1 // SPDX-License-Identifier: GPL-2.0
5 * Portions Copyright (C) 2020 Christoph Hellwig
8 #include <linux/module.h>
9 #include <linux/ctype.h>
11 #include <linux/kdev_t.h>
12 #include <linux/kernel.h>
13 #include <linux/blkdev.h>
14 #include <linux/backing-dev.h>
15 #include <linux/init.h>
16 #include <linux/spinlock.h>
17 #include <linux/proc_fs.h>
18 #include <linux/seq_file.h>
19 #include <linux/slab.h>
20 #include <linux/kmod.h>
21 #include <linux/major.h>
22 #include <linux/mutex.h>
23 #include <linux/idr.h>
24 #include <linux/log2.h>
25 #include <linux/pm_runtime.h>
26 #include <linux/badblocks.h>
27 #include <linux/part_stat.h>
28 #include "blk-throttle.h"
31 #include "blk-mq-sched.h"
32 #include "blk-rq-qos.h"
33 #include "blk-cgroup.h"
35 static struct kobject *block_depr;
38 * Unique, monotonically increasing sequential number associated with block
39 * devices instances (i.e. incremented each time a device is attached).
40 * Associating uevents with block devices in userspace is difficult and racy:
41 * the uevent netlink socket is lossy, and on slow and overloaded systems has
42 * a very high latency.
43 * Block devices do not have exclusive owners in userspace, any process can set
44 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
45 * can be reused again and again).
46 * A userspace process setting up a block device and watching for its events
47 * cannot thus reliably tell whether an event relates to the device it just set
48 * up or another earlier instance with the same name.
49 * This sequential number allows userspace processes to solve this problem, and
50 * uniquely associate an uevent to the lifetime to a device.
52 static atomic64_t diskseq;
54 /* for extended dynamic devt allocation, currently only one major is used */
55 #define NR_EXT_DEVT (1 << MINORBITS)
56 static DEFINE_IDA(ext_devt_ida);
58 void set_capacity(struct gendisk *disk, sector_t sectors)
60 struct block_device *bdev = disk->part0;
62 spin_lock(&bdev->bd_size_lock);
63 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
64 bdev->bd_nr_sectors = sectors;
65 spin_unlock(&bdev->bd_size_lock);
67 EXPORT_SYMBOL(set_capacity);
70 * Set disk capacity and notify if the size is not currently zero and will not
71 * be set to zero. Returns true if a uevent was sent, otherwise false.
73 bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
75 sector_t capacity = get_capacity(disk);
76 char *envp[] = { "RESIZE=1", NULL };
78 set_capacity(disk, size);
81 * Only print a message and send a uevent if the gendisk is user visible
82 * and alive. This avoids spamming the log and udev when setting the
83 * initial capacity during probing.
85 if (size == capacity ||
87 (disk->flags & GENHD_FL_HIDDEN))
90 pr_info("%s: detected capacity change from %lld to %lld\n",
91 disk->disk_name, capacity, size);
94 * Historically we did not send a uevent for changes to/from an empty
97 if (!capacity || !size)
99 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
102 EXPORT_SYMBOL_GPL(set_capacity_and_notify);
104 static void part_stat_read_all(struct block_device *part,
105 struct disk_stats *stat)
109 memset(stat, 0, sizeof(struct disk_stats));
110 for_each_possible_cpu(cpu) {
111 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
114 for (group = 0; group < NR_STAT_GROUPS; group++) {
115 stat->nsecs[group] += ptr->nsecs[group];
116 stat->sectors[group] += ptr->sectors[group];
117 stat->ios[group] += ptr->ios[group];
118 stat->merges[group] += ptr->merges[group];
121 stat->io_ticks += ptr->io_ticks;
125 static unsigned int part_in_flight(struct block_device *part)
127 unsigned int inflight = 0;
130 for_each_possible_cpu(cpu) {
131 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
132 part_stat_local_read_cpu(part, in_flight[1], cpu);
134 if ((int)inflight < 0)
140 static void part_in_flight_rw(struct block_device *part,
141 unsigned int inflight[2])
147 for_each_possible_cpu(cpu) {
148 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
149 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
151 if ((int)inflight[0] < 0)
153 if ((int)inflight[1] < 0)
158 * Can be deleted altogether. Later.
