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 NR_EXT_DEVT (1 << MINORBITS)
31 /* For extended devt allocation. ext_devt_lock prevents look up
32 * results from going away underneath its user.
34 static DEFINE_SPINLOCK(ext_devt_lock);
35 static DEFINE_IDR(ext_devt_idr);
37 static struct device_type disk_type;
39 static void disk_alloc_events(struct gendisk *disk);
40 static void disk_add_events(struct gendisk *disk);
41 static void disk_del_events(struct gendisk *disk);
42 static void disk_release_events(struct gendisk *disk);
45 * disk_get_part - get partition
46 * @disk: disk to look partition from
47 * @partno: partition number
49 * Look for partition @partno from @disk. If found, increment
50 * reference count and return it.
56 * Pointer to the found partition on success, NULL if not found.
58 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
60 struct hd_struct *part = NULL;
61 struct disk_part_tbl *ptbl;
63 if (unlikely(partno < 0))
68 ptbl = rcu_dereference(disk->part_tbl);
69 if (likely(partno < ptbl->len)) {
70 part = rcu_dereference(ptbl->part[partno]);
72 get_device(part_to_dev(part));
79 EXPORT_SYMBOL_GPL(disk_get_part);
82 * disk_part_iter_init - initialize partition iterator
83 * @piter: iterator to initialize
84 * @disk: disk to iterate over
85 * @flags: DISK_PITER_* flags
87 * Initialize @piter so that it iterates over partitions of @disk.
92 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
95 struct disk_part_tbl *ptbl;
98 ptbl = rcu_dereference(disk->part_tbl);
103 if (flags & DISK_PITER_REVERSE)
104 piter->idx = ptbl->len - 1;
105 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
110 piter->flags = flags;
114 EXPORT_SYMBOL_GPL(disk_part_iter_init);
117 * disk_part_iter_next - proceed iterator to the next partition and return it
118 * @piter: iterator of interest
120 * Proceed @piter to the next partition and return it.
125 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
127 struct disk_part_tbl *ptbl;
130 /* put the last partition */
131 disk_put_part(piter->part);
136 ptbl = rcu_dereference(piter->disk->part_tbl);
138 /* determine iteration parameters */
139 if (piter->flags & DISK_PITER_REVERSE) {
141 if (piter->flags & (DISK_PITER_INCL_PART0 |
142 DISK_PITER_INCL_EMPTY_PART0))
151 /* iterate to the next partition */
152 for (; piter->idx != end; piter->idx += inc) {
153 struct hd_struct *part;
155 part = rcu_dereference(ptbl->part[piter->idx]);
158 if (!part->nr_sects &&
159 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
160 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
164 get_device(part_to_dev(part));
174 EXPORT_SYMBOL_GPL(disk_part_iter_next);
177 * disk_part_iter_exit - finish up partition iteration
178 * @piter: iter of interest
180 * Called when iteration is over. Cleans up @piter.
185 void disk_part_iter_exit(struct disk_part_iter *piter)
187 disk_put_part(piter->part);
190 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
192 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
194 return part->start_sect <= sector &&
195 sector < part->start_sect + part->nr_sects;
199 * disk_map_sector_rcu - map sector to partition
200 * @disk: gendisk of interest
201 * @sector: sector to map
203 * Find out which partition @sector maps to on @disk. This is
204 * primarily used for stats accounting.
207 * RCU read locked. The returned partition pointer is valid only
208 * while preemption is disabled.
211 * Found partition on success, part0 is returned if no partition matches
213 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
215 struct disk_part_tbl *ptbl;
216 struct hd_struct *part;
219 ptbl = rcu_dereference(disk->part_tbl);
221 part = rcu_dereference(ptbl->last_lookup);
222 if (part && sector_in_part(part, sector))
225 for (i = 1; i < ptbl->len; i++) {
226 part = rcu_dereference(ptbl->part[i]);
228 if (part && sector_in_part(part, sector)) {
229 rcu_assign_pointer(ptbl->last_lookup, part);
235 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
238 * Can be deleted altogether. Later.
