5 #include <linux/module.h>
7 #include <linux/genhd.h>
8 #include <linux/kdev_t.h>
9 #include <linux/kernel.h>
10 #include <linux/blkdev.h>
11 #include <linux/init.h>
12 #include <linux/spinlock.h>
13 #include <linux/proc_fs.h>
14 #include <linux/seq_file.h>
15 #include <linux/slab.h>
16 #include <linux/kmod.h>
17 #include <linux/kobj_map.h>
18 #include <linux/buffer_head.h>
19 #include <linux/mutex.h>
20 #include <linux/idr.h>
21 #include <linux/log2.h>
25 static DEFINE_MUTEX(block_class_lock);
26 struct kobject *block_depr;
28 /* for extended dynamic devt allocation, currently only one major is used */
29 #define MAX_EXT_DEVT (1 << MINORBITS)
31 /* For extended devt allocation. ext_devt_mutex prevents look up
32 * results from going away underneath its user.
34 static DEFINE_MUTEX(ext_devt_mutex);
35 static DEFINE_IDR(ext_devt_idr);
37 static struct device_type disk_type;
39 static void disk_add_events(struct gendisk *disk);
40 static void disk_del_events(struct gendisk *disk);
41 static void disk_release_events(struct gendisk *disk);
44 * disk_get_part - get partition
45 * @disk: disk to look partition from
46 * @partno: partition number
48 * Look for partition @partno from @disk. If found, increment
49 * reference count and return it.
55 * Pointer to the found partition on success, NULL if not found.
57 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
59 struct hd_struct *part = NULL;
60 struct disk_part_tbl *ptbl;
62 if (unlikely(partno < 0))
67 ptbl = rcu_dereference(disk->part_tbl);
68 if (likely(partno < ptbl->len)) {
69 part = rcu_dereference(ptbl->part[partno]);
71 get_device(part_to_dev(part));
78 EXPORT_SYMBOL_GPL(disk_get_part);
81 * disk_part_iter_init - initialize partition iterator
82 * @piter: iterator to initialize
83 * @disk: disk to iterate over
84 * @flags: DISK_PITER_* flags
86 * Initialize @piter so that it iterates over partitions of @disk.
91 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
94 struct disk_part_tbl *ptbl;
97 ptbl = rcu_dereference(disk->part_tbl);
102 if (flags & DISK_PITER_REVERSE)
103 piter->idx = ptbl->len - 1;
104 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
109 piter->flags = flags;
113 EXPORT_SYMBOL_GPL(disk_part_iter_init);
116 * disk_part_iter_next - proceed iterator to the next partition and return it
117 * @piter: iterator of interest
119 * Proceed @piter to the next partition and return it.
124 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
126 struct disk_part_tbl *ptbl;
129 /* put the last partition */
130 disk_put_part(piter->part);
135 ptbl = rcu_dereference(piter->disk->part_tbl);
137 /* determine iteration parameters */
138 if (piter->flags & DISK_PITER_REVERSE) {
140 if (piter->flags & (DISK_PITER_INCL_PART0 |
141 DISK_PITER_INCL_EMPTY_PART0))
150 /* iterate to the next partition */
151 for (; piter->idx != end; piter->idx += inc) {
152 struct hd_struct *part;
154 part = rcu_dereference(ptbl->part[piter->idx]);
157 if (!part->nr_sects &&
158 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
159 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
163 get_device(part_to_dev(part));
173 EXPORT_SYMBOL_GPL(disk_part_iter_next);
176 * disk_part_iter_exit - finish up partition iteration
177 * @piter: iter of interest
179 * Called when iteration is over. Cleans up @piter.
184 void disk_part_iter_exit(struct disk_part_iter *piter)
186 disk_put_part(piter->part);
189 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
191 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
193 return part->start_sect <= sector &&
194 sector < part->start_sect + part->nr_sects;
198 * disk_map_sector_rcu - map sector to partition
199 * @disk: gendisk of interest
200 * @sector: sector to map
202 * Find out which partition @sector maps to on @disk. This is
203 * primarily used for stats accounting.
206 * RCU read locked. The returned partition pointer is valid only
207 * while preemption is disabled.
210 * Found partition on success, part0 is returned if no partition matches
212 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
214 struct disk_part_tbl *ptbl;
215 struct hd_struct *part;
218 ptbl = rcu_dereference(disk->part_tbl);
220 part = rcu_dereference(ptbl->last_lookup);
221 if (part && sector_in_part(part, sector))
224 for (i = 1; i < ptbl->len; i++) {
225 part = rcu_dereference(ptbl->part[i]);
227 if (part && sector_in_part(part, sector)) {
228 rcu_assign_pointer(ptbl->last_lookup, part);
234 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
237 * Can be deleted altogether. Later.
