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 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 unsigned int part_in_flight(struct request_queue *q, struct hd_struct *part)
145 unsigned int inflight;
147 if (queue_is_mq(q)) {
148 return blk_mq_in_flight(q, part);
152 for_each_possible_cpu(cpu) {
153 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
154 part_stat_local_read_cpu(part, in_flight[1], cpu);
156 if ((int)inflight < 0)
162 void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
163 unsigned int inflight[2])
167 if (queue_is_mq(q)) {
168 blk_mq_in_flight_rw(q, part, inflight);
174 for_each_possible_cpu(cpu) {
175 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
176 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
178 if ((int)inflight[0] < 0)
180 if ((int)inflight[1] < 0)
184 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
186 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
188 if (unlikely(partno < 0 || partno >= ptbl->len))
190 return rcu_dereference(ptbl->part[partno]);
194 * disk_get_part - get partition
195 * @disk: disk to look partition from
196 * @partno: partition number
198 * Look for partition @partno from @disk. If found, increment
199 * reference count and return it.
205 * Pointer to the found partition on success, NULL if not found.
207 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
209 struct hd_struct *part;
212 part = __disk_get_part(disk, partno);
214 get_device(part_to_dev(part));
219 EXPORT_SYMBOL_GPL(disk_get_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 valid only
348 * while preemption is disabled.
351 * Found partition on success, part0 is returned if no partition matches
353 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
355 struct disk_part_tbl *ptbl;
356 struct hd_struct *part;
359 ptbl = rcu_dereference(disk->part_tbl);
361 part = rcu_dereference(ptbl->last_lookup);
362 if (part && sector_in_part(part, sector))
365 for (i = 1; i < ptbl->len; i++) {
366 part = rcu_dereference(ptbl->part[i]);
368 if (part && sector_in_part(part, sector)) {
369 rcu_assign_pointer(ptbl->last_lookup, part);
375 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
378 * Can be deleted altogether. Later.
381 #define BLKDEV_MAJOR_HASH_SIZE 255
382 static struct blk_major_name {
383 struct blk_major_name *next;
386 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
388 /* index in the above - for now: assume no multimajor ranges */
389 static inline int major_to_index(unsigned major)
391 return major % BLKDEV_MAJOR_HASH_SIZE;
394 #ifdef CONFIG_PROC_FS
395 void blkdev_show(struct seq_file *seqf, off_t offset)
397 struct blk_major_name *dp;
399 mutex_lock(&block_class_lock);
400 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
401 if (dp->major == offset)
402 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
403 mutex_unlock(&block_class_lock);
405 #endif /* CONFIG_PROC_FS */
408 * register_blkdev - register a new block device
410 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
411 * @major = 0, try to allocate any unused major number.
412 * @name: the name of the new block device as a zero terminated string
414 * The @name must be unique within the system.
416 * The return value depends on the @major input parameter:
418 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
419 * then the function returns zero on success, or a negative error code
420 * - if any unused major number was requested with @major = 0 parameter
421 * then the return value is the allocated major number in range
422 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
424 * See Documentation/admin-guide/devices.txt for the list of allocated
427 int register_blkdev(unsigned int major, const char *name)
429 struct blk_major_name **n, *p;
432 mutex_lock(&block_class_lock);
436 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
437 if (major_names[index] == NULL)
442 printk("%s: failed to get major for %s\n",
451 if (major >= BLKDEV_MAJOR_MAX) {
452 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
453 __func__, major, BLKDEV_MAJOR_MAX-1, name);
459 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
466 strlcpy(p->name, name, sizeof(p->name));
468 index = major_to_index(major);
470 for (n = &major_names[index]; *n; n = &(*n)->next) {
471 if ((*n)->major == major)
480 printk("register_blkdev: cannot get major %u for %s\n",
485 mutex_unlock(&block_class_lock);
489 EXPORT_SYMBOL(register_blkdev);
491 void unregister_blkdev(unsigned int major, const char *name)
493 struct blk_major_name **n;
494 struct blk_major_name *p = NULL;
495 int index = major_to_index(major);
497 mutex_lock(&block_class_lock);
498 for (n = &major_names[index]; *n; n = &(*n)->next)
499 if ((*n)->major == major)
501 if (!*n || strcmp((*n)->name, name)) {
507 mutex_unlock(&block_class_lock);
511 EXPORT_SYMBOL(unregister_blkdev);
513 static struct kobj_map *bdev_map;
516 * blk_mangle_minor - scatter minor numbers apart
517 * @minor: minor number to mangle
519 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
520 * is enabled. Mangling twice gives the original value.
528 static int blk_mangle_minor(int minor)
530 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
533 for (i = 0; i < MINORBITS / 2; i++) {
534 int low = minor & (1 << i);
535 int high = minor & (1 << (MINORBITS - 1 - i));
536 int distance = MINORBITS - 1 - 2 * i;
538 minor ^= low | high; /* clear both bits */
539 low <<= distance; /* swap the positions */
541 minor |= low | high; /* and set */
548 * blk_alloc_devt - allocate a dev_t for a partition
549 * @part: partition to allocate dev_t for
550 * @devt: out parameter for resulting dev_t
552 * Allocate a dev_t for block device.
555 * 0 on success, allocated dev_t is returned in *@devt. -errno on
561 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
563 struct gendisk *disk = part_to_disk(part);
566 /* in consecutive minor range? */
567 if (part->partno < disk->minors) {
568 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
572 /* allocate ext devt */
573 idr_preload(GFP_KERNEL);
575 spin_lock_bh(&ext_devt_lock);
576 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
577 spin_unlock_bh(&ext_devt_lock);
581 return idx == -ENOSPC ? -EBUSY : idx;
583 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
588 * blk_free_devt - free a dev_t
589 * @devt: dev_t to free
591 * Free @devt which was allocated using blk_alloc_devt().