241 static struct blk_major_name {
242 struct blk_major_name *next;
245 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
247 /* index in the above - for now: assume no multimajor ranges */
248 static inline int major_to_index(unsigned major)
250 return major % BLKDEV_MAJOR_HASH_SIZE;
253 #ifdef CONFIG_PROC_FS
254 void blkdev_show(struct seq_file *seqf, off_t offset)
256 struct blk_major_name *dp;
258 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
259 mutex_lock(&block_class_lock);
260 for (dp = major_names[offset]; dp; dp = dp->next)
261 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
262 mutex_unlock(&block_class_lock);
265 #endif /* CONFIG_PROC_FS */
268 * register_blkdev - register a new block device
270 * @major: the requested major device number [1..255]. If @major=0, try to
271 * allocate any unused major number.
272 * @name: the name of the new block device as a zero terminated string
274 * The @name must be unique within the system.
276 * The return value depends on the @major input parameter.
277 * - if a major device number was requested in range [1..255] then the
278 * function returns zero on success, or a negative error code
279 * - if any unused major number was requested with @major=0 parameter
280 * then the return value is the allocated major number in range
281 * [1..255] or a negative error code otherwise
283 int register_blkdev(unsigned int major, const char *name)
285 struct blk_major_name **n, *p;
288 mutex_lock(&block_class_lock);
292 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
293 if (major_names[index] == NULL)
298 printk("register_blkdev: failed to get major for %s\n",
307 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
314 strlcpy(p->name, name, sizeof(p->name));
316 index = major_to_index(major);
318 for (n = &major_names[index]; *n; n = &(*n)->next) {
319 if ((*n)->major == major)
328 printk("register_blkdev: cannot get major %d for %s\n",
333 mutex_unlock(&block_class_lock);
337 EXPORT_SYMBOL(register_blkdev);
339 void unregister_blkdev(unsigned int major, const char *name)
341 struct blk_major_name **n;
342 struct blk_major_name *p = NULL;
343 int index = major_to_index(major);
345 mutex_lock(&block_class_lock);
346 for (n = &major_names[index]; *n; n = &(*n)->next)
347 if ((*n)->major == major)
349 if (!*n || strcmp((*n)->name, name)) {
355 mutex_unlock(&block_class_lock);
359 EXPORT_SYMBOL(unregister_blkdev);
361 static struct kobj_map *bdev_map;
364 * blk_mangle_minor - scatter minor numbers apart
365 * @minor: minor number to mangle
367 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
368 * is enabled. Mangling twice gives the original value.
376 static int blk_mangle_minor(int minor)
378 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
381 for (i = 0; i < MINORBITS / 2; i++) {
382 int low = minor & (1 << i);
383 int high = minor & (1 << (MINORBITS - 1 - i));
384 int distance = MINORBITS - 1 - 2 * i;
386 minor ^= low | high; /* clear both bits */
387 low <<= distance; /* swap the positions */
389 minor |= low | high; /* and set */
396 * blk_alloc_devt - allocate a dev_t for a partition
397 * @part: partition to allocate dev_t for
398 * @devt: out parameter for resulting dev_t
400 * Allocate a dev_t for block device.
403 * 0 on success, allocated dev_t is returned in *@devt. -errno on
409 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
411 struct gendisk *disk = part_to_disk(part);
414 /* in consecutive minor range? */
415 if (part->partno < disk->minors) {
416 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
420 /* allocate ext devt */
422 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
424 spin_lock(&ext_devt_lock);
425 rc = idr_get_new(&ext_devt_idr, part, &idx);
426 if (!rc && idx >= NR_EXT_DEVT) {
427 idr_remove(&ext_devt_idr, idx);
430 spin_unlock(&ext_devt_lock);
431 } while (rc == -EAGAIN);
436 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
441 * blk_free_devt - free a dev_t
442 * @devt: dev_t to free
444 * Free @devt which was allocated using blk_alloc_devt().
449 void blk_free_devt(dev_t devt)
453 if (devt == MKDEV(0, 0))
456 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
457 spin_lock(&ext_devt_lock);
458 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
459 spin_unlock(&ext_devt_lock);
463 static char *bdevt_str(dev_t devt, char *buf)
465 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
466 char tbuf[BDEVT_SIZE];
467 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
468 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
470 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
476 * Register device numbers dev..(dev+range-1)
477 * range must be nonzero
478 * The hash chain is sorted on range, so that subranges can override.
