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 bdev_set_nr_sectors(disk->part0, sectors);
62 EXPORT_SYMBOL(set_capacity);
65 * Set disk capacity and notify if the size is not currently zero and will not
66 * be set to zero. Returns true if a uevent was sent, otherwise false.
68 bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
70 sector_t capacity = get_capacity(disk);
71 char *envp[] = { "RESIZE=1", NULL };
73 set_capacity(disk, size);
76 * Only print a message and send a uevent if the gendisk is user visible
77 * and alive. This avoids spamming the log and udev when setting the
78 * initial capacity during probing.
80 if (size == capacity ||
82 (disk->flags & GENHD_FL_HIDDEN))
85 pr_info("%s: detected capacity change from %lld to %lld\n",
86 disk->disk_name, capacity, size);
89 * Historically we did not send a uevent for changes to/from an empty
92 if (!capacity || !size)
94 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
97 EXPORT_SYMBOL_GPL(set_capacity_and_notify);
99 static void part_stat_read_all(struct block_device *part,
100 struct disk_stats *stat)
104 memset(stat, 0, sizeof(struct disk_stats));
105 for_each_possible_cpu(cpu) {
106 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
109 for (group = 0; group < NR_STAT_GROUPS; group++) {
110 stat->nsecs[group] += ptr->nsecs[group];
111 stat->sectors[group] += ptr->sectors[group];
112 stat->ios[group] += ptr->ios[group];
113 stat->merges[group] += ptr->merges[group];
116 stat->io_ticks += ptr->io_ticks;
120 static unsigned int part_in_flight(struct block_device *part)
122 unsigned int inflight = 0;
125 for_each_possible_cpu(cpu) {
126 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
127 part_stat_local_read_cpu(part, in_flight[1], cpu);
129 if ((int)inflight < 0)
135 static void part_in_flight_rw(struct block_device *part,
136 unsigned int inflight[2])
142 for_each_possible_cpu(cpu) {
143 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
144 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
146 if ((int)inflight[0] < 0)
148 if ((int)inflight[1] < 0)
153 * Can be deleted altogether. Later.
156 #define BLKDEV_MAJOR_HASH_SIZE 255
157 static struct blk_major_name {
158 struct blk_major_name *next;
161 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
162 void (*probe)(dev_t devt);
164 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
165 static DEFINE_MUTEX(major_names_lock);
166 static DEFINE_SPINLOCK(major_names_spinlock);
168 /* index in the above - for now: assume no multimajor ranges */
169 static inline int major_to_index(unsigned major)
171 return major % BLKDEV_MAJOR_HASH_SIZE;
174 #ifdef CONFIG_PROC_FS
175 void blkdev_show(struct seq_file *seqf, off_t offset)
177 struct blk_major_name *dp;
179 spin_lock(&major_names_spinlock);
180 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
181 if (dp->major == offset)
182 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
183 spin_unlock(&major_names_spinlock);
185 #endif /* CONFIG_PROC_FS */
188 * __register_blkdev - register a new block device
190 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
191 * @major = 0, try to allocate any unused major number.
192 * @name: the name of the new block device as a zero terminated string
193 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
194 * pre-created device node is accessed. When a probe call uses
195 * add_disk() and it fails the driver must cleanup resources. This
196 * interface may soon be removed.
198 * The @name must be unique within the system.
200 * The return value depends on the @major input parameter:
202 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
203 * then the function returns zero on success, or a negative error code
204 * - if any unused major number was requested with @major = 0 parameter
205 * then the return value is the allocated major number in range
206 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
208 * See Documentation/admin-guide/devices.txt for the list of allocated
211 * Use register_blkdev instead for any new code.
