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_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 rc = idr_get_new(&ext_devt_idr, part, &idx);
425 } while (rc == -EAGAIN);
430 if (idx > MAX_EXT_DEVT) {
431 idr_remove(&ext_devt_idr, idx);
435 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
440 * blk_free_devt - free a dev_t
441 * @devt: dev_t to free
443 * Free @devt which was allocated using blk_alloc_devt().
448 void blk_free_devt(dev_t devt)
452 if (devt == MKDEV(0, 0))
455 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
456 mutex_lock(&ext_devt_mutex);
457 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
458 mutex_unlock(&ext_devt_mutex);
462 static char *bdevt_str(dev_t devt, char *buf)
464 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
465 char tbuf[BDEVT_SIZE];
466 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
467 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
469 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
475 * Register device numbers dev..(dev+range-1)
476 * range must be nonzero
477 * The hash chain is sorted on range, so that subranges can override.
479 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
480 struct kobject *(*probe)(dev_t, int *, void *),
481 int (*lock)(dev_t, void *), void *data)
483 kobj_map(bdev_map, devt, range, module, probe, lock, data);
486 EXPORT_SYMBOL(blk_register_region);
488 void blk_unregister_region(dev_t devt, unsigned long range)
490 kobj_unmap(bdev_map, devt, range);
493 EXPORT_SYMBOL(blk_unregister_region);
495 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
497 struct gendisk *p = data;
499 return &disk_to_dev(p)->kobj;
502 static int exact_lock(dev_t devt, void *data)
504 struct gendisk *p = data;
511 void register_disk(struct gendisk *disk)
513 struct device *ddev = disk_to_dev(disk);
514 struct block_device *bdev;
515 struct disk_part_iter piter;
516 struct hd_struct *part;
519 ddev->parent = disk->driverfs_dev;
521 dev_set_name(ddev, disk->disk_name);
523 /* delay uevents, until we scanned partition table */
524 dev_set_uevent_suppress(ddev, 1);
526 if (device_add(ddev))
528 if (!sysfs_deprecated) {
529 err = sysfs_create_link(block_depr, &ddev->kobj,
530 kobject_name(&ddev->kobj));
536 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
537 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
539 /* No minors to use for partitions */
540 if (!disk_partitionable(disk))
543 /* No such device (e.g., media were just removed) */
544 if (!get_capacity(disk))
547 bdev = bdget_disk(disk, 0);
551 bdev->bd_invalidated = 1;
552 err = blkdev_get(bdev, FMODE_READ, NULL);
555 blkdev_put(bdev, FMODE_READ);
558 /* announce disk after possible partitions are created */
559 dev_set_uevent_suppress(ddev, 0);
560 kobject_uevent(&ddev->kobj, KOBJ_ADD);
562 /* announce possible partitions */
563 disk_part_iter_init(&piter, disk, 0);
564 while ((part = disk_part_iter_next(&piter)))
565 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
566 disk_part_iter_exit(&piter);
570 * add_disk - add partitioning information to kernel list
571 * @disk: per-device partitioning information
573 * This function registers the partitioning information in @disk
576 * FIXME: error handling
578 void add_disk(struct gendisk *disk)
580 struct backing_dev_info *bdi;
584 /* minors == 0 indicates to use ext devt from part0 and should
585 * be accompanied with EXT_DEVT flag. Make sure all
586 * parameters make sense.
588 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
589 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
591 disk->flags |= GENHD_FL_UP;
593 retval = blk_alloc_devt(&disk->part0, &devt);
598 disk_to_dev(disk)->devt = devt;
600 /* ->major and ->first_minor aren't supposed to be
601 * dereferenced from here on, but set them just in case.
603 disk->major = MAJOR(devt);
604 disk->first_minor = MINOR(devt);
606 disk_alloc_events(disk);
608 /* Register BDI before referencing it from bdev */
609 bdi = &disk->queue->backing_dev_info;
610 bdi_register_dev(bdi, disk_devt(disk));
612 blk_register_region(disk_devt(disk), disk->minors, NULL,
613 exact_match, exact_lock, disk);
615 blk_register_queue(disk);
618 * Take an extra ref on queue which will be put on disk_release()
619 * so that it sticks around as long as @disk is there.
