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/genhd.h>
12 #include <linux/kdev_t.h>
13 #include <linux/kernel.h>
14 #include <linux/blkdev.h>
15 #include <linux/backing-dev.h>
16 #include <linux/init.h>
17 #include <linux/spinlock.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/slab.h>
21 #include <linux/kmod.h>
22 #include <linux/major.h>
23 #include <linux/mutex.h>
24 #include <linux/idr.h>
25 #include <linux/log2.h>
26 #include <linux/pm_runtime.h>
27 #include <linux/badblocks.h>
28 #include <linux/part_stat.h>
31 #include "blk-mq-sched.h"
32 #include "blk-rq-qos.h"
34 static struct kobject *block_depr;
37 * Unique, monotonically increasing sequential number associated with block
38 * devices instances (i.e. incremented each time a device is attached).
39 * Associating uevents with block devices in userspace is difficult and racy:
40 * the uevent netlink socket is lossy, and on slow and overloaded systems has
41 * a very high latency.
42 * Block devices do not have exclusive owners in userspace, any process can set
43 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
44 * can be reused again and again).
45 * A userspace process setting up a block device and watching for its events
46 * cannot thus reliably tell whether an event relates to the device it just set
47 * up or another earlier instance with the same name.
48 * This sequential number allows userspace processes to solve this problem, and
49 * uniquely associate an uevent to the lifetime to a device.
51 static atomic64_t diskseq;
53 /* for extended dynamic devt allocation, currently only one major is used */
54 #define NR_EXT_DEVT (1 << MINORBITS)
55 static DEFINE_IDA(ext_devt_ida);
57 void set_capacity(struct gendisk *disk, sector_t sectors)
59 struct block_device *bdev = disk->part0;
61 spin_lock(&bdev->bd_size_lock);
62 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
63 bdev->bd_nr_sectors = sectors;
64 spin_unlock(&bdev->bd_size_lock);
66 EXPORT_SYMBOL(set_capacity);
69 * Set disk capacity and notify if the size is not currently zero and will not
70 * be set to zero. Returns true if a uevent was sent, otherwise false.
72 bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
74 sector_t capacity = get_capacity(disk);
75 char *envp[] = { "RESIZE=1", NULL };
77 set_capacity(disk, size);
80 * Only print a message and send a uevent if the gendisk is user visible
81 * and alive. This avoids spamming the log and udev when setting the
82 * initial capacity during probing.
84 if (size == capacity ||
86 (disk->flags & GENHD_FL_HIDDEN))
89 pr_info("%s: detected capacity change from %lld to %lld\n",
90 disk->disk_name, capacity, size);
93 * Historically we did not send a uevent for changes to/from an empty
96 if (!capacity || !size)
98 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
101 EXPORT_SYMBOL_GPL(set_capacity_and_notify);
104 * Format the device name of the indicated block device into the supplied buffer
105 * and return a pointer to that same buffer for convenience.
107 * Note: do not use this in new code, use the %pg specifier to sprintf and
110 const char *bdevname(struct block_device *bdev, char *buf)
112 struct gendisk *hd = bdev->bd_disk;
113 int partno = bdev->bd_partno;
116 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
117 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
118 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
120 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
124 EXPORT_SYMBOL(bdevname);
126 static void part_stat_read_all(struct block_device *part,
127 struct disk_stats *stat)
131 memset(stat, 0, sizeof(struct disk_stats));
132 for_each_possible_cpu(cpu) {
133 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
136 for (group = 0; group < NR_STAT_GROUPS; group++) {
137 stat->nsecs[group] += ptr->nsecs[group];
138 stat->sectors[group] += ptr->sectors[group];
139 stat->ios[group] += ptr->ios[group];
140 stat->merges[group] += ptr->merges[group];
143 stat->io_ticks += ptr->io_ticks;
147 static unsigned int part_in_flight(struct block_device *part)
149 unsigned int inflight = 0;
152 for_each_possible_cpu(cpu) {
153 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
154 part_stat_local_read_cpu(part, in_flight[1], cpu);
156 if ((int)inflight < 0)
162 static void part_in_flight_rw(struct block_device *part,
163 unsigned int inflight[2])
169 for_each_possible_cpu(cpu) {
170 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
171 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
173 if ((int)inflight[0] < 0)
175 if ((int)inflight[1] < 0)
180 * Can be deleted altogether. Later.