213 int __register_blkdev(unsigned int major, const char *name,
214 void (*probe)(dev_t devt))
216 struct blk_major_name **n, *p;
219 mutex_lock(&major_names_lock);
223 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
224 if (major_names[index] == NULL)
229 printk("%s: failed to get major for %s\n",
238 if (major >= BLKDEV_MAJOR_MAX) {
239 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
240 __func__, major, BLKDEV_MAJOR_MAX-1, name);
246 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
253 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
256 strlcpy(p->name, name, sizeof(p->name));
258 index = major_to_index(major);
260 spin_lock(&major_names_spinlock);
261 for (n = &major_names[index]; *n; n = &(*n)->next) {
262 if ((*n)->major == major)
269 spin_unlock(&major_names_spinlock);
272 printk("register_blkdev: cannot get major %u for %s\n",
277 mutex_unlock(&major_names_lock);
280 EXPORT_SYMBOL(__register_blkdev);
282 void unregister_blkdev(unsigned int major, const char *name)
284 struct blk_major_name **n;
285 struct blk_major_name *p = NULL;
286 int index = major_to_index(major);
288 mutex_lock(&major_names_lock);
289 spin_lock(&major_names_spinlock);
290 for (n = &major_names[index]; *n; n = &(*n)->next)
291 if ((*n)->major == major)
293 if (!*n || strcmp((*n)->name, name)) {
299 spin_unlock(&major_names_spinlock);
300 mutex_unlock(&major_names_lock);
304 EXPORT_SYMBOL(unregister_blkdev);
306 int blk_alloc_ext_minor(void)
310 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
316 void blk_free_ext_minor(unsigned int minor)
318 ida_free(&ext_devt_ida, minor);
321 static char *bdevt_str(dev_t devt, char *buf)
323 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
324 char tbuf[BDEVT_SIZE];
325 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
326 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
328 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
333 void disk_uevent(struct gendisk *disk, enum kobject_action action)
335 struct block_device *part;
339 xa_for_each(&disk->part_tbl, idx, part) {
340 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
342 if (!kobject_get_unless_zero(&part->bd_device.kobj))
346 kobject_uevent(bdev_kobj(part), action);
347 put_device(&part->bd_device);
352 EXPORT_SYMBOL_GPL(disk_uevent);
354 int disk_scan_partitions(struct gendisk *disk, fmode_t mode)
356 struct block_device *bdev;
359 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
361 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
363 if (disk->open_partitions)
366 set_bit(GD_NEED_PART_SCAN, &disk->state);
368 * If the device is opened exclusively by current thread already, it's
369 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
370 * synchronize with other exclusive openers and other partition
373 if (!(mode & FMODE_EXCL)) {
374 ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions);
379 bdev = blkdev_get_by_dev(disk_devt(disk), mode & ~FMODE_EXCL, NULL);
383 blkdev_put(bdev, mode & ~FMODE_EXCL);
385 if (!(mode & FMODE_EXCL))
386 bd_abort_claiming(disk->part0, disk_scan_partitions);
391 * device_add_disk - add disk information to kernel list
392 * @parent: parent device for the disk
393 * @disk: per-device partitioning information
394 * @groups: Additional per-device sysfs groups
396 * This function registers the partitioning information in @disk
399 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
400 const struct attribute_group **groups)
403 struct device *ddev = disk_to_dev(disk);
406 /* Only makes sense for bio-based to set ->poll_bio */
407 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
411 * The disk queue should now be all set with enough information about
412 * the device for the elevator code to pick an adequate default
413 * elevator if one is needed, that is, for devices requesting queue
416 elevator_init_mq(disk->queue);
418 /* Mark bdev as having a submit_bio, if needed */
419 disk->part0->bd_has_submit_bio = disk->fops->submit_bio != NULL;
422 * If the driver provides an explicit major number it also must provide
423 * the number of minors numbers supported, and those will be used to
425 * Otherwise just allocate the device numbers for both the whole device
426 * and all partitions from the extended dev_t space.
