5 #include <linux/module.h>
7 #include <linux/genhd.h>
8 #include <linux/kdev_t.h>
9 #include <linux/kernel.h>
10 #include <linux/blkdev.h>
11 #include <linux/init.h>
12 #include <linux/spinlock.h>
13 #include <linux/proc_fs.h>
14 #include <linux/seq_file.h>
15 #include <linux/slab.h>
16 #include <linux/kmod.h>
17 #include <linux/kobj_map.h>
18 #include <linux/buffer_head.h>
19 #include <linux/mutex.h>
20 #include <linux/idr.h>
21 #include <linux/log2.h>
25 static DEFINE_MUTEX(block_class_lock);
26 struct kobject *block_depr;
28 /* for extended dynamic devt allocation, currently only one major is used */
29 #define MAX_EXT_DEVT (1 << MINORBITS)
31 /* For extended devt allocation. ext_devt_mutex prevents look up
32 * results from going away underneath its user.
34 static DEFINE_MUTEX(ext_devt_mutex);
35 static DEFINE_IDR(ext_devt_idr);
37 static struct device_type disk_type;
39 static void disk_add_events(struct gendisk *disk);
40 static void disk_del_events(struct gendisk *disk);
41 static void disk_release_events(struct gendisk *disk);
44 * disk_get_part - get partition
45 * @disk: disk to look partition from
46 * @partno: partition number
48 * Look for partition @partno from @disk. If found, increment
49 * reference count and return it.
55 * Pointer to the found partition on success, NULL if not found.
57 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
59 struct hd_struct *part = NULL;
60 struct disk_part_tbl *ptbl;
62 if (unlikely(partno < 0))
67 ptbl = rcu_dereference(disk->part_tbl);
68 if (likely(partno < ptbl->len)) {
69 part = rcu_dereference(ptbl->part[partno]);
71 get_device(part_to_dev(part));
78 EXPORT_SYMBOL_GPL(disk_get_part);
81 * disk_part_iter_init - initialize partition iterator
82 * @piter: iterator to initialize
83 * @disk: disk to iterate over
84 * @flags: DISK_PITER_* flags
86 * Initialize @piter so that it iterates over partitions of @disk.
91 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
94 struct disk_part_tbl *ptbl;
97 ptbl = rcu_dereference(disk->part_tbl);
102 if (flags & DISK_PITER_REVERSE)
103 piter->idx = ptbl->len - 1;
104 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
109 piter->flags = flags;
113 EXPORT_SYMBOL_GPL(disk_part_iter_init);
116 * disk_part_iter_next - proceed iterator to the next partition and return it
117 * @piter: iterator of interest
119 * Proceed @piter to the next partition and return it.
124 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
126 struct disk_part_tbl *ptbl;
129 /* put the last partition */
130 disk_put_part(piter->part);
135 ptbl = rcu_dereference(piter->disk->part_tbl);
137 /* determine iteration parameters */
138 if (piter->flags & DISK_PITER_REVERSE) {
140 if (piter->flags & (DISK_PITER_INCL_PART0 |
141 DISK_PITER_INCL_EMPTY_PART0))
150 /* iterate to the next partition */
151 for (; piter->idx != end; piter->idx += inc) {
152 struct hd_struct *part;
154 part = rcu_dereference(ptbl->part[piter->idx]);
157 if (!part->nr_sects &&
158 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
159 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
163 get_device(part_to_dev(part));
173 EXPORT_SYMBOL_GPL(disk_part_iter_next);
176 * disk_part_iter_exit - finish up partition iteration
177 * @piter: iter of interest
179 * Called when iteration is over. Cleans up @piter.
184 void disk_part_iter_exit(struct disk_part_iter *piter)
186 disk_put_part(piter->part);
189 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
191 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
193 return part->start_sect <= sector &&
194 sector < part->start_sect + part->nr_sects;
198 * disk_map_sector_rcu - map sector to partition
199 * @disk: gendisk of interest
200 * @sector: sector to map
202 * Find out which partition @sector maps to on @disk. This is
203 * primarily used for stats accounting.
206 * RCU read locked. The returned partition pointer is valid only
207 * while preemption is disabled.
210 * Found partition on success, part0 is returned if no partition matches
212 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
214 struct disk_part_tbl *ptbl;
215 struct hd_struct *part;
218 ptbl = rcu_dereference(disk->part_tbl);
220 part = rcu_dereference(ptbl->last_lookup);
221 if (part && sector_in_part(part, sector))
224 for (i = 1; i < ptbl->len; i++) {
225 part = rcu_dereference(ptbl->part[i]);
227 if (part && sector_in_part(part, sector)) {
228 rcu_assign_pointer(ptbl->last_lookup, part);
234 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
237 * Can be deleted altogether. Later.