596 void blk_free_devt(dev_t devt)
598 if (devt == MKDEV(0, 0))
601 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
602 spin_lock_bh(&ext_devt_lock);
603 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
604 spin_unlock_bh(&ext_devt_lock);
609 * We invalidate devt by assigning NULL pointer for devt in idr.
611 void blk_invalidate_devt(dev_t devt)
613 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
614 spin_lock_bh(&ext_devt_lock);
615 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
616 spin_unlock_bh(&ext_devt_lock);
620 static char *bdevt_str(dev_t devt, char *buf)
622 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
623 char tbuf[BDEVT_SIZE];
624 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
625 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
627 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
633 * Register device numbers dev..(dev+range-1)
634 * range must be nonzero
635 * The hash chain is sorted on range, so that subranges can override.
637 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
638 struct kobject *(*probe)(dev_t, int *, void *),
639 int (*lock)(dev_t, void *), void *data)
641 kobj_map(bdev_map, devt, range, module, probe, lock, data);
644 EXPORT_SYMBOL(blk_register_region);
646 void blk_unregister_region(dev_t devt, unsigned long range)
648 kobj_unmap(bdev_map, devt, range);
651 EXPORT_SYMBOL(blk_unregister_region);
653 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
655 struct gendisk *p = data;
657 return &disk_to_dev(p)->kobj;
660 static int exact_lock(dev_t devt, void *data)
662 struct gendisk *p = data;
664 if (!get_disk_and_module(p))
669 static void register_disk(struct device *parent, struct gendisk *disk,
670 const struct attribute_group **groups)
672 struct device *ddev = disk_to_dev(disk);
673 struct block_device *bdev;
674 struct disk_part_iter piter;
675 struct hd_struct *part;
678 ddev->parent = parent;
680 dev_set_name(ddev, "%s", disk->disk_name);
682 /* delay uevents, until we scanned partition table */
683 dev_set_uevent_suppress(ddev, 1);
686 WARN_ON(ddev->groups);
687 ddev->groups = groups;
689 if (device_add(ddev))
691 if (!sysfs_deprecated) {
692 err = sysfs_create_link(block_depr, &ddev->kobj,
693 kobject_name(&ddev->kobj));
701 * avoid probable deadlock caused by allocating memory with
702 * GFP_KERNEL in runtime_resume callback of its all ancestor
705 pm_runtime_set_memalloc_noio(ddev, true);
707 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
708 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
710 if (disk->flags & GENHD_FL_HIDDEN) {
711 dev_set_uevent_suppress(ddev, 0);
715 /* No minors to use for partitions */
716 if (!disk_part_scan_enabled(disk))
719 /* No such device (e.g., media were just removed) */
720 if (!get_capacity(disk))
723 bdev = bdget_disk(disk, 0);
727 bdev->bd_invalidated = 1;
728 err = blkdev_get(bdev, FMODE_READ, NULL);
731 blkdev_put(bdev, FMODE_READ);
734 /* announce disk after possible partitions are created */
735 dev_set_uevent_suppress(ddev, 0);
736 kobject_uevent(&ddev->kobj, KOBJ_ADD);
738 /* announce possible partitions */
739 disk_part_iter_init(&piter, disk, 0);
740 while ((part = disk_part_iter_next(&piter)))
741 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
742 disk_part_iter_exit(&piter);
744 if (disk->queue->backing_dev_info->dev) {
745 err = sysfs_create_link(&ddev->kobj,
746 &disk->queue->backing_dev_info->dev->kobj,
753 * __device_add_disk - add disk information to kernel list
754 * @parent: parent device for the disk
755 * @disk: per-device partitioning information
756 * @groups: Additional per-device sysfs groups
757 * @register_queue: register the queue if set to true
759 * This function registers the partitioning information in @disk
762 * FIXME: error handling
764 static void __device_add_disk(struct device *parent, struct gendisk *disk,
765 const struct attribute_group **groups,
772 * The disk queue should now be all set with enough information about
773 * the device for the elevator code to pick an adequate default
774 * elevator if one is needed, that is, for devices requesting queue
778 elevator_init_mq(disk->queue);
780 /* minors == 0 indicates to use ext devt from part0 and should
781 * be accompanied with EXT_DEVT flag. Make sure all
782 * parameters make sense.
784 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
785 WARN_ON(!disk->minors &&
786 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
788 disk->flags |= GENHD_FL_UP;
790 retval = blk_alloc_devt(&disk->part0, &devt);
795 disk->major = MAJOR(devt);
796 disk->first_minor = MINOR(devt);
798 disk_alloc_events(disk);
800 if (disk->flags & GENHD_FL_HIDDEN) {
802 * Don't let hidden disks show up in /proc/partitions,
803 * and don't bother scanning for partitions either.
805 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
806 disk->flags |= GENHD_FL_NO_PART_SCAN;
810 /* Register BDI before referencing it from bdev */
811 disk_to_dev(disk)->devt = devt;
812 ret = bdi_register_owner(disk->queue->backing_dev_info,
815 blk_register_region(disk_devt(disk), disk->minors, NULL,
816 exact_match, exact_lock, disk);
818 register_disk(parent, disk, groups);
820 blk_register_queue(disk);
823 * Take an extra ref on queue which will be put on disk_release()
824 * so that it sticks around as long as @disk is there.