621 WARN_ON_ONCE(blk_get_queue(disk->queue));
623 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
627 disk_add_events(disk);
629 EXPORT_SYMBOL(add_disk);
631 void del_gendisk(struct gendisk *disk)
633 struct disk_part_iter piter;
634 struct hd_struct *part;
636 disk_del_events(disk);
638 /* invalidate stuff */
639 disk_part_iter_init(&piter, disk,
640 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
641 while ((part = disk_part_iter_next(&piter))) {
642 invalidate_partition(disk, part->partno);
643 delete_partition(disk, part->partno);
645 disk_part_iter_exit(&piter);
647 invalidate_partition(disk, 0);
648 blk_free_devt(disk_to_dev(disk)->devt);
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 mutex_lock(&ext_devt_mutex);
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 mutex_unlock(&ext_devt_mutex);
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 u8 uuid[PARTITION_META_INFO_UUIDLTH * 2 + 1];
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 part_unpack_uuid(part->info->uuid, uuid);
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), uuid);
775 if (disk->driverfs_dev != NULL &&
776 disk->driverfs_dev->driver != NULL)
777 printk(" driver: %s\n",
778 disk->driverfs_dev->driver->name);
780 printk(" (driver?)\n");
784 disk_part_iter_exit(&piter);
786 class_dev_iter_exit(&iter);
789 #ifdef CONFIG_PROC_FS
791 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
794 struct class_dev_iter *iter;
797 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
799 return ERR_PTR(-ENOMEM);
801 seqf->private = iter;
802 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
804 dev = class_dev_iter_next(iter);
809 return dev_to_disk(dev);
812 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
817 dev = class_dev_iter_next(seqf->private);
819 return dev_to_disk(dev);
824 static void disk_seqf_stop(struct seq_file *seqf, void *v)
826 struct class_dev_iter *iter = seqf->private;
828 /* stop is called even after start failed :-( */
830 class_dev_iter_exit(iter);
835 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
839 p = disk_seqf_start(seqf, pos);
840 if (!IS_ERR_OR_NULL(p) && !*pos)
841 seq_puts(seqf, "major minor #blocks name\n\n");
845 static int show_partition(struct seq_file *seqf, void *v)
847 struct gendisk *sgp = v;
848 struct disk_part_iter piter;
849 struct hd_struct *part;
850 char buf[BDEVNAME_SIZE];
852 /* Don't show non-partitionable removeable devices or empty devices */
853 if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
854 (sgp->flags & GENHD_FL_REMOVABLE)))
856 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
859 /* show the full disk and all non-0 size partitions of it */
860 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
861 while ((part = disk_part_iter_next(&piter)))
862 seq_printf(seqf, "%4d %7d %10llu %s\n",
863 MAJOR(part_devt(part)), MINOR(part_devt(part)),
864 (unsigned long long)part->nr_sects >> 1,
865 disk_name(sgp, part->partno, buf));
866 disk_part_iter_exit(&piter);
871 static const struct seq_operations partitions_op = {
872 .start = show_partition_start,
873 .next = disk_seqf_next,
874 .stop = disk_seqf_stop,
875 .show = show_partition
878 static int partitions_open(struct inode *inode, struct file *file)
880 return seq_open(file, &partitions_op);
883 static const struct file_operations proc_partitions_operations = {
884 .open = partitions_open,
887 .release = seq_release,
892 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
894 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
895 /* Make old-style 2.4 aliases work */
896 request_module("block-major-%d", MAJOR(devt));
900 static int __init genhd_device_init(void)
904 block_class.dev_kobj = sysfs_dev_block_kobj;
905 error = class_register(&block_class);
908 bdev_map = kobj_map_init(base_probe, &block_class_lock);
911 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
913 /* create top-level block dir */
914 if (!sysfs_deprecated)
915 block_depr = kobject_create_and_add("block", NULL);
919 subsys_initcall(genhd_device_init);
921 static ssize_t disk_range_show(struct device *dev,
922 struct device_attribute *attr, char *buf)
924 struct gendisk *disk = dev_to_disk(dev);
926 return sprintf(buf, "%d\n", disk->minors);
929 static ssize_t disk_ext_range_show(struct device *dev,
930 struct device_attribute *attr, char *buf)
932 struct gendisk *disk = dev_to_disk(dev);
934 return sprintf(buf, "%d\n", disk_max_parts(disk));
937 static ssize_t disk_removable_show(struct device *dev,
938 struct device_attribute *attr, char *buf)
940 struct gendisk *disk = dev_to_disk(dev);
942 return sprintf(buf, "%d\n",
943 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
946 static ssize_t disk_ro_show(struct device *dev,
947 struct device_attribute *attr, char *buf)