240 static struct blk_major_name {
241 struct blk_major_name *next;
244 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
246 /* index in the above - for now: assume no multimajor ranges */
247 static inline int major_to_index(unsigned major)
249 return major % BLKDEV_MAJOR_HASH_SIZE;
252 #ifdef CONFIG_PROC_FS
253 void blkdev_show(struct seq_file *seqf, off_t offset)
255 struct blk_major_name *dp;
257 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
258 mutex_lock(&block_class_lock);
259 for (dp = major_names[offset]; dp; dp = dp->next)
260 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
261 mutex_unlock(&block_class_lock);
264 #endif /* CONFIG_PROC_FS */
267 * register_blkdev - register a new block device
269 * @major: the requested major device number [1..255]. If @major=0, try to
270 * allocate any unused major number.
271 * @name: the name of the new block device as a zero terminated string
273 * The @name must be unique within the system.
275 * The return value depends on the @major input parameter.
276 * - if a major device number was requested in range [1..255] then the
277 * function returns zero on success, or a negative error code
278 * - if any unused major number was requested with @major=0 parameter
279 * then the return value is the allocated major number in range
280 * [1..255] or a negative error code otherwise
282 int register_blkdev(unsigned int major, const char *name)
284 struct blk_major_name **n, *p;
287 mutex_lock(&block_class_lock);
291 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
292 if (major_names[index] == NULL)
297 printk("register_blkdev: failed to get major for %s\n",
306 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
313 strlcpy(p->name, name, sizeof(p->name));
315 index = major_to_index(major);
317 for (n = &major_names[index]; *n; n = &(*n)->next) {
318 if ((*n)->major == major)
327 printk("register_blkdev: cannot get major %d for %s\n",
332 mutex_unlock(&block_class_lock);
336 EXPORT_SYMBOL(register_blkdev);
338 void unregister_blkdev(unsigned int major, const char *name)
340 struct blk_major_name **n;
341 struct blk_major_name *p = NULL;
342 int index = major_to_index(major);
344 mutex_lock(&block_class_lock);
345 for (n = &major_names[index]; *n; n = &(*n)->next)
346 if ((*n)->major == major)
348 if (!*n || strcmp((*n)->name, name)) {
354 mutex_unlock(&block_class_lock);
358 EXPORT_SYMBOL(unregister_blkdev);
360 static struct kobj_map *bdev_map;
363 * blk_mangle_minor - scatter minor numbers apart
364 * @minor: minor number to mangle
366 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
367 * is enabled. Mangling twice gives the original value.
375 static int blk_mangle_minor(int minor)
377 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
380 for (i = 0; i < MINORBITS / 2; i++) {
381 int low = minor & (1 << i);
382 int high = minor & (1 << (MINORBITS - 1 - i));
383 int distance = MINORBITS - 1 - 2 * i;
385 minor ^= low | high; /* clear both bits */
386 low <<= distance; /* swap the positions */
388 minor |= low | high; /* and set */
395 * blk_alloc_devt - allocate a dev_t for a partition
396 * @part: partition to allocate dev_t for
397 * @devt: out parameter for resulting dev_t
399 * Allocate a dev_t for block device.
402 * 0 on success, allocated dev_t is returned in *@devt. -errno on
408 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
410 struct gendisk *disk = part_to_disk(part);
413 /* in consecutive minor range? */
414 if (part->partno < disk->minors) {
415 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
419 /* allocate ext devt */
421 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
423 rc = idr_get_new(&ext_devt_idr, part, &idx);
424 } while (rc == -EAGAIN);
429 if (idx > MAX_EXT_DEVT) {
430 idr_remove(&ext_devt_idr, idx);
434 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
439 * blk_free_devt - free a dev_t
440 * @devt: dev_t to free
442 * Free @devt which was allocated using blk_alloc_devt().
447 void blk_free_devt(dev_t devt)
451 if (devt == MKDEV(0, 0))
454 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
455 mutex_lock(&ext_devt_mutex);
456 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
457 mutex_unlock(&ext_devt_mutex);
461 static char *bdevt_str(dev_t devt, char *buf)
463 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
464 char tbuf[BDEVT_SIZE];
465 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
466 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
468 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
474 * Register device numbers dev..(dev+range-1)
475 * range must be nonzero
476 * The hash chain is sorted on range, so that subranges can override.