430 if (WARN_ON(!disk->minors))
431 goto out_exit_elevator;
433 if (disk->minors > DISK_MAX_PARTS) {
434 pr_err("block: can't allocate more than %d partitions\n",
436 disk->minors = DISK_MAX_PARTS;
438 if (disk->first_minor + disk->minors > MINORMASK + 1)
439 goto out_exit_elevator;
441 if (WARN_ON(disk->minors))
442 goto out_exit_elevator;
444 ret = blk_alloc_ext_minor();
446 goto out_exit_elevator;
447 disk->major = BLOCK_EXT_MAJOR;
448 disk->first_minor = ret;
451 /* delay uevents, until we scanned partition table */
452 dev_set_uevent_suppress(ddev, 1);
454 ddev->parent = parent;
455 ddev->groups = groups;
456 dev_set_name(ddev, "%s", disk->disk_name);
457 if (!(disk->flags & GENHD_FL_HIDDEN))
458 ddev->devt = MKDEV(disk->major, disk->first_minor);
459 ret = device_add(ddev);
461 goto out_free_ext_minor;
463 ret = disk_alloc_events(disk);
467 if (!sysfs_deprecated) {
468 ret = sysfs_create_link(block_depr, &ddev->kobj,
469 kobject_name(&ddev->kobj));
475 * avoid probable deadlock caused by allocating memory with
476 * GFP_KERNEL in runtime_resume callback of its all ancestor
479 pm_runtime_set_memalloc_noio(ddev, true);
481 ret = blk_integrity_add(disk);
483 goto out_del_block_link;
485 disk->part0->bd_holder_dir =
486 kobject_create_and_add("holders", &ddev->kobj);
487 if (!disk->part0->bd_holder_dir) {
489 goto out_del_integrity;
491 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
492 if (!disk->slave_dir) {
494 goto out_put_holder_dir;
497 ret = blk_register_queue(disk);
499 goto out_put_slave_dir;
501 if (!(disk->flags & GENHD_FL_HIDDEN)) {
502 ret = bdi_register(disk->bdi, "%u:%u",
503 disk->major, disk->first_minor);
505 goto out_unregister_queue;
506 bdi_set_owner(disk->bdi, ddev);
507 ret = sysfs_create_link(&ddev->kobj,
508 &disk->bdi->dev->kobj, "bdi");
510 goto out_unregister_bdi;
512 /* Make sure the first partition scan will be proceed */
513 if (get_capacity(disk) && !(disk->flags & GENHD_FL_NO_PART) &&
514 !test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
515 set_bit(GD_NEED_PART_SCAN, &disk->state);
517 bdev_add(disk->part0, ddev->devt);
518 if (get_capacity(disk))
519 disk_scan_partitions(disk, FMODE_READ);
522 * Announce the disk and partitions after all partitions are
523 * created. (for hidden disks uevents remain suppressed forever)
525 dev_set_uevent_suppress(ddev, 0);
526 disk_uevent(disk, KOBJ_ADD);
529 * Even if the block_device for a hidden gendisk is not
530 * registered, it needs to have a valid bd_dev so that the
531 * freeing of the dynamic major works.
533 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
536 disk_update_readahead(disk);
537 disk_add_events(disk);
538 set_bit(GD_ADDED, &disk->state);
542 if (!(disk->flags & GENHD_FL_HIDDEN))
543 bdi_unregister(disk->bdi);
544 out_unregister_queue:
545 blk_unregister_queue(disk);
546 rq_qos_exit(disk->queue);
548 kobject_put(disk->slave_dir);
549 disk->slave_dir = NULL;
551 kobject_put(disk->part0->bd_holder_dir);
553 blk_integrity_del(disk);
555 if (!sysfs_deprecated)
556 sysfs_remove_link(block_depr, dev_name(ddev));
560 if (disk->major == BLOCK_EXT_MAJOR)
561 blk_free_ext_minor(disk->first_minor);
563 if (disk->queue->elevator)
564 elevator_exit(disk->queue);
567 EXPORT_SYMBOL(device_add_disk);
570 * blk_mark_disk_dead - mark a disk as dead
571 * @disk: disk to mark as dead
573 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
576 void blk_mark_disk_dead(struct gendisk *disk)
578 set_bit(GD_DEAD, &disk->state);
579 blk_queue_start_drain(disk->queue);
582 * Stop buffered writers from dirtying pages that can't be written out.
584 set_capacity_and_notify(disk, 0);
586 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
589 * del_gendisk - remove the gendisk
590 * @disk: the struct gendisk to remove
592 * Removes the gendisk and all its associated resources. This deletes the
593 * partitions associated with the gendisk, and unregisters the associated
596 * This is the counter to the respective __device_add_disk() call.
598 * The final removal of the struct gendisk happens when its refcount reaches 0
599 * with put_disk(), which should be called after del_gendisk(), if
600 * __device_add_disk() was used.