949 struct gendisk *disk = dev_to_disk(dev);
951 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
954 static ssize_t disk_capability_show(struct device *dev,
955 struct device_attribute *attr, char *buf)
957 struct gendisk *disk = dev_to_disk(dev);
959 return sprintf(buf, "%x\n", disk->flags);
962 static ssize_t disk_alignment_offset_show(struct device *dev,
963 struct device_attribute *attr,
966 struct gendisk *disk = dev_to_disk(dev);
968 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
971 static ssize_t disk_discard_alignment_show(struct device *dev,
972 struct device_attribute *attr,
975 struct gendisk *disk = dev_to_disk(dev);
977 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
980 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
981 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
982 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
983 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
984 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
985 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
986 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
988 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
989 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
990 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
991 #ifdef CONFIG_FAIL_MAKE_REQUEST
992 static struct device_attribute dev_attr_fail =
993 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
995 #ifdef CONFIG_FAIL_IO_TIMEOUT
996 static struct device_attribute dev_attr_fail_timeout =
997 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
1001 static struct attribute *disk_attrs[] = {
1002 &dev_attr_range.attr,
1003 &dev_attr_ext_range.attr,
1004 &dev_attr_removable.attr,
1006 &dev_attr_size.attr,
1007 &dev_attr_alignment_offset.attr,
1008 &dev_attr_discard_alignment.attr,
1009 &dev_attr_capability.attr,
1010 &dev_attr_stat.attr,
1011 &dev_attr_inflight.attr,
1012 #ifdef CONFIG_FAIL_MAKE_REQUEST
1013 &dev_attr_fail.attr,
1015 #ifdef CONFIG_FAIL_IO_TIMEOUT
1016 &dev_attr_fail_timeout.attr,
1021 static struct attribute_group disk_attr_group = {
1022 .attrs = disk_attrs,
1025 static const struct attribute_group *disk_attr_groups[] = {
1030 static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
1032 struct disk_part_tbl *ptbl =
1033 container_of(head, struct disk_part_tbl, rcu_head);
1039 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1040 * @disk: disk to replace part_tbl for
1041 * @new_ptbl: new part_tbl to install
1043 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1044 * original ptbl is freed using RCU callback.
1047 * Matching bd_mutx locked.
1049 static void disk_replace_part_tbl(struct gendisk *disk,
1050 struct disk_part_tbl *new_ptbl)
1052 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1054 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1057 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1058 call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
1063 * disk_expand_part_tbl - expand disk->part_tbl
1064 * @disk: disk to expand part_tbl for
1065 * @partno: expand such that this partno can fit in
1067 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1068 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1071 * Matching bd_mutex locked, might sleep.
1074 * 0 on success, -errno on failure.
1076 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1078 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1079 struct disk_part_tbl *new_ptbl;
1080 int len = old_ptbl ? old_ptbl->len : 0;
1081 int target = partno + 1;
1085 /* disk_max_parts() is zero during initialization, ignore if so */
1086 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1092 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1093 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1097 new_ptbl->len = target;
1099 for (i = 0; i < len; i++)
1100 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1102 disk_replace_part_tbl(disk, new_ptbl);
1106 static void disk_release(struct device *dev)
1108 struct gendisk *disk = dev_to_disk(dev);
1110 disk_release_events(disk);
1111 kfree(disk->random);
1112 disk_replace_part_tbl(disk, NULL);
1113 free_part_stats(&disk->part0);
1114 free_part_info(&disk->part0);
1116 blk_put_queue(disk->queue);
1119 struct class block_class = {
1123 static char *block_devnode(struct device *dev, mode_t *mode)
1125 struct gendisk *disk = dev_to_disk(dev);
1128 return disk->devnode(disk, mode);
1132 static struct device_type disk_type = {
1134 .groups = disk_attr_groups,
1135 .release = disk_release,
1136 .devnode = block_devnode,
1139 #ifdef CONFIG_PROC_FS
1141 * aggregate disk stat collector. Uses the same stats that the sysfs
1142 * entries do, above, but makes them available through one seq_file.
