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_check_events(struct disk_events *ev,
39 unsigned int *clearing_ptr);
40 static void disk_alloc_events(struct gendisk *disk);
41 static void disk_add_events(struct gendisk *disk);
42 static void disk_del_events(struct gendisk *disk);
43 static void disk_release_events(struct gendisk *disk);
46 * disk_get_part - get partition
47 * @disk: disk to look partition from
48 * @partno: partition number
50 * Look for partition @partno from @disk. If found, increment
51 * reference count and return it.
57 * Pointer to the found partition on success, NULL if not found.
59 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
61 struct hd_struct *part = NULL;
62 struct disk_part_tbl *ptbl;
64 if (unlikely(partno < 0))
69 ptbl = rcu_dereference(disk->part_tbl);
70 if (likely(partno < ptbl->len)) {
71 part = rcu_dereference(ptbl->part[partno]);
73 get_device(part_to_dev(part));
80 EXPORT_SYMBOL_GPL(disk_get_part);
83 * disk_part_iter_init - initialize partition iterator
84 * @piter: iterator to initialize
85 * @disk: disk to iterate over
86 * @flags: DISK_PITER_* flags
88 * Initialize @piter so that it iterates over partitions of @disk.
93 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
96 struct disk_part_tbl *ptbl;
99 ptbl = rcu_dereference(disk->part_tbl);
104 if (flags & DISK_PITER_REVERSE)
105 piter->idx = ptbl->len - 1;
106 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
111 piter->flags = flags;
115 EXPORT_SYMBOL_GPL(disk_part_iter_init);
118 * disk_part_iter_next - proceed iterator to the next partition and return it
119 * @piter: iterator of interest
121 * Proceed @piter to the next partition and return it.
126 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
128 struct disk_part_tbl *ptbl;
131 /* put the last partition */
132 disk_put_part(piter->part);
137 ptbl = rcu_dereference(piter->disk->part_tbl);
139 /* determine iteration parameters */
140 if (piter->flags & DISK_PITER_REVERSE) {
142 if (piter->flags & (DISK_PITER_INCL_PART0 |
143 DISK_PITER_INCL_EMPTY_PART0))
152 /* iterate to the next partition */
153 for (; piter->idx != end; piter->idx += inc) {
154 struct hd_struct *part;
156 part = rcu_dereference(ptbl->part[piter->idx]);
159 if (!part_nr_sects_read(part) &&
160 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
161 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
165 get_device(part_to_dev(part));
175 EXPORT_SYMBOL_GPL(disk_part_iter_next);
178 * disk_part_iter_exit - finish up partition iteration
179 * @piter: iter of interest
181 * Called when iteration is over. Cleans up @piter.
186 void disk_part_iter_exit(struct disk_part_iter *piter)
188 disk_put_part(piter->part);
191 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
193 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
195 return part->start_sect <= sector &&
196 sector < part->start_sect + part_nr_sects_read(part);
200 * disk_map_sector_rcu - map sector to partition
201 * @disk: gendisk of interest
202 * @sector: sector to map
204 * Find out which partition @sector maps to on @disk. This is
205 * primarily used for stats accounting.
208 * RCU read locked. The returned partition pointer is valid only
209 * while preemption is disabled.
212 * Found partition on success, part0 is returned if no partition matches
214 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
216 struct disk_part_tbl *ptbl;
217 struct hd_struct *part;
220 ptbl = rcu_dereference(disk->part_tbl);
222 part = rcu_dereference(ptbl->last_lookup);
223 if (part && sector_in_part(part, sector))
226 for (i = 1; i < ptbl->len; i++) {
227 part = rcu_dereference(ptbl->part[i]);
229 if (part && sector_in_part(part, sector)) {
230 rcu_assign_pointer(ptbl->last_lookup, part);
236 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
239 * Can be deleted altogether. Later.
242 static struct blk_major_name {
243 struct blk_major_name *next;
246 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
248 /* index in the above - for now: assume no multimajor ranges */
249 static inline int major_to_index(unsigned major)
251 return major % BLKDEV_MAJOR_HASH_SIZE;
254 #ifdef CONFIG_PROC_FS
255 void blkdev_show(struct seq_file *seqf, off_t offset)
257 struct blk_major_name *dp;
259 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
260 mutex_lock(&block_class_lock);
261 for (dp = major_names[offset]; dp; dp = dp->next)
262 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
263 mutex_unlock(&block_class_lock);
266 #endif /* CONFIG_PROC_FS */
269 * register_blkdev - register a new block device
271 * @major: the requested major device number [1..255]. If @major=0, try to
272 * allocate any unused major number.
