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);
568 static int __read_mostly no_partition_scan;
570 static int __init no_partition_scan_setup(char *str)
572 no_partition_scan = 1;
573 printk(KERN_INFO "genhd: omit partition scan.\n");
578 __setup("no_partition_scan", no_partition_scan_setup);
581 * add_disk - add partitioning information to kernel list
582 * @disk: per-device partitioning information
584 * This function registers the partitioning information in @disk
587 * FIXME: error handling
589 void add_disk(struct gendisk *disk)
591 struct backing_dev_info *bdi;
595 /* minors == 0 indicates to use ext devt from part0 and should
596 * be accompanied with EXT_DEVT flag. Make sure all
597 * parameters make sense.
599 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
600 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
602 disk->flags |= GENHD_FL_UP;
604 if (no_partition_scan)
605 disk->flags |= GENHD_FL_NO_PARTITION_SCAN;
607 retval = blk_alloc_devt(&disk->part0, &devt);
612 disk_to_dev(disk)->devt = devt;
614 /* ->major and ->first_minor aren't supposed to be
615 * dereferenced from here on, but set them just in case.
617 disk->major = MAJOR(devt);
618 disk->first_minor = MINOR(devt);
620 /* Register BDI before referencing it from bdev */
621 bdi = &disk->queue->backing_dev_info;
622 bdi_register_dev(bdi, disk_devt(disk));
624 blk_register_region(disk_devt(disk), disk->minors, NULL,
625 exact_match, exact_lock, disk);
627 blk_register_queue(disk);
629 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
633 disk_add_events(disk);
635 EXPORT_SYMBOL(add_disk);
637 void del_gendisk(struct gendisk *disk)
639 struct disk_part_iter piter;
640 struct hd_struct *part;
642 disk_del_events(disk);
644 /* invalidate stuff */
645 disk_part_iter_init(&piter, disk,
646 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
647 while ((part = disk_part_iter_next(&piter))) {
648 invalidate_partition(disk, part->partno);
649 delete_partition(disk, part->partno);
651 disk_part_iter_exit(&piter);
653 invalidate_partition(disk, 0);
654 blk_free_devt(disk_to_dev(disk)->devt);
655 set_capacity(disk, 0);
656 disk->flags &= ~GENHD_FL_UP;
658 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
659 bdi_unregister(&disk->queue->backing_dev_info);
660 blk_unregister_queue(disk);
661 blk_unregister_region(disk_devt(disk), disk->minors);
663 part_stat_set_all(&disk->part0, 0);
664 disk->part0.stamp = 0;
666 kobject_put(disk->part0.holder_dir);
667 kobject_put(disk->slave_dir);
668 disk->driverfs_dev = NULL;
669 if (!sysfs_deprecated)
670 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
671 device_del(disk_to_dev(disk));
673 EXPORT_SYMBOL(del_gendisk);
676 * get_gendisk - get partitioning information for a given device
677 * @devt: device to get partitioning information for
678 * @partno: returned partition index
680 * This function gets the structure containing partitioning
681 * information for the given device @devt.
683 struct gendisk *get_gendisk(dev_t devt, int *partno)
685 struct gendisk *disk = NULL;
687 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
688 struct kobject *kobj;
690 kobj = kobj_lookup(bdev_map, devt, partno);
692 disk = dev_to_disk(kobj_to_dev(kobj));
694 struct hd_struct *part;
696 mutex_lock(&ext_devt_mutex);
697 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
698 if (part && get_disk(part_to_disk(part))) {
699 *partno = part->partno;
700 disk = part_to_disk(part);
702 mutex_unlock(&ext_devt_mutex);
707 EXPORT_SYMBOL(get_gendisk);
710 * bdget_disk - do bdget() by gendisk and partition number
711 * @disk: gendisk of interest
712 * @partno: partition number
714 * Find partition @partno from @disk, do bdget() on it.
720 * Resulting block_device on success, NULL on failure.