240 static struct blk_major_name {
241 struct blk_major_name *next;
244 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
246 /* index in the above - for now: assume no multimajor ranges */
247 static inline int major_to_index(unsigned major)
249 return major % BLKDEV_MAJOR_HASH_SIZE;
252 #ifdef CONFIG_PROC_FS
253 void blkdev_show(struct seq_file *seqf, off_t offset)
255 struct blk_major_name *dp;
257 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
258 mutex_lock(&block_class_lock);
259 for (dp = major_names[offset]; dp; dp = dp->next)
260 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
261 mutex_unlock(&block_class_lock);
264 #endif /* CONFIG_PROC_FS */
267 * register_blkdev - register a new block device
269 * @major: the requested major device number [1..255]. If @major=0, try to
270 * allocate any unused major number.
271 * @name: the name of the new block device as a zero terminated string
273 * The @name must be unique within the system.
275 * The return value depends on the @major input parameter.
276 * - if a major device number was requested in range [1..255] then the
277 * function returns zero on success, or a negative error code
278 * - if any unused major number was requested with @major=0 parameter
279 * then the return value is the allocated major number in range
280 * [1..255] or a negative error code otherwise
282 int register_blkdev(unsigned int major, const char *name)
284 struct blk_major_name **n, *p;
287 mutex_lock(&block_class_lock);
291 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
292 if (major_names[index] == NULL)
297 printk("register_blkdev: failed to get major for %s\n",
306 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
313 strlcpy(p->name, name, sizeof(p->name));
315 index = major_to_index(major);
317 for (n = &major_names[index]; *n; n = &(*n)->next) {
318 if ((*n)->major == major)
327 printk("register_blkdev: cannot get major %d for %s\n",
332 mutex_unlock(&block_class_lock);
336 EXPORT_SYMBOL(register_blkdev);
338 void unregister_blkdev(unsigned int major, const char *name)
340 struct blk_major_name **n;
341 struct blk_major_name *p = NULL;
342 int index = major_to_index(major);
344 mutex_lock(&block_class_lock);
345 for (n = &major_names[index]; *n; n = &(*n)->next)
346 if ((*n)->major == major)
348 if (!*n || strcmp((*n)->name, name)) {
354 mutex_unlock(&block_class_lock);
358 EXPORT_SYMBOL(unregister_blkdev);
360 static struct kobj_map *bdev_map;
363 * blk_mangle_minor - scatter minor numbers apart
364 * @minor: minor number to mangle
366 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
367 * is enabled. Mangling twice gives the original value.
375 static int blk_mangle_minor(int minor)
377 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
380 for (i = 0; i < MINORBITS / 2; i++) {
381 int low = minor & (1 << i);
382 int high = minor & (1 << (MINORBITS - 1 - i));
383 int distance = MINORBITS - 1 - 2 * i;
385 minor ^= low | high; /* clear both bits */
386 low <<= distance; /* swap the positions */
388 minor |= low | high; /* and set */
395 * blk_alloc_devt - allocate a dev_t for a partition
396 * @part: partition to allocate dev_t for
397 * @devt: out parameter for resulting dev_t
399 * Allocate a dev_t for block device.
402 * 0 on success, allocated dev_t is returned in *@devt. -errno on
408 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
410 struct gendisk *disk = part_to_disk(part);
413 /* in consecutive minor range? */
414 if (part->partno < disk->minors) {
415 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
419 /* allocate ext devt */
421 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
423 rc = idr_get_new(&ext_devt_idr, part, &idx);
424 } while (rc == -EAGAIN);
429 if (idx > MAX_EXT_DEVT) {
430 idr_remove(&ext_devt_idr, idx);
434 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
439 * blk_free_devt - free a dev_t
440 * @devt: dev_t to free
442 * Free @devt which was allocated using blk_alloc_devt().
447 void blk_free_devt(dev_t devt)
451 if (devt == MKDEV(0, 0))
454 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
455 mutex_lock(&ext_devt_mutex);
456 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
457 mutex_unlock(&ext_devt_mutex);
461 static char *bdevt_str(dev_t devt, char *buf)
463 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
464 char tbuf[BDEVT_SIZE];
465 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
466 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
468 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
474 * Register device numbers dev..(dev+range-1)
475 * range must be nonzero
476 * The hash chain is sorted on range, so that subranges can override.