273 * @name: the name of the new block device as a zero terminated string
275 * The @name must be unique within the system.
277 * The return value depends on the @major input parameter.
278 * - if a major device number was requested in range [1..255] then the
279 * function returns zero on success, or a negative error code
280 * - if any unused major number was requested with @major=0 parameter
281 * then the return value is the allocated major number in range
282 * [1..255] or a negative error code otherwise
284 int register_blkdev(unsigned int major, const char *name)
286 struct blk_major_name **n, *p;
289 mutex_lock(&block_class_lock);
293 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
294 if (major_names[index] == NULL)
299 printk("register_blkdev: failed to get major for %s\n",
308 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
315 strlcpy(p->name, name, sizeof(p->name));
317 index = major_to_index(major);
319 for (n = &major_names[index]; *n; n = &(*n)->next) {
320 if ((*n)->major == major)
329 printk("register_blkdev: cannot get major %d for %s\n",
334 mutex_unlock(&block_class_lock);
338 EXPORT_SYMBOL(register_blkdev);
340 void unregister_blkdev(unsigned int major, const char *name)
342 struct blk_major_name **n;
343 struct blk_major_name *p = NULL;
344 int index = major_to_index(major);
346 mutex_lock(&block_class_lock);
347 for (n = &major_names[index]; *n; n = &(*n)->next)
348 if ((*n)->major == major)
350 if (!*n || strcmp((*n)->name, name)) {
356 mutex_unlock(&block_class_lock);
360 EXPORT_SYMBOL(unregister_blkdev);
362 static struct kobj_map *bdev_map;
365 * blk_mangle_minor - scatter minor numbers apart
366 * @minor: minor number to mangle
368 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
369 * is enabled. Mangling twice gives the original value.
377 static int blk_mangle_minor(int minor)
379 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
382 for (i = 0; i < MINORBITS / 2; i++) {
383 int low = minor & (1 << i);
384 int high = minor & (1 << (MINORBITS - 1 - i));
385 int distance = MINORBITS - 1 - 2 * i;
387 minor ^= low | high; /* clear both bits */
388 low <<= distance; /* swap the positions */
390 minor |= low | high; /* and set */
397 * blk_alloc_devt - allocate a dev_t for a partition
398 * @part: partition to allocate dev_t for
399 * @devt: out parameter for resulting dev_t
401 * Allocate a dev_t for block device.
404 * 0 on success, allocated dev_t is returned in *@devt. -errno on
410 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
412 struct gendisk *disk = part_to_disk(part);
415 /* in consecutive minor range? */
416 if (part->partno < disk->minors) {
417 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
421 /* allocate ext devt */
423 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
425 rc = idr_get_new(&ext_devt_idr, part, &idx);
426 } while (rc == -EAGAIN);
431 if (idx > MAX_EXT_DEVT) {
432 idr_remove(&ext_devt_idr, idx);
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 mutex_lock(&ext_devt_mutex);
458 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
459 mutex_unlock(&ext_devt_mutex);
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 static 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, 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 blk_free_devt(disk_to_dev(disk)->devt);
650 set_capacity(disk, 0);
651 disk->flags &= ~GENHD_FL_UP;
653 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
654 bdi_unregister(&disk->queue->backing_dev_info);
655 blk_unregister_queue(disk);
656 blk_unregister_region(disk_devt(disk), disk->minors);
658 part_stat_set_all(&disk->part0, 0);
659 disk->part0.stamp = 0;
661 kobject_put(disk->part0.holder_dir);
662 kobject_put(disk->slave_dir);
663 disk->driverfs_dev = NULL;
664 if (!sysfs_deprecated)
665 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
666 device_del(disk_to_dev(disk));
668 EXPORT_SYMBOL(del_gendisk);
671 * get_gendisk - get partitioning information for a given device
672 * @devt: device to get partitioning information for
673 * @partno: returned partition index
675 * This function gets the structure containing partitioning
676 * information for the given device @devt.