722 struct block_device *bdget_disk(struct gendisk *disk, int partno)
724 struct hd_struct *part;
725 struct block_device *bdev = NULL;
727 part = disk_get_part(disk, partno);
729 bdev = bdget(part_devt(part));
734 EXPORT_SYMBOL(bdget_disk);
737 * print a full list of all partitions - intended for places where the root
738 * filesystem can't be mounted and thus to give the victim some idea of what
741 void __init printk_all_partitions(void)
743 struct class_dev_iter iter;
746 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
747 while ((dev = class_dev_iter_next(&iter))) {
748 struct gendisk *disk = dev_to_disk(dev);
749 struct disk_part_iter piter;
750 struct hd_struct *part;
751 char name_buf[BDEVNAME_SIZE];
752 char devt_buf[BDEVT_SIZE];
753 u8 uuid[PARTITION_META_INFO_UUIDLTH * 2 + 1];
756 * Don't show empty devices or things that have been
759 if (get_capacity(disk) == 0 ||
760 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
764 * Note, unlike /proc/partitions, I am showing the
765 * numbers in hex - the same format as the root=
768 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
769 while ((part = disk_part_iter_next(&piter))) {
770 bool is_part0 = part == &disk->part0;
774 part_unpack_uuid(part->info->uuid, uuid);
776 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
777 bdevt_str(part_devt(part), devt_buf),
778 (unsigned long long)part->nr_sects >> 1,
779 disk_name(disk, part->partno, name_buf), uuid);
781 if (disk->driverfs_dev != NULL &&
782 disk->driverfs_dev->driver != NULL)
783 printk(" driver: %s\n",
784 disk->driverfs_dev->driver->name);
786 printk(" (driver?)\n");
790 disk_part_iter_exit(&piter);
792 class_dev_iter_exit(&iter);
795 #ifdef CONFIG_PROC_FS
797 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
800 struct class_dev_iter *iter;
803 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
805 return ERR_PTR(-ENOMEM);
807 seqf->private = iter;
808 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
810 dev = class_dev_iter_next(iter);
815 return dev_to_disk(dev);
818 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
823 dev = class_dev_iter_next(seqf->private);
825 return dev_to_disk(dev);
830 static void disk_seqf_stop(struct seq_file *seqf, void *v)
832 struct class_dev_iter *iter = seqf->private;
834 /* stop is called even after start failed :-( */
836 class_dev_iter_exit(iter);
841 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
845 p = disk_seqf_start(seqf, pos);
846 if (!IS_ERR_OR_NULL(p) && !*pos)
847 seq_puts(seqf, "major minor #blocks name\n\n");
851 static int show_partition(struct seq_file *seqf, void *v)
853 struct gendisk *sgp = v;
854 struct disk_part_iter piter;
855 struct hd_struct *part;
856 char buf[BDEVNAME_SIZE];
858 /* Don't show non-partitionable removeable devices or empty devices */
859 if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
860 (sgp->flags & GENHD_FL_REMOVABLE)))
862 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
865 /* show the full disk and all non-0 size partitions of it */
866 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
867 while ((part = disk_part_iter_next(&piter)))
868 seq_printf(seqf, "%4d %7d %10llu %s\n",
869 MAJOR(part_devt(part)), MINOR(part_devt(part)),
870 (unsigned long long)part->nr_sects >> 1,
871 disk_name(sgp, part->partno, buf));
872 disk_part_iter_exit(&piter);
877 static const struct seq_operations partitions_op = {
878 .start = show_partition_start,
879 .next = disk_seqf_next,
880 .stop = disk_seqf_stop,
881 .show = show_partition
884 static int partitions_open(struct inode *inode, struct file *file)
886 return seq_open(file, &partitions_op);
889 static const struct file_operations proc_partitions_operations = {
890 .open = partitions_open,
893 .release = seq_release,
898 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
900 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
901 /* Make old-style 2.4 aliases work */
902 request_module("block-major-%d", MAJOR(devt));
906 static int __init genhd_device_init(void)
910 block_class.dev_kobj = sysfs_dev_block_kobj;
911 error = class_register(&block_class);
914 bdev_map = kobj_map_init(base_probe, &block_class_lock);
917 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
919 /* create top-level block dir */
920 if (!sysfs_deprecated)
921 block_depr = kobject_create_and_add("block", NULL);