478 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
479 struct kobject *(*probe)(dev_t, int *, void *),
480 int (*lock)(dev_t, void *), void *data)
482 kobj_map(bdev_map, devt, range, module, probe, lock, data);
485 EXPORT_SYMBOL(blk_register_region);
487 void blk_unregister_region(dev_t devt, unsigned long range)
489 kobj_unmap(bdev_map, devt, range);
492 EXPORT_SYMBOL(blk_unregister_region);
494 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
496 struct gendisk *p = data;
498 return &disk_to_dev(p)->kobj;
501 static int exact_lock(dev_t devt, void *data)
503 struct gendisk *p = data;
510 void register_disk(struct gendisk *disk)
512 struct device *ddev = disk_to_dev(disk);
513 struct block_device *bdev;
514 struct disk_part_iter piter;
515 struct hd_struct *part;
518 ddev->parent = disk->driverfs_dev;
520 dev_set_name(ddev, disk->disk_name);
522 /* delay uevents, until we scanned partition table */
523 dev_set_uevent_suppress(ddev, 1);
525 if (device_add(ddev))
527 if (!sysfs_deprecated) {
528 err = sysfs_create_link(block_depr, &ddev->kobj,
529 kobject_name(&ddev->kobj));
535 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
536 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
538 /* No minors to use for partitions */
539 if (!disk_partitionable(disk))
542 /* No such device (e.g., media were just removed) */
543 if (!get_capacity(disk))
546 bdev = bdget_disk(disk, 0);
550 bdev->bd_invalidated = 1;
551 err = blkdev_get(bdev, FMODE_READ, NULL);
554 blkdev_put(bdev, FMODE_READ);
557 /* announce disk after possible partitions are created */
558 dev_set_uevent_suppress(ddev, 0);
559 kobject_uevent(&ddev->kobj, KOBJ_ADD);
561 /* announce possible partitions */
562 disk_part_iter_init(&piter, disk, 0);
563 while ((part = disk_part_iter_next(&piter)))
564 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
565 disk_part_iter_exit(&piter);
569 * add_disk - add partitioning information to kernel list
570 * @disk: per-device partitioning information
572 * This function registers the partitioning information in @disk
575 * FIXME: error handling
577 void add_disk(struct gendisk *disk)
579 struct backing_dev_info *bdi;
583 /* minors == 0 indicates to use ext devt from part0 and should
584 * be accompanied with EXT_DEVT flag. Make sure all
585 * parameters make sense.
587 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
588 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
590 disk->flags |= GENHD_FL_UP;
592 retval = blk_alloc_devt(&disk->part0, &devt);
597 disk_to_dev(disk)->devt = devt;
599 /* ->major and ->first_minor aren't supposed to be
600 * dereferenced from here on, but set them just in case.
602 disk->major = MAJOR(devt);
603 disk->first_minor = MINOR(devt);
605 /* Register BDI before referencing it from bdev */
606 bdi = &disk->queue->backing_dev_info;
607 bdi_register_dev(bdi, disk_devt(disk));
609 blk_register_region(disk_devt(disk), disk->minors, NULL,
610 exact_match, exact_lock, disk);
612 blk_register_queue(disk);
615 * Take an extra ref on queue which will be put on disk_release()
616 * so that it sticks around as long as @disk is there.
618 WARN_ON_ONCE(blk_get_queue(disk->queue));
620 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
624 disk_add_events(disk);
626 EXPORT_SYMBOL(add_disk);
628 void del_gendisk(struct gendisk *disk)
630 struct disk_part_iter piter;
631 struct hd_struct *part;
633 disk_del_events(disk);
635 /* invalidate stuff */
636 disk_part_iter_init(&piter, disk,
637 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
638 while ((part = disk_part_iter_next(&piter))) {
639 invalidate_partition(disk, part->partno);
640 delete_partition(disk, part->partno);
642 disk_part_iter_exit(&piter);
644 invalidate_partition(disk, 0);
645 blk_free_devt(disk_to_dev(disk)->devt);
646 set_capacity(disk, 0);
647 disk->flags &= ~GENHD_FL_UP;
649 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
650 bdi_unregister(&disk->queue->backing_dev_info);
651 blk_unregister_queue(disk);
652 blk_unregister_region(disk_devt(disk), disk->minors);
654 part_stat_set_all(&disk->part0, 0);
655 disk->part0.stamp = 0;
657 kobject_put(disk->part0.holder_dir);
658 kobject_put(disk->slave_dir);
659 disk->driverfs_dev = NULL;
660 if (!sysfs_deprecated)
661 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
662 device_del(disk_to_dev(disk));
664 EXPORT_SYMBOL(del_gendisk);
667 * get_gendisk - get partitioning information for a given device
668 * @devt: device to get partitioning information for
669 * @partno: returned partition index
671 * This function gets the structure containing partitioning
672 * information for the given device @devt.
674 struct gendisk *get_gendisk(dev_t devt, int *partno)
676 struct gendisk *disk = NULL;
678 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
679 struct kobject *kobj;
681 kobj = kobj_lookup(bdev_map, devt, partno);
683 disk = dev_to_disk(kobj_to_dev(kobj));
685 struct hd_struct *part;
687 mutex_lock(&ext_devt_mutex);
688 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
689 if (part && get_disk(part_to_disk(part))) {
690 *partno = part->partno;
691 disk = part_to_disk(part);
693 mutex_unlock(&ext_devt_mutex);
698 EXPORT_SYMBOL(get_gendisk);
701 * bdget_disk - do bdget() by gendisk and partition number
702 * @disk: gendisk of interest
703 * @partno: partition number
705 * Find partition @partno from @disk, do bdget() on it.
711 * Resulting block_device on success, NULL on failure.