161 #define BLKDEV_MAJOR_HASH_SIZE 255
162 static struct blk_major_name {
163 struct blk_major_name *next;
166 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
167 void (*probe)(dev_t devt);
169 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
170 static DEFINE_MUTEX(major_names_lock);
171 static DEFINE_SPINLOCK(major_names_spinlock);
173 /* index in the above - for now: assume no multimajor ranges */
174 static inline int major_to_index(unsigned major)
176 return major % BLKDEV_MAJOR_HASH_SIZE;
179 #ifdef CONFIG_PROC_FS
180 void blkdev_show(struct seq_file *seqf, off_t offset)
182 struct blk_major_name *dp;
184 spin_lock(&major_names_spinlock);
185 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
186 if (dp->major == offset)
187 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
188 spin_unlock(&major_names_spinlock);
190 #endif /* CONFIG_PROC_FS */
193 * __register_blkdev - register a new block device
195 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
196 * @major = 0, try to allocate any unused major number.
197 * @name: the name of the new block device as a zero terminated string
198 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
199 * pre-created device node is accessed. When a probe call uses
200 * add_disk() and it fails the driver must cleanup resources. This
201 * interface may soon be removed.
203 * The @name must be unique within the system.
205 * The return value depends on the @major input parameter:
207 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
208 * then the function returns zero on success, or a negative error code
209 * - if any unused major number was requested with @major = 0 parameter
210 * then the return value is the allocated major number in range
211 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
213 * See Documentation/admin-guide/devices.txt for the list of allocated
216 * Use register_blkdev instead for any new code.
218 int __register_blkdev(unsigned int major, const char *name,
219 void (*probe)(dev_t devt))
221 struct blk_major_name **n, *p;
224 mutex_lock(&major_names_lock);
228 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
229 if (major_names[index] == NULL)
234 printk("%s: failed to get major for %s\n",
243 if (major >= BLKDEV_MAJOR_MAX) {
244 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
245 __func__, major, BLKDEV_MAJOR_MAX-1, name);
251 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
258 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
261 strlcpy(p->name, name, sizeof(p->name));
263 index = major_to_index(major);
265 spin_lock(&major_names_spinlock);
266 for (n = &major_names[index]; *n; n = &(*n)->next) {
267 if ((*n)->major == major)
274 spin_unlock(&major_names_spinlock);
277 printk("register_blkdev: cannot get major %u for %s\n",
282 mutex_unlock(&major_names_lock);
285 EXPORT_SYMBOL(__register_blkdev);
287 void unregister_blkdev(unsigned int major, const char *name)
289 struct blk_major_name **n;
290 struct blk_major_name *p = NULL;
291 int index = major_to_index(major);
293 mutex_lock(&major_names_lock);
294 spin_lock(&major_names_spinlock);
295 for (n = &major_names[index]; *n; n = &(*n)->next)
296 if ((*n)->major == major)
298 if (!*n || strcmp((*n)->name, name)) {
304 spin_unlock(&major_names_spinlock);
305 mutex_unlock(&major_names_lock);
309 EXPORT_SYMBOL(unregister_blkdev);
311 int blk_alloc_ext_minor(void)
315 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
321 void blk_free_ext_minor(unsigned int minor)
323 ida_free(&ext_devt_ida, minor);
326 static char *bdevt_str(dev_t devt, char *buf)
328 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
329 char tbuf[BDEVT_SIZE];
330 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
331 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
333 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
338 void disk_uevent(struct gendisk *disk, enum kobject_action action)
340 struct block_device *part;
344 xa_for_each(&disk->part_tbl, idx, part) {
345 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
347 if (!kobject_get_unless_zero(&part->bd_device.kobj))
351 kobject_uevent(bdev_kobj(part), action);
352 put_device(&part->bd_device);
357 EXPORT_SYMBOL_GPL(disk_uevent);
359 int disk_scan_partitions(struct gendisk *disk, fmode_t mode)
361 struct block_device *bdev;
364 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
366 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
368 if (disk->open_partitions)
372 * If the device is opened exclusively by current thread already, it's
373 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
374 * synchronize with other exclusive openers and other partition
377 if (!(mode & FMODE_EXCL)) {
378 ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions);
383 set_bit(GD_NEED_PART_SCAN, &disk->state);
384 bdev = blkdev_get_by_dev(disk_devt(disk), mode & ~FMODE_EXCL, NULL);
388 blkdev_put(bdev, mode & ~FMODE_EXCL);
391 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
392 * and this will cause that re-assemble partitioned raid device will
393 * creat partition for underlying disk.