478 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
479 struct kobject *(*probe)(dev_t, int *, void *),
480 int (*lock)(dev_t, void *), void *data)
482 kobj_map(bdev_map, devt, range, module, probe, lock, data);
485 EXPORT_SYMBOL(blk_register_region);
487 void blk_unregister_region(dev_t devt, unsigned long range)
489 kobj_unmap(bdev_map, devt, range);
492 EXPORT_SYMBOL(blk_unregister_region);
494 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
496 struct gendisk *p = data;
498 return &disk_to_dev(p)->kobj;
501 static int exact_lock(dev_t devt, void *data)
503 struct gendisk *p = data;
510 void register_disk(struct gendisk *disk)
512 struct device *ddev = disk_to_dev(disk);
513 struct block_device *bdev;
514 struct disk_part_iter piter;
515 struct hd_struct *part;
518 ddev->parent = disk->driverfs_dev;
520 dev_set_name(ddev, disk->disk_name);
522 /* delay uevents, until we scanned partition table */
523 dev_set_uevent_suppress(ddev, 1);
525 if (device_add(ddev))
527 if (!sysfs_deprecated) {
528 err = sysfs_create_link(block_depr, &ddev->kobj,
529 kobject_name(&ddev->kobj));
535 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
536 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
538 /* No minors to use for partitions */
539 if (!disk_partitionable(disk))
542 /* No such device (e.g., media were just removed) */
543 if (!get_capacity(disk))
546 bdev = bdget_disk(disk, 0);
550 bdev->bd_invalidated = 1;
551 err = blkdev_get(bdev, FMODE_READ, NULL);
554 blkdev_put(bdev, FMODE_READ);
557 /* announce disk after possible partitions are created */
558 dev_set_uevent_suppress(ddev, 0);
559 kobject_uevent(&ddev->kobj, KOBJ_ADD);
561 /* announce possible partitions */
562 disk_part_iter_init(&piter, disk, 0);
563 while ((part = disk_part_iter_next(&piter)))
564 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
565 disk_part_iter_exit(&piter);
569 * add_disk - add partitioning information to kernel list
570 * @disk: per-device partitioning information
572 * This function registers the partitioning information in @disk
575 * FIXME: error handling
577 void add_disk(struct gendisk *disk)
579 struct backing_dev_info *bdi;
583 /* minors == 0 indicates to use ext devt from part0 and should
584 * be accompanied with EXT_DEVT flag. Make sure all
585 * parameters make sense.
587 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
588 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
590 disk->flags |= GENHD_FL_UP;
592 retval = blk_alloc_devt(&disk->part0, &devt);
597 disk_to_dev(disk)->devt = devt;
599 /* ->major and ->first_minor aren't supposed to be
600 * dereferenced from here on, but set them just in case.
602 disk->major = MAJOR(devt);
603 disk->first_minor = MINOR(devt);
605 /* Register BDI before referencing it from bdev */
606 bdi = &disk->queue->backing_dev_info;
607 bdi_register_dev(bdi, disk_devt(disk));
609 blk_register_region(disk_devt(disk), disk->minors, NULL,
610 exact_match, exact_lock, disk);
612 blk_register_queue(disk);
614 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
618 disk_add_events(disk);
620 EXPORT_SYMBOL(add_disk);
622 void del_gendisk(struct gendisk *disk)
624 struct disk_part_iter piter;
625 struct hd_struct *part;
627 disk_del_events(disk);
629 /* invalidate stuff */
630 disk_part_iter_init(&piter, disk,
631 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
632 while ((part = disk_part_iter_next(&piter))) {
633 invalidate_partition(disk, part->partno);
634 delete_partition(disk, part->partno);
636 disk_part_iter_exit(&piter);
638 invalidate_partition(disk, 0);
639 blk_free_devt(disk_to_dev(disk)->devt);
640 set_capacity(disk, 0);
641 disk->flags &= ~GENHD_FL_UP;
643 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
644 bdi_unregister(&disk->queue->backing_dev_info);
645 blk_unregister_queue(disk);
646 blk_unregister_region(disk_devt(disk), disk->minors);
648 part_stat_set_all(&disk->part0, 0);
649 disk->part0.stamp = 0;
651 kobject_put(disk->part0.holder_dir);
652 kobject_put(disk->slave_dir);
653 disk->driverfs_dev = NULL;
654 if (!sysfs_deprecated)
655 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
656 device_del(disk_to_dev(disk));
658 EXPORT_SYMBOL(del_gendisk);
661 * get_gendisk - get partitioning information for a given device
662 * @devt: device to get partitioning information for
663 * @partno: returned partition index
665 * This function gets the structure containing partitioning
666 * information for the given device @devt.
