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/mutex.h>
19 #include <linux/idr.h>
20 #include <linux/log2.h>
24 static DEFINE_MUTEX(block_class_lock);
25 struct kobject *block_depr;
27 /* for extended dynamic devt allocation, currently only one major is used */
28 #define MAX_EXT_DEVT (1 << MINORBITS)
30 /* For extended devt allocation. ext_devt_mutex prevents look up
31 * results from going away underneath its user.
33 static DEFINE_MUTEX(ext_devt_mutex);
34 static DEFINE_IDR(ext_devt_idr);
36 static struct device_type disk_type;
38 static void disk_alloc_events(struct gendisk *disk);
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 static 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_part_scan_enabled(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 disk_alloc_events(disk);
607 /* Register BDI before referencing it from bdev */
608 bdi = &disk->queue->backing_dev_info;
609 bdi_register_dev(bdi, disk_devt(disk));
611 blk_register_region(disk_devt(disk), disk->minors, NULL,
612 exact_match, exact_lock, disk);
614 blk_register_queue(disk);
617 * Take an extra ref on queue which will be put on disk_release()
618 * so that it sticks around as long as @disk is there.
620 WARN_ON_ONCE(!blk_get_queue(disk->queue));
622 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
626 disk_add_events(disk);
628 EXPORT_SYMBOL(add_disk);
630 void del_gendisk(struct gendisk *disk)
632 struct disk_part_iter piter;
633 struct hd_struct *part;
635 disk_del_events(disk);
637 /* invalidate stuff */
638 disk_part_iter_init(&piter, disk,
639 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
640 while ((part = disk_part_iter_next(&piter))) {
641 invalidate_partition(disk, part->partno);
642 delete_partition(disk, part->partno);
644 disk_part_iter_exit(&piter);
646 invalidate_partition(disk, 0);
647 blk_free_devt(disk_to_dev(disk)->devt);
648 set_capacity(disk, 0);
649 disk->flags &= ~GENHD_FL_UP;
651 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
652 bdi_unregister(&disk->queue->backing_dev_info);
653 blk_unregister_queue(disk);
654 blk_unregister_region(disk_devt(disk), disk->minors);
656 part_stat_set_all(&disk->part0, 0);
657 disk->part0.stamp = 0;
659 kobject_put(disk->part0.holder_dir);
660 kobject_put(disk->slave_dir);
661 disk->driverfs_dev = NULL;
662 if (!sysfs_deprecated)
663 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
664 device_del(disk_to_dev(disk));
666 EXPORT_SYMBOL(del_gendisk);
669 * get_gendisk - get partitioning information for a given device
670 * @devt: device to get partitioning information for
671 * @partno: returned partition index
673 * This function gets the structure containing partitioning
674 * information for the given device @devt.
676 struct gendisk *get_gendisk(dev_t devt, int *partno)
678 struct gendisk *disk = NULL;
680 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
681 struct kobject *kobj;
683 kobj = kobj_lookup(bdev_map, devt, partno);
685 disk = dev_to_disk(kobj_to_dev(kobj));
687 struct hd_struct *part;
689 mutex_lock(&ext_devt_mutex);
690 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
691 if (part && get_disk(part_to_disk(part))) {
692 *partno = part->partno;
693 disk = part_to_disk(part);
695 mutex_unlock(&ext_devt_mutex);
700 EXPORT_SYMBOL(get_gendisk);
703 * bdget_disk - do bdget() by gendisk and partition number
704 * @disk: gendisk of interest
705 * @partno: partition number
707 * Find partition @partno from @disk, do bdget() on it.
713 * Resulting block_device on success, NULL on failure.