480 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
481 struct kobject *(*probe)(dev_t, int *, void *),
482 int (*lock)(dev_t, void *), void *data)
484 kobj_map(bdev_map, devt, range, module, probe, lock, data);
487 EXPORT_SYMBOL(blk_register_region);
489 void blk_unregister_region(dev_t devt, unsigned long range)
491 kobj_unmap(bdev_map, devt, range);
494 EXPORT_SYMBOL(blk_unregister_region);
496 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
498 struct gendisk *p = data;
500 return &disk_to_dev(p)->kobj;
503 static int exact_lock(dev_t devt, void *data)
505 struct gendisk *p = data;
512 void register_disk(struct gendisk *disk)
514 struct device *ddev = disk_to_dev(disk);
515 struct block_device *bdev;
516 struct disk_part_iter piter;
517 struct hd_struct *part;
520 ddev->parent = disk->driverfs_dev;
522 dev_set_name(ddev, "%s", disk->disk_name);
524 /* delay uevents, until we scanned partition table */
525 dev_set_uevent_suppress(ddev, 1);
527 if (device_add(ddev))
529 if (!sysfs_deprecated) {
530 err = sysfs_create_link(block_depr, &ddev->kobj,
531 kobject_name(&ddev->kobj));
537 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
538 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
540 /* No minors to use for partitions */
541 if (!disk_part_scan_enabled(disk))
544 /* No such device (e.g., media were just removed) */
545 if (!get_capacity(disk))
548 bdev = bdget_disk(disk, 0);
552 bdev->bd_invalidated = 1;
553 err = blkdev_get(bdev, FMODE_READ, NULL);
556 blkdev_put(bdev, FMODE_READ);
559 /* announce disk after possible partitions are created */
560 dev_set_uevent_suppress(ddev, 0);
561 kobject_uevent(&ddev->kobj, KOBJ_ADD);
563 /* announce possible partitions */
564 disk_part_iter_init(&piter, disk, 0);
565 while ((part = disk_part_iter_next(&piter)))
566 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
567 disk_part_iter_exit(&piter);
571 * add_disk - add partitioning information to kernel list
572 * @disk: per-device partitioning information
574 * This function registers the partitioning information in @disk
577 * FIXME: error handling
579 void add_disk(struct gendisk *disk)
581 struct backing_dev_info *bdi;
585 /* minors == 0 indicates to use ext devt from part0 and should
586 * be accompanied with EXT_DEVT flag. Make sure all
587 * parameters make sense.
589 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
590 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
592 disk->flags |= GENHD_FL_UP;
594 retval = blk_alloc_devt(&disk->part0, &devt);
599 disk_to_dev(disk)->devt = devt;
601 /* ->major and ->first_minor aren't supposed to be
602 * dereferenced from here on, but set them just in case.
604 disk->major = MAJOR(devt);
605 disk->first_minor = MINOR(devt);
607 disk_alloc_events(disk);
609 /* Register BDI before referencing it from bdev */
610 bdi = &disk->queue->backing_dev_info;
611 bdi_register_dev(bdi, disk_devt(disk));
613 blk_register_region(disk_devt(disk), disk->minors, NULL,
614 exact_match, exact_lock, disk);
616 blk_register_queue(disk);
619 * Take an extra ref on queue which will be put on disk_release()
620 * so that it sticks around as long as @disk is there.
622 WARN_ON_ONCE(blk_get_queue(disk->queue));
624 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
628 disk_add_events(disk);
630 EXPORT_SYMBOL(add_disk);
632 void del_gendisk(struct gendisk *disk)
634 struct disk_part_iter piter;
635 struct hd_struct *part;
637 disk_del_events(disk);
639 /* invalidate stuff */
640 disk_part_iter_init(&piter, disk,
641 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
642 while ((part = disk_part_iter_next(&piter))) {
643 invalidate_partition(disk, part->partno);
644 delete_partition(disk, part->partno);
646 disk_part_iter_exit(&piter);
648 invalidate_partition(disk, 0);
649 set_capacity(disk, 0);
650 disk->flags &= ~GENHD_FL_UP;
652 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
653 bdi_unregister(&disk->queue->backing_dev_info);
654 blk_unregister_queue(disk);
655 blk_unregister_region(disk_devt(disk), disk->minors);
657 part_stat_set_all(&disk->part0, 0);
658 disk->part0.stamp = 0;
660 kobject_put(disk->part0.holder_dir);
661 kobject_put(disk->slave_dir);
662 disk->driverfs_dev = NULL;
663 if (!sysfs_deprecated)
664 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
665 device_del(disk_to_dev(disk));
667 EXPORT_SYMBOL(del_gendisk);
670 * get_gendisk - get partitioning information for a given device
671 * @devt: device to get partitioning information for
672 * @partno: returned partition index
674 * This function gets the structure containing partitioning
675 * information for the given device @devt.