395 clear_bit(GD_NEED_PART_SCAN, &disk->state);
396 if (!(mode & FMODE_EXCL))
397 bd_abort_claiming(disk->part0, disk_scan_partitions);
402 * device_add_disk - add disk information to kernel list
403 * @parent: parent device for the disk
404 * @disk: per-device partitioning information
405 * @groups: Additional per-device sysfs groups
407 * This function registers the partitioning information in @disk
410 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
411 const struct attribute_group **groups)
414 struct device *ddev = disk_to_dev(disk);
417 /* Only makes sense for bio-based to set ->poll_bio */
418 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
422 * The disk queue should now be all set with enough information about
423 * the device for the elevator code to pick an adequate default
424 * elevator if one is needed, that is, for devices requesting queue
427 elevator_init_mq(disk->queue);
430 * If the driver provides an explicit major number it also must provide
431 * the number of minors numbers supported, and those will be used to
433 * Otherwise just allocate the device numbers for both the whole device
434 * and all partitions from the extended dev_t space.
438 if (WARN_ON(!disk->minors))
439 goto out_exit_elevator;
441 if (disk->minors > DISK_MAX_PARTS) {
442 pr_err("block: can't allocate more than %d partitions\n",
444 disk->minors = DISK_MAX_PARTS;
446 if (disk->first_minor + disk->minors > MINORMASK + 1)
447 goto out_exit_elevator;
449 if (WARN_ON(disk->minors))
450 goto out_exit_elevator;
452 ret = blk_alloc_ext_minor();
454 goto out_exit_elevator;
455 disk->major = BLOCK_EXT_MAJOR;
456 disk->first_minor = ret;
459 /* delay uevents, until we scanned partition table */
460 dev_set_uevent_suppress(ddev, 1);
462 ddev->parent = parent;
463 ddev->groups = groups;
464 dev_set_name(ddev, "%s", disk->disk_name);
465 if (!(disk->flags & GENHD_FL_HIDDEN))
466 ddev->devt = MKDEV(disk->major, disk->first_minor);
467 ret = device_add(ddev);
469 goto out_free_ext_minor;
471 ret = disk_alloc_events(disk);
475 if (!sysfs_deprecated) {
476 ret = sysfs_create_link(block_depr, &ddev->kobj,
477 kobject_name(&ddev->kobj));
483 * avoid probable deadlock caused by allocating memory with
484 * GFP_KERNEL in runtime_resume callback of its all ancestor
487 pm_runtime_set_memalloc_noio(ddev, true);
489 ret = blk_integrity_add(disk);
491 goto out_del_block_link;
493 disk->part0->bd_holder_dir =
494 kobject_create_and_add("holders", &ddev->kobj);
495 if (!disk->part0->bd_holder_dir) {
497 goto out_del_integrity;
499 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
500 if (!disk->slave_dir) {
502 goto out_put_holder_dir;
505 ret = blk_register_queue(disk);
507 goto out_put_slave_dir;
509 if (!(disk->flags & GENHD_FL_HIDDEN)) {
510 ret = bdi_register(disk->bdi, "%u:%u",
511 disk->major, disk->first_minor);
513 goto out_unregister_queue;
514 bdi_set_owner(disk->bdi, ddev);
515 ret = sysfs_create_link(&ddev->kobj,
516 &disk->bdi->dev->kobj, "bdi");
518 goto out_unregister_bdi;
520 /* Make sure the first partition scan will be proceed */
521 if (get_capacity(disk) && !(disk->flags & GENHD_FL_NO_PART) &&
522 !test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
523 set_bit(GD_NEED_PART_SCAN, &disk->state);
525 bdev_add(disk->part0, ddev->devt);
526 if (get_capacity(disk))
527 disk_scan_partitions(disk, FMODE_READ);
530 * Announce the disk and partitions after all partitions are
531 * created. (for hidden disks uevents remain suppressed forever)
533 dev_set_uevent_suppress(ddev, 0);
534 disk_uevent(disk, KOBJ_ADD);
537 * Even if the block_device for a hidden gendisk is not
538 * registered, it needs to have a valid bd_dev so that the
539 * freeing of the dynamic major works.