183 #define BLKDEV_MAJOR_HASH_SIZE 255
184 static struct blk_major_name {
185 struct blk_major_name *next;
188 void (*probe)(dev_t devt);
189 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
190 static DEFINE_MUTEX(major_names_lock);
191 static DEFINE_SPINLOCK(major_names_spinlock);
193 /* index in the above - for now: assume no multimajor ranges */
194 static inline int major_to_index(unsigned major)
196 return major % BLKDEV_MAJOR_HASH_SIZE;
199 #ifdef CONFIG_PROC_FS
200 void blkdev_show(struct seq_file *seqf, off_t offset)
202 struct blk_major_name *dp;
204 spin_lock(&major_names_spinlock);
205 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
206 if (dp->major == offset)
207 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
208 spin_unlock(&major_names_spinlock);
210 #endif /* CONFIG_PROC_FS */
213 * __register_blkdev - register a new block device
215 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
216 * @major = 0, try to allocate any unused major number.
217 * @name: the name of the new block device as a zero terminated string
218 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
219 * pre-created device node is accessed. When a probe call uses
220 * add_disk() and it fails the driver must cleanup resources. This
221 * interface may soon be removed.
223 * The @name must be unique within the system.
225 * The return value depends on the @major input parameter:
227 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
228 * then the function returns zero on success, or a negative error code
229 * - if any unused major number was requested with @major = 0 parameter
230 * then the return value is the allocated major number in range
231 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
233 * See Documentation/admin-guide/devices.txt for the list of allocated
236 * Use register_blkdev instead for any new code.
238 int __register_blkdev(unsigned int major, const char *name,
239 void (*probe)(dev_t devt))
241 struct blk_major_name **n, *p;
244 mutex_lock(&major_names_lock);
248 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
249 if (major_names[index] == NULL)
254 printk("%s: failed to get major for %s\n",
263 if (major >= BLKDEV_MAJOR_MAX) {
264 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
265 __func__, major, BLKDEV_MAJOR_MAX-1, name);
271 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
279 strlcpy(p->name, name, sizeof(p->name));
281 index = major_to_index(major);
283 spin_lock(&major_names_spinlock);
284 for (n = &major_names[index]; *n; n = &(*n)->next) {
285 if ((*n)->major == major)
292 spin_unlock(&major_names_spinlock);
295 printk("register_blkdev: cannot get major %u for %s\n",
300 mutex_unlock(&major_names_lock);
303 EXPORT_SYMBOL(__register_blkdev);
305 void unregister_blkdev(unsigned int major, const char *name)
307 struct blk_major_name **n;
308 struct blk_major_name *p = NULL;
309 int index = major_to_index(major);
311 mutex_lock(&major_names_lock);
312 spin_lock(&major_names_spinlock);
313 for (n = &major_names[index]; *n; n = &(*n)->next)
314 if ((*n)->major == major)
316 if (!*n || strcmp((*n)->name, name)) {
322 spin_unlock(&major_names_spinlock);
323 mutex_unlock(&major_names_lock);
327 EXPORT_SYMBOL(unregister_blkdev);
329 int blk_alloc_ext_minor(void)
333 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT, GFP_KERNEL);
339 void blk_free_ext_minor(unsigned int minor)
341 ida_free(&ext_devt_ida, minor);
344 static char *bdevt_str(dev_t devt, char *buf)
346 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
347 char tbuf[BDEVT_SIZE];
348 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
349 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
351 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
356 void disk_uevent(struct gendisk *disk, enum kobject_action action)
358 struct block_device *part;
362 xa_for_each(&disk->part_tbl, idx, part) {
363 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
365 if (!kobject_get_unless_zero(&part->bd_device.kobj))
369 kobject_uevent(bdev_kobj(part), action);
370 put_device(&part->bd_device);
375 EXPORT_SYMBOL_GPL(disk_uevent);
377 int disk_scan_partitions(struct gendisk *disk, fmode_t mode)
379 struct block_device *bdev;
381 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
383 if (disk->open_partitions)
386 set_bit(GD_NEED_PART_SCAN, &disk->state);
387 bdev = blkdev_get_by_dev(disk_devt(disk), mode, NULL);
389 return PTR_ERR(bdev);
390 blkdev_put(bdev, mode);
395 * device_add_disk - add disk information to kernel list
396 * @parent: parent device for the disk
397 * @disk: per-device partitioning information
398 * @groups: Additional per-device sysfs groups
400 * This function registers the partitioning information in @disk
403 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
404 const struct attribute_group **groups)
407 struct device *ddev = disk_to_dev(disk);
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);
419 * If the driver provides an explicit major number it also must provide
420 * the number of minors numbers supported, and those will be used to
422 * Otherwise just allocate the device numbers for both the whole device
423 * and all partitions from the extended dev_t space.