678 struct gendisk *get_gendisk(dev_t devt, int *partno)
680 struct gendisk *disk = NULL;
682 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
683 struct kobject *kobj;
685 kobj = kobj_lookup(bdev_map, devt, partno);
687 disk = dev_to_disk(kobj_to_dev(kobj));
689 struct hd_struct *part;
691 mutex_lock(&ext_devt_mutex);
692 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
693 if (part && get_disk(part_to_disk(part))) {
694 *partno = part->partno;
695 disk = part_to_disk(part);
697 mutex_unlock(&ext_devt_mutex);
702 EXPORT_SYMBOL(get_gendisk);
705 * bdget_disk - do bdget() by gendisk and partition number
706 * @disk: gendisk of interest
707 * @partno: partition number
709 * Find partition @partno from @disk, do bdget() on it.
715 * Resulting block_device on success, NULL on failure.
717 struct block_device *bdget_disk(struct gendisk *disk, int partno)
719 struct hd_struct *part;
720 struct block_device *bdev = NULL;
722 part = disk_get_part(disk, partno);
724 bdev = bdget(part_devt(part));
729 EXPORT_SYMBOL(bdget_disk);
732 * print a full list of all partitions - intended for places where the root
733 * filesystem can't be mounted and thus to give the victim some idea of what
736 void __init printk_all_partitions(void)
738 struct class_dev_iter iter;
741 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
742 while ((dev = class_dev_iter_next(&iter))) {
743 struct gendisk *disk = dev_to_disk(dev);
744 struct disk_part_iter piter;
745 struct hd_struct *part;
746 char name_buf[BDEVNAME_SIZE];
747 char devt_buf[BDEVT_SIZE];
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;
766 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
767 bdevt_str(part_devt(part), devt_buf),
768 (unsigned long long)part_nr_sects_read(part) >> 1
769 , disk_name(disk, part->partno, name_buf),
770 part->info ? part->info->uuid : "");
772 if (disk->driverfs_dev != NULL &&
773 disk->driverfs_dev->driver != NULL)
774 printk(" driver: %s\n",
775 disk->driverfs_dev->driver->name);
777 printk(" (driver?)\n");
781 disk_part_iter_exit(&piter);
783 class_dev_iter_exit(&iter);
786 #ifdef CONFIG_PROC_FS
788 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
791 struct class_dev_iter *iter;
794 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
796 return ERR_PTR(-ENOMEM);
798 seqf->private = iter;
799 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
801 dev = class_dev_iter_next(iter);
806 return dev_to_disk(dev);
809 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
814 dev = class_dev_iter_next(seqf->private);
816 return dev_to_disk(dev);
821 static void disk_seqf_stop(struct seq_file *seqf, void *v)
823 struct class_dev_iter *iter = seqf->private;
825 /* stop is called even after start failed :-( */
827 class_dev_iter_exit(iter);
832 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
836 p = disk_seqf_start(seqf, pos);
837 if (!IS_ERR_OR_NULL(p) && !*pos)
838 seq_puts(seqf, "major minor #blocks name\n\n");
842 static int show_partition(struct seq_file *seqf, void *v)
844 struct gendisk *sgp = v;
845 struct disk_part_iter piter;
846 struct hd_struct *part;
847 char buf[BDEVNAME_SIZE];
849 /* Don't show non-partitionable removeable devices or empty devices */
850 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
851 (sgp->flags & GENHD_FL_REMOVABLE)))
853 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
856 /* show the full disk and all non-0 size partitions of it */
857 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
858 while ((part = disk_part_iter_next(&piter)))
859 seq_printf(seqf, "%4d %7d %10llu %s\n",
860 MAJOR(part_devt(part)), MINOR(part_devt(part)),
861 (unsigned long long)part_nr_sects_read(part) >> 1,
862 disk_name(sgp, part->partno, buf));
863 disk_part_iter_exit(&piter);
868 static const struct seq_operations partitions_op = {
869 .start = show_partition_start,
870 .next = disk_seqf_next,
871 .stop = disk_seqf_stop,
872 .show = show_partition
875 static int partitions_open(struct inode *inode, struct file *file)
877 return seq_open(file, &partitions_op);
880 static const struct file_operations proc_partitions_operations = {
881 .open = partitions_open,
884 .release = seq_release,
889 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
891 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
892 /* Make old-style 2.4 aliases work */
893 request_module("block-major-%d", MAJOR(devt));
897 static int __init genhd_device_init(void)
901 block_class.dev_kobj = sysfs_dev_block_kobj;
902 error = class_register(&block_class);
905 bdev_map = kobj_map_init(base_probe, &block_class_lock);
908 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
910 /* create top-level block dir */
911 if (!sysfs_deprecated)
912 block_depr = kobject_create_and_add("block", NULL);
916 subsys_initcall(genhd_device_init);
918 static ssize_t disk_range_show(struct device *dev,
919 struct device_attribute *attr, char *buf)
921 struct gendisk *disk = dev_to_disk(dev);
923 return sprintf(buf, "%d\n", disk->minors);
926 static ssize_t disk_ext_range_show(struct device *dev,