925 subsys_initcall(genhd_device_init);
927 static ssize_t disk_range_show(struct device *dev,
928 struct device_attribute *attr, char *buf)
930 struct gendisk *disk = dev_to_disk(dev);
932 return sprintf(buf, "%d\n",
933 (disk->flags & GENHD_FL_NO_PARTITION_SCAN ? 0 : disk->minors));
936 static ssize_t disk_range_store(struct device *dev,
937 struct device_attribute *attr,
938 const char *buf, size_t count)
940 struct gendisk *disk = dev_to_disk(dev);
943 if (count > 0 && sscanf(buf, "%d", &i) > 0) {
945 disk->flags |= GENHD_FL_NO_PARTITION_SCAN;
946 else if (i <= disk->minors)
947 disk->flags &= ~GENHD_FL_NO_PARTITION_SCAN;
955 static ssize_t disk_ext_range_show(struct device *dev,
956 struct device_attribute *attr, char *buf)
958 struct gendisk *disk = dev_to_disk(dev);
960 return sprintf(buf, "%d\n", disk_max_parts(disk));
963 static ssize_t disk_removable_show(struct device *dev,
964 struct device_attribute *attr, char *buf)
966 struct gendisk *disk = dev_to_disk(dev);
968 return sprintf(buf, "%d\n",
969 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
972 static ssize_t disk_ro_show(struct device *dev,
973 struct device_attribute *attr, char *buf)
975 struct gendisk *disk = dev_to_disk(dev);
977 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
980 static ssize_t disk_capability_show(struct device *dev,
981 struct device_attribute *attr, char *buf)
983 struct gendisk *disk = dev_to_disk(dev);
985 return sprintf(buf, "%x\n", disk->flags);
988 static ssize_t disk_alignment_offset_show(struct device *dev,
989 struct device_attribute *attr,
992 struct gendisk *disk = dev_to_disk(dev);
994 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
997 static ssize_t disk_discard_alignment_show(struct device *dev,
998 struct device_attribute *attr,
1001 struct gendisk *disk = dev_to_disk(dev);
1003 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1006 static DEVICE_ATTR(range, S_IRUGO|S_IWUSR, disk_range_show, disk_range_store);
1007 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
1008 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
1009 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
1010 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
1011 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
1012 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
1014 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
1015 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
1016 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
1017 #ifdef CONFIG_FAIL_MAKE_REQUEST
1018 static struct device_attribute dev_attr_fail =
1019 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
1021 #ifdef CONFIG_FAIL_IO_TIMEOUT
1022 static struct device_attribute dev_attr_fail_timeout =
1023 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
1024 part_timeout_store);
1027 static struct attribute *disk_attrs[] = {
1028 &dev_attr_range.attr,
1029 &dev_attr_ext_range.attr,
1030 &dev_attr_removable.attr,
1032 &dev_attr_size.attr,
1033 &dev_attr_alignment_offset.attr,
1034 &dev_attr_discard_alignment.attr,
1035 &dev_attr_capability.attr,
1036 &dev_attr_stat.attr,
1037 &dev_attr_inflight.attr,
1038 #ifdef CONFIG_FAIL_MAKE_REQUEST
1039 &dev_attr_fail.attr,
1041 #ifdef CONFIG_FAIL_IO_TIMEOUT
1042 &dev_attr_fail_timeout.attr,
1047 static struct attribute_group disk_attr_group = {
1048 .attrs = disk_attrs,
1051 static const struct attribute_group *disk_attr_groups[] = {
1057 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1058 * @disk: disk to replace part_tbl for
1059 * @new_ptbl: new part_tbl to install
1061 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1062 * original ptbl is freed using RCU callback.
1065 * Matching bd_mutx locked.
1067 static void disk_replace_part_tbl(struct gendisk *disk,
1068 struct disk_part_tbl *new_ptbl)
1070 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1072 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1075 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1076 kfree_rcu(old_ptbl, rcu_head);
1081 * disk_expand_part_tbl - expand disk->part_tbl
1082 * @disk: disk to expand part_tbl for
1083 * @partno: expand such that this partno can fit in
1085 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1086 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1089 * Matching bd_mutex locked, might sleep.
1092 * 0 on success, -errno on failure.