668 struct gendisk *get_gendisk(dev_t devt, int *partno)
670 struct gendisk *disk = NULL;
672 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
673 struct kobject *kobj;
675 kobj = kobj_lookup(bdev_map, devt, partno);
677 disk = dev_to_disk(kobj_to_dev(kobj));
679 struct hd_struct *part;
681 mutex_lock(&ext_devt_mutex);
682 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
683 if (part && get_disk(part_to_disk(part))) {
684 *partno = part->partno;
685 disk = part_to_disk(part);
687 mutex_unlock(&ext_devt_mutex);
692 EXPORT_SYMBOL(get_gendisk);
695 * bdget_disk - do bdget() by gendisk and partition number
696 * @disk: gendisk of interest
697 * @partno: partition number
699 * Find partition @partno from @disk, do bdget() on it.
705 * Resulting block_device on success, NULL on failure.
707 struct block_device *bdget_disk(struct gendisk *disk, int partno)
709 struct hd_struct *part;
710 struct block_device *bdev = NULL;
712 part = disk_get_part(disk, partno);
714 bdev = bdget(part_devt(part));
719 EXPORT_SYMBOL(bdget_disk);
722 * print a full list of all partitions - intended for places where the root
723 * filesystem can't be mounted and thus to give the victim some idea of what
726 void __init printk_all_partitions(void)
728 struct class_dev_iter iter;
731 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
732 while ((dev = class_dev_iter_next(&iter))) {
733 struct gendisk *disk = dev_to_disk(dev);
734 struct disk_part_iter piter;
735 struct hd_struct *part;
736 char name_buf[BDEVNAME_SIZE];
737 char devt_buf[BDEVT_SIZE];
738 u8 uuid[PARTITION_META_INFO_UUIDLTH * 2 + 1];
741 * Don't show empty devices or things that have been
744 if (get_capacity(disk) == 0 ||
745 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
749 * Note, unlike /proc/partitions, I am showing the
750 * numbers in hex - the same format as the root=
753 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
754 while ((part = disk_part_iter_next(&piter))) {
755 bool is_part0 = part == &disk->part0;
759 part_unpack_uuid(part->info->uuid, uuid);
761 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
762 bdevt_str(part_devt(part), devt_buf),
763 (unsigned long long)part->nr_sects >> 1,
764 disk_name(disk, part->partno, name_buf), uuid);
766 if (disk->driverfs_dev != NULL &&
767 disk->driverfs_dev->driver != NULL)
768 printk(" driver: %s\n",
769 disk->driverfs_dev->driver->name);
771 printk(" (driver?)\n");
775 disk_part_iter_exit(&piter);
777 class_dev_iter_exit(&iter);
780 #ifdef CONFIG_PROC_FS
782 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
785 struct class_dev_iter *iter;
788 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
790 return ERR_PTR(-ENOMEM);
792 seqf->private = iter;
793 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
795 dev = class_dev_iter_next(iter);
800 return dev_to_disk(dev);
803 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
808 dev = class_dev_iter_next(seqf->private);
810 return dev_to_disk(dev);
815 static void disk_seqf_stop(struct seq_file *seqf, void *v)
817 struct class_dev_iter *iter = seqf->private;
819 /* stop is called even after start failed :-( */
821 class_dev_iter_exit(iter);
826 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
830 p = disk_seqf_start(seqf, pos);
831 if (!IS_ERR_OR_NULL(p) && !*pos)
832 seq_puts(seqf, "major minor #blocks name\n\n");
836 static int show_partition(struct seq_file *seqf, void *v)
838 struct gendisk *sgp = v;
839 struct disk_part_iter piter;
840 struct hd_struct *part;
841 char buf[BDEVNAME_SIZE];
843 /* Don't show non-partitionable removeable devices or empty devices */
844 if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
845 (sgp->flags & GENHD_FL_REMOVABLE)))
847 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
850 /* show the full disk and all non-0 size partitions of it */
851 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
852 while ((part = disk_part_iter_next(&piter)))
853 seq_printf(seqf, "%4d %7d %10llu %s\n",
854 MAJOR(part_devt(part)), MINOR(part_devt(part)),
855 (unsigned long long)part->nr_sects >> 1,
856 disk_name(sgp, part->partno, buf));
857 disk_part_iter_exit(&piter);
862 static const struct seq_operations partitions_op = {
863 .start = show_partition_start,
864 .next = disk_seqf_next,
865 .stop = disk_seqf_stop,
866 .show = show_partition
869 static int partitions_open(struct inode *inode, struct file *file)
871 return seq_open(file, &partitions_op);
874 static const struct file_operations proc_partitions_operations = {
875 .open = partitions_open,
878 .release = seq_release,