927 struct device_attribute *attr, char *buf)
929 struct gendisk *disk = dev_to_disk(dev);
931 return sprintf(buf, "%d\n", disk_max_parts(disk));
934 static ssize_t disk_removable_show(struct device *dev,
935 struct device_attribute *attr, char *buf)
937 struct gendisk *disk = dev_to_disk(dev);
939 return sprintf(buf, "%d\n",
940 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
943 static ssize_t disk_ro_show(struct device *dev,
944 struct device_attribute *attr, char *buf)
946 struct gendisk *disk = dev_to_disk(dev);
948 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
951 static ssize_t disk_capability_show(struct device *dev,
952 struct device_attribute *attr, char *buf)
954 struct gendisk *disk = dev_to_disk(dev);
956 return sprintf(buf, "%x\n", disk->flags);
959 static ssize_t disk_alignment_offset_show(struct device *dev,
960 struct device_attribute *attr,
963 struct gendisk *disk = dev_to_disk(dev);
965 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
968 static ssize_t disk_discard_alignment_show(struct device *dev,
969 struct device_attribute *attr,
972 struct gendisk *disk = dev_to_disk(dev);
974 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
977 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
978 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
979 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
980 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
981 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
982 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
983 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
985 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
986 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
987 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
988 #ifdef CONFIG_FAIL_MAKE_REQUEST
989 static struct device_attribute dev_attr_fail =
990 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
992 #ifdef CONFIG_FAIL_IO_TIMEOUT
993 static struct device_attribute dev_attr_fail_timeout =
994 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
998 static struct attribute *disk_attrs[] = {
999 &dev_attr_range.attr,
1000 &dev_attr_ext_range.attr,
1001 &dev_attr_removable.attr,
1003 &dev_attr_size.attr,
1004 &dev_attr_alignment_offset.attr,
1005 &dev_attr_discard_alignment.attr,
1006 &dev_attr_capability.attr,
1007 &dev_attr_stat.attr,
1008 &dev_attr_inflight.attr,
1009 #ifdef CONFIG_FAIL_MAKE_REQUEST
1010 &dev_attr_fail.attr,
1012 #ifdef CONFIG_FAIL_IO_TIMEOUT
1013 &dev_attr_fail_timeout.attr,
1018 static struct attribute_group disk_attr_group = {
1019 .attrs = disk_attrs,
1022 static const struct attribute_group *disk_attr_groups[] = {
1028 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1029 * @disk: disk to replace part_tbl for
1030 * @new_ptbl: new part_tbl to install
1032 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1033 * original ptbl is freed using RCU callback.
1036 * Matching bd_mutx locked.
1038 static void disk_replace_part_tbl(struct gendisk *disk,
1039 struct disk_part_tbl *new_ptbl)
1041 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1043 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1046 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1047 kfree_rcu(old_ptbl, rcu_head);
1052 * disk_expand_part_tbl - expand disk->part_tbl
1053 * @disk: disk to expand part_tbl for
1054 * @partno: expand such that this partno can fit in
1056 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1057 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1060 * Matching bd_mutex locked, might sleep.
1063 * 0 on success, -errno on failure.
1065 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1067 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1068 struct disk_part_tbl *new_ptbl;
1069 int len = old_ptbl ? old_ptbl->len : 0;
1070 int target = partno + 1;
1074 /* disk_max_parts() is zero during initialization, ignore if so */
1075 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1081 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1082 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1086 new_ptbl->len = target;
1088 for (i = 0; i < len; i++)
1089 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1091 disk_replace_part_tbl(disk, new_ptbl);
1095 static void disk_release(struct device *dev)
1097 struct gendisk *disk = dev_to_disk(dev);
1099 disk_release_events(disk);
1100 kfree(disk->random);
1101 disk_replace_part_tbl(disk, NULL);
1102 free_part_stats(&disk->part0);
1103 free_part_info(&disk->part0);
1105 blk_put_queue(disk->queue);
1108 struct class block_class = {
1112 static char *block_devnode(struct device *dev, umode_t *mode)
1114 struct gendisk *disk = dev_to_disk(dev);
1117 return disk->devnode(disk, mode);
1121 static struct device_type disk_type = {
1123 .groups = disk_attr_groups,
1124 .release = disk_release,
1125 .devnode = block_devnode,
1128 #ifdef CONFIG_PROC_FS
1130 * aggregate disk stat collector. Uses the same stats that the sysfs
1131 * entries do, above, but makes them available through one seq_file.