715 struct block_device *bdget_disk(struct gendisk *disk, int partno)
717 struct hd_struct *part;
718 struct block_device *bdev = NULL;
720 part = disk_get_part(disk, partno);
722 bdev = bdget(part_devt(part));
727 EXPORT_SYMBOL(bdget_disk);
730 * print a full list of all partitions - intended for places where the root
731 * filesystem can't be mounted and thus to give the victim some idea of what
734 void __init printk_all_partitions(void)
736 struct class_dev_iter iter;
739 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
740 while ((dev = class_dev_iter_next(&iter))) {
741 struct gendisk *disk = dev_to_disk(dev);
742 struct disk_part_iter piter;
743 struct hd_struct *part;
744 char name_buf[BDEVNAME_SIZE];
745 char devt_buf[BDEVT_SIZE];
746 char uuid_buf[PARTITION_META_INFO_UUIDLTH * 2 + 5];
749 * Don't show empty devices or things that have been
752 if (get_capacity(disk) == 0 ||
753 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
757 * Note, unlike /proc/partitions, I am showing the
758 * numbers in hex - the same format as the root=
761 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
762 while ((part = disk_part_iter_next(&piter))) {
763 bool is_part0 = part == &disk->part0;
767 snprintf(uuid_buf, sizeof(uuid_buf), "%pU",
770 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
771 bdevt_str(part_devt(part), devt_buf),
772 (unsigned long long)part->nr_sects >> 1,
773 disk_name(disk, part->partno, name_buf),
776 if (disk->driverfs_dev != NULL &&
777 disk->driverfs_dev->driver != NULL)
778 printk(" driver: %s\n",
779 disk->driverfs_dev->driver->name);
781 printk(" (driver?)\n");
785 disk_part_iter_exit(&piter);
787 class_dev_iter_exit(&iter);
790 #ifdef CONFIG_PROC_FS
792 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
795 struct class_dev_iter *iter;
798 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
800 return ERR_PTR(-ENOMEM);
802 seqf->private = iter;
803 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
805 dev = class_dev_iter_next(iter);
810 return dev_to_disk(dev);
813 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
818 dev = class_dev_iter_next(seqf->private);
820 return dev_to_disk(dev);
825 static void disk_seqf_stop(struct seq_file *seqf, void *v)
827 struct class_dev_iter *iter = seqf->private;
829 /* stop is called even after start failed :-( */
831 class_dev_iter_exit(iter);
836 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
840 p = disk_seqf_start(seqf, pos);
841 if (!IS_ERR_OR_NULL(p) && !*pos)
842 seq_puts(seqf, "major minor #blocks name\n\n");
846 static int show_partition(struct seq_file *seqf, void *v)
848 struct gendisk *sgp = v;
849 struct disk_part_iter piter;
850 struct hd_struct *part;
851 char buf[BDEVNAME_SIZE];
853 /* Don't show non-partitionable removeable devices or empty devices */
854 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
855 (sgp->flags & GENHD_FL_REMOVABLE)))
857 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
860 /* show the full disk and all non-0 size partitions of it */
861 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
862 while ((part = disk_part_iter_next(&piter)))
863 seq_printf(seqf, "%4d %7d %10llu %s\n",
864 MAJOR(part_devt(part)), MINOR(part_devt(part)),
865 (unsigned long long)part->nr_sects >> 1,
866 disk_name(sgp, part->partno, buf));
867 disk_part_iter_exit(&piter);
872 static const struct seq_operations partitions_op = {
873 .start = show_partition_start,
874 .next = disk_seqf_next,
875 .stop = disk_seqf_stop,
876 .show = show_partition
879 static int partitions_open(struct inode *inode, struct file *file)
881 return seq_open(file, &partitions_op);
884 static const struct file_operations proc_partitions_operations = {
885 .open = partitions_open,
888 .release = seq_release,
893 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
895 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
896 /* Make old-style 2.4 aliases work */
897 request_module("block-major-%d", MAJOR(devt));
901 static int __init genhd_device_init(void)
905 block_class.dev_kobj = sysfs_dev_block_kobj;
906 error = class_register(&block_class);
909 bdev_map = kobj_map_init(base_probe, &block_class_lock);
912 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
914 /* create top-level block dir */
915 if (!sysfs_deprecated)
916 block_depr = kobject_create_and_add("block", NULL);
920 subsys_initcall(genhd_device_init);
922 static ssize_t disk_range_show(struct device *dev,
923 struct device_attribute *attr, char *buf)
925 struct gendisk *disk = dev_to_disk(dev);
927 return sprintf(buf, "%d\n", disk->minors);
930 static ssize_t disk_ext_range_show(struct device *dev,
931 struct device_attribute *attr, char *buf)
933 struct gendisk *disk = dev_to_disk(dev);
935 return sprintf(buf, "%d\n", disk_max_parts(disk));
938 static ssize_t disk_removable_show(struct device *dev,