1144 * The output looks suspiciously like /proc/partitions with a bunch of
1147 static int diskstats_show(struct seq_file *seqf, void *v)
1149 struct gendisk *gp = v;
1150 struct disk_part_iter piter;
1151 struct hd_struct *hd;
1152 char buf[BDEVNAME_SIZE];
1156 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1157 seq_puts(seqf, "major minor name"
1158 " rio rmerge rsect ruse wio wmerge "
1159 "wsect wuse running use aveq"
1163 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1164 while ((hd = disk_part_iter_next(&piter))) {
1165 cpu = part_stat_lock();
1166 part_round_stats(cpu, hd);
1168 seq_printf(seqf, "%4d %7d %s %lu %lu %llu "
1169 "%u %lu %lu %llu %u %u %u %u\n",
1170 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1171 disk_name(gp, hd->partno, buf),
1172 part_stat_read(hd, ios[READ]),
1173 part_stat_read(hd, merges[READ]),
1174 (unsigned long long)part_stat_read(hd, sectors[READ]),
1175 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1176 part_stat_read(hd, ios[WRITE]),
1177 part_stat_read(hd, merges[WRITE]),
1178 (unsigned long long)part_stat_read(hd, sectors[WRITE]),
1179 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1181 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1182 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1185 disk_part_iter_exit(&piter);
1190 static const struct seq_operations diskstats_op = {
1191 .start = disk_seqf_start,
1192 .next = disk_seqf_next,
1193 .stop = disk_seqf_stop,
1194 .show = diskstats_show
1197 static int diskstats_open(struct inode *inode, struct file *file)
1199 return seq_open(file, &diskstats_op);
1202 static const struct file_operations proc_diskstats_operations = {
1203 .open = diskstats_open,
1205 .llseek = seq_lseek,
1206 .release = seq_release,
1209 static int __init proc_genhd_init(void)
1211 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1212 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1215 module_init(proc_genhd_init);
1216 #endif /* CONFIG_PROC_FS */
1218 dev_t blk_lookup_devt(const char *name, int partno)
1220 dev_t devt = MKDEV(0, 0);
1221 struct class_dev_iter iter;
1224 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1225 while ((dev = class_dev_iter_next(&iter))) {
1226 struct gendisk *disk = dev_to_disk(dev);
1227 struct hd_struct *part;
1229 if (strcmp(dev_name(dev), name))
1232 if (partno < disk->minors) {
1233 /* We need to return the right devno, even
1234 * if the partition doesn't exist yet.