1133 * The output looks suspiciously like /proc/partitions with a bunch of
1136 static int diskstats_show(struct seq_file *seqf, void *v)
1138 struct gendisk *gp = v;
1139 struct disk_part_iter piter;
1140 struct hd_struct *hd;
1141 char buf[BDEVNAME_SIZE];
1145 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1146 seq_puts(seqf, "major minor name"
1147 " rio rmerge rsect ruse wio wmerge "
1148 "wsect wuse running use aveq"
1152 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1153 while ((hd = disk_part_iter_next(&piter))) {
1154 cpu = part_stat_lock();
1155 part_round_stats(cpu, hd);
1157 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1158 "%u %lu %lu %lu %u %u %u %u\n",
1159 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1160 disk_name(gp, hd->partno, buf),
1161 part_stat_read(hd, ios[READ]),
1162 part_stat_read(hd, merges[READ]),
1163 part_stat_read(hd, sectors[READ]),
1164 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1165 part_stat_read(hd, ios[WRITE]),
1166 part_stat_read(hd, merges[WRITE]),
1167 part_stat_read(hd, sectors[WRITE]),
1168 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1170 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1171 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1174 disk_part_iter_exit(&piter);
1179 static const struct seq_operations diskstats_op = {
1180 .start = disk_seqf_start,
1181 .next = disk_seqf_next,
1182 .stop = disk_seqf_stop,
1183 .show = diskstats_show
1186 static int diskstats_open(struct inode *inode, struct file *file)
1188 return seq_open(file, &diskstats_op);
1191 static const struct file_operations proc_diskstats_operations = {
1192 .open = diskstats_open,
1194 .llseek = seq_lseek,
1195 .release = seq_release,
1198 static int __init proc_genhd_init(void)
1200 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1201 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1204 module_init(proc_genhd_init);
1205 #endif /* CONFIG_PROC_FS */
1207 dev_t blk_lookup_devt(const char *name, int partno)
1209 dev_t devt = MKDEV(0, 0);
1210 struct class_dev_iter iter;
1213 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1214 while ((dev = class_dev_iter_next(&iter))) {
1215 struct gendisk *disk = dev_to_disk(dev);
1216 struct hd_struct *part;
1218 if (strcmp(dev_name(dev), name))
1221 if (partno < disk->minors) {
1222 /* We need to return the right devno, even
1223 * if the partition doesn't exist yet.
1225 devt = MKDEV(MAJOR(dev->devt),
1226 MINOR(dev->devt) + partno);
1229 part = disk_get_part(disk, partno);
1231 devt = part_devt(part);
1232 disk_put_part(part);
1235 disk_put_part(part);
1237 class_dev_iter_exit(&iter);
1240 EXPORT_SYMBOL(blk_lookup_devt);
1242 struct gendisk *alloc_disk(int minors)
1244 return alloc_disk_node(minors, NUMA_NO_NODE);
1246 EXPORT_SYMBOL(alloc_disk);
1248 struct gendisk *alloc_disk_node(int minors, int node_id)
1250 struct gendisk *disk;
1252 disk = kmalloc_node(sizeof(struct gendisk),
1253 GFP_KERNEL | __GFP_ZERO, node_id);
1255 if (!init_part_stats(&disk->part0)) {
1259 disk->node_id = node_id;
1260 if (disk_expand_part_tbl(disk, 0)) {
1261 free_part_stats(&disk->part0);
1265 disk->part_tbl->part[0] = &disk->part0;
1268 * set_capacity() and get_capacity() currently don't use
1269 * seqcounter to read/update the part0->nr_sects. Still init
1270 * the counter as we can read the sectors in IO submission
1271 * patch using seqence counters.
1273 * TODO: Ideally set_capacity() and get_capacity() should be
1274 * converted to make use of bd_mutex and sequence counters.