826 WARN_ON_ONCE(!blk_get_queue(disk->queue));
828 disk_add_events(disk);
829 blk_integrity_add(disk);
832 void device_add_disk(struct device *parent, struct gendisk *disk,
833 const struct attribute_group **groups)
836 __device_add_disk(parent, disk, groups, true);
838 EXPORT_SYMBOL(device_add_disk);
840 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
842 __device_add_disk(parent, disk, NULL, false);
844 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
846 void del_gendisk(struct gendisk *disk)
848 struct disk_part_iter piter;
849 struct hd_struct *part;
851 blk_integrity_del(disk);
852 disk_del_events(disk);
855 * Block lookups of the disk until all bdevs are unhashed and the
856 * disk is marked as dead (GENHD_FL_UP cleared).
858 down_write(&disk->lookup_sem);
859 /* invalidate stuff */
860 disk_part_iter_init(&piter, disk,
861 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
862 while ((part = disk_part_iter_next(&piter))) {
863 invalidate_partition(disk, part->partno);
864 bdev_unhash_inode(part_devt(part));
865 delete_partition(disk, part->partno);
867 disk_part_iter_exit(&piter);
869 invalidate_partition(disk, 0);
870 bdev_unhash_inode(disk_devt(disk));
871 set_capacity(disk, 0);
872 disk->flags &= ~GENHD_FL_UP;
873 up_write(&disk->lookup_sem);
875 if (!(disk->flags & GENHD_FL_HIDDEN))
876 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
879 * Unregister bdi before releasing device numbers (as they can
880 * get reused and we'd get clashes in sysfs).
882 if (!(disk->flags & GENHD_FL_HIDDEN))
883 bdi_unregister(disk->queue->backing_dev_info);
884 blk_unregister_queue(disk);
889 if (!(disk->flags & GENHD_FL_HIDDEN))
890 blk_unregister_region(disk_devt(disk), disk->minors);
892 * Remove gendisk pointer from idr so that it cannot be looked up
893 * while RCU period before freeing gendisk is running to prevent
894 * use-after-free issues. Note that the device number stays
895 * "in-use" until we really free the gendisk.
897 blk_invalidate_devt(disk_devt(disk));
899 kobject_put(disk->part0.holder_dir);
900 kobject_put(disk->slave_dir);
902 part_stat_set_all(&disk->part0, 0);
903 disk->part0.stamp = 0;
904 if (!sysfs_deprecated)
905 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
906 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
907 device_del(disk_to_dev(disk));
909 EXPORT_SYMBOL(del_gendisk);
911 /* sysfs access to bad-blocks list. */
912 static ssize_t disk_badblocks_show(struct device *dev,
913 struct device_attribute *attr,
916 struct gendisk *disk = dev_to_disk(dev);
919 return sprintf(page, "\n");
921 return badblocks_show(disk->bb, page, 0);
924 static ssize_t disk_badblocks_store(struct device *dev,
925 struct device_attribute *attr,
926 const char *page, size_t len)
928 struct gendisk *disk = dev_to_disk(dev);
933 return badblocks_store(disk->bb, page, len, 0);
937 * get_gendisk - get partitioning information for a given device
938 * @devt: device to get partitioning information for
939 * @partno: returned partition index
941 * This function gets the structure containing partitioning
942 * information for the given device @devt.
944 struct gendisk *get_gendisk(dev_t devt, int *partno)
946 struct gendisk *disk = NULL;
948 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
949 struct kobject *kobj;
951 kobj = kobj_lookup(bdev_map, devt, partno);
953 disk = dev_to_disk(kobj_to_dev(kobj));
955 struct hd_struct *part;
957 spin_lock_bh(&ext_devt_lock);
958 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
959 if (part && get_disk_and_module(part_to_disk(part))) {
960 *partno = part->partno;
961 disk = part_to_disk(part);
963 spin_unlock_bh(&ext_devt_lock);
970 * Synchronize with del_gendisk() to not return disk that is being
973 down_read(&disk->lookup_sem);
974 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
975 !(disk->flags & GENHD_FL_UP))) {
976 up_read(&disk->lookup_sem);
977 put_disk_and_module(disk);
980 up_read(&disk->lookup_sem);
984 EXPORT_SYMBOL(get_gendisk);
987 * bdget_disk - do bdget() by gendisk and partition number
988 * @disk: gendisk of interest
989 * @partno: partition number
991 * Find partition @partno from @disk, do bdget() on it.
997 * Resulting block_device on success, NULL on failure.