713 struct block_device *bdget_disk(struct gendisk *disk, int partno)
715 struct hd_struct *part;
716 struct block_device *bdev = NULL;
718 part = disk_get_part(disk, partno);
720 bdev = bdget(part_devt(part));
725 EXPORT_SYMBOL(bdget_disk);
728 * print a full list of all partitions - intended for places where the root
729 * filesystem can't be mounted and thus to give the victim some idea of what
732 void __init printk_all_partitions(void)
734 struct class_dev_iter iter;
737 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
738 while ((dev = class_dev_iter_next(&iter))) {
739 struct gendisk *disk = dev_to_disk(dev);
740 struct disk_part_iter piter;
741 struct hd_struct *part;
742 char name_buf[BDEVNAME_SIZE];
743 char devt_buf[BDEVT_SIZE];
744 u8 uuid[PARTITION_META_INFO_UUIDLTH * 2 + 1];
747 * Don't show empty devices or things that have been
750 if (get_capacity(disk) == 0 ||
751 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
755 * Note, unlike /proc/partitions, I am showing the
756 * numbers in hex - the same format as the root=
759 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
760 while ((part = disk_part_iter_next(&piter))) {
761 bool is_part0 = part == &disk->part0;
765 part_unpack_uuid(part->info->uuid, uuid);
767 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
768 bdevt_str(part_devt(part), devt_buf),
769 (unsigned long long)part->nr_sects >> 1,
770 disk_name(disk, part->partno, name_buf), uuid);
772 if (disk->driverfs_dev != NULL &&
773 disk->driverfs_dev->driver != NULL)
774 printk(" driver: %s\n",
775 disk->driverfs_dev->driver->name);
777 printk(" (driver?)\n");
781 disk_part_iter_exit(&piter);
783 class_dev_iter_exit(&iter);
786 #ifdef CONFIG_PROC_FS
788 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
791 struct class_dev_iter *iter;
794 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
796 return ERR_PTR(-ENOMEM);
798 seqf->private = iter;
799 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
801 dev = class_dev_iter_next(iter);
806 return dev_to_disk(dev);
809 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
814 dev = class_dev_iter_next(seqf->private);
816 return dev_to_disk(dev);
821 static void disk_seqf_stop(struct seq_file *seqf, void *v)
823 struct class_dev_iter *iter = seqf->private;
825 /* stop is called even after start failed :-( */
827 class_dev_iter_exit(iter);
832 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
836 p = disk_seqf_start(seqf, pos);
837 if (!IS_ERR_OR_NULL(p) && !*pos)
838 seq_puts(seqf, "major minor #blocks name\n\n");
842 static int show_partition(struct seq_file *seqf, void *v)
844 struct gendisk *sgp = v;
845 struct disk_part_iter piter;
846 struct hd_struct *part;
847 char buf[BDEVNAME_SIZE];
849 /* Don't show non-partitionable removeable devices or empty devices */
850 if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
851 (sgp->flags & GENHD_FL_REMOVABLE)))
853 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
856 /* show the full disk and all non-0 size partitions of it */
857 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
858 while ((part = disk_part_iter_next(&piter)))
859 seq_printf(seqf, "%4d %7d %10llu %s\n",
860 MAJOR(part_devt(part)), MINOR(part_devt(part)),
861 (unsigned long long)part->nr_sects >> 1,
862 disk_name(sgp, part->partno, buf));
863 disk_part_iter_exit(&piter);
868 static const struct seq_operations partitions_op = {
869 .start = show_partition_start,
870 .next = disk_seqf_next,
871 .stop = disk_seqf_stop,
872 .show = show_partition
875 static int partitions_open(struct inode *inode, struct file *file)
877 return seq_open(file, &partitions_op);
880 static const struct file_operations proc_partitions_operations = {
881 .open = partitions_open,
884 .release = seq_release,
889 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
891 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
892 /* Make old-style 2.4 aliases work */
893 request_module("block-major-%d", MAJOR(devt));
897 static int __init genhd_device_init(void)
901 block_class.dev_kobj = sysfs_dev_block_kobj;
902 error = class_register(&block_class);
905 bdev_map = kobj_map_init(base_probe, &block_class_lock);
908 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
910 /* create top-level block dir */
911 if (!sysfs_deprecated)
912 block_depr = kobject_create_and_add("block", NULL);
916 subsys_initcall(genhd_device_init);
918 static ssize_t disk_range_show(struct device *dev,
919 struct device_attribute *attr, char *buf)
921 struct gendisk *disk = dev_to_disk(dev);
923 return sprintf(buf, "%d\n", disk->minors);
926 static ssize_t disk_ext_range_show(struct device *dev,
927 struct device_attribute *attr, char *buf)
929 struct gendisk *disk = dev_to_disk(dev);
931 return sprintf(buf, "%d\n", disk_max_parts(disk));
934 static ssize_t disk_removable_show(struct device *dev,
935 struct device_attribute *attr, char *buf)
937 struct gendisk *disk = dev_to_disk(dev);
