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, void *owner)
361 struct block_device *bdev;
363 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
365 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
367 if (disk->open_partitions)
369 /* Someone else has bdev exclusively open? */
370 if (disk->part0->bd_holder && disk->part0->bd_holder != owner)
373 set_bit(GD_NEED_PART_SCAN, &disk->state);
374 bdev = blkdev_get_by_dev(disk_devt(disk), mode, NULL);
376 return PTR_ERR(bdev);
377 blkdev_put(bdev, mode);
382 * device_add_disk - add disk information to kernel list
383 * @parent: parent device for the disk
384 * @disk: per-device partitioning information
385 * @groups: Additional per-device sysfs groups
387 * This function registers the partitioning information in @disk
390 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
391 const struct attribute_group **groups)
394 struct device *ddev = disk_to_dev(disk);
397 /* Only makes sense for bio-based to set ->poll_bio */
398 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
402 * The disk queue should now be all set with enough information about
403 * the device for the elevator code to pick an adequate default
404 * elevator if one is needed, that is, for devices requesting queue
407 elevator_init_mq(disk->queue);
410 * If the driver provides an explicit major number it also must provide
411 * the number of minors numbers supported, and those will be used to
413 * Otherwise just allocate the device numbers for both the whole device
414 * and all partitions from the extended dev_t space.
418 if (WARN_ON(!disk->minors))
419 goto out_exit_elevator;
421 if (disk->minors > DISK_MAX_PARTS) {
422 pr_err("block: can't allocate more than %d partitions\n",
424 disk->minors = DISK_MAX_PARTS;
426 if (disk->first_minor + disk->minors > MINORMASK + 1)
427 goto out_exit_elevator;
429 if (WARN_ON(disk->minors))
430 goto out_exit_elevator;
432 ret = blk_alloc_ext_minor();
434 goto out_exit_elevator;
435 disk->major = BLOCK_EXT_MAJOR;
436 disk->first_minor = ret;
439 /* delay uevents, until we scanned partition table */
440 dev_set_uevent_suppress(ddev, 1);
442 ddev->parent = parent;
443 ddev->groups = groups;
444 dev_set_name(ddev, "%s", disk->disk_name);
445 if (!(disk->flags & GENHD_FL_HIDDEN))
446 ddev->devt = MKDEV(disk->major, disk->first_minor);
447 ret = device_add(ddev);
449 goto out_free_ext_minor;
451 ret = disk_alloc_events(disk);
455 if (!sysfs_deprecated) {
456 ret = sysfs_create_link(block_depr, &ddev->kobj,
457 kobject_name(&ddev->kobj));
463 * avoid probable deadlock caused by allocating memory with
464 * GFP_KERNEL in runtime_resume callback of its all ancestor
467 pm_runtime_set_memalloc_noio(ddev, true);
469 ret = blk_integrity_add(disk);
471 goto out_del_block_link;
473 disk->part0->bd_holder_dir =
474 kobject_create_and_add("holders", &ddev->kobj);
475 if (!disk->part0->bd_holder_dir) {
477 goto out_del_integrity;
479 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
480 if (!disk->slave_dir) {
482 goto out_put_holder_dir;
485 ret = blk_register_queue(disk);
487 goto out_put_slave_dir;
489 if (!(disk->flags & GENHD_FL_HIDDEN)) {
490 ret = bdi_register(disk->bdi, "%u:%u",
491 disk->major, disk->first_minor);
493 goto out_unregister_queue;
494 bdi_set_owner(disk->bdi, ddev);
495 ret = sysfs_create_link(&ddev->kobj,
496 &disk->bdi->dev->kobj, "bdi");
498 goto out_unregister_bdi;
500 bdev_add(disk->part0, ddev->devt);
501 if (get_capacity(disk))
502 disk_scan_partitions(disk, FMODE_READ, NULL);
505 * Announce the disk and partitions after all partitions are
506 * created. (for hidden disks uevents remain suppressed forever)
508 dev_set_uevent_suppress(ddev, 0);
509 disk_uevent(disk, KOBJ_ADD);
512 * Even if the block_device for a hidden gendisk is not
513 * registered, it needs to have a valid bd_dev so that the
514 * freeing of the dynamic major works.