999 struct block_device *bdget_disk(struct gendisk *disk, int partno)
1001 struct hd_struct *part;
1002 struct block_device *bdev = NULL;
1004 part = disk_get_part(disk, partno);
1006 bdev = bdget(part_devt(part));
1007 disk_put_part(part);
1011 EXPORT_SYMBOL(bdget_disk);
1014 * print a full list of all partitions - intended for places where the root
1015 * filesystem can't be mounted and thus to give the victim some idea of what
1018 void __init printk_all_partitions(void)
1020 struct class_dev_iter iter;
1023 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1024 while ((dev = class_dev_iter_next(&iter))) {
1025 struct gendisk *disk = dev_to_disk(dev);
1026 struct disk_part_iter piter;
1027 struct hd_struct *part;
1028 char name_buf[BDEVNAME_SIZE];
1029 char devt_buf[BDEVT_SIZE];
1032 * Don't show empty devices or things that have been
1035 if (get_capacity(disk) == 0 ||
1036 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1040 * Note, unlike /proc/partitions, I am showing the
1041 * numbers in hex - the same format as the root=
1044 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1045 while ((part = disk_part_iter_next(&piter))) {
1046 bool is_part0 = part == &disk->part0;
1048 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1049 bdevt_str(part_devt(part), devt_buf),
1050 (unsigned long long)part_nr_sects_read(part) >> 1
1051 , disk_name(disk, part->partno, name_buf),
1052 part->info ? part->info->uuid : "");
1054 if (dev->parent && dev->parent->driver)
1055 printk(" driver: %s\n",
1056 dev->parent->driver->name);
1058 printk(" (driver?)\n");
1062 disk_part_iter_exit(&piter);
1064 class_dev_iter_exit(&iter);
1067 #ifdef CONFIG_PROC_FS
1069 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1072 struct class_dev_iter *iter;
1075 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1077 return ERR_PTR(-ENOMEM);
1079 seqf->private = iter;
1080 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1082 dev = class_dev_iter_next(iter);
1087 return dev_to_disk(dev);
1090 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1095 dev = class_dev_iter_next(seqf->private);
1097 return dev_to_disk(dev);
1102 static void disk_seqf_stop(struct seq_file *seqf, void *v)
1104 struct class_dev_iter *iter = seqf->private;
1106 /* stop is called even after start failed :-( */
1108 class_dev_iter_exit(iter);
1110 seqf->private = NULL;
1114 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1118 p = disk_seqf_start(seqf, pos);
1119 if (!IS_ERR_OR_NULL(p) && !*pos)
1120 seq_puts(seqf, "major minor #blocks name\n\n");
1124 static int show_partition(struct seq_file *seqf, void *v)
1126 struct gendisk *sgp = v;
1127 struct disk_part_iter piter;
1128 struct hd_struct *part;
1129 char buf[BDEVNAME_SIZE];
1131 /* Don't show non-partitionable removeable devices or empty devices */
1132 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1133 (sgp->flags & GENHD_FL_REMOVABLE)))
1135 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1138 /* show the full disk and all non-0 size partitions of it */
1139 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1140 while ((part = disk_part_iter_next(&piter)))
1141 seq_printf(seqf, "%4d %7d %10llu %s\n",
1142 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1143 (unsigned long long)part_nr_sects_read(part) >> 1,
1144 disk_name(sgp, part->partno, buf));
1145 disk_part_iter_exit(&piter);
1150 static const struct seq_operations partitions_op = {
1151 .start = show_partition_start,
1152 .next = disk_seqf_next,
1153 .stop = disk_seqf_stop,
1154 .show = show_partition
1159 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1161 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1162 /* Make old-style 2.4 aliases work */
1163 request_module("block-major-%d", MAJOR(devt));
1167 static int __init genhd_device_init(void)
1171 block_class.dev_kobj = sysfs_dev_block_kobj;
1172 error = class_register(&block_class);
1173 if (unlikely(error))
1175 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1178 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1180 /* create top-level block dir */
1181 if (!sysfs_deprecated)
1182 block_depr = kobject_create_and_add("block", NULL);
1186 subsys_initcall(genhd_device_init);
1188 static ssize_t disk_range_show(struct device *dev,
1189 struct device_attribute *attr, char *buf)
1191 struct gendisk *disk = dev_to_disk(dev);
1193 return sprintf(buf, "%d\n", disk->minors);
1196 static ssize_t disk_ext_range_show(struct device *dev,
1197 struct device_attribute *attr, char *buf)
1199 struct gendisk *disk = dev_to_disk(dev);
1201 return sprintf(buf, "%d\n", disk_max_parts(disk));
1204 static ssize_t disk_removable_show(struct device *dev,
1205 struct device_attribute *attr, char *buf)
1207 struct gendisk *disk = dev_to_disk(dev);
1209 return sprintf(buf, "%d\n",
1210 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1213 static ssize_t disk_hidden_show(struct device *dev,
1214 struct device_attribute *attr, char *buf)
1216 struct gendisk *disk = dev_to_disk(dev);
1218 return sprintf(buf, "%d\n",
1219 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1222 static ssize_t disk_ro_show(struct device *dev,
1223 struct device_attribute *attr, char *buf)
1225 struct gendisk *disk = dev_to_disk(dev);
1227 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1230 ssize_t part_size_show(struct device *dev,
1231 struct device_attribute *attr, char *buf)
1233 struct hd_struct *p = dev_to_part(dev);
1235 return sprintf(buf, "%llu\n",
1236 (unsigned long long)part_nr_sects_read(p));
1239 ssize_t part_stat_show(struct device *dev,
1240 struct device_attribute *attr, char *buf)
1242 struct hd_struct *p = dev_to_part(dev);
1243 struct request_queue *q = part_to_disk(p)->queue;
1244 struct disk_stats stat;
1245 unsigned int inflight;
1247 part_stat_read_all(p, &stat);
1248 inflight = part_in_flight(q, p);
1251 "%8lu %8lu %8llu %8u "
1252 "%8lu %8lu %8llu %8u "
1254 "%8lu %8lu %8llu %8u "
1257 stat.ios[STAT_READ],
1258 stat.merges[STAT_READ],
1259 (unsigned long long)stat.sectors[STAT_READ],
1260 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1261 stat.ios[STAT_WRITE],
1262 stat.merges[STAT_WRITE],
1263 (unsigned long long)stat.sectors[STAT_WRITE],
1264 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1266 jiffies_to_msecs(stat.io_ticks),