677 struct gendisk *get_gendisk(dev_t devt, int *partno)
679 struct gendisk *disk = NULL;
681 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
682 struct kobject *kobj;
684 kobj = kobj_lookup(bdev_map, devt, partno);
686 disk = dev_to_disk(kobj_to_dev(kobj));
688 struct hd_struct *part;
690 spin_lock(&ext_devt_lock);
691 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
692 if (part && get_disk(part_to_disk(part))) {
693 *partno = part->partno;
694 disk = part_to_disk(part);
696 spin_unlock(&ext_devt_lock);
701 EXPORT_SYMBOL(get_gendisk);
704 * bdget_disk - do bdget() by gendisk and partition number
705 * @disk: gendisk of interest
706 * @partno: partition number
708 * Find partition @partno from @disk, do bdget() on it.
714 * Resulting block_device on success, NULL on failure.
716 struct block_device *bdget_disk(struct gendisk *disk, int partno)
718 struct hd_struct *part;
719 struct block_device *bdev = NULL;
721 part = disk_get_part(disk, partno);
723 bdev = bdget(part_devt(part));
728 EXPORT_SYMBOL(bdget_disk);
731 * print a full list of all partitions - intended for places where the root
732 * filesystem can't be mounted and thus to give the victim some idea of what
735 void __init printk_all_partitions(void)
737 struct class_dev_iter iter;
740 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
741 while ((dev = class_dev_iter_next(&iter))) {
742 struct gendisk *disk = dev_to_disk(dev);
743 struct disk_part_iter piter;
744 struct hd_struct *part;
745 char name_buf[BDEVNAME_SIZE];
746 char devt_buf[BDEVT_SIZE];
747 char uuid_buf[PARTITION_META_INFO_UUIDLTH * 2 + 5];
750 * Don't show empty devices or things that have been
753 if (get_capacity(disk) == 0 ||
754 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
758 * Note, unlike /proc/partitions, I am showing the
759 * numbers in hex - the same format as the root=
762 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
763 while ((part = disk_part_iter_next(&piter))) {
764 bool is_part0 = part == &disk->part0;
768 snprintf(uuid_buf, sizeof(uuid_buf), "%pU",
771 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
772 bdevt_str(part_devt(part), devt_buf),
773 (unsigned long long)part->nr_sects >> 1,
774 disk_name(disk, part->partno, name_buf),
777 if (disk->driverfs_dev != NULL &&
778 disk->driverfs_dev->driver != NULL)
779 printk(" driver: %s\n",
780 disk->driverfs_dev->driver->name);
782 printk(" (driver?)\n");
786 disk_part_iter_exit(&piter);
788 class_dev_iter_exit(&iter);
791 #ifdef CONFIG_PROC_FS
793 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
796 struct class_dev_iter *iter;
799 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
801 return ERR_PTR(-ENOMEM);
803 seqf->private = iter;
804 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
806 dev = class_dev_iter_next(iter);
811 return dev_to_disk(dev);
814 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
819 dev = class_dev_iter_next(seqf->private);
821 return dev_to_disk(dev);
826 static void disk_seqf_stop(struct seq_file *seqf, void *v)
828 struct class_dev_iter *iter = seqf->private;
830 /* stop is called even after start failed :-( */
832 class_dev_iter_exit(iter);
837 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
841 p = disk_seqf_start(seqf, pos);
842 if (!IS_ERR_OR_NULL(p) && !*pos)
843 seq_puts(seqf, "major minor #blocks name\n\n");
847 static int show_partition(struct seq_file *seqf, void *v)
849 struct gendisk *sgp = v;
850 struct disk_part_iter piter;
851 struct hd_struct *part;
852 char buf[BDEVNAME_SIZE];
854 /* Don't show non-partitionable removeable devices or empty devices */
855 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
856 (sgp->flags & GENHD_FL_REMOVABLE)))
858 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
861 /* show the full disk and all non-0 size partitions of it */
862 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
863 while ((part = disk_part_iter_next(&piter)))
864 seq_printf(seqf, "%4d %7d %10llu %s\n",
865 MAJOR(part_devt(part)), MINOR(part_devt(part)),
866 (unsigned long long)part->nr_sects >> 1,
867 disk_name(sgp, part->partno, buf));
868 disk_part_iter_exit(&piter);
873 static const struct seq_operations partitions_op = {