516 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
519 disk_update_readahead(disk);
520 disk_add_events(disk);
521 set_bit(GD_ADDED, &disk->state);
525 if (!(disk->flags & GENHD_FL_HIDDEN))
526 bdi_unregister(disk->bdi);
527 out_unregister_queue:
528 blk_unregister_queue(disk);
529 rq_qos_exit(disk->queue);
531 kobject_put(disk->slave_dir);
532 disk->slave_dir = NULL;
534 kobject_put(disk->part0->bd_holder_dir);
536 blk_integrity_del(disk);
538 if (!sysfs_deprecated)
539 sysfs_remove_link(block_depr, dev_name(ddev));
543 if (disk->major == BLOCK_EXT_MAJOR)
544 blk_free_ext_minor(disk->first_minor);
546 if (disk->queue->elevator)
547 elevator_exit(disk->queue);
550 EXPORT_SYMBOL(device_add_disk);
553 * blk_mark_disk_dead - mark a disk as dead
554 * @disk: disk to mark as dead
556 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
559 void blk_mark_disk_dead(struct gendisk *disk)
561 set_bit(GD_DEAD, &disk->state);
562 blk_queue_start_drain(disk->queue);
565 * Stop buffered writers from dirtying pages that can't be written out.
567 set_capacity_and_notify(disk, 0);
569 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
572 * del_gendisk - remove the gendisk
573 * @disk: the struct gendisk to remove
575 * Removes the gendisk and all its associated resources. This deletes the
576 * partitions associated with the gendisk, and unregisters the associated
579 * This is the counter to the respective __device_add_disk() call.
581 * The final removal of the struct gendisk happens when its refcount reaches 0
582 * with put_disk(), which should be called after del_gendisk(), if
583 * __device_add_disk() was used.
585 * Drivers exist which depend on the release of the gendisk to be synchronous,
586 * it should not be deferred.
590 void del_gendisk(struct gendisk *disk)
592 struct request_queue *q = disk->queue;
596 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
599 blk_integrity_del(disk);
600 disk_del_events(disk);
602 mutex_lock(&disk->open_mutex);
603 remove_inode_hash(disk->part0->bd_inode);
604 blk_drop_partitions(disk);
605 mutex_unlock(&disk->open_mutex);
607 fsync_bdev(disk->part0);
608 __invalidate_device(disk->part0, true);
613 set_bit(GD_DEAD, &disk->state);
614 if (test_bit(GD_OWNS_QUEUE, &disk->state))
615 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
616 set_capacity(disk, 0);
619 * Prevent new I/O from crossing bio_queue_enter().
621 blk_queue_start_drain(q);
623 if (!(disk->flags & GENHD_FL_HIDDEN)) {
624 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
627 * Unregister bdi before releasing device numbers (as they can
628 * get reused and we'd get clashes in sysfs).
630 bdi_unregister(disk->bdi);
633 blk_unregister_queue(disk);
635 kobject_put(disk->part0->bd_holder_dir);
636 kobject_put(disk->slave_dir);
637 disk->slave_dir = NULL;
639 part_stat_set_all(disk->part0, 0);
640 disk->part0->bd_stamp = 0;
641 if (!sysfs_deprecated)
642 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
643 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
644 device_del(disk_to_dev(disk));
646 blk_mq_freeze_queue_wait(q);
648 blk_throtl_cancel_bios(disk);
651 blk_flush_integrity();
654 blk_mq_cancel_work_sync(q);
656 blk_mq_quiesce_queue(q);
658 mutex_lock(&q->sysfs_lock);
660 mutex_unlock(&q->sysfs_lock);
663 blk_mq_unquiesce_queue(q);
666 * If the disk does not own the queue, allow using passthrough requests
667 * again. Else leave the queue frozen to fail all I/O.
669 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
670 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
671 __blk_mq_unfreeze_queue(q, true);
674 blk_mq_exit_queue(q);
677 EXPORT_SYMBOL(del_gendisk);
680 * invalidate_disk - invalidate the disk
681 * @disk: the struct gendisk to invalidate
683 * A helper to invalidates the disk. It will clean the disk's associated
684 * buffer/page caches and reset its internal states so that the disk
685 * can be reused by the drivers.