939 return sprintf(buf, "%d\n",
940 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
943 static ssize_t disk_ro_show(struct device *dev,
944 struct device_attribute *attr, char *buf)
946 struct gendisk *disk = dev_to_disk(dev);
948 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
951 static ssize_t disk_capability_show(struct device *dev,
952 struct device_attribute *attr, char *buf)
954 struct gendisk *disk = dev_to_disk(dev);
956 return sprintf(buf, "%x\n", disk->flags);
959 static ssize_t disk_alignment_offset_show(struct device *dev,
960 struct device_attribute *attr,
963 struct gendisk *disk = dev_to_disk(dev);
965 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
968 static ssize_t disk_discard_alignment_show(struct device *dev,
969 struct device_attribute *attr,
972 struct gendisk *disk = dev_to_disk(dev);
974 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
977 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
978 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
979 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
980 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
981 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
982 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
983 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
985 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
986 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
987 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
988 #ifdef CONFIG_FAIL_MAKE_REQUEST
989 static struct device_attribute dev_attr_fail =
990 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
992 #ifdef CONFIG_FAIL_IO_TIMEOUT
993 static struct device_attribute dev_attr_fail_timeout =
994 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
998 static struct attribute *disk_attrs[] = {
999 &dev_attr_range.attr,
1000 &dev_attr_ext_range.attr,
1001 &dev_attr_removable.attr,
1003 &dev_attr_size.attr,
1004 &dev_attr_alignment_offset.attr,
1005 &dev_attr_discard_alignment.attr,
1006 &dev_attr_capability.attr,
1007 &dev_attr_stat.attr,
1008 &dev_attr_inflight.attr,
1009 #ifdef CONFIG_FAIL_MAKE_REQUEST
1010 &dev_attr_fail.attr,
1012 #ifdef CONFIG_FAIL_IO_TIMEOUT
1013 &dev_attr_fail_timeout.attr,
1018 static struct attribute_group disk_attr_group = {
1019 .attrs = disk_attrs,
1022 static const struct attribute_group *disk_attr_groups[] = {
1027 static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
1029 struct disk_part_tbl *ptbl =
1030 container_of(head, struct disk_part_tbl, rcu_head);
1036 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1037 * @disk: disk to replace part_tbl for
1038 * @new_ptbl: new part_tbl to install
1040 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1041 * original ptbl is freed using RCU callback.
1044 * Matching bd_mutx locked.
1046 static void disk_replace_part_tbl(struct gendisk *disk,
1047 struct disk_part_tbl *new_ptbl)
1049 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1051 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1054 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1055 call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
1060 * disk_expand_part_tbl - expand disk->part_tbl
1061 * @disk: disk to expand part_tbl for
1062 * @partno: expand such that this partno can fit in
1064 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1065 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1068 * Matching bd_mutex locked, might sleep.
1071 * 0 on success, -errno on failure.
1073 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1075 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1076 struct disk_part_tbl *new_ptbl;
1077 int len = old_ptbl ? old_ptbl->len : 0;
1078 int target = partno + 1;
1082 /* disk_max_parts() is zero during initialization, ignore if so */
1083 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1089 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1090 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1094 new_ptbl->len = target;
1096 for (i = 0; i < len; i++)
1097 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1099 disk_replace_part_tbl(disk, new_ptbl);
1103 static void disk_release(struct device *dev)
1105 struct gendisk *disk = dev_to_disk(dev);
1107 disk_release_events(disk);
1108 kfree(disk->random);
1109 disk_replace_part_tbl(disk, NULL);
1110 free_part_stats(&disk->part0);
1111 free_part_info(&disk->part0);
1113 blk_put_queue(disk->queue);
1116 struct class block_class = {
1120 static char *block_devnode(struct device *dev, mode_t *mode)
1122 struct gendisk *disk = dev_to_disk(dev);
1125 return disk->devnode(disk, mode);
1129 static struct device_type disk_type = {
1131 .groups = disk_attr_groups,
1132 .release = disk_release,
1133 .devnode = block_devnode,
1136 #ifdef CONFIG_PROC_FS
1138 * aggregate disk stat collector. Uses the same stats that the sysfs
1139 * entries do, above, but makes them available through one seq_file.