874 .start = show_partition_start,
875 .next = disk_seqf_next,
876 .stop = disk_seqf_stop,
877 .show = show_partition
880 static int partitions_open(struct inode *inode, struct file *file)
882 return seq_open(file, &partitions_op);
885 static const struct file_operations proc_partitions_operations = {
886 .open = partitions_open,
889 .release = seq_release,
894 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
896 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
897 /* Make old-style 2.4 aliases work */
898 request_module("block-major-%d", MAJOR(devt));
902 static int __init genhd_device_init(void)
906 block_class.dev_kobj = sysfs_dev_block_kobj;
907 error = class_register(&block_class);
910 bdev_map = kobj_map_init(base_probe, &block_class_lock);
913 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
915 /* create top-level block dir */
916 if (!sysfs_deprecated)
917 block_depr = kobject_create_and_add("block", NULL);
921 subsys_initcall(genhd_device_init);
923 static ssize_t disk_range_show(struct device *dev,
924 struct device_attribute *attr, char *buf)
926 struct gendisk *disk = dev_to_disk(dev);
928 return sprintf(buf, "%d\n", disk->minors);
931 static ssize_t disk_ext_range_show(struct device *dev,
932 struct device_attribute *attr, char *buf)
934 struct gendisk *disk = dev_to_disk(dev);
936 return sprintf(buf, "%d\n", disk_max_parts(disk));
939 static ssize_t disk_removable_show(struct device *dev,
940 struct device_attribute *attr, char *buf)
942 struct gendisk *disk = dev_to_disk(dev);
944 return sprintf(buf, "%d\n",
945 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
948 static ssize_t disk_ro_show(struct device *dev,
949 struct device_attribute *attr, char *buf)
951 struct gendisk *disk = dev_to_disk(dev);
953 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
956 static ssize_t disk_capability_show(struct device *dev,
957 struct device_attribute *attr, char *buf)
959 struct gendisk *disk = dev_to_disk(dev);
961 return sprintf(buf, "%x\n", disk->flags);
964 static ssize_t disk_alignment_offset_show(struct device *dev,
965 struct device_attribute *attr,
968 struct gendisk *disk = dev_to_disk(dev);
970 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
973 static ssize_t disk_discard_alignment_show(struct device *dev,
974 struct device_attribute *attr,
977 struct gendisk *disk = dev_to_disk(dev);
979 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
982 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
983 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
984 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
985 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
986 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
987 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
988 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
990 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
991 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
992 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
993 #ifdef CONFIG_FAIL_MAKE_REQUEST
994 static struct device_attribute dev_attr_fail =
995 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
997 #ifdef CONFIG_FAIL_IO_TIMEOUT
998 static struct device_attribute dev_attr_fail_timeout =
999 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
1000 part_timeout_store);
1003 static struct attribute *disk_attrs[] = {
1004 &dev_attr_range.attr,
1005 &dev_attr_ext_range.attr,
1006 &dev_attr_removable.attr,
1008 &dev_attr_size.attr,
1009 &dev_attr_alignment_offset.attr,
1010 &dev_attr_discard_alignment.attr,
1011 &dev_attr_capability.attr,
1012 &dev_attr_stat.attr,
1013 &dev_attr_inflight.attr,
1014 #ifdef CONFIG_FAIL_MAKE_REQUEST
1015 &dev_attr_fail.attr,
1017 #ifdef CONFIG_FAIL_IO_TIMEOUT
1018 &dev_attr_fail_timeout.attr,
1023 static struct attribute_group disk_attr_group = {
1024 .attrs = disk_attrs,
1027 static const struct attribute_group *disk_attr_groups[] = {
1033 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1034 * @disk: disk to replace part_tbl for
1035 * @new_ptbl: new part_tbl to install
1037 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1038 * original ptbl is freed using RCU callback.