541 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
544 disk_update_readahead(disk);
545 disk_add_events(disk);
546 set_bit(GD_ADDED, &disk->state);
550 if (!(disk->flags & GENHD_FL_HIDDEN))
551 bdi_unregister(disk->bdi);
552 out_unregister_queue:
553 blk_unregister_queue(disk);
554 rq_qos_exit(disk->queue);
556 kobject_put(disk->slave_dir);
557 disk->slave_dir = NULL;
559 kobject_put(disk->part0->bd_holder_dir);
561 blk_integrity_del(disk);
563 if (!sysfs_deprecated)
564 sysfs_remove_link(block_depr, dev_name(ddev));
568 if (disk->major == BLOCK_EXT_MAJOR)
569 blk_free_ext_minor(disk->first_minor);
571 if (disk->queue->elevator)
572 elevator_exit(disk->queue);
575 EXPORT_SYMBOL(device_add_disk);
578 * blk_mark_disk_dead - mark a disk as dead
579 * @disk: disk to mark as dead
581 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
584 void blk_mark_disk_dead(struct gendisk *disk)
586 set_bit(GD_DEAD, &disk->state);
587 blk_queue_start_drain(disk->queue);
590 * Stop buffered writers from dirtying pages that can't be written out.
592 set_capacity_and_notify(disk, 0);
594 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
597 * del_gendisk - remove the gendisk
598 * @disk: the struct gendisk to remove
600 * Removes the gendisk and all its associated resources. This deletes the
601 * partitions associated with the gendisk, and unregisters the associated
604 * This is the counter to the respective __device_add_disk() call.
606 * The final removal of the struct gendisk happens when its refcount reaches 0
607 * with put_disk(), which should be called after del_gendisk(), if
608 * __device_add_disk() was used.
610 * Drivers exist which depend on the release of the gendisk to be synchronous,
611 * it should not be deferred.
615 void del_gendisk(struct gendisk *disk)
617 struct request_queue *q = disk->queue;
621 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
624 blk_integrity_del(disk);
625 disk_del_events(disk);
627 mutex_lock(&disk->open_mutex);
628 remove_inode_hash(disk->part0->bd_inode);
629 blk_drop_partitions(disk);
630 mutex_unlock(&disk->open_mutex);
632 fsync_bdev(disk->part0);
633 __invalidate_device(disk->part0, true);
638 set_bit(GD_DEAD, &disk->state);
639 if (test_bit(GD_OWNS_QUEUE, &disk->state))
640 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
641 set_capacity(disk, 0);
644 * Prevent new I/O from crossing bio_queue_enter().
646 blk_queue_start_drain(q);
648 if (!(disk->flags & GENHD_FL_HIDDEN)) {
649 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
652 * Unregister bdi before releasing device numbers (as they can
653 * get reused and we'd get clashes in sysfs).
655 bdi_unregister(disk->bdi);
658 blk_unregister_queue(disk);
660 kobject_put(disk->part0->bd_holder_dir);
661 kobject_put(disk->slave_dir);
662 disk->slave_dir = NULL;
664 part_stat_set_all(disk->part0, 0);
665 disk->part0->bd_stamp = 0;
666 if (!sysfs_deprecated)
667 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
668 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
669 device_del(disk_to_dev(disk));
671 blk_mq_freeze_queue_wait(q);
673 blk_throtl_cancel_bios(disk);
676 blk_flush_integrity();
679 blk_mq_cancel_work_sync(q);
681 blk_mq_quiesce_queue(q);
683 mutex_lock(&q->sysfs_lock);
685 mutex_unlock(&q->sysfs_lock);
688 blk_mq_unquiesce_queue(q);
691 * If the disk does not own the queue, allow using passthrough requests
692 * again. Else leave the queue frozen to fail all I/O.
694 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
695 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
696 __blk_mq_unfreeze_queue(q, true);
699 blk_mq_exit_queue(q);
702 EXPORT_SYMBOL(del_gendisk);
705 * invalidate_disk - invalidate the disk
706 * @disk: the struct gendisk to invalidate
708 * A helper to invalidates the disk. It will clean the disk's associated
709 * buffer/page caches and reset its internal states so that the disk
710 * can be reused by the drivers.