883 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
885 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
886 /* Make old-style 2.4 aliases work */
887 request_module("block-major-%d", MAJOR(devt));
891 static int __init genhd_device_init(void)
895 block_class.dev_kobj = sysfs_dev_block_kobj;
896 error = class_register(&block_class);
899 bdev_map = kobj_map_init(base_probe, &block_class_lock);
902 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
904 /* create top-level block dir */
905 if (!sysfs_deprecated)
906 block_depr = kobject_create_and_add("block", NULL);
910 subsys_initcall(genhd_device_init);
912 static ssize_t disk_range_show(struct device *dev,
913 struct device_attribute *attr, char *buf)
915 struct gendisk *disk = dev_to_disk(dev);
917 return sprintf(buf, "%d\n", disk->minors);
920 static ssize_t disk_ext_range_show(struct device *dev,
921 struct device_attribute *attr, char *buf)
923 struct gendisk *disk = dev_to_disk(dev);
925 return sprintf(buf, "%d\n", disk_max_parts(disk));
928 static ssize_t disk_removable_show(struct device *dev,
929 struct device_attribute *attr, char *buf)
931 struct gendisk *disk = dev_to_disk(dev);
933 return sprintf(buf, "%d\n",
934 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
937 static ssize_t disk_ro_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", get_disk_ro(disk) ? 1 : 0);
945 static ssize_t disk_capability_show(struct device *dev,
946 struct device_attribute *attr, char *buf)
948 struct gendisk *disk = dev_to_disk(dev);
950 return sprintf(buf, "%x\n", disk->flags);
953 static ssize_t disk_alignment_offset_show(struct device *dev,
954 struct device_attribute *attr,
957 struct gendisk *disk = dev_to_disk(dev);
959 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
962 static ssize_t disk_discard_alignment_show(struct device *dev,
963 struct device_attribute *attr,
966 struct gendisk *disk = dev_to_disk(dev);
968 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
971 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
972 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
973 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
974 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
975 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
976 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
977 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
979 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
980 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
981 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
982 #ifdef CONFIG_FAIL_MAKE_REQUEST
983 static struct device_attribute dev_attr_fail =
984 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
986 #ifdef CONFIG_FAIL_IO_TIMEOUT
987 static struct device_attribute dev_attr_fail_timeout =
988 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
992 static struct attribute *disk_attrs[] = {
993 &dev_attr_range.attr,
994 &dev_attr_ext_range.attr,
995 &dev_attr_removable.attr,
998 &dev_attr_alignment_offset.attr,
999 &dev_attr_discard_alignment.attr,
1000 &dev_attr_capability.attr,
1001 &dev_attr_stat.attr,
1002 &dev_attr_inflight.attr,
1003 #ifdef CONFIG_FAIL_MAKE_REQUEST
1004 &dev_attr_fail.attr,
1006 #ifdef CONFIG_FAIL_IO_TIMEOUT
1007 &dev_attr_fail_timeout.attr,
1012 static struct attribute_group disk_attr_group = {
1013 .attrs = disk_attrs,
1016 static const struct attribute_group *disk_attr_groups[] = {
1021 static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
1023 struct disk_part_tbl *ptbl =
1024 container_of(head, struct disk_part_tbl, rcu_head);
1030 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1031 * @disk: disk to replace part_tbl for
1032 * @new_ptbl: new part_tbl to install
1034 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1035 * original ptbl is freed using RCU callback.
1038 * Matching bd_mutx locked.
1040 static void disk_replace_part_tbl(struct gendisk *disk,
1041 struct disk_part_tbl *new_ptbl)
1043 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1045 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1048 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1049 call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
1054 * disk_expand_part_tbl - expand disk->part_tbl
1055 * @disk: disk to expand part_tbl for
1056 * @partno: expand such that this partno can fit in
1058 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1059 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1062 * Matching bd_mutex locked, might sleep.
1065 * 0 on success, -errno on failure.