1041 * Matching bd_mutx locked.
1043 static void disk_replace_part_tbl(struct gendisk *disk,
1044 struct disk_part_tbl *new_ptbl)
1046 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1048 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1051 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1052 kfree_rcu(old_ptbl, rcu_head);
1057 * disk_expand_part_tbl - expand disk->part_tbl
1058 * @disk: disk to expand part_tbl for
1059 * @partno: expand such that this partno can fit in
1061 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1062 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1065 * Matching bd_mutex locked, might sleep.
1068 * 0 on success, -errno on failure.
1070 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1072 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1073 struct disk_part_tbl *new_ptbl;
1074 int len = old_ptbl ? old_ptbl->len : 0;
1075 int target = partno + 1;
1079 /* disk_max_parts() is zero during initialization, ignore if so */
1080 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1086 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1087 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1091 new_ptbl->len = target;
1093 for (i = 0; i < len; i++)
1094 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1096 disk_replace_part_tbl(disk, new_ptbl);
1100 static void disk_release(struct device *dev)
1102 struct gendisk *disk = dev_to_disk(dev);
1104 blk_free_devt(dev->devt);
1105 disk_release_events(disk);
1106 kfree(disk->random);
1107 disk_replace_part_tbl(disk, NULL);
1108 free_part_stats(&disk->part0);
1109 free_part_info(&disk->part0);
1111 blk_put_queue(disk->queue);
1114 struct class block_class = {
1118 static char *block_devnode(struct device *dev, mode_t *mode)
1120 struct gendisk *disk = dev_to_disk(dev);
1123 return disk->devnode(disk, mode);
1127 static struct device_type disk_type = {
1129 .groups = disk_attr_groups,
1130 .release = disk_release,
1131 .devnode = block_devnode,
1134 #ifdef CONFIG_PROC_FS
1136 * aggregate disk stat collector. Uses the same stats that the sysfs
1137 * entries do, above, but makes them available through one seq_file.
1139 * The output looks suspiciously like /proc/partitions with a bunch of
1142 static int diskstats_show(struct seq_file *seqf, void *v)
1144 struct gendisk *gp = v;
1145 struct disk_part_iter piter;
1146 struct hd_struct *hd;
1147 char buf[BDEVNAME_SIZE];
1151 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1152 seq_puts(seqf, "major minor name"
1153 " rio rmerge rsect ruse wio wmerge "
1154 "wsect wuse running use aveq"
1158 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1159 while ((hd = disk_part_iter_next(&piter))) {
1160 cpu = part_stat_lock();
1161 part_round_stats(cpu, hd);
1163 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1164 "%u %lu %lu %lu %u %u %u %u\n",
1165 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1166 disk_name(gp, hd->partno, buf),
1167 part_stat_read(hd, ios[READ]),
1168 part_stat_read(hd, merges[READ]),
1169 part_stat_read(hd, sectors[READ]),
1170 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1171 part_stat_read(hd, ios[WRITE]),
1172 part_stat_read(hd, merges[WRITE]),
1173 part_stat_read(hd, sectors[WRITE]),
1174 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1176 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1177 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1180 disk_part_iter_exit(&piter);
1185 static const struct seq_operations diskstats_op = {
1186 .start = disk_seqf_start,
1187 .next = disk_seqf_next,
1188 .stop = disk_seqf_stop,
1189 .show = diskstats_show
1192 static int diskstats_open(struct inode *inode, struct file *file)
1194 return seq_open(file, &diskstats_op);
1197 static const struct file_operations proc_diskstats_operations = {
1198 .open = diskstats_open,
1200 .llseek = seq_lseek,
1201 .release = seq_release,
1204 static int __init proc_genhd_init(void)
1206 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1207 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1210 module_init(proc_genhd_init);
1211 #endif /* CONFIG_PROC_FS */
1213 dev_t blk_lookup_devt(const char *name, int partno)
1215 dev_t devt = MKDEV(0, 0);
1216 struct class_dev_iter iter;
1219 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1220 while ((dev = class_dev_iter_next(&iter))) {
1221 struct gendisk *disk = dev_to_disk(dev);
1222 struct hd_struct *part;
1224 if (strcmp(dev_name(dev), name))
1227 if (partno < disk->minors) {
1228 /* We need to return the right devno, even
1229 * if the partition doesn't exist yet.