602 * Drivers exist which depend on the release of the gendisk to be synchronous,
603 * it should not be deferred.
607 void del_gendisk(struct gendisk *disk)
609 struct request_queue *q = disk->queue;
613 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
616 blk_integrity_del(disk);
617 disk_del_events(disk);
619 mutex_lock(&disk->open_mutex);
620 remove_inode_hash(disk->part0->bd_inode);
621 blk_drop_partitions(disk);
622 mutex_unlock(&disk->open_mutex);
624 fsync_bdev(disk->part0);
625 __invalidate_device(disk->part0, true);
630 set_bit(GD_DEAD, &disk->state);
631 if (test_bit(GD_OWNS_QUEUE, &disk->state))
632 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
633 set_capacity(disk, 0);
636 * Prevent new I/O from crossing bio_queue_enter().
638 blk_queue_start_drain(q);
640 if (!(disk->flags & GENHD_FL_HIDDEN)) {
641 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
644 * Unregister bdi before releasing device numbers (as they can
645 * get reused and we'd get clashes in sysfs).
647 bdi_unregister(disk->bdi);
650 blk_unregister_queue(disk);
652 kobject_put(disk->part0->bd_holder_dir);
653 kobject_put(disk->slave_dir);
654 disk->slave_dir = NULL;
656 part_stat_set_all(disk->part0, 0);
657 disk->part0->bd_stamp = 0;
658 if (!sysfs_deprecated)
659 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
660 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
661 device_del(disk_to_dev(disk));
663 blk_mq_freeze_queue_wait(q);
665 blk_throtl_cancel_bios(disk);
668 blk_flush_integrity();
671 blk_mq_cancel_work_sync(q);
673 blk_mq_quiesce_queue(q);
675 mutex_lock(&q->sysfs_lock);
677 mutex_unlock(&q->sysfs_lock);
680 blk_mq_unquiesce_queue(q);
683 * If the disk does not own the queue, allow using passthrough requests
684 * again. Else leave the queue frozen to fail all I/O.
686 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
687 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
688 __blk_mq_unfreeze_queue(q, true);
691 blk_mq_exit_queue(q);
694 EXPORT_SYMBOL(del_gendisk);
697 * invalidate_disk - invalidate the disk
698 * @disk: the struct gendisk to invalidate
700 * A helper to invalidates the disk. It will clean the disk's associated
701 * buffer/page caches and reset its internal states so that the disk
702 * can be reused by the drivers.
706 void invalidate_disk(struct gendisk *disk)
708 struct block_device *bdev = disk->part0;
710 invalidate_bdev(bdev);
711 bdev->bd_inode->i_mapping->wb_err = 0;
712 set_capacity(disk, 0);
714 EXPORT_SYMBOL(invalidate_disk);
716 /* sysfs access to bad-blocks list. */
717 static ssize_t disk_badblocks_show(struct device *dev,
718 struct device_attribute *attr,
721 struct gendisk *disk = dev_to_disk(dev);
724 return sprintf(page, "\n");
726 return badblocks_show(disk->bb, page, 0);
729 static ssize_t disk_badblocks_store(struct device *dev,
730 struct device_attribute *attr,
731 const char *page, size_t len)
733 struct gendisk *disk = dev_to_disk(dev);
738 return badblocks_store(disk->bb, page, len, 0);
741 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
742 void blk_request_module(dev_t devt)
744 unsigned int major = MAJOR(devt);
745 struct blk_major_name **n;
747 mutex_lock(&major_names_lock);
748 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
749 if ((*n)->major == major && (*n)->probe) {
751 mutex_unlock(&major_names_lock);
755 mutex_unlock(&major_names_lock);
757 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
758 /* Make old-style 2.4 aliases work */
759 request_module("block-major-%d", MAJOR(devt));
761 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
764 * print a full list of all partitions - intended for places where the root
765 * filesystem can't be mounted and thus to give the victim some idea of what
768 void __init printk_all_partitions(void)
770 struct class_dev_iter iter;
773 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
774 while ((dev = class_dev_iter_next(&iter))) {
775 struct gendisk *disk = dev_to_disk(dev);
776 struct block_device *part;
777 char devt_buf[BDEVT_SIZE];
781 * Don't show empty devices or things that have been
784 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
788 * Note, unlike /proc/partitions, I am showing the numbers in
789 * hex - the same format as the root= option takes.