1067 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1069 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1070 struct disk_part_tbl *new_ptbl;
1071 int len = old_ptbl ? old_ptbl->len : 0;
1072 int target = partno + 1;
1076 /* disk_max_parts() is zero during initialization, ignore if so */
1077 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1083 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1084 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1088 new_ptbl->len = target;
1090 for (i = 0; i < len; i++)
1091 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1093 disk_replace_part_tbl(disk, new_ptbl);
1097 static void disk_release(struct device *dev)
1099 struct gendisk *disk = dev_to_disk(dev);
1101 disk_release_events(disk);
1102 kfree(disk->random);
1103 disk_replace_part_tbl(disk, NULL);
1104 free_part_stats(&disk->part0);
1105 free_part_info(&disk->part0);
1108 struct class block_class = {
1112 static char *block_devnode(struct device *dev, mode_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 %llu "
1158 "%u %lu %lu %llu %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[0]),
1162 part_stat_read(hd, merges[0]),
1163 (unsigned long long)part_stat_read(hd, sectors[0]),
1164 jiffies_to_msecs(part_stat_read(hd, ticks[0])),
1165 part_stat_read(hd, ios[1]),
1166 part_stat_read(hd, merges[1]),
1167 (unsigned long long)part_stat_read(hd, sectors[1]),
1168 jiffies_to_msecs(part_stat_read(hd, ticks[1])),
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, -1);
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;
1267 hd_ref_init(&disk->part0);
1269 disk->minors = minors;
1270 rand_initialize_disk(disk);
1271 disk_to_dev(disk)->class = &block_class;
1272 disk_to_dev(disk)->type = &disk_type;
1273 device_initialize(disk_to_dev(disk));
1277 EXPORT_SYMBOL(alloc_disk_node);
1279 struct kobject *get_disk(struct gendisk *disk)
1281 struct module *owner;
1282 struct kobject *kobj;
1286 owner = disk->fops->owner;
1287 if (owner && !try_module_get(owner))
1289 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1298 EXPORT_SYMBOL(get_disk);
1300 void put_disk(struct gendisk *disk)
1303 kobject_put(&disk_to_dev(disk)->kobj);
1306 EXPORT_SYMBOL(put_disk);
1308 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1310 char event[] = "DISK_RO=1";
1311 char *envp[] = { event, NULL };
1315 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1318 void set_device_ro(struct block_device *bdev, int flag)
1320 bdev->bd_part->policy = flag;
1323 EXPORT_SYMBOL(set_device_ro);
1325 void set_disk_ro(struct gendisk *disk, int flag)
1327 struct disk_part_iter piter;
1328 struct hd_struct *part;
1330 if (disk->part0.policy != flag) {
1331 set_disk_ro_uevent(disk, flag);
1332 disk->part0.policy = flag;
1335 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1336 while ((part = disk_part_iter_next(&piter)))
1337 part->policy = flag;
1338 disk_part_iter_exit(&piter);
1341 EXPORT_SYMBOL(set_disk_ro);
1343 int bdev_read_only(struct block_device *bdev)
1347 return bdev->bd_part->policy;
1350 EXPORT_SYMBOL(bdev_read_only);
1352 int invalidate_partition(struct gendisk *disk, int partno)
1355 struct block_device *bdev = bdget_disk(disk, partno);
1358 res = __invalidate_device(bdev);
1364 EXPORT_SYMBOL(invalidate_partition);
1367 * Disk events - monitor disk events like media change and eject request.
1369 struct disk_events {
1370 struct list_head node; /* all disk_event's */
1371 struct gendisk *disk; /* the associated disk */
1374 int block; /* event blocking depth */
1375 unsigned int pending; /* events already sent out */
1376 unsigned int clearing; /* events being cleared */
1378 long poll_msecs; /* interval, -1 for default */
1379 struct delayed_work dwork;
1382 static const char *disk_events_strs[] = {
1383 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1384 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1387 static char *disk_uevents[] = {
1388 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1389 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1392 /* list of all disk_events */
1393 static DEFINE_MUTEX(disk_events_mutex);
1394 static LIST_HEAD(disk_events);
1396 /* disable in-kernel polling by default */
1397 static unsigned long disk_events_dfl_poll_msecs = 0;
1399 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1401 struct disk_events *ev = disk->ev;
1402 long intv_msecs = 0;
1405 * If device-specific poll interval is set, always use it. If
1406 * the default is being used, poll iff there are events which
1407 * can't be monitored asynchronously.