1276 seqcount_init(&disk->part0.nr_sects_seq);
1277 hd_ref_init(&disk->part0);
1279 disk->minors = minors;
1280 rand_initialize_disk(disk);
1281 disk_to_dev(disk)->class = &block_class;
1282 disk_to_dev(disk)->type = &disk_type;
1283 device_initialize(disk_to_dev(disk));
1287 EXPORT_SYMBOL(alloc_disk_node);
1289 struct kobject *get_disk(struct gendisk *disk)
1291 struct module *owner;
1292 struct kobject *kobj;
1296 owner = disk->fops->owner;
1297 if (owner && !try_module_get(owner))
1299 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1308 EXPORT_SYMBOL(get_disk);
1310 void put_disk(struct gendisk *disk)
1313 kobject_put(&disk_to_dev(disk)->kobj);
1316 EXPORT_SYMBOL(put_disk);
1318 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1320 char event[] = "DISK_RO=1";
1321 char *envp[] = { event, NULL };
1325 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1328 void set_device_ro(struct block_device *bdev, int flag)
1330 bdev->bd_part->policy = flag;
1333 EXPORT_SYMBOL(set_device_ro);
1335 void set_disk_ro(struct gendisk *disk, int flag)
1337 struct disk_part_iter piter;
1338 struct hd_struct *part;
1340 if (disk->part0.policy != flag) {
1341 set_disk_ro_uevent(disk, flag);
1342 disk->part0.policy = flag;
1345 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1346 while ((part = disk_part_iter_next(&piter)))
1347 part->policy = flag;
1348 disk_part_iter_exit(&piter);
1351 EXPORT_SYMBOL(set_disk_ro);
1353 int bdev_read_only(struct block_device *bdev)
1357 return bdev->bd_part->policy;
1360 EXPORT_SYMBOL(bdev_read_only);
1362 int invalidate_partition(struct gendisk *disk, int partno)
1365 struct block_device *bdev = bdget_disk(disk, partno);
1368 res = __invalidate_device(bdev, true);
1374 EXPORT_SYMBOL(invalidate_partition);
1377 * Disk events - monitor disk events like media change and eject request.
1379 struct disk_events {
1380 struct list_head node; /* all disk_event's */
1381 struct gendisk *disk; /* the associated disk */
1384 struct mutex block_mutex; /* protects blocking */
1385 int block; /* event blocking depth */
1386 unsigned int pending; /* events already sent out */
1387 unsigned int clearing; /* events being cleared */
1389 long poll_msecs; /* interval, -1 for default */
1390 struct delayed_work dwork;
1393 static const char *disk_events_strs[] = {
1394 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1395 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1398 static char *disk_uevents[] = {
1399 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1400 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1403 /* list of all disk_events */
1404 static DEFINE_MUTEX(disk_events_mutex);
1405 static LIST_HEAD(disk_events);
1407 /* disable in-kernel polling by default */
1408 static unsigned long disk_events_dfl_poll_msecs = 0;
1410 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1412 struct disk_events *ev = disk->ev;
1413 long intv_msecs = 0;
1416 * If device-specific poll interval is set, always use it. If
1417 * the default is being used, poll iff there are events which
1418 * can't be monitored asynchronously.
1420 if (ev->poll_msecs >= 0)
1421 intv_msecs = ev->poll_msecs;
1422 else if (disk->events & ~disk->async_events)
1423 intv_msecs = disk_events_dfl_poll_msecs;
1425 return msecs_to_jiffies(intv_msecs);
1429 * disk_block_events - block and flush disk event checking
1430 * @disk: disk to block events for
1432 * On return from this function, it is guaranteed that event checking
1433 * isn't in progress and won't happen until unblocked by
1434 * disk_unblock_events(). Events blocking is counted and the actual
1435 * unblocking happens after the matching number of unblocks are done.
1437 * Note that this intentionally does not block event checking from
1438 * disk_clear_events().
1443 void disk_block_events(struct gendisk *disk)
1445 struct disk_events *ev = disk->ev;
1446 unsigned long flags;
1453 * Outer mutex ensures that the first blocker completes canceling
1454 * the event work before further blockers are allowed to finish.