939 struct device_attribute *attr, char *buf)
941 struct gendisk *disk = dev_to_disk(dev);
943 return sprintf(buf, "%d\n",
944 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
947 static ssize_t disk_ro_show(struct device *dev,
948 struct device_attribute *attr, char *buf)
950 struct gendisk *disk = dev_to_disk(dev);
952 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
955 static ssize_t disk_capability_show(struct device *dev,
956 struct device_attribute *attr, char *buf)
958 struct gendisk *disk = dev_to_disk(dev);
960 return sprintf(buf, "%x\n", disk->flags);
963 static ssize_t disk_alignment_offset_show(struct device *dev,
964 struct device_attribute *attr,
967 struct gendisk *disk = dev_to_disk(dev);
969 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
972 static ssize_t disk_discard_alignment_show(struct device *dev,
973 struct device_attribute *attr,
976 struct gendisk *disk = dev_to_disk(dev);
978 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
981 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
982 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
983 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
984 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
985 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
986 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
987 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
989 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
990 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
991 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
992 #ifdef CONFIG_FAIL_MAKE_REQUEST
993 static struct device_attribute dev_attr_fail =
994 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
996 #ifdef CONFIG_FAIL_IO_TIMEOUT
997 static struct device_attribute dev_attr_fail_timeout =
998 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
1002 static struct attribute *disk_attrs[] = {
1003 &dev_attr_range.attr,
1004 &dev_attr_ext_range.attr,
1005 &dev_attr_removable.attr,
1007 &dev_attr_size.attr,
1008 &dev_attr_alignment_offset.attr,
1009 &dev_attr_discard_alignment.attr,
1010 &dev_attr_capability.attr,
1011 &dev_attr_stat.attr,
1012 &dev_attr_inflight.attr,
1013 #ifdef CONFIG_FAIL_MAKE_REQUEST
1014 &dev_attr_fail.attr,
1016 #ifdef CONFIG_FAIL_IO_TIMEOUT
1017 &dev_attr_fail_timeout.attr,
1022 static struct attribute_group disk_attr_group = {
1023 .attrs = disk_attrs,
1026 static const struct attribute_group *disk_attr_groups[] = {
1032 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1033 * @disk: disk to replace part_tbl for
1034 * @new_ptbl: new part_tbl to install
1036 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1037 * original ptbl is freed using RCU callback.
1040 * Matching bd_mutx locked.
1042 static void disk_replace_part_tbl(struct gendisk *disk,
1043 struct disk_part_tbl *new_ptbl)
1045 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1047 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1050 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1051 kfree_rcu(old_ptbl, rcu_head);
1056 * disk_expand_part_tbl - expand disk->part_tbl
1057 * @disk: disk to expand part_tbl for
1058 * @partno: expand such that this partno can fit in
1060 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1061 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1064 * Matching bd_mutex locked, might sleep.
1067 * 0 on success, -errno on failure.
1069 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1071 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1072 struct disk_part_tbl *new_ptbl;
1073 int len = old_ptbl ? old_ptbl->len : 0;
1074 int target = partno + 1;
1078 /* disk_max_parts() is zero during initialization, ignore if so */
1079 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1085 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1086 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1090 new_ptbl->len = target;
1092 for (i = 0; i < len; i++)
1093 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1095 disk_replace_part_tbl(disk, new_ptbl);
1099 static void disk_release(struct device *dev)
1101 struct gendisk *disk = dev_to_disk(dev);
1103 disk_release_events(disk);
1104 kfree(disk->random);
1105 disk_replace_part_tbl(disk, NULL);
1106 free_part_stats(&disk->part0);
1107 free_part_info(&disk->part0);
1109 blk_put_queue(disk->queue);
1112 struct class block_class = {
1116 static char *block_devnode(struct device *dev, umode_t *mode)
1118 struct gendisk *disk = dev_to_disk(dev);
1121 return disk->devnode(disk, mode);
1125 static struct device_type disk_type = {
1127 .groups = disk_attr_groups,
1128 .release = disk_release,
1129 .devnode = block_devnode,
1132 #ifdef CONFIG_PROC_FS
1134 * aggregate disk stat collector. Uses the same stats that the sysfs
1135 * entries do, above, but makes them available through one seq_file.