1236 devt = MKDEV(MAJOR(dev->devt),
1237 MINOR(dev->devt) + partno);
1240 part = disk_get_part(disk, partno);
1242 devt = part_devt(part);
1243 disk_put_part(part);
1246 disk_put_part(part);
1248 class_dev_iter_exit(&iter);
1251 EXPORT_SYMBOL(blk_lookup_devt);
1253 struct gendisk *alloc_disk(int minors)
1255 return alloc_disk_node(minors, -1);
1257 EXPORT_SYMBOL(alloc_disk);
1259 struct gendisk *alloc_disk_node(int minors, int node_id)
1261 struct gendisk *disk;
1263 disk = kmalloc_node(sizeof(struct gendisk),
1264 GFP_KERNEL | __GFP_ZERO, node_id);
1266 if (!init_part_stats(&disk->part0)) {
1270 disk->node_id = node_id;
1271 if (disk_expand_part_tbl(disk, 0)) {
1272 free_part_stats(&disk->part0);
1276 disk->part_tbl->part[0] = &disk->part0;
1278 hd_ref_init(&disk->part0);
1280 disk->minors = minors;
1281 rand_initialize_disk(disk);
1282 disk_to_dev(disk)->class = &block_class;
1283 disk_to_dev(disk)->type = &disk_type;
1284 device_initialize(disk_to_dev(disk));
1288 EXPORT_SYMBOL(alloc_disk_node);
1290 struct kobject *get_disk(struct gendisk *disk)
1292 struct module *owner;
1293 struct kobject *kobj;
1297 owner = disk->fops->owner;
1298 if (owner && !try_module_get(owner))
1300 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1309 EXPORT_SYMBOL(get_disk);
1311 void put_disk(struct gendisk *disk)
1314 kobject_put(&disk_to_dev(disk)->kobj);
1317 EXPORT_SYMBOL(put_disk);
1319 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1321 char event[] = "DISK_RO=1";
1322 char *envp[] = { event, NULL };
1326 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1329 void set_device_ro(struct block_device *bdev, int flag)
1331 bdev->bd_part->policy = flag;
1334 EXPORT_SYMBOL(set_device_ro);
1336 void set_disk_ro(struct gendisk *disk, int flag)
1338 struct disk_part_iter piter;
1339 struct hd_struct *part;
1341 if (disk->part0.policy != flag) {
1342 set_disk_ro_uevent(disk, flag);
1343 disk->part0.policy = flag;
1346 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1347 while ((part = disk_part_iter_next(&piter)))
1348 part->policy = flag;
1349 disk_part_iter_exit(&piter);
1352 EXPORT_SYMBOL(set_disk_ro);
1354 int bdev_read_only(struct block_device *bdev)
1358 return bdev->bd_part->policy;
1361 EXPORT_SYMBOL(bdev_read_only);
1363 int invalidate_partition(struct gendisk *disk, int partno)
1366 struct block_device *bdev = bdget_disk(disk, partno);
1369 res = __invalidate_device(bdev, true);
1375 EXPORT_SYMBOL(invalidate_partition);
1378 * Disk events - monitor disk events like media change and eject request.
1380 struct disk_events {
1381 struct list_head node; /* all disk_event's */
1382 struct gendisk *disk; /* the associated disk */
1385 struct mutex block_mutex; /* protects blocking */
1386 int block; /* event blocking depth */
1387 unsigned int pending; /* events already sent out */
1388 unsigned int clearing; /* events being cleared */
1390 long poll_msecs; /* interval, -1 for default */
1391 struct delayed_work dwork;
1394 static const char *disk_events_strs[] = {
1395 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1396 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1399 static char *disk_uevents[] = {
1400 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1401 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1404 /* list of all disk_events */
1405 static DEFINE_MUTEX(disk_events_mutex);
1406 static LIST_HEAD(disk_events);
1408 /* disable in-kernel polling by default */
1409 static unsigned long disk_events_dfl_poll_msecs = 0;
1411 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1413 struct disk_events *ev = disk->ev;
1414 long intv_msecs = 0;
1417 * If device-specific poll interval is set, always use it. If
1418 * the default is being used, poll iff there are events which
1419 * can't be monitored asynchronously.
1421 if (ev->poll_msecs >= 0)
1422 intv_msecs = ev->poll_msecs;
1423 else if (disk->events & ~disk->async_events)
1424 intv_msecs = disk_events_dfl_poll_msecs;
1426 return msecs_to_jiffies(intv_msecs);
1430 * disk_block_events - block and flush disk event checking
1431 * @disk: disk to block events for
1433 * On return from this function, it is guaranteed that event checking
1434 * isn't in progress and won't happen until unblocked by
1435 * disk_unblock_events(). Events blocking is counted and the actual
1436 * unblocking happens after the matching number of unblocks are done.
1438 * Note that this intentionally does not block event checking from
1439 * disk_clear_events().
1444 void disk_block_events(struct gendisk *disk)
1446 struct disk_events *ev = disk->ev;
1447 unsigned long flags;
1454 * Outer mutex ensures that the first blocker completes canceling
1455 * the event work before further blockers are allowed to finish.