714 void invalidate_disk(struct gendisk *disk)
716 struct block_device *bdev = disk->part0;
718 invalidate_bdev(bdev);
719 bdev->bd_inode->i_mapping->wb_err = 0;
720 set_capacity(disk, 0);
722 EXPORT_SYMBOL(invalidate_disk);
724 /* sysfs access to bad-blocks list. */
725 static ssize_t disk_badblocks_show(struct device *dev,
726 struct device_attribute *attr,
729 struct gendisk *disk = dev_to_disk(dev);
732 return sprintf(page, "\n");
734 return badblocks_show(disk->bb, page, 0);
737 static ssize_t disk_badblocks_store(struct device *dev,
738 struct device_attribute *attr,
739 const char *page, size_t len)
741 struct gendisk *disk = dev_to_disk(dev);
746 return badblocks_store(disk->bb, page, len, 0);
749 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
750 void blk_request_module(dev_t devt)
752 unsigned int major = MAJOR(devt);
753 struct blk_major_name **n;
755 mutex_lock(&major_names_lock);
756 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
757 if ((*n)->major == major && (*n)->probe) {
759 mutex_unlock(&major_names_lock);
763 mutex_unlock(&major_names_lock);
765 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
766 /* Make old-style 2.4 aliases work */
767 request_module("block-major-%d", MAJOR(devt));
769 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
772 * print a full list of all partitions - intended for places where the root
773 * filesystem can't be mounted and thus to give the victim some idea of what
776 void __init printk_all_partitions(void)
778 struct class_dev_iter iter;
781 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
782 while ((dev = class_dev_iter_next(&iter))) {
783 struct gendisk *disk = dev_to_disk(dev);
784 struct block_device *part;
785 char devt_buf[BDEVT_SIZE];
789 * Don't show empty devices or things that have been
792 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
796 * Note, unlike /proc/partitions, I am showing the numbers in
797 * hex - the same format as the root= option takes.
800 xa_for_each(&disk->part_tbl, idx, part) {
801 if (!bdev_nr_sectors(part))
803 printk("%s%s %10llu %pg %s",
804 bdev_is_partition(part) ? " " : "",
805 bdevt_str(part->bd_dev, devt_buf),
806 bdev_nr_sectors(part) >> 1, part,
808 part->bd_meta_info->uuid : "");
809 if (bdev_is_partition(part))
811 else if (dev->parent && dev->parent->driver)
812 printk(" driver: %s\n",
813 dev->parent->driver->name);
815 printk(" (driver?)\n");
819 class_dev_iter_exit(&iter);
822 #ifdef CONFIG_PROC_FS
824 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
827 struct class_dev_iter *iter;
830 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
832 return ERR_PTR(-ENOMEM);
834 seqf->private = iter;
835 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
837 dev = class_dev_iter_next(iter);
842 return dev_to_disk(dev);
845 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
850 dev = class_dev_iter_next(seqf->private);
852 return dev_to_disk(dev);
857 static void disk_seqf_stop(struct seq_file *seqf, void *v)
859 struct class_dev_iter *iter = seqf->private;
861 /* stop is called even after start failed :-( */
863 class_dev_iter_exit(iter);
865 seqf->private = NULL;
869 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
873 p = disk_seqf_start(seqf, pos);
874 if (!IS_ERR_OR_NULL(p) && !*pos)
875 seq_puts(seqf, "major minor #blocks name\n\n");
879 static int show_partition(struct seq_file *seqf, void *v)
881 struct gendisk *sgp = v;
882 struct block_device *part;
885 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
889 xa_for_each(&sgp->part_tbl, idx, part) {
890 if (!bdev_nr_sectors(part))
892 seq_printf(seqf, "%4d %7d %10llu %pg\n",
893 MAJOR(part->bd_dev), MINOR(part->bd_dev),
894 bdev_nr_sectors(part) >> 1, part);
900 static const struct seq_operations partitions_op = {
901 .start = show_partition_start,
902 .next = disk_seqf_next,
903 .stop = disk_seqf_stop,
904 .show = show_partition
908 static int __init genhd_device_init(void)
912 block_class.dev_kobj = sysfs_dev_block_kobj;
913 error = class_register(&block_class);
918 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
920 /* create top-level block dir */
921 if (!sysfs_deprecated)
922 block_depr = kobject_create_and_add("block", NULL);
926 subsys_initcall(genhd_device_init);
928 static ssize_t disk_range_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", disk->minors);
936 static ssize_t disk_ext_range_show(struct device *dev,
937 struct device_attribute *attr, char *buf)
939 struct gendisk *disk = dev_to_disk(dev);
941 return sprintf(buf, "%d\n",
942 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
945 static ssize_t disk_removable_show(struct device *dev,
946 struct device_attribute *attr, char *buf)
948 struct gendisk *disk = dev_to_disk(dev);
950 return sprintf(buf, "%d\n",
951 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
954 static ssize_t disk_hidden_show(struct device *dev,
955 struct device_attribute *attr, char *buf)
957 struct gendisk *disk = dev_to_disk(dev);
959 return sprintf(buf, "%d\n",
960 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
963 static ssize_t disk_ro_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", get_disk_ro(disk) ? 1 : 0);
971 ssize_t part_size_show(struct device *dev,
972 struct device_attribute *attr, char *buf)