1137 * The output looks suspiciously like /proc/partitions with a bunch of
1140 static int diskstats_show(struct seq_file *seqf, void *v)
1142 struct gendisk *gp = v;
1143 struct disk_part_iter piter;
1144 struct hd_struct *hd;
1145 char buf[BDEVNAME_SIZE];
1149 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1150 seq_puts(seqf, "major minor name"
1151 " rio rmerge rsect ruse wio wmerge "
1152 "wsect wuse running use aveq"
1156 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1157 while ((hd = disk_part_iter_next(&piter))) {
1158 cpu = part_stat_lock();
1159 part_round_stats(cpu, hd);
1161 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1162 "%u %lu %lu %lu %u %u %u %u\n",
1163 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1164 disk_name(gp, hd->partno, buf),
1165 part_stat_read(hd, ios[READ]),
1166 part_stat_read(hd, merges[READ]),
1167 part_stat_read(hd, sectors[READ]),
1168 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1169 part_stat_read(hd, ios[WRITE]),
1170 part_stat_read(hd, merges[WRITE]),
1171 part_stat_read(hd, sectors[WRITE]),
1172 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1174 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1175 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1178 disk_part_iter_exit(&piter);
1183 static const struct seq_operations diskstats_op = {
1184 .start = disk_seqf_start,
1185 .next = disk_seqf_next,
1186 .stop = disk_seqf_stop,
1187 .show = diskstats_show
1190 static int diskstats_open(struct inode *inode, struct file *file)
1192 return seq_open(file, &diskstats_op);
1195 static const struct file_operations proc_diskstats_operations = {
1196 .open = diskstats_open,
1198 .llseek = seq_lseek,
1199 .release = seq_release,
1202 static int __init proc_genhd_init(void)
1204 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1205 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1208 module_init(proc_genhd_init);
1209 #endif /* CONFIG_PROC_FS */
1211 dev_t blk_lookup_devt(const char *name, int partno)
1213 dev_t devt = MKDEV(0, 0);
1214 struct class_dev_iter iter;
1217 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1218 while ((dev = class_dev_iter_next(&iter))) {
1219 struct gendisk *disk = dev_to_disk(dev);
1220 struct hd_struct *part;
1222 if (strcmp(dev_name(dev), name))
1225 if (partno < disk->minors) {
1226 /* We need to return the right devno, even
1227 * if the partition doesn't exist yet.
1229 devt = MKDEV(MAJOR(dev->devt),
1230 MINOR(dev->devt) + partno);
1233 part = disk_get_part(disk, partno);
1235 devt = part_devt(part);
1236 disk_put_part(part);
1239 disk_put_part(part);
1241 class_dev_iter_exit(&iter);
1244 EXPORT_SYMBOL(blk_lookup_devt);
1246 struct gendisk *alloc_disk(int minors)
1248 return alloc_disk_node(minors, -1);
1250 EXPORT_SYMBOL(alloc_disk);
1252 struct gendisk *alloc_disk_node(int minors, int node_id)
1254 struct gendisk *disk;
1256 disk = kmalloc_node(sizeof(struct gendisk),
1257 GFP_KERNEL | __GFP_ZERO, node_id);
1259 if (!init_part_stats(&disk->part0)) {
1263 disk->node_id = node_id;
1264 if (disk_expand_part_tbl(disk, 0)) {
1265 free_part_stats(&disk->part0);
1269 disk->part_tbl->part[0] = &disk->part0;
1271 hd_ref_init(&disk->part0);
1273 disk->minors = minors;
1274 rand_initialize_disk(disk);
1275 disk_to_dev(disk)->class = &block_class;
1276 disk_to_dev(disk)->type = &disk_type;
1277 device_initialize(disk_to_dev(disk));
1281 EXPORT_SYMBOL(alloc_disk_node);
1283 struct kobject *get_disk(struct gendisk *disk)
1285 struct module *owner;
1286 struct kobject *kobj;
1290 owner = disk->fops->owner;