1409 if (ev->poll_msecs >= 0)
1410 intv_msecs = ev->poll_msecs;
1411 else if (disk->events & ~disk->async_events)
1412 intv_msecs = disk_events_dfl_poll_msecs;
1414 return msecs_to_jiffies(intv_msecs);
1417 static void __disk_block_events(struct gendisk *disk, bool sync)
1419 struct disk_events *ev = disk->ev;
1420 unsigned long flags;
1423 spin_lock_irqsave(&ev->lock, flags);
1424 cancel = !ev->block++;
1425 spin_unlock_irqrestore(&ev->lock, flags);
1429 cancel_delayed_work_sync(&disk->ev->dwork);
1431 cancel_delayed_work(&disk->ev->dwork);
1435 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1437 struct disk_events *ev = disk->ev;
1439 unsigned long flags;
1441 spin_lock_irqsave(&ev->lock, flags);
1443 if (WARN_ON_ONCE(ev->block <= 0))
1450 * Not exactly a latency critical operation, set poll timer
1451 * slack to 25% and kick event check.
1453 intv = disk_events_poll_jiffies(disk);
1454 set_timer_slack(&ev->dwork.timer, intv / 4);
1456 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1458 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1460 spin_unlock_irqrestore(&ev->lock, flags);
1464 * disk_block_events - block and flush disk event checking
1465 * @disk: disk to block events for
1467 * On return from this function, it is guaranteed that event checking
1468 * isn't in progress and won't happen until unblocked by
1469 * disk_unblock_events(). Events blocking is counted and the actual
1470 * unblocking happens after the matching number of unblocks are done.
1472 * Note that this intentionally does not block event checking from
1473 * disk_clear_events().
1478 void disk_block_events(struct gendisk *disk)
1481 __disk_block_events(disk, true);
1485 * disk_unblock_events - unblock disk event checking
1486 * @disk: disk to unblock events for
1488 * Undo disk_block_events(). When the block count reaches zero, it
1489 * starts events polling if configured.
1492 * Don't care. Safe to call from irq context.
1494 void disk_unblock_events(struct gendisk *disk)
1497 __disk_unblock_events(disk, true);
1501 * disk_check_events - schedule immediate event checking
1502 * @disk: disk to check events for
1504 * Schedule immediate event checking on @disk if not blocked.
1507 * Don't care. Safe to call from irq context.
1509 void disk_check_events(struct gendisk *disk)
1512 __disk_block_events(disk, false);
1513 __disk_unblock_events(disk, true);
1516 EXPORT_SYMBOL_GPL(disk_check_events);
1519 * disk_clear_events - synchronously check, clear and return pending events
1520 * @disk: disk to fetch and clear events from
1521 * @mask: mask of events to be fetched and clearted
1523 * Disk events are synchronously checked and pending events in @mask
1524 * are cleared and returned. This ignores the block count.
1529 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1531 const struct block_device_operations *bdops = disk->fops;
1532 struct disk_events *ev = disk->ev;
1533 unsigned int pending;
1536 /* for drivers still using the old ->media_changed method */
1537 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1538 bdops->media_changed && bdops->media_changed(disk))
1539 return DISK_EVENT_MEDIA_CHANGE;
1543 /* tell the workfn about the events being cleared */
1544 spin_lock_irq(&ev->lock);
1545 ev->clearing |= mask;
1546 spin_unlock_irq(&ev->lock);
1548 /* uncondtionally schedule event check and wait for it to finish */
1549 __disk_block_events(disk, true);
1550 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1551 flush_delayed_work(&ev->dwork);
1552 __disk_unblock_events(disk, false);
1554 /* then, fetch and clear pending events */
1555 spin_lock_irq(&ev->lock);
1556 WARN_ON_ONCE(ev->clearing & mask); /* cleared by workfn */
1557 pending = ev->pending & mask;
1558 ev->pending &= ~mask;
1559 spin_unlock_irq(&ev->lock);
1564 static void disk_events_workfn(struct work_struct *work)
1566 struct delayed_work *dwork = to_delayed_work(work);
1567 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1568 struct gendisk *disk = ev->disk;
1569 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1570 unsigned int clearing = ev->clearing;
1571 unsigned int events;
1573 int nr_events = 0, i;
1576 events = disk->fops->check_events(disk, clearing);
1578 /* accumulate pending events and schedule next poll if necessary */
1579 spin_lock_irq(&ev->lock);
1581 events &= ~ev->pending;
1582 ev->pending |= events;
1583 ev->clearing &= ~clearing;
1585 intv = disk_events_poll_jiffies(disk);
1586 if (!ev->block && intv)
1587 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1589 spin_unlock_irq(&ev->lock);
1591 /* tell userland about new events */
1592 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1593 if (events & (1 << i))
1594 envp[nr_events++] = disk_uevents[i];
1597 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1601 * A disk events enabled device has the following sysfs nodes under
1602 * its /sys/block/X/ directory.