1456 mutex_lock(&ev->block_mutex);
1458 spin_lock_irqsave(&ev->lock, flags);
1459 cancel = !ev->block++;
1460 spin_unlock_irqrestore(&ev->lock, flags);
1463 cancel_delayed_work_sync(&disk->ev->dwork);
1465 mutex_unlock(&ev->block_mutex);
1468 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1470 struct disk_events *ev = disk->ev;
1472 unsigned long flags;
1474 spin_lock_irqsave(&ev->lock, flags);
1476 if (WARN_ON_ONCE(ev->block <= 0))
1483 * Not exactly a latency critical operation, set poll timer
1484 * slack to 25% and kick event check.
1486 intv = disk_events_poll_jiffies(disk);
1487 set_timer_slack(&ev->dwork.timer, intv / 4);
1489 queue_delayed_work(system_freezable_wq, &ev->dwork, 0);
1491 queue_delayed_work(system_freezable_wq, &ev->dwork, intv);
1493 spin_unlock_irqrestore(&ev->lock, flags);
1497 * disk_unblock_events - unblock disk event checking
1498 * @disk: disk to unblock events for
1500 * Undo disk_block_events(). When the block count reaches zero, it
1501 * starts events polling if configured.
1504 * Don't care. Safe to call from irq context.
1506 void disk_unblock_events(struct gendisk *disk)
1509 __disk_unblock_events(disk, false);
1513 * disk_flush_events - schedule immediate event checking and flushing
1514 * @disk: disk to check and flush events for
1515 * @mask: events to flush
1517 * Schedule immediate event checking on @disk if not blocked. Events in
1518 * @mask are scheduled to be cleared from the driver. Note that this
1519 * doesn't clear the events from @disk->ev.
1522 * If @mask is non-zero must be called with bdev->bd_mutex held.
1524 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1526 struct disk_events *ev = disk->ev;
1531 spin_lock_irq(&ev->lock);
1532 ev->clearing |= mask;
1534 mod_delayed_work(system_freezable_wq, &ev->dwork, 0);
1535 spin_unlock_irq(&ev->lock);
1539 * disk_clear_events - synchronously check, clear and return pending events
1540 * @disk: disk to fetch and clear events from
1541 * @mask: mask of events to be fetched and clearted
1543 * Disk events are synchronously checked and pending events in @mask
1544 * are cleared and returned. This ignores the block count.
1549 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1551 const struct block_device_operations *bdops = disk->fops;
1552 struct disk_events *ev = disk->ev;
1553 unsigned int pending;
1554 unsigned int clearing = mask;
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 disk_block_events(disk);
1567 * store the union of mask and ev->clearing on the stack so that the
1568 * race with disk_flush_events does not cause ambiguity (ev->clearing
1569 * can still be modified even if events are blocked).
1571 spin_lock_irq(&ev->lock);
1572 clearing |= ev->clearing;
1574 spin_unlock_irq(&ev->lock);
1576 disk_check_events(ev, &clearing);
1578 * if ev->clearing is not 0, the disk_flush_events got called in the
1579 * middle of this function, so we want to run the workfn without delay.
1581 __disk_unblock_events(disk, ev->clearing ? true : false);
1583 /* then, fetch and clear pending events */
1584 spin_lock_irq(&ev->lock);
1585 pending = ev->pending & mask;
1586 ev->pending &= ~mask;
1587 spin_unlock_irq(&ev->lock);
1588 WARN_ON_ONCE(clearing & mask);
1594 * Separate this part out so that a different pointer for clearing_ptr can be
1595 * passed in for disk_clear_events.
1597 static void disk_events_workfn(struct work_struct *work)
1599 struct delayed_work *dwork = to_delayed_work(work);
1600 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1602 disk_check_events(ev, &ev->clearing);
1605 static void disk_check_events(struct disk_events *ev,
1606 unsigned int *clearing_ptr)
1608 struct gendisk *disk = ev->disk;
1609 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1610 unsigned int clearing = *clearing_ptr;
1611 unsigned int events;
1613 int nr_events = 0, i;
1616 events = disk->fops->check_events(disk, clearing);
1618 /* accumulate pending events and schedule next poll if necessary */
1619 spin_lock_irq(&ev->lock);
1621 events &= ~ev->pending;
1622 ev->pending |= events;
1623 *clearing_ptr &= ~clearing;
1625 intv = disk_events_poll_jiffies(disk);
1626 if (!ev->block && intv)
1627 queue_delayed_work(system_freezable_wq, &ev->dwork, intv);
1629 spin_unlock_irq(&ev->lock);
1632 * Tell userland about new events. Only the events listed in
1633 * @disk->events are reported. Unlisted events are processed the
1634 * same internally but never get reported to userland.