1291 if (owner && !try_module_get(owner))
1293 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1302 EXPORT_SYMBOL(get_disk);
1304 void put_disk(struct gendisk *disk)
1307 kobject_put(&disk_to_dev(disk)->kobj);
1310 EXPORT_SYMBOL(put_disk);
1312 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1314 char event[] = "DISK_RO=1";
1315 char *envp[] = { event, NULL };
1319 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1322 void set_device_ro(struct block_device *bdev, int flag)
1324 bdev->bd_part->policy = flag;
1327 EXPORT_SYMBOL(set_device_ro);
1329 void set_disk_ro(struct gendisk *disk, int flag)
1331 struct disk_part_iter piter;
1332 struct hd_struct *part;
1334 if (disk->part0.policy != flag) {
1335 set_disk_ro_uevent(disk, flag);
1336 disk->part0.policy = flag;
1339 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1340 while ((part = disk_part_iter_next(&piter)))
1341 part->policy = flag;
1342 disk_part_iter_exit(&piter);
1345 EXPORT_SYMBOL(set_disk_ro);
1347 int bdev_read_only(struct block_device *bdev)
1351 return bdev->bd_part->policy;
1354 EXPORT_SYMBOL(bdev_read_only);
1356 int invalidate_partition(struct gendisk *disk, int partno)
1359 struct block_device *bdev = bdget_disk(disk, partno);
1362 res = __invalidate_device(bdev, true);
1368 EXPORT_SYMBOL(invalidate_partition);
1371 * Disk events - monitor disk events like media change and eject request.
1373 struct disk_events {
1374 struct list_head node; /* all disk_event's */
1375 struct gendisk *disk; /* the associated disk */
1378 struct mutex block_mutex; /* protects blocking */
1379 int block; /* event blocking depth */
1380 unsigned int pending; /* events already sent out */
1381 unsigned int clearing; /* events being cleared */
1383 long poll_msecs; /* interval, -1 for default */
1384 struct delayed_work dwork;
1387 static const char *disk_events_strs[] = {
1388 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1389 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1392 static char *disk_uevents[] = {
1393 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1394 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1397 /* list of all disk_events */
1398 static DEFINE_MUTEX(disk_events_mutex);
1399 static LIST_HEAD(disk_events);
1401 /* disable in-kernel polling by default */
1402 static unsigned long disk_events_dfl_poll_msecs = 0;
1404 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1406 struct disk_events *ev = disk->ev;
1407 long intv_msecs = 0;
1410 * If device-specific poll interval is set, always use it. If
1411 * the default is being used, poll iff there are events which
1412 * can't be monitored asynchronously.
1414 if (ev->poll_msecs >= 0)
1415 intv_msecs = ev->poll_msecs;
1416 else if (disk->events & ~disk->async_events)
1417 intv_msecs = disk_events_dfl_poll_msecs;
1419 return msecs_to_jiffies(intv_msecs);
1423 * disk_block_events - block and flush disk event checking
1424 * @disk: disk to block events for
1426 * On return from this function, it is guaranteed that event checking
1427 * isn't in progress and won't happen until unblocked by
1428 * disk_unblock_events(). Events blocking is counted and the actual
1429 * unblocking happens after the matching number of unblocks are done.
1431 * Note that this intentionally does not block event checking from
1432 * disk_clear_events().
1437 void disk_block_events(struct gendisk *disk)
1439 struct disk_events *ev = disk->ev;
1440 unsigned long flags;
1447 * Outer mutex ensures that the first blocker completes canceling
1448 * the event work before further blockers are allowed to finish.