1141 * The output looks suspiciously like /proc/partitions with a bunch of
1144 static int diskstats_show(struct seq_file *seqf, void *v)
1146 struct gendisk *gp = v;
1147 struct disk_part_iter piter;
1148 struct hd_struct *hd;
1149 char buf[BDEVNAME_SIZE];
1153 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1154 seq_puts(seqf, "major minor name"
1155 " rio rmerge rsect ruse wio wmerge "
1156 "wsect wuse running use aveq"
1160 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1161 while ((hd = disk_part_iter_next(&piter))) {
1162 cpu = part_stat_lock();
1163 part_round_stats(cpu, hd);
1165 seq_printf(seqf, "%4d %7d %s %lu %lu %llu "
1166 "%u %lu %lu %llu %u %u %u %u\n",
1167 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1168 disk_name(gp, hd->partno, buf),
1169 part_stat_read(hd, ios[READ]),
1170 part_stat_read(hd, merges[READ]),
1171 (unsigned long long)part_stat_read(hd, sectors[READ]),
1172 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1173 part_stat_read(hd, ios[WRITE]),
1174 part_stat_read(hd, merges[WRITE]),
1175 (unsigned long long)part_stat_read(hd, sectors[WRITE]),
1176 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1178 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1179 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1182 disk_part_iter_exit(&piter);
1187 static const struct seq_operations diskstats_op = {
1188 .start = disk_seqf_start,
1189 .next = disk_seqf_next,
1190 .stop = disk_seqf_stop,
1191 .show = diskstats_show
1194 static int diskstats_open(struct inode *inode, struct file *file)
1196 return seq_open(file, &diskstats_op);
1199 static const struct file_operations proc_diskstats_operations = {
1200 .open = diskstats_open,
1202 .llseek = seq_lseek,
1203 .release = seq_release,
1206 static int __init proc_genhd_init(void)
1208 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1209 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1212 module_init(proc_genhd_init);
1213 #endif /* CONFIG_PROC_FS */
1215 dev_t blk_lookup_devt(const char *name, int partno)
1217 dev_t devt = MKDEV(0, 0);
1218 struct class_dev_iter iter;
1221 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1222 while ((dev = class_dev_iter_next(&iter))) {
1223 struct gendisk *disk = dev_to_disk(dev);
1224 struct hd_struct *part;
1226 if (strcmp(dev_name(dev), name))
1229 if (partno < disk->minors) {
1230 /* We need to return the right devno, even
1231 * if the partition doesn't exist yet.
1233 devt = MKDEV(MAJOR(dev->devt),
1234 MINOR(dev->devt) + partno);
1237 part = disk_get_part(disk, partno);
1239 devt = part_devt(part);
1240 disk_put_part(part);
1243 disk_put_part(part);
1245 class_dev_iter_exit(&iter);
1248 EXPORT_SYMBOL(blk_lookup_devt);
1250 struct gendisk *alloc_disk(int minors)
1252 return alloc_disk_node(minors, -1);
1254 EXPORT_SYMBOL(alloc_disk);
1256 struct gendisk *alloc_disk_node(int minors, int node_id)
1258 struct gendisk *disk;
1260 disk = kmalloc_node(sizeof(struct gendisk),
1261 GFP_KERNEL | __GFP_ZERO, node_id);
1263 if (!init_part_stats(&disk->part0)) {
1267 disk->node_id = node_id;
1268 if (disk_expand_part_tbl(disk, 0)) {
1269 free_part_stats(&disk->part0);
1273 disk->part_tbl->part[0] = &disk->part0;
1275 hd_ref_init(&disk->part0);
1277 disk->minors = minors;
1278 rand_initialize_disk(disk);
1279 disk_to_dev(disk)->class = &block_class;
1280 disk_to_dev(disk)->type = &disk_type;
1281 device_initialize(disk_to_dev(disk));
1285 EXPORT_SYMBOL(alloc_disk_node);
1287 struct kobject *get_disk(struct gendisk *disk)
1289 struct module *owner;
1290 struct kobject *kobj;
1294 owner = disk->fops->owner;
1295 if (owner && !try_module_get(owner))
1297 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1306 EXPORT_SYMBOL(get_disk);
1308 void put_disk(struct gendisk *disk)
1311 kobject_put(&disk_to_dev(disk)->kobj);
1314 EXPORT_SYMBOL(put_disk);
1316 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1318 char event[] = "DISK_RO=1";
1319 char *envp[] = { event, NULL };
1323 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1326 void set_device_ro(struct block_device *bdev, int flag)
1328 bdev->bd_part->policy = flag;
1331 EXPORT_SYMBOL(set_device_ro);
1333 void set_disk_ro(struct gendisk *disk, int flag)
1335 struct disk_part_iter piter;
1336 struct hd_struct *part;
1338 if (disk->part0.policy != flag) {
1339 set_disk_ro_uevent(disk, flag);
1340 disk->part0.policy = flag;
1343 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1344 while ((part = disk_part_iter_next(&piter)))
1345 part->policy = flag;
1346 disk_part_iter_exit(&piter);
1349 EXPORT_SYMBOL(set_disk_ro);
1351 int bdev_read_only(struct block_device *bdev)
1355 return bdev->bd_part->policy;
1358 EXPORT_SYMBOL(bdev_read_only);
1360 int invalidate_partition(struct gendisk *disk, int partno)
1363 struct block_device *bdev = bdget_disk(disk, partno);
1366 res = __invalidate_device(bdev, true);
1372 EXPORT_SYMBOL(invalidate_partition);
1375 * Disk events - monitor disk events like media change and eject request.