792 xa_for_each(&disk->part_tbl, idx, part) {
793 if (!bdev_nr_sectors(part))
795 printk("%s%s %10llu %pg %s",
796 bdev_is_partition(part) ? " " : "",
797 bdevt_str(part->bd_dev, devt_buf),
798 bdev_nr_sectors(part) >> 1, part,
800 part->bd_meta_info->uuid : "");
801 if (bdev_is_partition(part))
803 else if (dev->parent && dev->parent->driver)
804 printk(" driver: %s\n",
805 dev->parent->driver->name);
807 printk(" (driver?)\n");
811 class_dev_iter_exit(&iter);
814 #ifdef CONFIG_PROC_FS
816 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
819 struct class_dev_iter *iter;
822 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
824 return ERR_PTR(-ENOMEM);
826 seqf->private = iter;
827 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
829 dev = class_dev_iter_next(iter);
834 return dev_to_disk(dev);
837 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
842 dev = class_dev_iter_next(seqf->private);
844 return dev_to_disk(dev);
849 static void disk_seqf_stop(struct seq_file *seqf, void *v)
851 struct class_dev_iter *iter = seqf->private;
853 /* stop is called even after start failed :-( */
855 class_dev_iter_exit(iter);
857 seqf->private = NULL;
861 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
865 p = disk_seqf_start(seqf, pos);
866 if (!IS_ERR_OR_NULL(p) && !*pos)
867 seq_puts(seqf, "major minor #blocks name\n\n");
871 static int show_partition(struct seq_file *seqf, void *v)
873 struct gendisk *sgp = v;
874 struct block_device *part;
877 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
881 xa_for_each(&sgp->part_tbl, idx, part) {
882 if (!bdev_nr_sectors(part))
884 seq_printf(seqf, "%4d %7d %10llu %pg\n",
885 MAJOR(part->bd_dev), MINOR(part->bd_dev),
886 bdev_nr_sectors(part) >> 1, part);
892 static const struct seq_operations partitions_op = {
893 .start = show_partition_start,
894 .next = disk_seqf_next,
895 .stop = disk_seqf_stop,
896 .show = show_partition
900 static int __init genhd_device_init(void)
904 block_class.dev_kobj = sysfs_dev_block_kobj;
905 error = class_register(&block_class);
910 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
912 /* create top-level block dir */
913 if (!sysfs_deprecated)
914 block_depr = kobject_create_and_add("block", NULL);
918 subsys_initcall(genhd_device_init);
920 static ssize_t disk_range_show(struct device *dev,
921 struct device_attribute *attr, char *buf)
923 struct gendisk *disk = dev_to_disk(dev);
925 return sprintf(buf, "%d\n", disk->minors);
928 static ssize_t disk_ext_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",
934 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
937 static ssize_t disk_removable_show(struct device *dev,
938 struct device_attribute *attr, char *buf)
940 struct gendisk *disk = dev_to_disk(dev);
942 return sprintf(buf, "%d\n",
943 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
946 static ssize_t disk_hidden_show(struct device *dev,
947 struct device_attribute *attr, char *buf)
949 struct gendisk *disk = dev_to_disk(dev);
951 return sprintf(buf, "%d\n",
952 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
955 static ssize_t disk_ro_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", get_disk_ro(disk) ? 1 : 0);
963 ssize_t part_size_show(struct device *dev,
964 struct device_attribute *attr, char *buf)
966 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
969 ssize_t part_stat_show(struct device *dev,
970 struct device_attribute *attr, char *buf)
972 struct block_device *bdev = dev_to_bdev(dev);
973 struct request_queue *q = bdev_get_queue(bdev);
974 struct disk_stats stat;
975 unsigned int inflight;
978 inflight = blk_mq_in_flight(q, bdev);
980 inflight = part_in_flight(bdev);
984 update_io_ticks(bdev, jiffies, true);
987 part_stat_read_all(bdev, &stat);
989 "%8lu %8lu %8llu %8u "
990 "%8lu %8lu %8llu %8u "
992 "%8lu %8lu %8llu %8u "
996 stat.merges[STAT_READ],
997 (unsigned long long)stat.sectors[STAT_READ],
998 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
999 stat.ios[STAT_WRITE],
1000 stat.merges[STAT_WRITE],
1001 (unsigned long long)stat.sectors[STAT_WRITE],
1002 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1004 jiffies_to_msecs(stat.io_ticks),
1005 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1006 stat.nsecs[STAT_WRITE] +