1267 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1268 stat.nsecs[STAT_WRITE] +
1269 stat.nsecs[STAT_DISCARD] +
1270 stat.nsecs[STAT_FLUSH],
1272 stat.ios[STAT_DISCARD],
1273 stat.merges[STAT_DISCARD],
1274 (unsigned long long)stat.sectors[STAT_DISCARD],
1275 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1276 stat.ios[STAT_FLUSH],
1277 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1280 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1283 struct hd_struct *p = dev_to_part(dev);
1284 struct request_queue *q = part_to_disk(p)->queue;
1285 unsigned int inflight[2];
1287 part_in_flight_rw(q, p, inflight);
1288 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1291 static ssize_t disk_capability_show(struct device *dev,
1292 struct device_attribute *attr, char *buf)
1294 struct gendisk *disk = dev_to_disk(dev);
1296 return sprintf(buf, "%x\n", disk->flags);
1299 static ssize_t disk_alignment_offset_show(struct device *dev,
1300 struct device_attribute *attr,
1303 struct gendisk *disk = dev_to_disk(dev);
1305 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1308 static ssize_t disk_discard_alignment_show(struct device *dev,
1309 struct device_attribute *attr,
1312 struct gendisk *disk = dev_to_disk(dev);
1314 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1317 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1318 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1319 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1320 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1321 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1322 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1323 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1324 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1325 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1326 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1327 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1328 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1330 #ifdef CONFIG_FAIL_MAKE_REQUEST
1331 ssize_t part_fail_show(struct device *dev,
1332 struct device_attribute *attr, char *buf)
1334 struct hd_struct *p = dev_to_part(dev);
1336 return sprintf(buf, "%d\n", p->make_it_fail);
1339 ssize_t part_fail_store(struct device *dev,
1340 struct device_attribute *attr,
1341 const char *buf, size_t count)
1343 struct hd_struct *p = dev_to_part(dev);
1346 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1347 p->make_it_fail = (i == 0) ? 0 : 1;
1352 static struct device_attribute dev_attr_fail =
1353 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1354 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1356 #ifdef CONFIG_FAIL_IO_TIMEOUT
1357 static struct device_attribute dev_attr_fail_timeout =
1358 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1361 static struct attribute *disk_attrs[] = {
1362 &dev_attr_range.attr,
1363 &dev_attr_ext_range.attr,
1364 &dev_attr_removable.attr,
1365 &dev_attr_hidden.attr,
1367 &dev_attr_size.attr,
1368 &dev_attr_alignment_offset.attr,
1369 &dev_attr_discard_alignment.attr,
1370 &dev_attr_capability.attr,
1371 &dev_attr_stat.attr,
1372 &dev_attr_inflight.attr,
1373 &dev_attr_badblocks.attr,
1374 #ifdef CONFIG_FAIL_MAKE_REQUEST
1375 &dev_attr_fail.attr,
1377 #ifdef CONFIG_FAIL_IO_TIMEOUT
1378 &dev_attr_fail_timeout.attr,
1383 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1385 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1386 struct gendisk *disk = dev_to_disk(dev);
1388 if (a == &dev_attr_badblocks.attr && !disk->bb)
1393 static struct attribute_group disk_attr_group = {
1394 .attrs = disk_attrs,
1395 .is_visible = disk_visible,
1398 static const struct attribute_group *disk_attr_groups[] = {
1404 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1405 * @disk: disk to replace part_tbl for
1406 * @new_ptbl: new part_tbl to install
1408 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1409 * original ptbl is freed using RCU callback.
1412 * Matching bd_mutex locked or the caller is the only user of @disk.
1414 static void disk_replace_part_tbl(struct gendisk *disk,
1415 struct disk_part_tbl *new_ptbl)
1417 struct disk_part_tbl *old_ptbl =
1418 rcu_dereference_protected(disk->part_tbl, 1);
1420 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1423 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1424 kfree_rcu(old_ptbl, rcu_head);
1429 * disk_expand_part_tbl - expand disk->part_tbl
1430 * @disk: disk to expand part_tbl for
1431 * @partno: expand such that this partno can fit in
1433 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1434 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1437 * Matching bd_mutex locked or the caller is the only user of @disk.
1441 * 0 on success, -errno on failure.
1443 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1445 struct disk_part_tbl *old_ptbl =
1446 rcu_dereference_protected(disk->part_tbl, 1);
1447 struct disk_part_tbl *new_ptbl;
1448 int len = old_ptbl ? old_ptbl->len : 0;
1452 * check for int overflow, since we can get here from blkpg_ioctl()
1453 * with a user passed 'partno'.
1455 target = partno + 1;
1459 /* disk_max_parts() is zero during initialization, ignore if so */
1460 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1466 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1471 new_ptbl->len = target;
1473 for (i = 0; i < len; i++)
1474 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1476 disk_replace_part_tbl(disk, new_ptbl);
1480 static void disk_release(struct device *dev)
1482 struct gendisk *disk = dev_to_disk(dev);
1484 blk_free_devt(dev->devt);
1485 disk_release_events(disk);
1486 kfree(disk->random);
1487 disk_replace_part_tbl(disk, NULL);
1488 hd_free_part(&disk->part0);
1490 blk_put_queue(disk->queue);
1493 struct class block_class = {
1497 static char *block_devnode(struct device *dev, umode_t *mode,
1498 kuid_t *uid, kgid_t *gid)
1500 struct gendisk *disk = dev_to_disk(dev);
1503 return disk->devnode(disk, mode);
1507 static const struct device_type disk_type = {
1509 .groups = disk_attr_groups,
1510 .release = disk_release,
1511 .devnode = block_devnode,
1514 #ifdef CONFIG_PROC_FS
1516 * aggregate disk stat collector. Uses the same stats that the sysfs
1517 * entries do, above, but makes them available through one seq_file.