426 if (WARN_ON(!disk->minors))
429 if (disk->minors > DISK_MAX_PARTS) {
430 pr_err("block: can't allocate more than %d partitions\n",
432 disk->minors = DISK_MAX_PARTS;
434 if (disk->first_minor + disk->minors > MINORMASK + 1)
437 if (WARN_ON(disk->minors))
440 ret = blk_alloc_ext_minor();
443 disk->major = BLOCK_EXT_MAJOR;
444 disk->first_minor = ret;
447 /* delay uevents, until we scanned partition table */
448 dev_set_uevent_suppress(ddev, 1);
450 ddev->parent = parent;
451 ddev->groups = groups;
452 dev_set_name(ddev, "%s", disk->disk_name);
453 if (!(disk->flags & GENHD_FL_HIDDEN))
454 ddev->devt = MKDEV(disk->major, disk->first_minor);
455 ret = device_add(ddev);
457 goto out_free_ext_minor;
459 ret = disk_alloc_events(disk);
463 if (!sysfs_deprecated) {
464 ret = sysfs_create_link(block_depr, &ddev->kobj,
465 kobject_name(&ddev->kobj));
471 * avoid probable deadlock caused by allocating memory with
472 * GFP_KERNEL in runtime_resume callback of its all ancestor
475 pm_runtime_set_memalloc_noio(ddev, true);
477 ret = blk_integrity_add(disk);
479 goto out_del_block_link;
481 disk->part0->bd_holder_dir =
482 kobject_create_and_add("holders", &ddev->kobj);
483 if (!disk->part0->bd_holder_dir) {
485 goto out_del_integrity;
487 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
488 if (!disk->slave_dir) {
490 goto out_put_holder_dir;
493 ret = bd_register_pending_holders(disk);
495 goto out_put_slave_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 bdev_add(disk->part0, ddev->devt);
513 if (get_capacity(disk))
514 disk_scan_partitions(disk, FMODE_READ);
517 * Announce the disk and partitions after all partitions are
518 * created. (for hidden disks uevents remain suppressed forever)
520 dev_set_uevent_suppress(ddev, 0);
521 disk_uevent(disk, KOBJ_ADD);
524 disk_update_readahead(disk);
525 disk_add_events(disk);
529 if (!(disk->flags & GENHD_FL_HIDDEN))
530 bdi_unregister(disk->bdi);
531 out_unregister_queue:
532 blk_unregister_queue(disk);
534 kobject_put(disk->slave_dir);
536 kobject_put(disk->part0->bd_holder_dir);
538 blk_integrity_del(disk);
540 if (!sysfs_deprecated)
541 sysfs_remove_link(block_depr, dev_name(ddev));
545 if (disk->major == BLOCK_EXT_MAJOR)
546 blk_free_ext_minor(disk->first_minor);
549 EXPORT_SYMBOL(device_add_disk);
552 * blk_mark_disk_dead - mark a disk as dead
553 * @disk: disk to mark as dead
555 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
558 void blk_mark_disk_dead(struct gendisk *disk)
560 set_bit(GD_DEAD, &disk->state);
561 blk_queue_start_drain(disk->queue);
563 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
566 * del_gendisk - remove the gendisk
567 * @disk: the struct gendisk to remove
569 * Removes the gendisk and all its associated resources. This deletes the
570 * partitions associated with the gendisk, and unregisters the associated
573 * This is the counter to the respective __device_add_disk() call.