1377 struct disk_events {
1378 struct list_head node; /* all disk_event's */
1379 struct gendisk *disk; /* the associated disk */
1382 struct mutex block_mutex; /* protects blocking */
1383 int block; /* event blocking depth */
1384 unsigned int pending; /* events already sent out */
1385 unsigned int clearing; /* events being cleared */
1387 long poll_msecs; /* interval, -1 for default */
1388 struct delayed_work dwork;
1391 static const char *disk_events_strs[] = {
1392 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1393 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1396 static char *disk_uevents[] = {
1397 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1398 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1401 /* list of all disk_events */
1402 static DEFINE_MUTEX(disk_events_mutex);
1403 static LIST_HEAD(disk_events);
1405 /* disable in-kernel polling by default */
1406 static unsigned long disk_events_dfl_poll_msecs = 0;
1408 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1410 struct disk_events *ev = disk->ev;
1411 long intv_msecs = 0;
1414 * If device-specific poll interval is set, always use it. If
1415 * the default is being used, poll iff there are events which
1416 * can't be monitored asynchronously.
1418 if (ev->poll_msecs >= 0)
1419 intv_msecs = ev->poll_msecs;
1420 else if (disk->events & ~disk->async_events)
1421 intv_msecs = disk_events_dfl_poll_msecs;
1423 return msecs_to_jiffies(intv_msecs);
1427 * disk_block_events - block and flush disk event checking
1428 * @disk: disk to block events for
1430 * On return from this function, it is guaranteed that event checking
1431 * isn't in progress and won't happen until unblocked by
1432 * disk_unblock_events(). Events blocking is counted and the actual
1433 * unblocking happens after the matching number of unblocks are done.
1435 * Note that this intentionally does not block event checking from
1436 * disk_clear_events().
1441 void disk_block_events(struct gendisk *disk)
1443 struct disk_events *ev = disk->ev;
1444 unsigned long flags;
1451 * Outer mutex ensures that the first blocker completes canceling
1452 * the event work before further blockers are allowed to finish.
1454 mutex_lock(&ev->block_mutex);
1456 spin_lock_irqsave(&ev->lock, flags);
1457 cancel = !ev->block++;
1458 spin_unlock_irqrestore(&ev->lock, flags);
1461 cancel_delayed_work_sync(&disk->ev->dwork);
1463 mutex_unlock(&ev->block_mutex);
1466 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1468 struct disk_events *ev = disk->ev;
1470 unsigned long flags;
1472 spin_lock_irqsave(&ev->lock, flags);
1474 if (WARN_ON_ONCE(ev->block <= 0))
1481 * Not exactly a latency critical operation, set poll timer
1482 * slack to 25% and kick event check.
1484 intv = disk_events_poll_jiffies(disk);
1485 set_timer_slack(&ev->dwork.timer, intv / 4);
1487 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1489 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1491 spin_unlock_irqrestore(&ev->lock, flags);
1495 * disk_unblock_events - unblock disk event checking
1496 * @disk: disk to unblock events for
1498 * Undo disk_block_events(). When the block count reaches zero, it
1499 * starts events polling if configured.
1502 * Don't care. Safe to call from irq context.
1504 void disk_unblock_events(struct gendisk *disk)
1507 __disk_unblock_events(disk, false);
1511 * disk_check_events - schedule immediate event checking
1512 * @disk: disk to check events for
1514 * Schedule immediate event checking on @disk if not blocked.
1517 * Don't care. Safe to call from irq context.
1519 void disk_check_events(struct gendisk *disk)
1521 struct disk_events *ev = disk->ev;
1522 unsigned long flags;
1527 spin_lock_irqsave(&ev->lock, flags);
1529 cancel_delayed_work(&ev->dwork);
1530 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1532 spin_unlock_irqrestore(&ev->lock, flags);
1534 EXPORT_SYMBOL_GPL(disk_check_events);
1537 * disk_clear_events - synchronously check, clear and return pending events
1538 * @disk: disk to fetch and clear events from
1539 * @mask: mask of events to be fetched and clearted
1541 * Disk events are synchronously checked and pending events in @mask
1542 * are cleared and returned. This ignores the block count.
1547 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1549 const struct block_device_operations *bdops = disk->fops;
1550 struct disk_events *ev = disk->ev;
1551 unsigned int pending;
1554 /* for drivers still using the old ->media_changed method */
1555 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1556 bdops->media_changed && bdops->media_changed(disk))
1557 return DISK_EVENT_MEDIA_CHANGE;
1561 /* tell the workfn about the events being cleared */
1562 spin_lock_irq(&ev->lock);
1563 ev->clearing |= mask;
1564 spin_unlock_irq(&ev->lock);
1566 /* uncondtionally schedule event check and wait for it to finish */
1567 disk_block_events(disk);
1568 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1569 flush_delayed_work(&ev->dwork);
1570 __disk_unblock_events(disk, false);
1572 /* then, fetch and clear pending events */
1573 spin_lock_irq(&ev->lock);
1574 WARN_ON_ONCE(ev->clearing & mask); /* cleared by workfn */
1575 pending = ev->pending & mask;
1576 ev->pending &= ~mask;
1577 spin_unlock_irq(&ev->lock);
1582 static void disk_events_workfn(struct work_struct *work)
1584 struct delayed_work *dwork = to_delayed_work(work);
1585 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1586 struct gendisk *disk = ev->disk;
1587 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1588 unsigned int clearing = ev->clearing;
1589 unsigned int events;
1591 int nr_events = 0, i;
1594 events = disk->fops->check_events(disk, clearing);
1596 /* accumulate pending events and schedule next poll if necessary */
1597 spin_lock_irq(&ev->lock);
1599 events &= ~ev->pending;
1600 ev->pending |= events;
1601 ev->clearing &= ~clearing;
1603 intv = disk_events_poll_jiffies(disk);
1604 if (!ev->block && intv)
1605 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1607 spin_unlock_irq(&ev->lock);
1610 * Tell userland about new events. Only the events listed in
1611 * @disk->events are reported. Unlisted events are processed the
1612 * same internally but never get reported to userland.