974 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
977 ssize_t part_stat_show(struct device *dev,
978 struct device_attribute *attr, char *buf)
980 struct block_device *bdev = dev_to_bdev(dev);
981 struct request_queue *q = bdev_get_queue(bdev);
982 struct disk_stats stat;
983 unsigned int inflight;
986 inflight = blk_mq_in_flight(q, bdev);
988 inflight = part_in_flight(bdev);
992 update_io_ticks(bdev, jiffies, true);
995 part_stat_read_all(bdev, &stat);
997 "%8lu %8lu %8llu %8u "
998 "%8lu %8lu %8llu %8u "
1000 "%8lu %8lu %8llu %8u "
1003 stat.ios[STAT_READ],
1004 stat.merges[STAT_READ],
1005 (unsigned long long)stat.sectors[STAT_READ],
1006 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1007 stat.ios[STAT_WRITE],
1008 stat.merges[STAT_WRITE],
1009 (unsigned long long)stat.sectors[STAT_WRITE],
1010 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1012 jiffies_to_msecs(stat.io_ticks),
1013 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1014 stat.nsecs[STAT_WRITE] +
1015 stat.nsecs[STAT_DISCARD] +
1016 stat.nsecs[STAT_FLUSH],
1018 stat.ios[STAT_DISCARD],
1019 stat.merges[STAT_DISCARD],
1020 (unsigned long long)stat.sectors[STAT_DISCARD],
1021 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1022 stat.ios[STAT_FLUSH],
1023 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1026 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1029 struct block_device *bdev = dev_to_bdev(dev);
1030 struct request_queue *q = bdev_get_queue(bdev);
1031 unsigned int inflight[2];
1034 blk_mq_in_flight_rw(q, bdev, inflight);
1036 part_in_flight_rw(bdev, inflight);
1038 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1041 static ssize_t disk_capability_show(struct device *dev,
1042 struct device_attribute *attr, char *buf)
1044 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1045 return sprintf(buf, "0\n");
1048 static ssize_t disk_alignment_offset_show(struct device *dev,
1049 struct device_attribute *attr,
1052 struct gendisk *disk = dev_to_disk(dev);
1054 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1057 static ssize_t disk_discard_alignment_show(struct device *dev,
1058 struct device_attribute *attr,
1061 struct gendisk *disk = dev_to_disk(dev);
1063 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1066 static ssize_t diskseq_show(struct device *dev,
1067 struct device_attribute *attr, char *buf)
1069 struct gendisk *disk = dev_to_disk(dev);
1071 return sprintf(buf, "%llu\n", disk->diskseq);
1074 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1075 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1076 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1077 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1078 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1079 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1080 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1081 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1082 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1083 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1084 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1085 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1086 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1088 #ifdef CONFIG_FAIL_MAKE_REQUEST
1089 ssize_t part_fail_show(struct device *dev,
1090 struct device_attribute *attr, char *buf)
1092 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1095 ssize_t part_fail_store(struct device *dev,
1096 struct device_attribute *attr,
1097 const char *buf, size_t count)
1101 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1102 dev_to_bdev(dev)->bd_make_it_fail = i;
1107 static struct device_attribute dev_attr_fail =
1108 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1109 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1111 #ifdef CONFIG_FAIL_IO_TIMEOUT
1112 static struct device_attribute dev_attr_fail_timeout =
1113 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1116 static struct attribute *disk_attrs[] = {
1117 &dev_attr_range.attr,
1118 &dev_attr_ext_range.attr,
1119 &dev_attr_removable.attr,
1120 &dev_attr_hidden.attr,
1122 &dev_attr_size.attr,
1123 &dev_attr_alignment_offset.attr,
1124 &dev_attr_discard_alignment.attr,
1125 &dev_attr_capability.attr,
1126 &dev_attr_stat.attr,
1127 &dev_attr_inflight.attr,
1128 &dev_attr_badblocks.attr,
1129 &dev_attr_events.attr,
1130 &dev_attr_events_async.attr,
1131 &dev_attr_events_poll_msecs.attr,
1132 &dev_attr_diskseq.attr,
1133 #ifdef CONFIG_FAIL_MAKE_REQUEST
1134 &dev_attr_fail.attr,
1136 #ifdef CONFIG_FAIL_IO_TIMEOUT
1137 &dev_attr_fail_timeout.attr,
1142 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1144 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1145 struct gendisk *disk = dev_to_disk(dev);
1147 if (a == &dev_attr_badblocks.attr && !disk->bb)
1152 static struct attribute_group disk_attr_group = {
1153 .attrs = disk_attrs,
1154 .is_visible = disk_visible,
1157 static const struct attribute_group *disk_attr_groups[] = {
1159 #ifdef CONFIG_BLK_DEV_IO_TRACE
1160 &blk_trace_attr_group,
1166 * disk_release - releases all allocated resources of the gendisk
1167 * @dev: the device representing this disk
1169 * This function releases all allocated resources of the gendisk.