575 * The final removal of the struct gendisk happens when its refcount reaches 0
576 * with put_disk(), which should be called after del_gendisk(), if
577 * __device_add_disk() was used.
579 * Drivers exist which depend on the release of the gendisk to be synchronous,
580 * it should not be deferred.
584 void del_gendisk(struct gendisk *disk)
586 struct request_queue *q = disk->queue;
590 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
593 blk_integrity_del(disk);
594 disk_del_events(disk);
596 mutex_lock(&disk->open_mutex);
597 remove_inode_hash(disk->part0->bd_inode);
598 blk_drop_partitions(disk);
599 mutex_unlock(&disk->open_mutex);
601 fsync_bdev(disk->part0);
602 __invalidate_device(disk->part0, true);
607 set_bit(GD_DEAD, &disk->state);
608 set_capacity(disk, 0);
611 * Prevent new I/O from crossing bio_queue_enter().
613 blk_queue_start_drain(q);
615 if (!(disk->flags & GENHD_FL_HIDDEN)) {
616 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
619 * Unregister bdi before releasing device numbers (as they can
620 * get reused and we'd get clashes in sysfs).
622 bdi_unregister(disk->bdi);
625 blk_unregister_queue(disk);
627 kobject_put(disk->part0->bd_holder_dir);
628 kobject_put(disk->slave_dir);
630 part_stat_set_all(disk->part0, 0);
631 disk->part0->bd_stamp = 0;
632 if (!sysfs_deprecated)
633 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
634 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
635 device_del(disk_to_dev(disk));
637 blk_mq_freeze_queue_wait(q);
641 blk_flush_integrity();
643 * Allow using passthrough request again after the queue is torn down.
645 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
646 __blk_mq_unfreeze_queue(q, true);
649 EXPORT_SYMBOL(del_gendisk);
652 * invalidate_disk - invalidate the disk
653 * @disk: the struct gendisk to invalidate
655 * A helper to invalidates the disk. It will clean the disk's associated
656 * buffer/page caches and reset its internal states so that the disk
657 * can be reused by the drivers.
661 void invalidate_disk(struct gendisk *disk)
663 struct block_device *bdev = disk->part0;
665 invalidate_bdev(bdev);
666 bdev->bd_inode->i_mapping->wb_err = 0;
667 set_capacity(disk, 0);
669 EXPORT_SYMBOL(invalidate_disk);
671 /* sysfs access to bad-blocks list. */
672 static ssize_t disk_badblocks_show(struct device *dev,
673 struct device_attribute *attr,
676 struct gendisk *disk = dev_to_disk(dev);
679 return sprintf(page, "\n");
681 return badblocks_show(disk->bb, page, 0);
684 static ssize_t disk_badblocks_store(struct device *dev,
685 struct device_attribute *attr,
686 const char *page, size_t len)
688 struct gendisk *disk = dev_to_disk(dev);
693 return badblocks_store(disk->bb, page, len, 0);
696 void blk_request_module(dev_t devt)
698 unsigned int major = MAJOR(devt);
699 struct blk_major_name **n;
701 mutex_lock(&major_names_lock);
702 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
703 if ((*n)->major == major && (*n)->probe) {
705 mutex_unlock(&major_names_lock);
709 mutex_unlock(&major_names_lock);
711 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
712 /* Make old-style 2.4 aliases work */
713 request_module("block-major-%d", MAJOR(devt));
717 * print a full list of all partitions - intended for places where the root
718 * filesystem can't be mounted and thus to give the victim some idea of what
721 void __init printk_all_partitions(void)
723 struct class_dev_iter iter;
726 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
727 while ((dev = class_dev_iter_next(&iter))) {
728 struct gendisk *disk = dev_to_disk(dev);
729 struct block_device *part;
730 char devt_buf[BDEVT_SIZE];
734 * Don't show empty devices or things that have been
737 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
741 * Note, unlike /proc/partitions, I am showing the numbers in
742 * hex - the same format as the root= option takes.
