {
2190 &dev_attr_events.attr,
2191 &dev_attr_events_async.attr,
2192 &dev_attr_events_poll_msecs.attr,
2197 * The default polling interval can be specified by the kernel
2198 * parameter block.events_dfl_poll_msecs which defaults to 0
2199 * (disable). This can also be modified runtime by writing to
2200 * /sys/module/block/parameters/events_dfl_poll_msecs.
2202 static int disk_events_set_dfl_poll_msecs(const char *val,
2203 const struct kernel_param *kp)
2205 struct disk_events *ev;
2208 ret = param_set_ulong(val, kp);
2212 mutex_lock(&disk_events_mutex);
2214 list_for_each_entry(ev, &disk_events, node)
2215 disk_flush_events(ev->disk, 0);
2217 mutex_unlock(&disk_events_mutex);
2222 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2223 .set = disk_events_set_dfl_poll_msecs,
2224 .get = param_get_ulong,
2227 #undef MODULE_PARAM_PREFIX
2228 #define MODULE_PARAM_PREFIX "block."
2230 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2231 &disk_events_dfl_poll_msecs, 0644);
2234 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2236 static void disk_alloc_events(struct gendisk *disk)
2238 struct disk_events *ev;
2240 if (!disk->fops->check_events || !disk->events)
2243 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2245 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2249 INIT_LIST_HEAD(&ev->node);
2251 spin_lock_init(&ev->lock);
2252 mutex_init(&ev->block_mutex);
2254 ev->poll_msecs = -1;
2255 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2260 static void disk_add_events(struct gendisk *disk)
2262 /* FIXME: error handling */
2263 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2264 pr_warn("%s: failed to create sysfs files for events\n",
2270 mutex_lock(&disk_events_mutex);
2271 list_add_tail(&disk->ev->node, &disk_events);
2272 mutex_unlock(&disk_events_mutex);
2275 * Block count is initialized to 1 and the following initial
2276 * unblock kicks it into action.
2278 __disk_unblock_events(disk, true);
2281 static void disk_del_events(struct gendisk *disk)
2284 disk_block_events(disk);
2286 mutex_lock(&disk_events_mutex);
2287 list_del_init(&disk->ev->node);
2288 mutex_unlock(&disk_events_mutex);
2291 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2294 static void disk_release_events(struct gendisk *disk)
2296 /* the block count should be 1 from disk_del_events() */
2297 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);