1457 mutex_lock(&ev->block_mutex);
1459 spin_lock_irqsave(&ev->lock, flags);
1460 cancel = !ev->block++;
1461 spin_unlock_irqrestore(&ev->lock, flags);
1464 cancel_delayed_work_sync(&disk->ev->dwork);
1466 mutex_unlock(&ev->block_mutex);
1469 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1471 struct disk_events *ev = disk->ev;
1473 unsigned long flags;
1475 spin_lock_irqsave(&ev->lock, flags);
1477 if (WARN_ON_ONCE(ev->block <= 0))
1484 * Not exactly a latency critical operation, set poll timer
1485 * slack to 25% and kick event check.
1487 intv = disk_events_poll_jiffies(disk);
1488 set_timer_slack(&ev->dwork.timer, intv / 4);
1490 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, 0);
1492 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, intv);
1494 spin_unlock_irqrestore(&ev->lock, flags);
1498 * disk_unblock_events - unblock disk event checking
1499 * @disk: disk to unblock events for
1501 * Undo disk_block_events(). When the block count reaches zero, it
1502 * starts events polling if configured.
1505 * Don't care. Safe to call from irq context.
1507 void disk_unblock_events(struct gendisk *disk)
1510 __disk_unblock_events(disk, false);
1514 * disk_check_events - schedule immediate event checking
1515 * @disk: disk to check events for
1517 * Schedule immediate event checking on @disk if not blocked.
1520 * Don't care. Safe to call from irq context.
1522 void disk_check_events(struct gendisk *disk)
1524 struct disk_events *ev = disk->ev;
1525 unsigned long flags;
1530 spin_lock_irqsave(&ev->lock, flags);
1532 cancel_delayed_work(&ev->dwork);
1533 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, 0);
1535 spin_unlock_irqrestore(&ev->lock, flags);
1537 EXPORT_SYMBOL_GPL(disk_check_events);
1540 * disk_clear_events - synchronously check, clear and return pending events
1541 * @disk: disk to fetch and clear events from
1542 * @mask: mask of events to be fetched and clearted
1544 * Disk events are synchronously checked and pending events in @mask
1545 * are cleared and returned. This ignores the block count.
1550 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1552 const struct block_device_operations *bdops = disk->fops;
1553 struct disk_events *ev = disk->ev;
1554 unsigned int pending;
1557 /* for drivers still using the old ->media_changed method */
1558 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1559 bdops->media_changed && bdops->media_changed(disk))
1560 return DISK_EVENT_MEDIA_CHANGE;
1564 /* tell the workfn about the events being cleared */
1565 spin_lock_irq(&ev->lock);
1566 ev->clearing |= mask;
1567 spin_unlock_irq(&ev->lock);
1569 /* uncondtionally schedule event check and wait for it to finish */
1570 disk_block_events(disk);
1571 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, 0);
1572 flush_delayed_work(&ev->dwork);
1573 __disk_unblock_events(disk, false);
1575 /* then, fetch and clear pending events */
1576 spin_lock_irq(&ev->lock);
1577 WARN_ON_ONCE(ev->clearing & mask); /* cleared by workfn */
1578 pending = ev->pending & mask;
1579 ev->pending &= ~mask;
1580 spin_unlock_irq(&ev->lock);
1585 static void disk_events_workfn(struct work_struct *work)
1587 struct delayed_work *dwork = to_delayed_work(work);
1588 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1589 struct gendisk *disk = ev->disk;
1590 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1591 unsigned int clearing = ev->clearing;
1592 unsigned int events;
1594 int nr_events = 0, i;
1597 events = disk->fops->check_events(disk, clearing);
1599 /* accumulate pending events and schedule next poll if necessary */
1600 spin_lock_irq(&ev->lock);
1602 events &= ~ev->pending;
1603 ev->pending |= events;
1604 ev->clearing &= ~clearing;
1606 intv = disk_events_poll_jiffies(disk);
1607 if (!ev->block && intv)
1608 queue_delayed_work(system_nrt_freezable_wq, &ev->dwork, intv);
1610 spin_unlock_irq(&ev->lock);
1613 * Tell userland about new events. Only the events listed in
1614 * @disk->events are reported. Unlisted events are processed the
1615 * same internally but never get reported to userland.