745 xa_for_each(&disk->part_tbl, idx, part) {
746 if (!bdev_nr_sectors(part))
748 printk("%s%s %10llu %pg %s",
749 bdev_is_partition(part) ? " " : "",
750 bdevt_str(part->bd_dev, devt_buf),
751 bdev_nr_sectors(part) >> 1, part,
753 part->bd_meta_info->uuid : "");
754 if (bdev_is_partition(part))
756 else if (dev->parent && dev->parent->driver)
757 printk(" driver: %s\n",
758 dev->parent->driver->name);
760 printk(" (driver?)\n");
764 class_dev_iter_exit(&iter);
767 #ifdef CONFIG_PROC_FS
769 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
772 struct class_dev_iter *iter;
775 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
777 return ERR_PTR(-ENOMEM);
779 seqf->private = iter;
780 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
782 dev = class_dev_iter_next(iter);
787 return dev_to_disk(dev);
790 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
795 dev = class_dev_iter_next(seqf->private);
797 return dev_to_disk(dev);
802 static void disk_seqf_stop(struct seq_file *seqf, void *v)
804 struct class_dev_iter *iter = seqf->private;
806 /* stop is called even after start failed :-( */
808 class_dev_iter_exit(iter);
810 seqf->private = NULL;
814 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
818 p = disk_seqf_start(seqf, pos);
819 if (!IS_ERR_OR_NULL(p) && !*pos)
820 seq_puts(seqf, "major minor #blocks name\n\n");
824 static int show_partition(struct seq_file *seqf, void *v)
826 struct gendisk *sgp = v;
827 struct block_device *part;
830 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
834 xa_for_each(&sgp->part_tbl, idx, part) {
835 if (!bdev_nr_sectors(part))
837 seq_printf(seqf, "%4d %7d %10llu %pg\n",
838 MAJOR(part->bd_dev), MINOR(part->bd_dev),
839 bdev_nr_sectors(part) >> 1, part);
845 static const struct seq_operations partitions_op = {
846 .start = show_partition_start,
847 .next = disk_seqf_next,
848 .stop = disk_seqf_stop,
849 .show = show_partition
853 static int __init genhd_device_init(void)
857 block_class.dev_kobj = sysfs_dev_block_kobj;
858 error = class_register(&block_class);
863 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
865 /* create top-level block dir */
866 if (!sysfs_deprecated)
867 block_depr = kobject_create_and_add("block", NULL);
871 subsys_initcall(genhd_device_init);
873 static ssize_t disk_range_show(struct device *dev,
874 struct device_attribute *attr, char *buf)
876 struct gendisk *disk = dev_to_disk(dev);
878 return sprintf(buf, "%d\n", disk->minors);
881 static ssize_t disk_ext_range_show(struct device *dev,
882 struct device_attribute *attr, char *buf)
884 struct gendisk *disk = dev_to_disk(dev);
886 return sprintf(buf, "%d\n",
887 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
890 static ssize_t disk_removable_show(struct device *dev,
891 struct device_attribute *attr, char *buf)
893 struct gendisk *disk = dev_to_disk(dev);
895 return sprintf(buf, "%d\n",
896 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
899 static ssize_t disk_hidden_show(struct device *dev,
900 struct device_attribute *attr, char *buf)
902 struct gendisk *disk = dev_to_disk(dev);
904 return sprintf(buf, "%d\n",
905 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
908 static ssize_t disk_ro_show(struct device *dev,
909 struct device_attribute *attr, char *buf)
911 struct gendisk *disk = dev_to_disk(dev);
913 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
916 ssize_t part_size_show(struct device *dev,
917 struct device_attribute *attr, char *buf)
919 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
922 ssize_t part_stat_show(struct device *dev,
923 struct device_attribute *attr, char *buf)
925 struct block_device *bdev = dev_to_bdev(dev);
926 struct request_queue *q = bdev_get_queue(bdev);
927 struct disk_stats stat;
928 unsigned int inflight;
930 part_stat_read_all(bdev, &stat);
932 inflight = blk_mq_in_flight(q, bdev);
934 inflight = part_in_flight(bdev);
937 "%8lu %8lu %8llu %8u "
938 "%8lu %8lu %8llu %8u "
940 "%8lu %8lu %8llu %8u "
944 stat.merges[STAT_READ],
945 (unsigned long long)stat.sectors[STAT_READ],
946 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
947 stat.ios[STAT_WRITE],
948 stat.merges[STAT_WRITE],
949 (unsigned long long)stat.sectors[STAT_WRITE],
950 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