689 void invalidate_disk(struct gendisk *disk)
691 struct block_device *bdev = disk->part0;
693 invalidate_bdev(bdev);
694 bdev->bd_inode->i_mapping->wb_err = 0;
695 set_capacity(disk, 0);
697 EXPORT_SYMBOL(invalidate_disk);
699 /* sysfs access to bad-blocks list. */
700 static ssize_t disk_badblocks_show(struct device *dev,
701 struct device_attribute *attr,
704 struct gendisk *disk = dev_to_disk(dev);
707 return sprintf(page, "\n");
709 return badblocks_show(disk->bb, page, 0);
712 static ssize_t disk_badblocks_store(struct device *dev,
713 struct device_attribute *attr,
714 const char *page, size_t len)
716 struct gendisk *disk = dev_to_disk(dev);
721 return badblocks_store(disk->bb, page, len, 0);
724 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
725 void blk_request_module(dev_t devt)
727 unsigned int major = MAJOR(devt);
728 struct blk_major_name **n;
730 mutex_lock(&major_names_lock);
731 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
732 if ((*n)->major == major && (*n)->probe) {
734 mutex_unlock(&major_names_lock);
738 mutex_unlock(&major_names_lock);
740 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
741 /* Make old-style 2.4 aliases work */
742 request_module("block-major-%d", MAJOR(devt));
744 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
747 * print a full list of all partitions - intended for places where the root
748 * filesystem can't be mounted and thus to give the victim some idea of what
751 void __init printk_all_partitions(void)
753 struct class_dev_iter iter;
756 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
757 while ((dev = class_dev_iter_next(&iter))) {
758 struct gendisk *disk = dev_to_disk(dev);
759 struct block_device *part;
760 char devt_buf[BDEVT_SIZE];
764 * Don't show empty devices or things that have been
767 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
771 * Note, unlike /proc/partitions, I am showing the numbers in
772 * hex - the same format as the root= option takes.
775 xa_for_each(&disk->part_tbl, idx, part) {
776 if (!bdev_nr_sectors(part))
778 printk("%s%s %10llu %pg %s",
779 bdev_is_partition(part) ? " " : "",
780 bdevt_str(part->bd_dev, devt_buf),
781 bdev_nr_sectors(part) >> 1, part,
783 part->bd_meta_info->uuid : "");
784 if (bdev_is_partition(part))
786 else if (dev->parent && dev->parent->driver)
787 printk(" driver: %s\n",
788 dev->parent->driver->name);
790 printk(" (driver?)\n");
794 class_dev_iter_exit(&iter);
797 #ifdef CONFIG_PROC_FS
799 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
802 struct class_dev_iter *iter;
805 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
807 return ERR_PTR(-ENOMEM);
809 seqf->private = iter;
810 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
812 dev = class_dev_iter_next(iter);
817 return dev_to_disk(dev);
820 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
825 dev = class_dev_iter_next(seqf->private);
827 return dev_to_disk(dev);
832 static void disk_seqf_stop(struct seq_file *seqf, void *v)
834 struct class_dev_iter *iter = seqf->private;
836 /* stop is called even after start failed :-( */
838 class_dev_iter_exit(iter);
840 seqf->private = NULL;
844 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
848 p = disk_seqf_start(seqf, pos);
849 if (!IS_ERR_OR_NULL(p) && !*pos)
850 seq_puts(seqf, "major minor #blocks name\n\n");
854 static int show_partition(struct seq_file *seqf, void *v)
856 struct gendisk *sgp = v;
857 struct block_device *part;
860 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
864 xa_for_each(&sgp->part_tbl, idx, part) {
865 if (!bdev_nr_sectors(part))
867 seq_printf(seqf, "%4d %7d %10llu %pg\n",
868 MAJOR(part->bd_dev), MINOR(part->bd_dev),
869 bdev_nr_sectors(part) >> 1, part);
875 static const struct seq_operations partitions_op = {
876 .start = show_partition_start,
877 .next = disk_seqf_next,
878 .stop = disk_seqf_stop,
879 .show = show_partition
883 static int __init genhd_device_init(void)
887 block_class.dev_kobj = sysfs_dev_block_kobj;
888 error = class_register(&block_class);
893 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
895 /* create top-level block dir */
896 if (!sysfs_deprecated)
897 block_depr = kobject_create_and_add("block", NULL);
901 subsys_initcall(genhd_device_init);
903 static ssize_t disk_range_show(struct device *dev,
904 struct device_attribute *attr, char *buf)
906 struct gendisk *disk = dev_to_disk(dev);
908 return sprintf(buf, "%d\n", disk->minors);
911 static ssize_t disk_ext_range_show(struct device *dev,
912 struct device_attribute *attr, char *buf)
914 struct gendisk *disk = dev_to_disk(dev);
916 return sprintf(buf, "%d\n",
917 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
920 static ssize_t disk_removable_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",
926 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
929 static ssize_t disk_hidden_show(struct device *dev,
930 struct device_attribute *attr, char *buf)
932 struct gendisk *disk = dev_to_disk(dev);
934 return sprintf(buf, "%d\n",
935 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
938 static ssize_t disk_ro_show(struct device *dev,
939 struct device_attribute *attr, char *buf)
941 struct gendisk *disk = dev_to_disk(dev);
943 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