1450 mutex_lock(&ev->block_mutex);
1452 spin_lock_irqsave(&ev->lock, flags);
1453 cancel = !ev->block++;
1454 spin_unlock_irqrestore(&ev->lock, flags);
1457 cancel_delayed_work_sync(&disk->ev->dwork);
1459 mutex_unlock(&ev->block_mutex);
1462 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1464 struct disk_events *ev = disk->ev;
1466 unsigned long flags;
1468 spin_lock_irqsave(&ev->lock, flags);
1470 if (WARN_ON_ONCE(ev->block <= 0))
1477 * Not exactly a latency critical operation, set poll timer
1478 * slack to 25% and kick event check.
1480 intv = disk_events_poll_jiffies(disk);
1481 set_timer_slack(&ev->dwork.timer, intv / 4);
1483 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, 0);
1485 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, intv);
1487 spin_unlock_irqrestore(&ev->lock, flags);
1491 * disk_unblock_events - unblock disk event checking
1492 * @disk: disk to unblock events for
1494 * Undo disk_block_events(). When the block count reaches zero, it
1495 * starts events polling if configured.
1498 * Don't care. Safe to call from irq context.
1500 void disk_unblock_events(struct gendisk *disk)
1503 __disk_unblock_events(disk, false);
1507 * disk_flush_events - schedule immediate event checking and flushing
1508 * @disk: disk to check and flush events for
1509 * @mask: events to flush
1511 * Schedule immediate event checking on @disk if not blocked. Events in
1512 * @mask are scheduled to be cleared from the driver. Note that this
1513 * doesn't clear the events from @disk->ev.
1516 * If @mask is non-zero must be called with bdev->bd_mutex held.
1518 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1520 struct disk_events *ev = disk->ev;
1525 spin_lock_irq(&ev->lock);
1526 ev->clearing |= mask;
1528 cancel_delayed_work(&ev->dwork);
1529 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, 0);
1531 spin_unlock_irq(&ev->lock);
1535 * disk_clear_events - synchronously check, clear and return pending events
1536 * @disk: disk to fetch and clear events from
1537 * @mask: mask of events to be fetched and clearted
1539 * Disk events are synchronously checked and pending events in @mask
1540 * are cleared and returned. This ignores the block count.
1545 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1547 const struct block_device_operations *bdops = disk->fops;
1548 struct disk_events *ev = disk->ev;
1549 unsigned int pending;
1552 /* for drivers still using the old ->media_changed method */
1553 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1554 bdops->media_changed && bdops->media_changed(disk))
1555 return DISK_EVENT_MEDIA_CHANGE;
1559 /* tell the workfn about the events being cleared */
1560 spin_lock_irq(&ev->lock);
1561 ev->clearing |= mask;
1562 spin_unlock_irq(&ev->lock);
1564 /* uncondtionally schedule event check and wait for it to finish */
1565 disk_block_events(disk);
1566 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, 0);
1567 flush_delayed_work(&ev->dwork);
1568 __disk_unblock_events(disk, false);
1570 /* then, fetch and clear pending events */
1571 spin_lock_irq(&ev->lock);
1572 WARN_ON_ONCE(ev->clearing & mask); /* cleared by workfn */
1573 pending = ev->pending & mask;
1574 ev->pending &= ~mask;
1575 spin_unlock_irq(&ev->lock);
1580 static void disk_events_workfn(struct work_struct *work)
1582 struct delayed_work *dwork = to_delayed_work(work);
1583 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1584 struct gendisk *disk = ev->disk;
1585 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1586 unsigned int clearing = ev->clearing;
1587 unsigned int events;
1589 int nr_events = 0, i;
1592 events = disk->fops->check_events(disk, clearing);
1594 /* accumulate pending events and schedule next poll if necessary */
1595 spin_lock_irq(&ev->lock);
1597 events &= ~ev->pending;
1598 ev->pending |= events;
1599 ev->clearing &= ~clearing;
1601 intv = disk_events_poll_jiffies(disk);
1602 if (!ev->block && intv)
1603 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, intv);
1605 spin_unlock_irq(&ev->lock);
1608 * Tell userland about new events. Only the events listed in
1609 * @disk->events are reported. Unlisted events are processed the
1610 * same internally but never get reported to userland.
1612 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1613 if (events & disk->events & (1 << i))
1614 envp[nr_events++] = disk_uevents[i];
1617 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1621 * A disk events enabled device has the following sysfs nodes under
1622 * its /sys/block/X/ directory.