952 jiffies_to_msecs(stat.io_ticks),
953 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
954 stat.nsecs[STAT_WRITE] +
955 stat.nsecs[STAT_DISCARD] +
956 stat.nsecs[STAT_FLUSH],
958 stat.ios[STAT_DISCARD],
959 stat.merges[STAT_DISCARD],
960 (unsigned long long)stat.sectors[STAT_DISCARD],
961 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
962 stat.ios[STAT_FLUSH],
963 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
966 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
969 struct block_device *bdev = dev_to_bdev(dev);
970 struct request_queue *q = bdev_get_queue(bdev);
971 unsigned int inflight[2];
974 blk_mq_in_flight_rw(q, bdev, inflight);
976 part_in_flight_rw(bdev, inflight);
978 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
981 static ssize_t disk_capability_show(struct device *dev,
982 struct device_attribute *attr, char *buf)
984 struct gendisk *disk = dev_to_disk(dev);
986 return sprintf(buf, "%x\n", disk->flags);
989 static ssize_t disk_alignment_offset_show(struct device *dev,
990 struct device_attribute *attr,
993 struct gendisk *disk = dev_to_disk(dev);
995 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
998 static ssize_t disk_discard_alignment_show(struct device *dev,
999 struct device_attribute *attr,
1002 struct gendisk *disk = dev_to_disk(dev);
1004 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1007 static ssize_t diskseq_show(struct device *dev,
1008 struct device_attribute *attr, char *buf)
1010 struct gendisk *disk = dev_to_disk(dev);
1012 return sprintf(buf, "%llu\n", disk->diskseq);
1015 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1016 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1017 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1018 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1019 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1020 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1021 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1022 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1023 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1024 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1025 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1026 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1027 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1029 #ifdef CONFIG_FAIL_MAKE_REQUEST
1030 ssize_t part_fail_show(struct device *dev,
1031 struct device_attribute *attr, char *buf)
1033 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1036 ssize_t part_fail_store(struct device *dev,
1037 struct device_attribute *attr,
1038 const char *buf, size_t count)
1042 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1043 dev_to_bdev(dev)->bd_make_it_fail = i;
1048 static struct device_attribute dev_attr_fail =
1049 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1050 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1052 #ifdef CONFIG_FAIL_IO_TIMEOUT
1053 static struct device_attribute dev_attr_fail_timeout =
1054 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1057 static struct attribute *disk_attrs[] = {
1058 &dev_attr_range.attr,
1059 &dev_attr_ext_range.attr,
1060 &dev_attr_removable.attr,
1061 &dev_attr_hidden.attr,
1063 &dev_attr_size.attr,
1064 &dev_attr_alignment_offset.attr,
1065 &dev_attr_discard_alignment.attr,
1066 &dev_attr_capability.attr,
1067 &dev_attr_stat.attr,
1068 &dev_attr_inflight.attr,
1069 &dev_attr_badblocks.attr,
1070 &dev_attr_events.attr,
1071 &dev_attr_events_async.attr,
1072 &dev_attr_events_poll_msecs.attr,
1073 &dev_attr_diskseq.attr,
1074 #ifdef CONFIG_FAIL_MAKE_REQUEST
1075 &dev_attr_fail.attr,
1077 #ifdef CONFIG_FAIL_IO_TIMEOUT
1078 &dev_attr_fail_timeout.attr,
1083 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1085 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1086 struct gendisk *disk = dev_to_disk(dev);
1088 if (a == &dev_attr_badblocks.attr && !disk->bb)
1093 static struct attribute_group disk_attr_group = {
1094 .attrs = disk_attrs,
1095 .is_visible = disk_visible,
1098 static const struct attribute_group *disk_attr_groups[] = {
1104 * disk_release - releases all allocated resources of the gendisk
1105 * @dev: the device representing this disk
1107 * This function releases all allocated resources of the gendisk.