946 ssize_t part_size_show(struct device *dev,
947 struct device_attribute *attr, char *buf)
949 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
952 ssize_t part_stat_show(struct device *dev,
953 struct device_attribute *attr, char *buf)
955 struct block_device *bdev = dev_to_bdev(dev);
956 struct request_queue *q = bdev_get_queue(bdev);
957 struct disk_stats stat;
958 unsigned int inflight;
961 inflight = blk_mq_in_flight(q, bdev);
963 inflight = part_in_flight(bdev);
967 update_io_ticks(bdev, jiffies, true);
970 part_stat_read_all(bdev, &stat);
972 "%8lu %8lu %8llu %8u "
973 "%8lu %8lu %8llu %8u "
975 "%8lu %8lu %8llu %8u "
979 stat.merges[STAT_READ],
980 (unsigned long long)stat.sectors[STAT_READ],
981 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
982 stat.ios[STAT_WRITE],
983 stat.merges[STAT_WRITE],
984 (unsigned long long)stat.sectors[STAT_WRITE],
985 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
987 jiffies_to_msecs(stat.io_ticks),
988 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
989 stat.nsecs[STAT_WRITE] +
990 stat.nsecs[STAT_DISCARD] +
991 stat.nsecs[STAT_FLUSH],
993 stat.ios[STAT_DISCARD],
994 stat.merges[STAT_DISCARD],
995 (unsigned long long)stat.sectors[STAT_DISCARD],
996 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
997 stat.ios[STAT_FLUSH],
998 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1001 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1004 struct block_device *bdev = dev_to_bdev(dev);
1005 struct request_queue *q = bdev_get_queue(bdev);
1006 unsigned int inflight[2];
1009 blk_mq_in_flight_rw(q, bdev, inflight);
1011 part_in_flight_rw(bdev, inflight);
1013 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1016 static ssize_t disk_capability_show(struct device *dev,
1017 struct device_attribute *attr, char *buf)
1019 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1020 return sprintf(buf, "0\n");
1023 static ssize_t disk_alignment_offset_show(struct device *dev,
1024 struct device_attribute *attr,
1027 struct gendisk *disk = dev_to_disk(dev);
1029 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1032 static ssize_t disk_discard_alignment_show(struct device *dev,
1033 struct device_attribute *attr,
1036 struct gendisk *disk = dev_to_disk(dev);
1038 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1041 static ssize_t diskseq_show(struct device *dev,
1042 struct device_attribute *attr, char *buf)
1044 struct gendisk *disk = dev_to_disk(dev);
1046 return sprintf(buf, "%llu\n", disk->diskseq);
1049 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1050 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1051 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1052 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1053 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1054 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1055 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1056 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1057 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1058 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1059 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1060 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1061 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1063 #ifdef CONFIG_FAIL_MAKE_REQUEST
1064 ssize_t part_fail_show(struct device *dev,
1065 struct device_attribute *attr, char *buf)
1067 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1070 ssize_t part_fail_store(struct device *dev,
1071 struct device_attribute *attr,
1072 const char *buf, size_t count)
1076 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1077 dev_to_bdev(dev)->bd_make_it_fail = i;
1082 static struct device_attribute dev_attr_fail =
1083 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1084 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1086 #ifdef CONFIG_FAIL_IO_TIMEOUT
1087 static struct device_attribute dev_attr_fail_timeout =
1088 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1091 static struct attribute *disk_attrs[] = {
1092 &dev_attr_range.attr,
1093 &dev_attr_ext_range.attr,
1094 &dev_attr_removable.attr,
1095 &dev_attr_hidden.attr,
1097 &dev_attr_size.attr,
1098 &dev_attr_alignment_offset.attr,
1099 &dev_attr_discard_alignment.attr,
1100 &dev_attr_capability.attr,
1101 &dev_attr_stat.attr,
1102 &dev_attr_inflight.attr,
1103 &dev_attr_badblocks.attr,
1104 &dev_attr_events.attr,
1105 &dev_attr_events_async.attr,
1106 &dev_attr_events_poll_msecs.attr,
1107 &dev_attr_diskseq.attr,
1108 #ifdef CONFIG_FAIL_MAKE_REQUEST
1109 &dev_attr_fail.attr,
1111 #ifdef CONFIG_FAIL_IO_TIMEOUT
1112 &dev_attr_fail_timeout.attr,
1117 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1119 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1120 struct gendisk *disk = dev_to_disk(dev);
1122 if (a == &dev_attr_badblocks.attr && !disk->bb)
1127 static struct attribute_group disk_attr_group = {
1128 .attrs = disk_attrs,
1129 .is_visible = disk_visible,
1132 static const struct attribute_group *disk_attr_groups[] = {
1134 #ifdef CONFIG_BLK_DEV_IO_TRACE
1135 &blk_trace_attr_group,
1141 * disk_release - releases all allocated resources of the gendisk
1142 * @dev: the device representing this disk
1144 * This function releases all allocated resources of the gendisk.