1604 * events : list of all supported events
1605 * events_async : list of events which can be detected w/o polling
1606 * events_poll_msecs : polling interval, 0: disable, -1: system default
1608 static ssize_t __disk_events_show(unsigned int events, char *buf)
1610 const char *delim = "";
1614 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1615 if (events & (1 << i)) {
1616 pos += sprintf(buf + pos, "%s%s",
1617 delim, disk_events_strs[i]);
1621 pos += sprintf(buf + pos, "\n");
1625 static ssize_t disk_events_show(struct device *dev,
1626 struct device_attribute *attr, char *buf)
1628 struct gendisk *disk = dev_to_disk(dev);
1630 return __disk_events_show(disk->events, buf);
1633 static ssize_t disk_events_async_show(struct device *dev,
1634 struct device_attribute *attr, char *buf)
1636 struct gendisk *disk = dev_to_disk(dev);
1638 return __disk_events_show(disk->async_events, buf);
1641 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1642 struct device_attribute *attr,
1645 struct gendisk *disk = dev_to_disk(dev);
1647 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1650 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1651 struct device_attribute *attr,
1652 const char *buf, size_t count)
1654 struct gendisk *disk = dev_to_disk(dev);
1657 if (!count || !sscanf(buf, "%ld", &intv))
1660 if (intv < 0 && intv != -1)
1663 __disk_block_events(disk, true);
1664 disk->ev->poll_msecs = intv;
1665 __disk_unblock_events(disk, true);
1670 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1671 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1672 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1673 disk_events_poll_msecs_show,
1674 disk_events_poll_msecs_store);
1676 static const struct attribute *disk_events_attrs[] = {
1677 &dev_attr_events.attr,
1678 &dev_attr_events_async.attr,
1679 &dev_attr_events_poll_msecs.attr,
1684 * The default polling interval can be specified by the kernel
1685 * parameter block.events_dfl_poll_msecs which defaults to 0
1686 * (disable). This can also be modified runtime by writing to
1687 * /sys/module/block/events_dfl_poll_msecs.
1689 static int disk_events_set_dfl_poll_msecs(const char *val,
1690 const struct kernel_param *kp)
1692 struct disk_events *ev;
1695 ret = param_set_ulong(val, kp);
1699 mutex_lock(&disk_events_mutex);
1701 list_for_each_entry(ev, &disk_events, node)
1702 disk_check_events(ev->disk);
1704 mutex_unlock(&disk_events_mutex);
1709 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1710 .set = disk_events_set_dfl_poll_msecs,
1711 .get = param_get_ulong,
1714 #undef MODULE_PARAM_PREFIX
1715 #define MODULE_PARAM_PREFIX "block."
1717 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1718 &disk_events_dfl_poll_msecs, 0644);
1721 * disk_{add|del|release}_events - initialize and destroy disk_events.
1723 static void disk_add_events(struct gendisk *disk)
1725 struct disk_events *ev;
1727 if (!disk->fops->check_events || !(disk->events | disk->async_events))
1730 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1732 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1736 if (sysfs_create_files(&disk_to_dev(disk)->kobj,
1737 disk_events_attrs) < 0) {
1738 pr_warn("%s: failed to create sysfs files for events\n",
1746 INIT_LIST_HEAD(&ev->node);
1748 spin_lock_init(&ev->lock);
1750 ev->poll_msecs = -1;
1751 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1753 mutex_lock(&disk_events_mutex);
1754 list_add_tail(&ev->node, &disk_events);
1755 mutex_unlock(&disk_events_mutex);
1758 * Block count is initialized to 1 and the following initial
1759 * unblock kicks it into action.
1761 __disk_unblock_events(disk, true);
1764 static void disk_del_events(struct gendisk *disk)
1769 __disk_block_events(disk, true);
1771 mutex_lock(&disk_events_mutex);
1772 list_del_init(&disk->ev->node);
1773 mutex_unlock(&disk_events_mutex);
1775 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1778 static void disk_release_events(struct gendisk *disk)
1780 /* the block count should be 1 from disk_del_events() */
1781 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);