1109 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1110 * assigned. Since the request_queue sits on top of the gendisk for these
1111 * drivers we also call blk_put_queue() for them, and we expect the
1112 * request_queue refcount to reach 0 at this point, and so the request_queue
1113 * will also be freed prior to the disk.
1115 * Context: can sleep
1117 static void disk_release(struct device *dev)
1119 struct gendisk *disk = dev_to_disk(dev);
1122 WARN_ON_ONCE(disk_live(disk));
1124 blk_mq_cancel_work_sync(disk->queue);
1126 disk_release_events(disk);
1127 kfree(disk->random);
1128 xa_destroy(&disk->part_tbl);
1129 disk->queue->disk = NULL;
1130 blk_put_queue(disk->queue);
1131 iput(disk->part0->bd_inode); /* frees the disk */
1134 static int block_uevent(struct device *dev, struct kobj_uevent_env *env)
1136 struct gendisk *disk = dev_to_disk(dev);
1138 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1141 struct class block_class = {
1143 .dev_uevent = block_uevent,
1146 static char *block_devnode(struct device *dev, umode_t *mode,
1147 kuid_t *uid, kgid_t *gid)
1149 struct gendisk *disk = dev_to_disk(dev);
1151 if (disk->fops->devnode)
1152 return disk->fops->devnode(disk, mode);
1156 const struct device_type disk_type = {
1158 .groups = disk_attr_groups,
1159 .release = disk_release,
1160 .devnode = block_devnode,
1163 #ifdef CONFIG_PROC_FS
1165 * aggregate disk stat collector. Uses the same stats that the sysfs
1166 * entries do, above, but makes them available through one seq_file.
1168 * The output looks suspiciously like /proc/partitions with a bunch of
1171 static int diskstats_show(struct seq_file *seqf, void *v)
1173 struct gendisk *gp = v;
1174 struct block_device *hd;
1175 unsigned int inflight;
1176 struct disk_stats stat;
1180 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1181 seq_puts(seqf, "major minor name"
1182 " rio rmerge rsect ruse wio wmerge "
1183 "wsect wuse running use aveq"
1188 xa_for_each(&gp->part_tbl, idx, hd) {
1189 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1191 part_stat_read_all(hd, &stat);
1192 if (queue_is_mq(gp->queue))
1193 inflight = blk_mq_in_flight(gp->queue, hd);
1195 inflight = part_in_flight(hd);
1197 seq_printf(seqf, "%4d %7d %pg "
1204 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1205 stat.ios[STAT_READ],
1206 stat.merges[STAT_READ],
1207 stat.sectors[STAT_READ],
1208 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1210 stat.ios[STAT_WRITE],
1211 stat.merges[STAT_WRITE],
1212 stat.sectors[STAT_WRITE],
1213 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1216 jiffies_to_msecs(stat.io_ticks),
1217 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1218 stat.nsecs[STAT_WRITE] +
1219 stat.nsecs[STAT_DISCARD] +
1220 stat.nsecs[STAT_FLUSH],
1222 stat.ios[STAT_DISCARD],
1223 stat.merges[STAT_DISCARD],
1224 stat.sectors[STAT_DISCARD],
1225 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1227 stat.ios[STAT_FLUSH],
1228 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1237 static const struct seq_operations diskstats_op = {
1238 .start = disk_seqf_start,
1239 .next = disk_seqf_next,
1240 .stop = disk_seqf_stop,
1241 .show = diskstats_show
1244 static int __init proc_genhd_init(void)