1146 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1147 * assigned. Since the request_queue sits on top of the gendisk for these
1148 * drivers we also call blk_put_queue() for them, and we expect the
1149 * request_queue refcount to reach 0 at this point, and so the request_queue
1150 * will also be freed prior to the disk.
1152 * Context: can sleep
1154 static void disk_release(struct device *dev)
1156 struct gendisk *disk = dev_to_disk(dev);
1159 WARN_ON_ONCE(disk_live(disk));
1162 * To undo the all initialization from blk_mq_init_allocated_queue in
1163 * case of a probe failure where add_disk is never called we have to
1164 * call blk_mq_exit_queue here. We can't do this for the more common
1165 * teardown case (yet) as the tagset can be gone by the time the disk
1166 * is released once it was added.
1168 if (queue_is_mq(disk->queue) &&
1169 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1170 !test_bit(GD_ADDED, &disk->state))
1171 blk_mq_exit_queue(disk->queue);
1173 blkcg_exit_disk(disk);
1175 bioset_exit(&disk->bio_split);
1177 disk_release_events(disk);
1178 kfree(disk->random);
1179 disk_free_zone_bitmaps(disk);
1180 xa_destroy(&disk->part_tbl);
1182 disk->queue->disk = NULL;
1183 blk_put_queue(disk->queue);
1185 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1186 disk->fops->free_disk(disk);
1188 iput(disk->part0->bd_inode); /* frees the disk */
1191 static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1193 const struct gendisk *disk = dev_to_disk(dev);
1195 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1198 struct class block_class = {
1200 .dev_uevent = block_uevent,
1203 static char *block_devnode(const struct device *dev, umode_t *mode,
1204 kuid_t *uid, kgid_t *gid)
1206 struct gendisk *disk = dev_to_disk(dev);
1208 if (disk->fops->devnode)
1209 return disk->fops->devnode(disk, mode);
1213 const struct device_type disk_type = {
1215 .groups = disk_attr_groups,
1216 .release = disk_release,
1217 .devnode = block_devnode,
1220 #ifdef CONFIG_PROC_FS
1222 * aggregate disk stat collector. Uses the same stats that the sysfs
1223 * entries do, above, but makes them available through one seq_file.
1225 * The output looks suspiciously like /proc/partitions with a bunch of
1228 static int diskstats_show(struct seq_file *seqf, void *v)
1230 struct gendisk *gp = v;
1231 struct block_device *hd;
1232 unsigned int inflight;
1233 struct disk_stats stat;
1237 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1238 seq_puts(seqf, "major minor name"
1239 " rio rmerge rsect ruse wio wmerge "
1240 "wsect wuse running use aveq"
1245 xa_for_each(&gp->part_tbl, idx, hd) {
1246 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1248 if (queue_is_mq(gp->queue))
1249 inflight = blk_mq_in_flight(gp->queue, hd);
1251 inflight = part_in_flight(hd);
1255 update_io_ticks(hd, jiffies, true);
1258 part_stat_read_all(hd, &stat);
1259 seq_printf(seqf, "%4d %7d %pg "
1266 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1267 stat.ios[STAT_READ],
1268 stat.merges[STAT_READ],
1269 stat.sectors[STAT_READ],
1270 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1272 stat.ios[STAT_WRITE],
1273 stat.merges[STAT_WRITE],
1274 stat.sectors[STAT_WRITE],
1275 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1278 jiffies_to_msecs(stat.io_ticks),
1279 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1280 stat.nsecs[STAT_WRITE] +
1281 stat.nsecs[STAT_DISCARD] +
1282 stat.nsecs[STAT_FLUSH],
1284 stat.ios[STAT_DISCARD],
1285 stat.merges[STAT_DISCARD],
1286 stat.sectors[STAT_DISCARD],
1287 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1289 stat.ios[STAT_FLUSH],
1290 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1299 static const struct seq_operations diskstats_op = {
1300 .start = disk_seqf_start,
1301 .next = disk_seqf_next,
1302 .stop = disk_seqf_stop,
1303 .show = diskstats_show