1617 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1618 if (events & disk->events & (1 << i))
1619 envp[nr_events++] = disk_uevents[i];
1622 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1626 * A disk events enabled device has the following sysfs nodes under
1627 * its /sys/block/X/ directory.
1629 * events : list of all supported events
1630 * events_async : list of events which can be detected w/o polling
1631 * events_poll_msecs : polling interval, 0: disable, -1: system default
1633 static ssize_t __disk_events_show(unsigned int events, char *buf)
1635 const char *delim = "";
1639 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1640 if (events & (1 << i)) {
1641 pos += sprintf(buf + pos, "%s%s",
1642 delim, disk_events_strs[i]);
1646 pos += sprintf(buf + pos, "\n");
1650 static ssize_t disk_events_show(struct device *dev,
1651 struct device_attribute *attr, char *buf)
1653 struct gendisk *disk = dev_to_disk(dev);
1655 return __disk_events_show(disk->events, buf);
1658 static ssize_t disk_events_async_show(struct device *dev,
1659 struct device_attribute *attr, char *buf)
1661 struct gendisk *disk = dev_to_disk(dev);
1663 return __disk_events_show(disk->async_events, buf);
1666 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1667 struct device_attribute *attr,
1670 struct gendisk *disk = dev_to_disk(dev);
1672 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1675 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1676 struct device_attribute *attr,
1677 const char *buf, size_t count)
1679 struct gendisk *disk = dev_to_disk(dev);
1682 if (!count || !sscanf(buf, "%ld", &intv))
1685 if (intv < 0 && intv != -1)
1688 disk_block_events(disk);
1689 disk->ev->poll_msecs = intv;
1690 __disk_unblock_events(disk, true);
1695 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1696 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1697 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1698 disk_events_poll_msecs_show,
1699 disk_events_poll_msecs_store);
1701 static const struct attribute *disk_events_attrs[] = {
1702 &dev_attr_events.attr,
1703 &dev_attr_events_async.attr,
1704 &dev_attr_events_poll_msecs.attr,
1709 * The default polling interval can be specified by the kernel
1710 * parameter block.events_dfl_poll_msecs which defaults to 0
1711 * (disable). This can also be modified runtime by writing to
1712 * /sys/module/block/events_dfl_poll_msecs.
1714 static int disk_events_set_dfl_poll_msecs(const char *val,
1715 const struct kernel_param *kp)
1717 struct disk_events *ev;
1720 ret = param_set_ulong(val, kp);
1724 mutex_lock(&disk_events_mutex);
1726 list_for_each_entry(ev, &disk_events, node)
1727 disk_check_events(ev->disk);
1729 mutex_unlock(&disk_events_mutex);
1734 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1735 .set = disk_events_set_dfl_poll_msecs,
1736 .get = param_get_ulong,
1739 #undef MODULE_PARAM_PREFIX
1740 #define MODULE_PARAM_PREFIX "block."
1742 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1743 &disk_events_dfl_poll_msecs, 0644);
1746 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1748 static void disk_alloc_events(struct gendisk *disk)
1750 struct disk_events *ev;
1752 if (!disk->fops->check_events)
1755 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1757 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1761 INIT_LIST_HEAD(&ev->node);
1763 spin_lock_init(&ev->lock);
1764 mutex_init(&ev->block_mutex);
1766 ev->poll_msecs = -1;
1767 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1772 static void disk_add_events(struct gendisk *disk)
1777 /* FIXME: error handling */
1778 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1779 pr_warn("%s: failed to create sysfs files for events\n",
1782 mutex_lock(&disk_events_mutex);
1783 list_add_tail(&disk->ev->node, &disk_events);
1784 mutex_unlock(&disk_events_mutex);
1787 * Block count is initialized to 1 and the following initial
1788 * unblock kicks it into action.
1790 __disk_unblock_events(disk, true);
1793 static void disk_del_events(struct gendisk *disk)
1798 disk_block_events(disk);
1800 mutex_lock(&disk_events_mutex);
1801 list_del_init(&disk->ev->node);
1802 mutex_unlock(&disk_events_mutex);
1804 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1807 static void disk_release_events(struct gendisk *disk)
1809 /* the block count should be 1 from disk_del_events() */
1810 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);