1231 devt = MKDEV(MAJOR(dev->devt),
1232 MINOR(dev->devt) + partno);
1235 part = disk_get_part(disk, partno);
1237 devt = part_devt(part);
1238 disk_put_part(part);
1241 disk_put_part(part);
1243 class_dev_iter_exit(&iter);
1246 EXPORT_SYMBOL(blk_lookup_devt);
1248 struct gendisk *alloc_disk(int minors)
1250 return alloc_disk_node(minors, -1);
1252 EXPORT_SYMBOL(alloc_disk);
1254 struct gendisk *alloc_disk_node(int minors, int node_id)
1256 struct gendisk *disk;
1258 disk = kmalloc_node(sizeof(struct gendisk),
1259 GFP_KERNEL | __GFP_ZERO, node_id);
1261 if (!init_part_stats(&disk->part0)) {
1265 disk->node_id = node_id;
1266 if (disk_expand_part_tbl(disk, 0)) {
1267 free_part_stats(&disk->part0);
1271 disk->part_tbl->part[0] = &disk->part0;
1273 hd_ref_init(&disk->part0);
1275 disk->minors = minors;
1276 rand_initialize_disk(disk);
1277 disk_to_dev(disk)->class = &block_class;
1278 disk_to_dev(disk)->type = &disk_type;
1279 device_initialize(disk_to_dev(disk));
1283 EXPORT_SYMBOL(alloc_disk_node);
1285 struct kobject *get_disk(struct gendisk *disk)
1287 struct module *owner;
1288 struct kobject *kobj;
1292 owner = disk->fops->owner;
1293 if (owner && !try_module_get(owner))
1295 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1304 EXPORT_SYMBOL(get_disk);
1306 void put_disk(struct gendisk *disk)
1309 kobject_put(&disk_to_dev(disk)->kobj);
1312 EXPORT_SYMBOL(put_disk);
1314 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1316 char event[] = "DISK_RO=1";
1317 char *envp[] = { event, NULL };
1321 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1324 void set_device_ro(struct block_device *bdev, int flag)
1326 bdev->bd_part->policy = flag;
1329 EXPORT_SYMBOL(set_device_ro);
1331 void set_disk_ro(struct gendisk *disk, int flag)
1333 struct disk_part_iter piter;
1334 struct hd_struct *part;
1336 if (disk->part0.policy != flag) {
1337 set_disk_ro_uevent(disk, flag);
1338 disk->part0.policy = flag;
1341 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1342 while ((part = disk_part_iter_next(&piter)))
1343 part->policy = flag;
1344 disk_part_iter_exit(&piter);
1347 EXPORT_SYMBOL(set_disk_ro);
1349 int bdev_read_only(struct block_device *bdev)
1353 return bdev->bd_part->policy;
1356 EXPORT_SYMBOL(bdev_read_only);
1358 int invalidate_partition(struct gendisk *disk, int partno)
1361 struct block_device *bdev = bdget_disk(disk, partno);
1364 res = __invalidate_device(bdev, true);
1370 EXPORT_SYMBOL(invalidate_partition);
1373 * Disk events - monitor disk events like media change and eject request.
1375 struct disk_events {
1376 struct list_head node; /* all disk_event's */
1377 struct gendisk *disk; /* the associated disk */
1380 struct mutex block_mutex; /* protects blocking */
1381 int block; /* event blocking depth */
1382 unsigned int pending; /* events already sent out */
1383 unsigned int clearing; /* events being cleared */
1385 long poll_msecs; /* interval, -1 for default */
1386 struct delayed_work dwork;
1389 static const char *disk_events_strs[] = {
1390 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1391 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1394 static char *disk_uevents[] = {
1395 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1396 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1399 /* list of all disk_events */
1400 static DEFINE_MUTEX(disk_events_mutex);
1401 static LIST_HEAD(disk_events);
1403 /* disable in-kernel polling by default */
1404 static unsigned long disk_events_dfl_poll_msecs = 0;
1406 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1408 struct disk_events *ev = disk->ev;
1409 long intv_msecs = 0;
1412 * If device-specific poll interval is set, always use it. If
1413 * the default is being used, poll iff there are events which
1414 * can't be monitored asynchronously.