1519 * The output looks suspiciously like /proc/partitions with a bunch of
1522 static int diskstats_show(struct seq_file *seqf, void *v)
1524 struct gendisk *gp = v;
1525 struct disk_part_iter piter;
1526 struct hd_struct *hd;
1527 char buf[BDEVNAME_SIZE];
1528 unsigned int inflight;
1529 struct disk_stats stat;
1532 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1533 seq_puts(seqf, "major minor name"
1534 " rio rmerge rsect ruse wio wmerge "
1535 "wsect wuse running use aveq"
1539 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1540 while ((hd = disk_part_iter_next(&piter))) {
1541 part_stat_read_all(hd, &stat);
1542 inflight = part_in_flight(gp->queue, hd);
1544 seq_printf(seqf, "%4d %7d %s "
1551 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1552 disk_name(gp, hd->partno, buf),
1553 stat.ios[STAT_READ],
1554 stat.merges[STAT_READ],
1555 stat.sectors[STAT_READ],
1556 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1558 stat.ios[STAT_WRITE],
1559 stat.merges[STAT_WRITE],
1560 stat.sectors[STAT_WRITE],
1561 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1564 jiffies_to_msecs(stat.io_ticks),
1565 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1566 stat.nsecs[STAT_WRITE] +
1567 stat.nsecs[STAT_DISCARD] +
1568 stat.nsecs[STAT_FLUSH],
1570 stat.ios[STAT_DISCARD],
1571 stat.merges[STAT_DISCARD],
1572 stat.sectors[STAT_DISCARD],
1573 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1575 stat.ios[STAT_FLUSH],
1576 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1580 disk_part_iter_exit(&piter);
1585 static const struct seq_operations diskstats_op = {
1586 .start = disk_seqf_start,
1587 .next = disk_seqf_next,
1588 .stop = disk_seqf_stop,
1589 .show = diskstats_show
1592 static int __init proc_genhd_init(void)
1594 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1595 proc_create_seq("partitions", 0, NULL, &partitions_op);
1598 module_init(proc_genhd_init);
1599 #endif /* CONFIG_PROC_FS */
1601 dev_t blk_lookup_devt(const char *name, int partno)
1603 dev_t devt = MKDEV(0, 0);
1604 struct class_dev_iter iter;
1607 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1608 while ((dev = class_dev_iter_next(&iter))) {
1609 struct gendisk *disk = dev_to_disk(dev);
1610 struct hd_struct *part;
1612 if (strcmp(dev_name(dev), name))
1615 if (partno < disk->minors) {
1616 /* We need to return the right devno, even
1617 * if the partition doesn't exist yet.
1619 devt = MKDEV(MAJOR(dev->devt),
1620 MINOR(dev->devt) + partno);
1623 part = disk_get_part(disk, partno);
1625 devt = part_devt(part);
1626 disk_put_part(part);
1629 disk_put_part(part);
1631 class_dev_iter_exit(&iter);
1635 struct gendisk *__alloc_disk_node(int minors, int node_id)
1637 struct gendisk *disk;
1638 struct disk_part_tbl *ptbl;
1640 if (minors > DISK_MAX_PARTS) {
1642 "block: can't allocate more than %d partitions\n",
1644 minors = DISK_MAX_PARTS;
1647 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1649 if (!init_part_stats(&disk->part0)) {
1653 init_rwsem(&disk->lookup_sem);
1654 disk->node_id = node_id;
1655 if (disk_expand_part_tbl(disk, 0)) {
1656 free_part_stats(&disk->part0);
1660 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1661 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1664 * set_capacity() and get_capacity() currently don't use
1665 * seqcounter to read/update the part0->nr_sects. Still init
1666 * the counter as we can read the sectors in IO submission
1667 * patch using seqence counters.
1669 * TODO: Ideally set_capacity() and get_capacity() should be
1670 * converted to make use of bd_mutex and sequence counters.
1672 seqcount_init(&disk->part0.nr_sects_seq);
1673 if (hd_ref_init(&disk->part0)) {
1674 hd_free_part(&disk->part0);
1679 disk->minors = minors;
1680 rand_initialize_disk(disk);
1681 disk_to_dev(disk)->class = &block_class;
1682 disk_to_dev(disk)->type = &disk_type;
1683 device_initialize(disk_to_dev(disk));
1687 EXPORT_SYMBOL(__alloc_disk_node);
1689 struct kobject *get_disk_and_module(struct gendisk *disk)
1691 struct module *owner;
1692 struct kobject *kobj;
1696 owner = disk->fops->owner;
1697 if (owner && !try_module_get(owner))
1699 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1707 EXPORT_SYMBOL(get_disk_and_module);
1709 void put_disk(struct gendisk *disk)
1712 kobject_put(&disk_to_dev(disk)->kobj);
1714 EXPORT_SYMBOL(put_disk);
1717 * This is a counterpart of get_disk_and_module() and thus also of
1720 void put_disk_and_module(struct gendisk *disk)
1723 struct module *owner = disk->fops->owner;
1729 EXPORT_SYMBOL(put_disk_and_module);
1731 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1733 char event[] = "DISK_RO=1";
1734 char *envp[] = { event, NULL };
1738 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1741 void set_device_ro(struct block_device *bdev, int flag)
1743 bdev->bd_part->policy = flag;
1746 EXPORT_SYMBOL(set_device_ro);
1748 void set_disk_ro(struct gendisk *disk, int flag)
1750 struct disk_part_iter piter;
1751 struct hd_struct *part;
1753 if (disk->part0.policy != flag) {
1754 set_disk_ro_uevent(disk, flag);
1755 disk->part0.policy = flag;
1758 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1759 while ((part = disk_part_iter_next(&piter)))
1760 part->policy = flag;
1761 disk_part_iter_exit(&piter);
1764 EXPORT_SYMBOL(set_disk_ro);
1766 int bdev_read_only(struct block_device *bdev)
1770 return bdev->bd_part->policy;
1773 EXPORT_SYMBOL(bdev_read_only);
1775 int invalidate_partition(struct gendisk *disk, int partno)
1778 struct block_device *bdev = bdget_disk(disk, partno);
1781 res = __invalidate_device(bdev, true);
1787 EXPORT_SYMBOL(invalidate_partition);
1790 * Disk events - monitor disk events like media change and eject request.