1171 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1172 * assigned. Since the request_queue sits on top of the gendisk for these
1173 * drivers we also call blk_put_queue() for them, and we expect the
1174 * request_queue refcount to reach 0 at this point, and so the request_queue
1175 * will also be freed prior to the disk.
1177 * Context: can sleep
1179 static void disk_release(struct device *dev)
1181 struct gendisk *disk = dev_to_disk(dev);
1184 WARN_ON_ONCE(disk_live(disk));
1187 * To undo the all initialization from blk_mq_init_allocated_queue in
1188 * case of a probe failure where add_disk is never called we have to
1189 * call blk_mq_exit_queue here. We can't do this for the more common
1190 * teardown case (yet) as the tagset can be gone by the time the disk
1191 * is released once it was added.
1193 if (queue_is_mq(disk->queue) &&
1194 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1195 !test_bit(GD_ADDED, &disk->state))
1196 blk_mq_exit_queue(disk->queue);
1198 blkcg_exit_disk(disk);
1200 bioset_exit(&disk->bio_split);
1202 disk_release_events(disk);
1203 kfree(disk->random);
1204 disk_free_zone_bitmaps(disk);
1205 xa_destroy(&disk->part_tbl);
1207 disk->queue->disk = NULL;
1208 blk_put_queue(disk->queue);
1210 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1211 disk->fops->free_disk(disk);
1213 iput(disk->part0->bd_inode); /* frees the disk */
1216 static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1218 const struct gendisk *disk = dev_to_disk(dev);
1220 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1223 struct class block_class = {
1225 .dev_uevent = block_uevent,
1228 static char *block_devnode(const struct device *dev, umode_t *mode,
1229 kuid_t *uid, kgid_t *gid)
1231 struct gendisk *disk = dev_to_disk(dev);
1233 if (disk->fops->devnode)
1234 return disk->fops->devnode(disk, mode);
1238 const struct device_type disk_type = {
1240 .groups = disk_attr_groups,
1241 .release = disk_release,
1242 .devnode = block_devnode,
1245 #ifdef CONFIG_PROC_FS
1247 * aggregate disk stat collector. Uses the same stats that the sysfs
1248 * entries do, above, but makes them available through one seq_file.
1250 * The output looks suspiciously like /proc/partitions with a bunch of
1253 static int diskstats_show(struct seq_file *seqf, void *v)
1255 struct gendisk *gp = v;
1256 struct block_device *hd;
1257 unsigned int inflight;
1258 struct disk_stats stat;
1262 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1263 seq_puts(seqf, "major minor name"
1264 " rio rmerge rsect ruse wio wmerge "
1265 "wsect wuse running use aveq"
1270 xa_for_each(&gp->part_tbl, idx, hd) {
1271 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1273 if (queue_is_mq(gp->queue))
1274 inflight = blk_mq_in_flight(gp->queue, hd);
1276 inflight = part_in_flight(hd);
1280 update_io_ticks(hd, jiffies, true);
1283 part_stat_read_all(hd, &stat);
1284 seq_printf(seqf, "%4d %7d %pg "
1291 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1292 stat.ios[STAT_READ],
1293 stat.merges[STAT_READ],
1294 stat.sectors[STAT_READ],
1295 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1297 stat.ios[STAT_WRITE],
1298 stat.merges[STAT_WRITE],
1299 stat.sectors[STAT_WRITE],
1300 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1303 jiffies_to_msecs(stat.io_ticks),
1304 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1305 stat.nsecs[STAT_WRITE] +
1306 stat.nsecs[STAT_DISCARD] +
1307 stat.nsecs[STAT_FLUSH],
1309 stat.ios[STAT_DISCARD],
1310 stat.merges[STAT_DISCARD],
1311 stat.sectors[STAT_DISCARD],
1312 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1314 stat.ios[STAT_FLUSH],
1315 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1324 static const struct seq_operations diskstats_op = {
1325 .start = disk_seqf_start,
1326 .next = disk_seqf_next,
1327 .stop = disk_seqf_stop,
1328 .show = diskstats_show