1094 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1096 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1097 struct disk_part_tbl *new_ptbl;
1098 int len = old_ptbl ? old_ptbl->len : 0;
1099 int target = partno + 1;
1103 /* disk_max_parts() is zero during initialization, ignore if so */
1104 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1110 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1111 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1115 new_ptbl->len = target;
1117 for (i = 0; i < len; i++)
1118 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1120 disk_replace_part_tbl(disk, new_ptbl);
1124 static void disk_release(struct device *dev)
1126 struct gendisk *disk = dev_to_disk(dev);
1128 disk_release_events(disk);
1129 kfree(disk->random);
1130 disk_replace_part_tbl(disk, NULL);
1131 free_part_stats(&disk->part0);
1132 free_part_info(&disk->part0);
1135 struct class block_class = {
1139 static char *block_devnode(struct device *dev, mode_t *mode)
1141 struct gendisk *disk = dev_to_disk(dev);
1144 return disk->devnode(disk, mode);
1148 static struct device_type disk_type = {
1150 .groups = disk_attr_groups,
1151 .release = disk_release,
1152 .devnode = block_devnode,
1155 #ifdef CONFIG_PROC_FS
1157 * aggregate disk stat collector. Uses the same stats that the sysfs
1158 * entries do, above, but makes them available through one seq_file.
1160 * The output looks suspiciously like /proc/partitions with a bunch of
1163 static int diskstats_show(struct seq_file *seqf, void *v)
1165 struct gendisk *gp = v;
1166 struct disk_part_iter piter;
1167 struct hd_struct *hd;
1168 char buf[BDEVNAME_SIZE];
1172 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1173 seq_puts(seqf, "major minor name"
1174 " rio rmerge rsect ruse wio wmerge "
1175 "wsect wuse running use aveq"
1179 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1180 while ((hd = disk_part_iter_next(&piter))) {
1181 cpu = part_stat_lock();
1182 part_round_stats(cpu, hd);
1184 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1185 "%u %lu %lu %lu %u %u %u %u\n",
1186 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1187 disk_name(gp, hd->partno, buf),
1188 part_stat_read(hd, ios[READ]),
1189 part_stat_read(hd, merges[READ]),
1190 part_stat_read(hd, sectors[READ]),
1191 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1192 part_stat_read(hd, ios[WRITE]),
1193 part_stat_read(hd, merges[WRITE]),
1194 part_stat_read(hd, sectors[WRITE]),
1195 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1197 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1198 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1201 disk_part_iter_exit(&piter);
1206 static const struct seq_operations diskstats_op = {
1207 .start = disk_seqf_start,
1208 .next = disk_seqf_next,
1209 .stop = disk_seqf_stop,
1210 .show = diskstats_show
1213 static int diskstats_open(struct inode *inode, struct file *file)
1215 return seq_open(file, &diskstats_op);
1218 static const struct file_operations proc_diskstats_operations = {
1219 .open = diskstats_open,
1221 .llseek = seq_lseek,
1222 .release = seq_release,
1225 static int __init proc_genhd_init(void)
1227 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1228 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1231 module_init(proc_genhd_init);
1232 #endif /* CONFIG_PROC_FS */
1234 dev_t blk_lookup_devt(const char *name, int partno)
1236 dev_t devt = MKDEV(0, 0);
1237 struct class_dev_iter iter;
1240 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1241 while ((dev = class_dev_iter_next(&iter))) {
1242 struct gendisk *disk = dev_to_disk(dev);
1243 struct hd_struct *part;
1245 if (strcmp(dev_name(dev), name))
1248 if (partno < disk->minors) {
1249 /* We need to return the right devno, even
1250 * if the partition doesn't exist yet.