1007 stat.nsecs[STAT_DISCARD] +
1008 stat.nsecs[STAT_FLUSH],
1010 stat.ios[STAT_DISCARD],
1011 stat.merges[STAT_DISCARD],
1012 (unsigned long long)stat.sectors[STAT_DISCARD],
1013 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1014 stat.ios[STAT_FLUSH],
1015 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1018 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1021 struct block_device *bdev = dev_to_bdev(dev);
1022 struct request_queue *q = bdev_get_queue(bdev);
1023 unsigned int inflight[2];
1026 blk_mq_in_flight_rw(q, bdev, inflight);
1028 part_in_flight_rw(bdev, inflight);
1030 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1033 static ssize_t disk_capability_show(struct device *dev,
1034 struct device_attribute *attr, char *buf)
1036 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1037 return sprintf(buf, "0\n");
1040 static ssize_t disk_alignment_offset_show(struct device *dev,
1041 struct device_attribute *attr,
1044 struct gendisk *disk = dev_to_disk(dev);
1046 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1049 static ssize_t disk_discard_alignment_show(struct device *dev,
1050 struct device_attribute *attr,
1053 struct gendisk *disk = dev_to_disk(dev);
1055 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1058 static ssize_t diskseq_show(struct device *dev,
1059 struct device_attribute *attr, char *buf)
1061 struct gendisk *disk = dev_to_disk(dev);
1063 return sprintf(buf, "%llu\n", disk->diskseq);
1066 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1067 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1068 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1069 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1070 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1071 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1072 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1073 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1074 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1075 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1076 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1077 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1078 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1080 #ifdef CONFIG_FAIL_MAKE_REQUEST
1081 ssize_t part_fail_show(struct device *dev,
1082 struct device_attribute *attr, char *buf)
1084 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1087 ssize_t part_fail_store(struct device *dev,
1088 struct device_attribute *attr,
1089 const char *buf, size_t count)
1093 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1094 dev_to_bdev(dev)->bd_make_it_fail = i;
1099 static struct device_attribute dev_attr_fail =
1100 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1101 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1103 #ifdef CONFIG_FAIL_IO_TIMEOUT
1104 static struct device_attribute dev_attr_fail_timeout =
1105 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1108 static struct attribute *disk_attrs[] = {
1109 &dev_attr_range.attr,
1110 &dev_attr_ext_range.attr,
1111 &dev_attr_removable.attr,
1112 &dev_attr_hidden.attr,
1114 &dev_attr_size.attr,
1115 &dev_attr_alignment_offset.attr,
1116 &dev_attr_discard_alignment.attr,
1117 &dev_attr_capability.attr,
1118 &dev_attr_stat.attr,
1119 &dev_attr_inflight.attr,
1120 &dev_attr_badblocks.attr,
1121 &dev_attr_events.attr,
1122 &dev_attr_events_async.attr,
1123 &dev_attr_events_poll_msecs.attr,
1124 &dev_attr_diskseq.attr,
1125 #ifdef CONFIG_FAIL_MAKE_REQUEST
1126 &dev_attr_fail.attr,
1128 #ifdef CONFIG_FAIL_IO_TIMEOUT
1129 &dev_attr_fail_timeout.attr,
1134 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1136 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1137 struct gendisk *disk = dev_to_disk(dev);
1139 if (a == &dev_attr_badblocks.attr && !disk->bb)
1144 static struct attribute_group disk_attr_group = {
1145 .attrs = disk_attrs,
1146 .is_visible = disk_visible,
1149 static const struct attribute_group *disk_attr_groups[] = {
1151 #ifdef CONFIG_BLK_DEV_IO_TRACE
1152 &blk_trace_attr_group,
1158 * disk_release - releases all allocated resources of the gendisk
1159 * @dev: the device representing this disk
1161 * This function releases all allocated resources of the gendisk.