1636 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1637 if (events & disk->events & (1 << i))
1638 envp[nr_events++] = disk_uevents[i];
1641 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1645 * A disk events enabled device has the following sysfs nodes under
1646 * its /sys/block/X/ directory.
1648 * events : list of all supported events
1649 * events_async : list of events which can be detected w/o polling
1650 * events_poll_msecs : polling interval, 0: disable, -1: system default
1652 static ssize_t __disk_events_show(unsigned int events, char *buf)
1654 const char *delim = "";
1658 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1659 if (events & (1 << i)) {
1660 pos += sprintf(buf + pos, "%s%s",
1661 delim, disk_events_strs[i]);
1665 pos += sprintf(buf + pos, "\n");
1669 static ssize_t disk_events_show(struct device *dev,
1670 struct device_attribute *attr, char *buf)
1672 struct gendisk *disk = dev_to_disk(dev);
1674 return __disk_events_show(disk->events, buf);
1677 static ssize_t disk_events_async_show(struct device *dev,
1678 struct device_attribute *attr, char *buf)
1680 struct gendisk *disk = dev_to_disk(dev);
1682 return __disk_events_show(disk->async_events, buf);
1685 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1686 struct device_attribute *attr,
1689 struct gendisk *disk = dev_to_disk(dev);
1691 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1694 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1695 struct device_attribute *attr,
1696 const char *buf, size_t count)
1698 struct gendisk *disk = dev_to_disk(dev);
1701 if (!count || !sscanf(buf, "%ld", &intv))
1704 if (intv < 0 && intv != -1)
1707 disk_block_events(disk);
1708 disk->ev->poll_msecs = intv;
1709 __disk_unblock_events(disk, true);
1714 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1715 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1716 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1717 disk_events_poll_msecs_show,
1718 disk_events_poll_msecs_store);
1720 static const struct attribute *disk_events_attrs[] = {
1721 &dev_attr_events.attr,
1722 &dev_attr_events_async.attr,
1723 &dev_attr_events_poll_msecs.attr,
1728 * The default polling interval can be specified by the kernel
1729 * parameter block.events_dfl_poll_msecs which defaults to 0
1730 * (disable). This can also be modified runtime by writing to
1731 * /sys/module/block/events_dfl_poll_msecs.
1733 static int disk_events_set_dfl_poll_msecs(const char *val,
1734 const struct kernel_param *kp)
1736 struct disk_events *ev;
1739 ret = param_set_ulong(val, kp);
1743 mutex_lock(&disk_events_mutex);
1745 list_for_each_entry(ev, &disk_events, node)
1746 disk_flush_events(ev->disk, 0);
1748 mutex_unlock(&disk_events_mutex);
1753 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1754 .set = disk_events_set_dfl_poll_msecs,
1755 .get = param_get_ulong,
1758 #undef MODULE_PARAM_PREFIX
1759 #define MODULE_PARAM_PREFIX "block."
1761 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1762 &disk_events_dfl_poll_msecs, 0644);
1765 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1767 static void disk_alloc_events(struct gendisk *disk)
1769 struct disk_events *ev;
1771 if (!disk->fops->check_events)
1774 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1776 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1780 INIT_LIST_HEAD(&ev->node);
1782 spin_lock_init(&ev->lock);
1783 mutex_init(&ev->block_mutex);
1785 ev->poll_msecs = -1;
1786 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1791 static void disk_add_events(struct gendisk *disk)
1796 /* FIXME: error handling */
1797 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1798 pr_warn("%s: failed to create sysfs files for events\n",
1801 mutex_lock(&disk_events_mutex);
1802 list_add_tail(&disk->ev->node, &disk_events);
1803 mutex_unlock(&disk_events_mutex);
1806 * Block count is initialized to 1 and the following initial
1807 * unblock kicks it into action.
1809 __disk_unblock_events(disk, true);
1812 static void disk_del_events(struct gendisk *disk)
1817 disk_block_events(disk);
1819 mutex_lock(&disk_events_mutex);
1820 list_del_init(&disk->ev->node);
1821 mutex_unlock(&disk_events_mutex);
1823 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1826 static void disk_release_events(struct gendisk *disk)
1828 /* the block count should be 1 from disk_del_events() */
1829 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);