1252 devt = MKDEV(MAJOR(dev->devt),
1253 MINOR(dev->devt) + partno);
1256 part = disk_get_part(disk, partno);
1258 devt = part_devt(part);
1259 disk_put_part(part);
1262 disk_put_part(part);
1264 class_dev_iter_exit(&iter);
1267 EXPORT_SYMBOL(blk_lookup_devt);
1269 struct gendisk *alloc_disk(int minors)
1271 return alloc_disk_node(minors, -1);
1273 EXPORT_SYMBOL(alloc_disk);
1275 struct gendisk *alloc_disk_node(int minors, int node_id)
1277 struct gendisk *disk;
1279 disk = kmalloc_node(sizeof(struct gendisk),
1280 GFP_KERNEL | __GFP_ZERO, node_id);
1282 if (!init_part_stats(&disk->part0)) {
1286 disk->node_id = node_id;
1287 if (disk_expand_part_tbl(disk, 0)) {
1288 free_part_stats(&disk->part0);
1292 disk->part_tbl->part[0] = &disk->part0;
1294 hd_ref_init(&disk->part0);
1296 disk->minors = minors;
1297 rand_initialize_disk(disk);
1298 disk_to_dev(disk)->class = &block_class;
1299 disk_to_dev(disk)->type = &disk_type;
1300 device_initialize(disk_to_dev(disk));
1304 EXPORT_SYMBOL(alloc_disk_node);
1306 struct kobject *get_disk(struct gendisk *disk)
1308 struct module *owner;
1309 struct kobject *kobj;
1313 owner = disk->fops->owner;
1314 if (owner && !try_module_get(owner))
1316 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1325 EXPORT_SYMBOL(get_disk);
1327 void put_disk(struct gendisk *disk)
1330 kobject_put(&disk_to_dev(disk)->kobj);
1333 EXPORT_SYMBOL(put_disk);
1335 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1337 char event[] = "DISK_RO=1";
1338 char *envp[] = { event, NULL };
1342 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1345 void set_device_ro(struct block_device *bdev, int flag)
1347 bdev->bd_part->policy = flag;
1350 EXPORT_SYMBOL(set_device_ro);
1352 void set_disk_ro(struct gendisk *disk, int flag)
1354 struct disk_part_iter piter;
1355 struct hd_struct *part;
1357 if (disk->part0.policy != flag) {
1358 set_disk_ro_uevent(disk, flag);
1359 disk->part0.policy = flag;
1362 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1363 while ((part = disk_part_iter_next(&piter)))
1364 part->policy = flag;
1365 disk_part_iter_exit(&piter);
1368 EXPORT_SYMBOL(set_disk_ro);
1370 int bdev_read_only(struct block_device *bdev)
1374 return bdev->bd_part->policy;
1377 EXPORT_SYMBOL(bdev_read_only);
1379 int invalidate_partition(struct gendisk *disk, int partno)
1382 struct block_device *bdev = bdget_disk(disk, partno);
1385 res = __invalidate_device(bdev, true);
1391 EXPORT_SYMBOL(invalidate_partition);
1394 * Disk events - monitor disk events like media change and eject request.
1396 struct disk_events {
1397 struct list_head node; /* all disk_event's */
1398 struct gendisk *disk; /* the associated disk */
1401 struct mutex block_mutex; /* protects blocking */
1402 int block; /* event blocking depth */
1403 unsigned int pending; /* events already sent out */
1404 unsigned int clearing; /* events being cleared */
1406 long poll_msecs; /* interval, -1 for default */
1407 struct delayed_work dwork;
1410 static const char *disk_events_strs[] = {
1411 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1412 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1415 static char *disk_uevents[] = {
1416 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1417 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1420 /* list of all disk_events */
1421 static DEFINE_MUTEX(disk_events_mutex);
1422 static LIST_HEAD(disk_events);
1424 /* disable in-kernel polling by default */
1425 static unsigned long disk_events_dfl_poll_msecs = 0;
1427 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1429 struct disk_events *ev = disk->ev;
1430 long intv_msecs = 0;
1433 * If device-specific poll interval is set, always use it. If
1434 * the default is being used, poll iff there are events which
1435 * can't be monitored asynchronously.
1437 if (ev->poll_msecs >= 0)
1438 intv_msecs = ev->poll_msecs;
1439 else if (disk->events & ~disk->async_events)
1440 intv_msecs = disk_events_dfl_poll_msecs;
1442 return msecs_to_jiffies(intv_msecs);
1446 * disk_block_events - block and flush disk event checking
1447 * @disk: disk to block events for
1449 * On return from this function, it is guaranteed that event checking
1450 * isn't in progress and won't happen until unblocked by
1451 * disk_unblock_events(). Events blocking is counted and the actual
1452 * unblocking happens after the matching number of unblocks are done.
1454 * Note that this intentionally does not block event checking from
1455 * disk_clear_events().
1460 void disk_block_events(struct gendisk *disk)
1462 struct disk_events *ev = disk->ev;
1463 unsigned long flags;
1470 * Outer mutex ensures that the first blocker completes canceling
1471 * the event work before further blockers are allowed to finish.