1614 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1615 if (events & disk->events & (1 << i))
1616 envp[nr_events++] = disk_uevents[i];
1619 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1623 * A disk events enabled device has the following sysfs nodes under
1624 * its /sys/block/X/ directory.
1626 * events : list of all supported events
1627 * events_async : list of events which can be detected w/o polling
1628 * events_poll_msecs : polling interval, 0: disable, -1: system default
1630 static ssize_t __disk_events_show(unsigned int events, char *buf)
1632 const char *delim = "";
1636 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1637 if (events & (1 << i)) {
1638 pos += sprintf(buf + pos, "%s%s",
1639 delim, disk_events_strs[i]);
1643 pos += sprintf(buf + pos, "\n");
1647 static ssize_t disk_events_show(struct device *dev,
1648 struct device_attribute *attr, char *buf)
1650 struct gendisk *disk = dev_to_disk(dev);
1652 return __disk_events_show(disk->events, buf);
1655 static ssize_t disk_events_async_show(struct device *dev,
1656 struct device_attribute *attr, char *buf)
1658 struct gendisk *disk = dev_to_disk(dev);
1660 return __disk_events_show(disk->async_events, buf);
1663 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1664 struct device_attribute *attr,
1667 struct gendisk *disk = dev_to_disk(dev);
1669 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1672 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1673 struct device_attribute *attr,
1674 const char *buf, size_t count)
1676 struct gendisk *disk = dev_to_disk(dev);
1679 if (!count || !sscanf(buf, "%ld", &intv))
1682 if (intv < 0 && intv != -1)
1685 disk_block_events(disk);
1686 disk->ev->poll_msecs = intv;
1687 __disk_unblock_events(disk, true);
1692 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1693 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1694 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1695 disk_events_poll_msecs_show,
1696 disk_events_poll_msecs_store);
1698 static const struct attribute *disk_events_attrs[] = {
1699 &dev_attr_events.attr,
1700 &dev_attr_events_async.attr,
1701 &dev_attr_events_poll_msecs.attr,
1706 * The default polling interval can be specified by the kernel
1707 * parameter block.events_dfl_poll_msecs which defaults to 0
1708 * (disable). This can also be modified runtime by writing to
1709 * /sys/module/block/events_dfl_poll_msecs.
1711 static int disk_events_set_dfl_poll_msecs(const char *val,
1712 const struct kernel_param *kp)
1714 struct disk_events *ev;
1717 ret = param_set_ulong(val, kp);
1721 mutex_lock(&disk_events_mutex);
1723 list_for_each_entry(ev, &disk_events, node)
1724 disk_check_events(ev->disk);
1726 mutex_unlock(&disk_events_mutex);
1731 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1732 .set = disk_events_set_dfl_poll_msecs,
1733 .get = param_get_ulong,
1736 #undef MODULE_PARAM_PREFIX
1737 #define MODULE_PARAM_PREFIX "block."
1739 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1740 &disk_events_dfl_poll_msecs, 0644);
1743 * disk_{add|del|release}_events - initialize and destroy disk_events.
1745 static void disk_add_events(struct gendisk *disk)
1747 struct disk_events *ev;
1749 if (!disk->fops->check_events)
1752 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1754 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1758 if (sysfs_create_files(&disk_to_dev(disk)->kobj,
1759 disk_events_attrs) < 0) {
1760 pr_warn("%s: failed to create sysfs files for events\n",
1768 INIT_LIST_HEAD(&ev->node);
1770 spin_lock_init(&ev->lock);
1771 mutex_init(&ev->block_mutex);
1773 ev->poll_msecs = -1;
1774 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1776 mutex_lock(&disk_events_mutex);
1777 list_add_tail(&ev->node, &disk_events);
1778 mutex_unlock(&disk_events_mutex);
1781 * Block count is initialized to 1 and the following initial
1782 * unblock kicks it into action.
1784 __disk_unblock_events(disk, true);
1787 static void disk_del_events(struct gendisk *disk)
1792 disk_block_events(disk);
1794 mutex_lock(&disk_events_mutex);
1795 list_del_init(&disk->ev->node);
1796 mutex_unlock(&disk_events_mutex);
1798 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1801 static void disk_release_events(struct gendisk *disk)
1803 /* the block count should be 1 from disk_del_events() */
1804 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);