1792 struct disk_events {
1793 struct list_head node; /* all disk_event's */
1794 struct gendisk *disk; /* the associated disk */
1797 struct mutex block_mutex; /* protects blocking */
1798 int block; /* event blocking depth */
1799 unsigned int pending; /* events already sent out */
1800 unsigned int clearing; /* events being cleared */
1802 long poll_msecs; /* interval, -1 for default */
1803 struct delayed_work dwork;
1806 static const char *disk_events_strs[] = {
1807 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1808 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1811 static char *disk_uevents[] = {
1812 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1813 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1816 /* list of all disk_events */
1817 static DEFINE_MUTEX(disk_events_mutex);
1818 static LIST_HEAD(disk_events);
1820 /* disable in-kernel polling by default */
1821 static unsigned long disk_events_dfl_poll_msecs;
1823 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1825 struct disk_events *ev = disk->ev;
1826 long intv_msecs = 0;
1829 * If device-specific poll interval is set, always use it. If
1830 * the default is being used, poll if the POLL flag is set.
1832 if (ev->poll_msecs >= 0)
1833 intv_msecs = ev->poll_msecs;
1834 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1835 intv_msecs = disk_events_dfl_poll_msecs;
1837 return msecs_to_jiffies(intv_msecs);
1841 * disk_block_events - block and flush disk event checking
1842 * @disk: disk to block events for
1844 * On return from this function, it is guaranteed that event checking
1845 * isn't in progress and won't happen until unblocked by
1846 * disk_unblock_events(). Events blocking is counted and the actual
1847 * unblocking happens after the matching number of unblocks are done.
1849 * Note that this intentionally does not block event checking from
1850 * disk_clear_events().
1855 void disk_block_events(struct gendisk *disk)
1857 struct disk_events *ev = disk->ev;
1858 unsigned long flags;
1865 * Outer mutex ensures that the first blocker completes canceling
1866 * the event work before further blockers are allowed to finish.
1868 mutex_lock(&ev->block_mutex);
1870 spin_lock_irqsave(&ev->lock, flags);
1871 cancel = !ev->block++;
1872 spin_unlock_irqrestore(&ev->lock, flags);
1875 cancel_delayed_work_sync(&disk->ev->dwork);
1877 mutex_unlock(&ev->block_mutex);
1880 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1882 struct disk_events *ev = disk->ev;
1884 unsigned long flags;
1886 spin_lock_irqsave(&ev->lock, flags);
1888 if (WARN_ON_ONCE(ev->block <= 0))
1894 intv = disk_events_poll_jiffies(disk);
1896 queue_delayed_work(system_freezable_power_efficient_wq,
1899 queue_delayed_work(system_freezable_power_efficient_wq,
1902 spin_unlock_irqrestore(&ev->lock, flags);
1906 * disk_unblock_events - unblock disk event checking
1907 * @disk: disk to unblock events for
1909 * Undo disk_block_events(). When the block count reaches zero, it
1910 * starts events polling if configured.
1913 * Don't care. Safe to call from irq context.
1915 void disk_unblock_events(struct gendisk *disk)
1918 __disk_unblock_events(disk, false);
1922 * disk_flush_events - schedule immediate event checking and flushing
1923 * @disk: disk to check and flush events for
1924 * @mask: events to flush
1926 * Schedule immediate event checking on @disk if not blocked. Events in
1927 * @mask are scheduled to be cleared from the driver. Note that this
1928 * doesn't clear the events from @disk->ev.
1931 * If @mask is non-zero must be called with bdev->bd_mutex held.
1933 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1935 struct disk_events *ev = disk->ev;
1940 spin_lock_irq(&ev->lock);
1941 ev->clearing |= mask;
1943 mod_delayed_work(system_freezable_power_efficient_wq,
1945 spin_unlock_irq(&ev->lock);
1949 * disk_clear_events - synchronously check, clear and return pending events
1950 * @disk: disk to fetch and clear events from
1951 * @mask: mask of events to be fetched and cleared
1953 * Disk events are synchronously checked and pending events in @mask
1954 * are cleared and returned. This ignores the block count.
1959 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1961 const struct block_device_operations *bdops = disk->fops;
1962 struct disk_events *ev = disk->ev;
1963 unsigned int pending;
1964 unsigned int clearing = mask;
1967 /* for drivers still using the old ->media_changed method */
1968 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1969 bdops->media_changed && bdops->media_changed(disk))
1970 return DISK_EVENT_MEDIA_CHANGE;
1974 disk_block_events(disk);
1977 * store the union of mask and ev->clearing on the stack so that the
1978 * race with disk_flush_events does not cause ambiguity (ev->clearing
1979 * can still be modified even if events are blocked).
1981 spin_lock_irq(&ev->lock);
1982 clearing |= ev->clearing;
1984 spin_unlock_irq(&ev->lock);
1986 disk_check_events(ev, &clearing);
1988 * if ev->clearing is not 0, the disk_flush_events got called in the
1989 * middle of this function, so we want to run the workfn without delay.
1991 __disk_unblock_events(disk, ev->clearing ? true : false);
1993 /* then, fetch and clear pending events */
1994 spin_lock_irq(&ev->lock);
1995 pending = ev->pending & mask;
1996 ev->pending &= ~mask;
1997 spin_unlock_irq(&ev->lock);
1998 WARN_ON_ONCE(clearing & mask);
2004 * Separate this part out so that a different pointer for clearing_ptr can be
2005 * passed in for disk_clear_events.