1473 mutex_lock(&ev->block_mutex);
1475 spin_lock_irqsave(&ev->lock, flags);
1476 cancel = !ev->block++;
1477 spin_unlock_irqrestore(&ev->lock, flags);
1480 cancel_delayed_work_sync(&disk->ev->dwork);
1482 mutex_unlock(&ev->block_mutex);
1485 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1487 struct disk_events *ev = disk->ev;
1489 unsigned long flags;
1491 spin_lock_irqsave(&ev->lock, flags);
1493 if (WARN_ON_ONCE(ev->block <= 0))
1500 * Not exactly a latency critical operation, set poll timer
1501 * slack to 25% and kick event check.
1503 intv = disk_events_poll_jiffies(disk);
1504 set_timer_slack(&ev->dwork.timer, intv / 4);
1506 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1508 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1510 spin_unlock_irqrestore(&ev->lock, flags);
1514 * disk_unblock_events - unblock disk event checking
1515 * @disk: disk to unblock events for
1517 * Undo disk_block_events(). When the block count reaches zero, it
1518 * starts events polling if configured.
1521 * Don't care. Safe to call from irq context.
1523 void disk_unblock_events(struct gendisk *disk)
1526 __disk_unblock_events(disk, false);
1530 * disk_flush_events - schedule immediate event checking and flushing
1531 * @disk: disk to check and flush events for
1532 * @mask: events to flush
1534 * Schedule immediate event checking on @disk if not blocked. Events in
1535 * @mask are scheduled to be cleared from the driver. Note that this
1536 * doesn't clear the events from @disk->ev.
1539 * If @mask is non-zero must be called with bdev->bd_mutex held.
1541 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1543 struct disk_events *ev = disk->ev;
1548 spin_lock_irq(&ev->lock);
1549 ev->clearing |= mask;
1551 cancel_delayed_work(&ev->dwork);
1552 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1554 spin_unlock_irq(&ev->lock);
1558 * disk_clear_events - synchronously check, clear and return pending events
1559 * @disk: disk to fetch and clear events from
1560 * @mask: mask of events to be fetched and clearted
1562 * Disk events are synchronously checked and pending events in @mask
1563 * are cleared and returned. This ignores the block count.
1568 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1570 const struct block_device_operations *bdops = disk->fops;
1571 struct disk_events *ev = disk->ev;
1572 unsigned int pending;
1575 /* for drivers still using the old ->media_changed method */
1576 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1577 bdops->media_changed && bdops->media_changed(disk))
1578 return DISK_EVENT_MEDIA_CHANGE;
1582 /* tell the workfn about the events being cleared */
1583 spin_lock_irq(&ev->lock);
1584 ev->clearing |= mask;
1585 spin_unlock_irq(&ev->lock);
1587 /* uncondtionally schedule event check and wait for it to finish */
1588 disk_block_events(disk);
1589 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1590 flush_delayed_work(&ev->dwork);
1591 __disk_unblock_events(disk, false);
1593 /* then, fetch and clear pending events */
1594 spin_lock_irq(&ev->lock);
1595 WARN_ON_ONCE(ev->clearing & mask); /* cleared by workfn */
1596 pending = ev->pending & mask;
1597 ev->pending &= ~mask;
1598 spin_unlock_irq(&ev->lock);
1603 static void disk_events_workfn(struct work_struct *work)
1605 struct delayed_work *dwork = to_delayed_work(work);
1606 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1607 struct gendisk *disk = ev->disk;
1608 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1609 unsigned int clearing = ev->clearing;
1610 unsigned int events;
1612 int nr_events = 0, i;
1615 events = disk->fops->check_events(disk, clearing);
1617 /* accumulate pending events and schedule next poll if necessary */
1618 spin_lock_irq(&ev->lock);
1620 events &= ~ev->pending;
1621 ev->pending |= events;
1622 ev->clearing &= ~clearing;
1624 intv = disk_events_poll_jiffies(disk);
1625 if (!ev->block && intv)
1626 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1628 spin_unlock_irq(&ev->lock);
1631 * Tell userland about new events. Only the events listed in
1632 * @disk->events are reported. Unlisted events are processed the
1633 * same internally but never get reported to userland.