1624 * events : list of all supported events
1625 * events_async : list of events which can be detected w/o polling
1626 * events_poll_msecs : polling interval, 0: disable, -1: system default
1628 static ssize_t __disk_events_show(unsigned int events, char *buf)
1630 const char *delim = "";
1634 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1635 if (events & (1 << i)) {
1636 pos += sprintf(buf + pos, "%s%s",
1637 delim, disk_events_strs[i]);
1641 pos += sprintf(buf + pos, "\n");
1645 static ssize_t disk_events_show(struct device *dev,
1646 struct device_attribute *attr, char *buf)
1648 struct gendisk *disk = dev_to_disk(dev);
1650 return __disk_events_show(disk->events, buf);
1653 static ssize_t disk_events_async_show(struct device *dev,
1654 struct device_attribute *attr, char *buf)
1656 struct gendisk *disk = dev_to_disk(dev);
1658 return __disk_events_show(disk->async_events, buf);
1661 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1662 struct device_attribute *attr,
1665 struct gendisk *disk = dev_to_disk(dev);
1667 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1670 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1671 struct device_attribute *attr,
1672 const char *buf, size_t count)
1674 struct gendisk *disk = dev_to_disk(dev);
1677 if (!count || !sscanf(buf, "%ld", &intv))
1680 if (intv < 0 && intv != -1)
1683 disk_block_events(disk);
1684 disk->ev->poll_msecs = intv;
1685 __disk_unblock_events(disk, true);
1690 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1691 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1692 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1693 disk_events_poll_msecs_show,
1694 disk_events_poll_msecs_store);
1696 static const struct attribute *disk_events_attrs[] = {
1697 &dev_attr_events.attr,
1698 &dev_attr_events_async.attr,
1699 &dev_attr_events_poll_msecs.attr,
1704 * The default polling interval can be specified by the kernel
1705 * parameter block.events_dfl_poll_msecs which defaults to 0
1706 * (disable). This can also be modified runtime by writing to
1707 * /sys/module/block/events_dfl_poll_msecs.
1709 static int disk_events_set_dfl_poll_msecs(const char *val,
1710 const struct kernel_param *kp)
1712 struct disk_events *ev;
1715 ret = param_set_ulong(val, kp);
1719 mutex_lock(&disk_events_mutex);
1721 list_for_each_entry(ev, &disk_events, node)
1722 disk_flush_events(ev->disk, 0);
1724 mutex_unlock(&disk_events_mutex);
1729 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1730 .set = disk_events_set_dfl_poll_msecs,
1731 .get = param_get_ulong,
1734 #undef MODULE_PARAM_PREFIX
1735 #define MODULE_PARAM_PREFIX "block."
1737 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1738 &disk_events_dfl_poll_msecs, 0644);
1741 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1743 static void disk_alloc_events(struct gendisk *disk)
1745 struct disk_events *ev;
1747 if (!disk->fops->check_events)
1750 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1752 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1756 INIT_LIST_HEAD(&ev->node);
1758 spin_lock_init(&ev->lock);
1759 mutex_init(&ev->block_mutex);
1761 ev->poll_msecs = -1;
1762 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1767 static void disk_add_events(struct gendisk *disk)
1772 /* FIXME: error handling */
1773 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1774 pr_warn("%s: failed to create sysfs files for events\n",
1777 mutex_lock(&disk_events_mutex);
1778 list_add_tail(&disk->ev->node, &disk_events);
1779 mutex_unlock(&disk_events_mutex);
1782 * Block count is initialized to 1 and the following initial
1783 * unblock kicks it into action.
1785 __disk_unblock_events(disk, true);
1788 static void disk_del_events(struct gendisk *disk)
1793 disk_block_events(disk);
1795 mutex_lock(&disk_events_mutex);
1796 list_del_init(&disk->ev->node);
1797 mutex_unlock(&disk_events_mutex);
1799 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1802 static void disk_release_events(struct gendisk *disk)
1804 /* the block count should be 1 from disk_del_events() */
1805 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);