2007 static void disk_events_workfn(struct work_struct *work)
2009 struct delayed_work *dwork = to_delayed_work(work);
2010 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2012 disk_check_events(ev, &ev->clearing);
2015 static void disk_check_events(struct disk_events *ev,
2016 unsigned int *clearing_ptr)
2018 struct gendisk *disk = ev->disk;
2019 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2020 unsigned int clearing = *clearing_ptr;
2021 unsigned int events;
2023 int nr_events = 0, i;
2026 events = disk->fops->check_events(disk, clearing);
2028 /* accumulate pending events and schedule next poll if necessary */
2029 spin_lock_irq(&ev->lock);
2031 events &= ~ev->pending;
2032 ev->pending |= events;
2033 *clearing_ptr &= ~clearing;
2035 intv = disk_events_poll_jiffies(disk);
2036 if (!ev->block && intv)
2037 queue_delayed_work(system_freezable_power_efficient_wq,
2040 spin_unlock_irq(&ev->lock);
2043 * Tell userland about new events. Only the events listed in
2044 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2045 * is set. Otherwise, events are processed internally but never
2046 * get reported to userland.
2048 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2049 if ((events & disk->events & (1 << i)) &&
2050 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2051 envp[nr_events++] = disk_uevents[i];
2054 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2058 * A disk events enabled device has the following sysfs nodes under
2059 * its /sys/block/X/ directory.
2061 * events : list of all supported events
2062 * events_async : list of events which can be detected w/o polling
2063 * (always empty, only for backwards compatibility)
2064 * events_poll_msecs : polling interval, 0: disable, -1: system default
2066 static ssize_t __disk_events_show(unsigned int events, char *buf)
2068 const char *delim = "";
2072 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2073 if (events & (1 << i)) {
2074 pos += sprintf(buf + pos, "%s%s",
2075 delim, disk_events_strs[i]);
2079 pos += sprintf(buf + pos, "\n");
2083 static ssize_t disk_events_show(struct device *dev,
2084 struct device_attribute *attr, char *buf)
2086 struct gendisk *disk = dev_to_disk(dev);
2088 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2091 return __disk_events_show(disk->events, buf);
2094 static ssize_t disk_events_async_show(struct device *dev,
2095 struct device_attribute *attr, char *buf)
2100 static ssize_t disk_events_poll_msecs_show(struct device *dev,
2101 struct device_attribute *attr,
2104 struct gendisk *disk = dev_to_disk(dev);
2107 return sprintf(buf, "-1\n");
2109 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2112 static ssize_t disk_events_poll_msecs_store(struct device *dev,
2113 struct device_attribute *attr,
2114 const char *buf, size_t count)
2116 struct gendisk *disk = dev_to_disk(dev);
2119 if (!count || !sscanf(buf, "%ld", &intv))
2122 if (intv < 0 && intv != -1)
2128 disk_block_events(disk);
2129 disk->ev->poll_msecs = intv;
2130 __disk_unblock_events(disk, true);
2135 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2136 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2137 static const DEVICE_ATTR(events_poll_msecs, 0644,
2138 disk_events_poll_msecs_show,
2139 disk_events_poll_msecs_store);
2141 static const struct attribute *disk_events_attrs[] = {
2142 &dev_attr_events.attr,
2143 &dev_attr_events_async.attr,
2144 &dev_attr_events_poll_msecs.attr,
2149 * The default polling interval can be specified by the kernel
2150 * parameter block.events_dfl_poll_msecs which defaults to 0
2151 * (disable). This can also be modified runtime by writing to
2152 * /sys/module/block/parameters/events_dfl_poll_msecs.
2154 static int disk_events_set_dfl_poll_msecs(const char *val,
2155 const struct kernel_param *kp)
2157 struct disk_events *ev;
2160 ret = param_set_ulong(val, kp);
2164 mutex_lock(&disk_events_mutex);
2166 list_for_each_entry(ev, &disk_events, node)
2167 disk_flush_events(ev->disk, 0);
2169 mutex_unlock(&disk_events_mutex);
2174 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2175 .set = disk_events_set_dfl_poll_msecs,
2176 .get = param_get_ulong,
2179 #undef MODULE_PARAM_PREFIX
2180 #define MODULE_PARAM_PREFIX "block."
2182 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2183 &disk_events_dfl_poll_msecs, 0644);
2186 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2188 static void disk_alloc_events(struct gendisk *disk)
2190 struct disk_events *ev;
2192 if (!disk->fops->check_events || !disk->events)
2195 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2197 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2201 INIT_LIST_HEAD(&ev->node);
2203 spin_lock_init(&ev->lock);
2204 mutex_init(&ev->block_mutex);
2206 ev->poll_msecs = -1;
2207 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2212 static void disk_add_events(struct gendisk *disk)
2214 /* FIXME: error handling */
2215 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2216 pr_warn("%s: failed to create sysfs files for events\n",
2222 mutex_lock(&disk_events_mutex);
2223 list_add_tail(&disk->ev->node, &disk_events);
2224 mutex_unlock(&disk_events_mutex);
2227 * Block count is initialized to 1 and the following initial
2228 * unblock kicks it into action.
2230 __disk_unblock_events(disk, true);
2233 static void disk_del_events(struct gendisk *disk)
2236 disk_block_events(disk);
2238 mutex_lock(&disk_events_mutex);
2239 list_del_init(&disk->ev->node);
2240 mutex_unlock(&disk_events_mutex);
2243 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2246 static void disk_release_events(struct gendisk *disk)
2248 /* the block count should be 1 from disk_del_events() */
2249 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);