1163 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1164 * assigned. Since the request_queue sits on top of the gendisk for these
1165 * drivers we also call blk_put_queue() for them, and we expect the
1166 * request_queue refcount to reach 0 at this point, and so the request_queue
1167 * will also be freed prior to the disk.
1169 * Context: can sleep
1171 static void disk_release(struct device *dev)
1173 struct gendisk *disk = dev_to_disk(dev);
1176 WARN_ON_ONCE(disk_live(disk));
1179 * To undo the all initialization from blk_mq_init_allocated_queue in
1180 * case of a probe failure where add_disk is never called we have to
1181 * call blk_mq_exit_queue here. We can't do this for the more common
1182 * teardown case (yet) as the tagset can be gone by the time the disk
1183 * is released once it was added.
1185 if (queue_is_mq(disk->queue) &&
1186 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1187 !test_bit(GD_ADDED, &disk->state))
1188 blk_mq_exit_queue(disk->queue);
1190 blkcg_exit_disk(disk);
1192 bioset_exit(&disk->bio_split);
1194 disk_release_events(disk);
1195 kfree(disk->random);
1196 disk_free_zone_bitmaps(disk);
1197 xa_destroy(&disk->part_tbl);
1199 disk->queue->disk = NULL;
1200 blk_put_queue(disk->queue);
1202 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1203 disk->fops->free_disk(disk);
1205 iput(disk->part0->bd_inode); /* frees the disk */
1208 static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1210 const struct gendisk *disk = dev_to_disk(dev);
1212 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1215 struct class block_class = {
1217 .dev_uevent = block_uevent,
1220 static char *block_devnode(const struct device *dev, umode_t *mode,
1221 kuid_t *uid, kgid_t *gid)
1223 struct gendisk *disk = dev_to_disk(dev);
1225 if (disk->fops->devnode)
1226 return disk->fops->devnode(disk, mode);
1230 const struct device_type disk_type = {
1232 .groups = disk_attr_groups,
1233 .release = disk_release,
1234 .devnode = block_devnode,
1237 #ifdef CONFIG_PROC_FS
1239 * aggregate disk stat collector. Uses the same stats that the sysfs
1240 * entries do, above, but makes them available through one seq_file.
1242 * The output looks suspiciously like /proc/partitions with a bunch of
1245 static int diskstats_show(struct seq_file *seqf, void *v)
1247 struct gendisk *gp = v;
1248 struct block_device *hd;
1249 unsigned int inflight;
1250 struct disk_stats stat;
1254 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1255 seq_puts(seqf, "major minor name"
1256 " rio rmerge rsect ruse wio wmerge "
1257 "wsect wuse running use aveq"
1262 xa_for_each(&gp->part_tbl, idx, hd) {
1263 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1265 if (queue_is_mq(gp->queue))
1266 inflight = blk_mq_in_flight(gp->queue, hd);
1268 inflight = part_in_flight(hd);
1272 update_io_ticks(hd, jiffies, true);
1275 part_stat_read_all(hd, &stat);
1276 seq_printf(seqf, "%4d %7d %pg "
1283 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1284 stat.ios[STAT_READ],
1285 stat.merges[STAT_READ],
1286 stat.sectors[STAT_READ],
1287 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1289 stat.ios[STAT_WRITE],
1290 stat.merges[STAT_WRITE],
1291 stat.sectors[STAT_WRITE],
1292 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1295 jiffies_to_msecs(stat.io_ticks),
1296 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1297 stat.nsecs[STAT_WRITE] +
1298 stat.nsecs[STAT_DISCARD] +
1299 stat.nsecs[STAT_FLUSH],
1301 stat.ios[STAT_DISCARD],
1302 stat.merges[STAT_DISCARD],
1303 stat.sectors[STAT_DISCARD],
1304 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1306 stat.ios[STAT_FLUSH],
1307 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1316 static const struct seq_operations diskstats_op = {
1317 .start = disk_seqf_start,
1318 .next = disk_seqf_next,
1319 .stop = disk_seqf_stop,
1320 .show = diskstats_show
1323 static int __init proc_genhd_init(void)
1325 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1326 proc_create_seq("partitions", 0, NULL, &partitions_op);
1329 module_init(proc_genhd_init);
1330 #endif /* CONFIG_PROC_FS */
1332 dev_t part_devt(struct gendisk *disk, u8 partno)
1334 struct block_device *part;
1338 part = xa_load(&disk->part_tbl, partno);
1340 devt = part->bd_dev;
1346 dev_t blk_lookup_devt(const char *name, int partno)
1348 dev_t devt = MKDEV(0, 0);
1349 struct class_dev_iter iter;
1352 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1353 while ((dev = class_dev_iter_next(&iter))) {
1354 struct gendisk *disk = dev_to_disk(dev);
1356 if (strcmp(dev_name(dev), name))
1359 if (partno < disk->minors) {
1360 /* We need to return the right devno, even
1361 * if the partition doesn't exist yet.