1331 static int __init proc_genhd_init(void)
1333 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1334 proc_create_seq("partitions", 0, NULL, &partitions_op);
1337 module_init(proc_genhd_init);
1338 #endif /* CONFIG_PROC_FS */
1340 dev_t part_devt(struct gendisk *disk, u8 partno)
1342 struct block_device *part;
1346 part = xa_load(&disk->part_tbl, partno);
1348 devt = part->bd_dev;
1354 dev_t blk_lookup_devt(const char *name, int partno)
1356 dev_t devt = MKDEV(0, 0);
1357 struct class_dev_iter iter;
1360 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1361 while ((dev = class_dev_iter_next(&iter))) {
1362 struct gendisk *disk = dev_to_disk(dev);
1364 if (strcmp(dev_name(dev), name))
1367 if (partno < disk->minors) {
1368 /* We need to return the right devno, even
1369 * if the partition doesn't exist yet.
1371 devt = MKDEV(MAJOR(dev->devt),
1372 MINOR(dev->devt) + partno);
1374 devt = part_devt(disk, partno);
1379 class_dev_iter_exit(&iter);
1383 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1384 struct lock_class_key *lkclass)
1386 struct gendisk *disk;
1388 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1392 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1395 disk->bdi = bdi_alloc(node_id);
1397 goto out_free_bioset;
1399 /* bdev_alloc() might need the queue, set before the first call */
1402 disk->part0 = bdev_alloc(disk, 0);
1406 disk->node_id = node_id;
1407 mutex_init(&disk->open_mutex);
1408 xa_init(&disk->part_tbl);
1409 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1410 goto out_destroy_part_tbl;
1412 if (blkcg_init_disk(disk))
1413 goto out_erase_part0;
1415 rand_initialize_disk(disk);
1416 disk_to_dev(disk)->class = &block_class;
1417 disk_to_dev(disk)->type = &disk_type;
1418 device_initialize(disk_to_dev(disk));
1421 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1422 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1423 INIT_LIST_HEAD(&disk->slave_bdevs);
1428 xa_erase(&disk->part_tbl, 0);
1429 out_destroy_part_tbl:
1430 xa_destroy(&disk->part_tbl);
1431 disk->part0->bd_disk = NULL;
1432 iput(disk->part0->bd_inode);
1436 bioset_exit(&disk->bio_split);
1442 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1444 struct request_queue *q;
1445 struct gendisk *disk;
1447 q = blk_alloc_queue(node);
1451 disk = __alloc_disk_node(q, node, lkclass);
1456 set_bit(GD_OWNS_QUEUE, &disk->state);
1459 EXPORT_SYMBOL(__blk_alloc_disk);
1462 * put_disk - decrements the gendisk refcount
1463 * @disk: the struct gendisk to decrement the refcount for
1465 * This decrements the refcount for the struct gendisk. When this reaches 0
1466 * we'll have disk_release() called.
1468 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1469 * when handling probe errors (that is before add_disk() is called).
1471 * Context: Any context, but the last reference must not be dropped from
1474 void put_disk(struct gendisk *disk)
1477 put_device(disk_to_dev(disk));
1479 EXPORT_SYMBOL(put_disk);
1481 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1483 char event[] = "DISK_RO=1";
1484 char *envp[] = { event, NULL };
1488 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1492 * set_disk_ro - set a gendisk read-only
1493 * @disk: gendisk to operate on
1494 * @read_only: %true to set the disk read-only, %false set the disk read/write
1496 * This function is used to indicate whether a given disk device should have its
1497 * read-only flag set. set_disk_ro() is typically used by device drivers to
1498 * indicate whether the underlying physical device is write-protected.
1500 void set_disk_ro(struct gendisk *disk, bool read_only)
1503 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1506 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1509 set_disk_ro_uevent(disk, read_only);
1511 EXPORT_SYMBOL(set_disk_ro);
1513 void inc_diskseq(struct gendisk *disk)
1515 disk->diskseq = atomic64_inc_return(&diskseq);