1306 static int __init proc_genhd_init(void)
1308 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1309 proc_create_seq("partitions", 0, NULL, &partitions_op);
1312 module_init(proc_genhd_init);
1313 #endif /* CONFIG_PROC_FS */
1315 dev_t part_devt(struct gendisk *disk, u8 partno)
1317 struct block_device *part;
1321 part = xa_load(&disk->part_tbl, partno);
1323 devt = part->bd_dev;
1329 dev_t blk_lookup_devt(const char *name, int partno)
1331 dev_t devt = MKDEV(0, 0);
1332 struct class_dev_iter iter;
1335 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1336 while ((dev = class_dev_iter_next(&iter))) {
1337 struct gendisk *disk = dev_to_disk(dev);
1339 if (strcmp(dev_name(dev), name))
1342 if (partno < disk->minors) {
1343 /* We need to return the right devno, even
1344 * if the partition doesn't exist yet.
1346 devt = MKDEV(MAJOR(dev->devt),
1347 MINOR(dev->devt) + partno);
1349 devt = part_devt(disk, partno);
1354 class_dev_iter_exit(&iter);
1358 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1359 struct lock_class_key *lkclass)
1361 struct gendisk *disk;
1363 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1367 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1370 disk->bdi = bdi_alloc(node_id);
1372 goto out_free_bioset;
1374 /* bdev_alloc() might need the queue, set before the first call */
1377 disk->part0 = bdev_alloc(disk, 0);
1381 disk->node_id = node_id;
1382 mutex_init(&disk->open_mutex);
1383 xa_init(&disk->part_tbl);
1384 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1385 goto out_destroy_part_tbl;
1387 if (blkcg_init_disk(disk))
1388 goto out_erase_part0;
1390 rand_initialize_disk(disk);
1391 disk_to_dev(disk)->class = &block_class;
1392 disk_to_dev(disk)->type = &disk_type;
1393 device_initialize(disk_to_dev(disk));
1396 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1397 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1398 INIT_LIST_HEAD(&disk->slave_bdevs);
1403 xa_erase(&disk->part_tbl, 0);
1404 out_destroy_part_tbl:
1405 xa_destroy(&disk->part_tbl);
1406 disk->part0->bd_disk = NULL;
1407 iput(disk->part0->bd_inode);
1411 bioset_exit(&disk->bio_split);
1417 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1419 struct request_queue *q;
1420 struct gendisk *disk;
1422 q = blk_alloc_queue(node);
1426 disk = __alloc_disk_node(q, node, lkclass);
1431 set_bit(GD_OWNS_QUEUE, &disk->state);
1434 EXPORT_SYMBOL(__blk_alloc_disk);
1437 * put_disk - decrements the gendisk refcount
1438 * @disk: the struct gendisk to decrement the refcount for
1440 * This decrements the refcount for the struct gendisk. When this reaches 0
1441 * we'll have disk_release() called.
1443 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1444 * when handling probe errors (that is before add_disk() is called).
1446 * Context: Any context, but the last reference must not be dropped from
1449 void put_disk(struct gendisk *disk)
1452 put_device(disk_to_dev(disk));
1454 EXPORT_SYMBOL(put_disk);
1456 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1458 char event[] = "DISK_RO=1";
1459 char *envp[] = { event, NULL };
1463 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1467 * set_disk_ro - set a gendisk read-only
1468 * @disk: gendisk to operate on
1469 * @read_only: %true to set the disk read-only, %false set the disk read/write
1471 * This function is used to indicate whether a given disk device should have its
1472 * read-only flag set. set_disk_ro() is typically used by device drivers to
1473 * indicate whether the underlying physical device is write-protected.
1475 void set_disk_ro(struct gendisk *disk, bool read_only)
1478 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1481 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1484 set_disk_ro_uevent(disk, read_only);
1486 EXPORT_SYMBOL(set_disk_ro);
1488 void inc_diskseq(struct gendisk *disk)
1490 disk->diskseq = atomic64_inc_return(&diskseq);