1363 devt = MKDEV(MAJOR(dev->devt),
1364 MINOR(dev->devt) + partno);
1366 devt = part_devt(disk, partno);
1371 class_dev_iter_exit(&iter);
1375 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1376 struct lock_class_key *lkclass)
1378 struct gendisk *disk;
1380 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1384 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1387 disk->bdi = bdi_alloc(node_id);
1389 goto out_free_bioset;
1391 /* bdev_alloc() might need the queue, set before the first call */
1394 disk->part0 = bdev_alloc(disk, 0);
1398 disk->node_id = node_id;
1399 mutex_init(&disk->open_mutex);
1400 xa_init(&disk->part_tbl);
1401 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1402 goto out_destroy_part_tbl;
1404 if (blkcg_init_disk(disk))
1405 goto out_erase_part0;
1407 rand_initialize_disk(disk);
1408 disk_to_dev(disk)->class = &block_class;
1409 disk_to_dev(disk)->type = &disk_type;
1410 device_initialize(disk_to_dev(disk));
1413 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1414 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1415 INIT_LIST_HEAD(&disk->slave_bdevs);
1420 xa_erase(&disk->part_tbl, 0);
1421 out_destroy_part_tbl:
1422 xa_destroy(&disk->part_tbl);
1423 disk->part0->bd_disk = NULL;
1424 iput(disk->part0->bd_inode);
1428 bioset_exit(&disk->bio_split);
1434 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1436 struct request_queue *q;
1437 struct gendisk *disk;
1439 q = blk_alloc_queue(node);
1443 disk = __alloc_disk_node(q, node, lkclass);
1448 set_bit(GD_OWNS_QUEUE, &disk->state);
1451 EXPORT_SYMBOL(__blk_alloc_disk);
1454 * put_disk - decrements the gendisk refcount
1455 * @disk: the struct gendisk to decrement the refcount for
1457 * This decrements the refcount for the struct gendisk. When this reaches 0
1458 * we'll have disk_release() called.
1460 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1461 * when handling probe errors (that is before add_disk() is called).
1463 * Context: Any context, but the last reference must not be dropped from
1466 void put_disk(struct gendisk *disk)
1469 put_device(disk_to_dev(disk));
1471 EXPORT_SYMBOL(put_disk);
1473 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1475 char event[] = "DISK_RO=1";
1476 char *envp[] = { event, NULL };
1480 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1484 * set_disk_ro - set a gendisk read-only
1485 * @disk: gendisk to operate on
1486 * @read_only: %true to set the disk read-only, %false set the disk read/write
1488 * This function is used to indicate whether a given disk device should have its
1489 * read-only flag set. set_disk_ro() is typically used by device drivers to
1490 * indicate whether the underlying physical device is write-protected.
1492 void set_disk_ro(struct gendisk *disk, bool read_only)
1495 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1498 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1501 set_disk_ro_uevent(disk, read_only);
1503 EXPORT_SYMBOL(set_disk_ro);
1505 void inc_diskseq(struct gendisk *disk)
1507 disk->diskseq = atomic64_inc_return(&diskseq);