1635 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1636 if (events & disk->events & (1 << i))
1637 envp[nr_events++] = disk_uevents[i];
1640 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1644 * A disk events enabled device has the following sysfs nodes under
1645 * its /sys/block/X/ directory.
1647 * events : list of all supported events
1648 * events_async : list of events which can be detected w/o polling
1649 * events_poll_msecs : polling interval, 0: disable, -1: system default
1651 static ssize_t __disk_events_show(unsigned int events, char *buf)
1653 const char *delim = "";
1657 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1658 if (events & (1 << i)) {
1659 pos += sprintf(buf + pos, "%s%s",
1660 delim, disk_events_strs[i]);
1664 pos += sprintf(buf + pos, "\n");
1668 static ssize_t disk_events_show(struct device *dev,
1669 struct device_attribute *attr, char *buf)
1671 struct gendisk *disk = dev_to_disk(dev);
1673 return __disk_events_show(disk->events, buf);
1676 static ssize_t disk_events_async_show(struct device *dev,
1677 struct device_attribute *attr, char *buf)
1679 struct gendisk *disk = dev_to_disk(dev);
1681 return __disk_events_show(disk->async_events, buf);
1684 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1685 struct device_attribute *attr,
1688 struct gendisk *disk = dev_to_disk(dev);
1690 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1693 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1694 struct device_attribute *attr,
1695 const char *buf, size_t count)
1697 struct gendisk *disk = dev_to_disk(dev);
1700 if (!count || !sscanf(buf, "%ld", &intv))
1703 if (intv < 0 && intv != -1)
1706 disk_block_events(disk);
1707 disk->ev->poll_msecs = intv;
1708 __disk_unblock_events(disk, true);
1713 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1714 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1715 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1716 disk_events_poll_msecs_show,
1717 disk_events_poll_msecs_store);
1719 static const struct attribute *disk_events_attrs[] = {
1720 &dev_attr_events.attr,
1721 &dev_attr_events_async.attr,
1722 &dev_attr_events_poll_msecs.attr,
1727 * The default polling interval can be specified by the kernel
1728 * parameter block.events_dfl_poll_msecs which defaults to 0
1729 * (disable). This can also be modified runtime by writing to
1730 * /sys/module/block/events_dfl_poll_msecs.
1732 static int disk_events_set_dfl_poll_msecs(const char *val,
1733 const struct kernel_param *kp)
1735 struct disk_events *ev;
1738 ret = param_set_ulong(val, kp);
1742 mutex_lock(&disk_events_mutex);
1744 list_for_each_entry(ev, &disk_events, node)
1745 disk_flush_events(ev->disk, 0);
1747 mutex_unlock(&disk_events_mutex);
1752 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1753 .set = disk_events_set_dfl_poll_msecs,
1754 .get = param_get_ulong,
1757 #undef MODULE_PARAM_PREFIX
1758 #define MODULE_PARAM_PREFIX "block."
1760 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1761 &disk_events_dfl_poll_msecs, 0644);
1764 * disk_{add|del|release}_events - initialize and destroy disk_events.
1766 static void disk_add_events(struct gendisk *disk)
1768 struct disk_events *ev;
1770 if (!disk->fops->check_events)
1773 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1775 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1779 if (sysfs_create_files(&disk_to_dev(disk)->kobj,
1780 disk_events_attrs) < 0) {
1781 pr_warn("%s: failed to create sysfs files for events\n",
1789 INIT_LIST_HEAD(&ev->node);
1791 spin_lock_init(&ev->lock);
1792 mutex_init(&ev->block_mutex);
1794 ev->poll_msecs = -1;
1795 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1797 mutex_lock(&disk_events_mutex);
1798 list_add_tail(&ev->node, &disk_events);
1799 mutex_unlock(&disk_events_mutex);
1802 * Block count is initialized to 1 and the following initial
1803 * unblock kicks it into action.
1805 __disk_unblock_events(disk, true);
1808 static void disk_del_events(struct gendisk *disk)
1813 disk_block_events(disk);
1815 mutex_lock(&disk_events_mutex);
1816 list_del_init(&disk->ev->node);
1817 mutex_unlock(&disk_events_mutex);
1819 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1822 static void disk_release_events(struct gendisk *disk)
1824 /* the block count should be 1 from disk_del_events() */
1825 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);