1416 if (ev->poll_msecs >= 0)
1417 intv_msecs = ev->poll_msecs;
1418 else if (disk->events & ~disk->async_events)
1419 intv_msecs = disk_events_dfl_poll_msecs;
1421 return msecs_to_jiffies(intv_msecs);
1425 * disk_block_events - block and flush disk event checking
1426 * @disk: disk to block events for
1428 * On return from this function, it is guaranteed that event checking
1429 * isn't in progress and won't happen until unblocked by
1430 * disk_unblock_events(). Events blocking is counted and the actual
1431 * unblocking happens after the matching number of unblocks are done.
1433 * Note that this intentionally does not block event checking from
1434 * disk_clear_events().
1439 void disk_block_events(struct gendisk *disk)
1441 struct disk_events *ev = disk->ev;
1442 unsigned long flags;
1449 * Outer mutex ensures that the first blocker completes canceling
1450 * the event work before further blockers are allowed to finish.
1452 mutex_lock(&ev->block_mutex);
1454 spin_lock_irqsave(&ev->lock, flags);
1455 cancel = !ev->block++;
1456 spin_unlock_irqrestore(&ev->lock, flags);
1459 cancel_delayed_work_sync(&disk->ev->dwork);
1461 mutex_unlock(&ev->block_mutex);
1464 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1466 struct disk_events *ev = disk->ev;
1468 unsigned long flags;
1470 spin_lock_irqsave(&ev->lock, flags);
1472 if (WARN_ON_ONCE(ev->block <= 0))
1479 * Not exactly a latency critical operation, set poll timer
1480 * slack to 25% and kick event check.
1482 intv = disk_events_poll_jiffies(disk);
1483 set_timer_slack(&ev->dwork.timer, intv / 4);
1485 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, 0);
1487 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, intv);
1489 spin_unlock_irqrestore(&ev->lock, flags);
1493 * disk_unblock_events - unblock disk event checking
1494 * @disk: disk to unblock events for
1496 * Undo disk_block_events(). When the block count reaches zero, it
1497 * starts events polling if configured.
1500 * Don't care. Safe to call from irq context.
1502 void disk_unblock_events(struct gendisk *disk)
1505 __disk_unblock_events(disk, false);
1509 * disk_flush_events - schedule immediate event checking and flushing
1510 * @disk: disk to check and flush events for
1511 * @mask: events to flush
1513 * Schedule immediate event checking on @disk if not blocked. Events in
1514 * @mask are scheduled to be cleared from the driver. Note that this
1515 * doesn't clear the events from @disk->ev.
1518 * If @mask is non-zero must be called with bdev->bd_mutex held.
1520 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1522 struct disk_events *ev = disk->ev;
1527 spin_lock_irq(&ev->lock);
1528 ev->clearing |= mask;
1530 cancel_delayed_work(&ev->dwork);
1531 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, 0);
1533 spin_unlock_irq(&ev->lock);
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_freezable_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_freezable_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_flush_events(ev->disk, 0);
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_{alloc|add|del|release}_events - initialize and destroy disk_events.
1745 static void disk_alloc_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 INIT_LIST_HEAD(&ev->node);
1760 spin_lock_init(&ev->lock);
1761 mutex_init(&ev->block_mutex);
1763 ev->poll_msecs = -1;
1764 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1769 static void disk_add_events(struct gendisk *disk)
1774 /* FIXME: error handling */
1775 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1776 pr_warn("%s: failed to create sysfs files for events\n",
1779 mutex_lock(&disk_events_mutex);
1780 list_add_tail(&disk->ev->node, &disk_events);
1781 mutex_unlock(&disk_events_mutex);
1784 * Block count is initialized to 1 and the following initial
1785 * unblock kicks it into action.
1787 __disk_unblock_events(disk, true);
1790 static void disk_del_events(struct gendisk *disk)
1795 disk_block_events(disk);
1797 mutex_lock(&disk_events_mutex);
1798 list_del_init(&disk->ev->node);
1799 mutex_unlock(&disk_events_mutex);
1801 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1804 static void disk_release_events(struct gendisk *disk)
1806 /* the block count should be 1 from disk_del_events() */
1807 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);