1246 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1247 proc_create_seq("partitions", 0, NULL, &partitions_op);
1250 module_init(proc_genhd_init);
1251 #endif /* CONFIG_PROC_FS */
1253 dev_t part_devt(struct gendisk *disk, u8 partno)
1255 struct block_device *part;
1259 part = xa_load(&disk->part_tbl, partno);
1261 devt = part->bd_dev;
1267 dev_t blk_lookup_devt(const char *name, int partno)
1269 dev_t devt = MKDEV(0, 0);
1270 struct class_dev_iter iter;
1273 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1274 while ((dev = class_dev_iter_next(&iter))) {
1275 struct gendisk *disk = dev_to_disk(dev);
1277 if (strcmp(dev_name(dev), name))
1280 if (partno < disk->minors) {
1281 /* We need to return the right devno, even
1282 * if the partition doesn't exist yet.
1284 devt = MKDEV(MAJOR(dev->devt),
1285 MINOR(dev->devt) + partno);
1287 devt = part_devt(disk, partno);
1292 class_dev_iter_exit(&iter);
1296 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1297 struct lock_class_key *lkclass)
1299 struct gendisk *disk;
1301 if (!blk_get_queue(q))
1304 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1308 disk->bdi = bdi_alloc(node_id);
1312 /* bdev_alloc() might need the queue, set before the first call */
1315 disk->part0 = bdev_alloc(disk, 0);
1319 disk->node_id = node_id;
1320 mutex_init(&disk->open_mutex);
1321 xa_init(&disk->part_tbl);
1322 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1323 goto out_destroy_part_tbl;
1325 rand_initialize_disk(disk);
1326 disk_to_dev(disk)->class = &block_class;
1327 disk_to_dev(disk)->type = &disk_type;
1328 device_initialize(disk_to_dev(disk));
1331 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1332 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1333 INIT_LIST_HEAD(&disk->slave_bdevs);
1337 out_destroy_part_tbl:
1338 xa_destroy(&disk->part_tbl);
1339 disk->part0->bd_disk = NULL;
1340 iput(disk->part0->bd_inode);
1349 EXPORT_SYMBOL(__alloc_disk_node);
1351 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1353 struct request_queue *q;
1354 struct gendisk *disk;
1356 q = blk_alloc_queue(node, false);
1360 disk = __alloc_disk_node(q, node, lkclass);
1362 blk_cleanup_queue(q);
1367 EXPORT_SYMBOL(__blk_alloc_disk);
1370 * put_disk - decrements the gendisk refcount
1371 * @disk: the struct gendisk to decrement the refcount for
1373 * This decrements the refcount for the struct gendisk. When this reaches 0
1374 * we'll have disk_release() called.
1376 * Context: Any context, but the last reference must not be dropped from
1379 void put_disk(struct gendisk *disk)
1382 put_device(disk_to_dev(disk));
1384 EXPORT_SYMBOL(put_disk);
1387 * blk_cleanup_disk - shutdown a gendisk allocated by blk_alloc_disk
1388 * @disk: gendisk to shutdown
1390 * Mark the queue hanging off @disk DYING, drain all pending requests, then mark
1391 * the queue DEAD, destroy and put it and the gendisk structure.
1393 * Context: can sleep
1395 void blk_cleanup_disk(struct gendisk *disk)
1397 blk_cleanup_queue(disk->queue);
1400 EXPORT_SYMBOL(blk_cleanup_disk);
1402 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1404 char event[] = "DISK_RO=1";
1405 char *envp[] = { event, NULL };
1409 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1413 * set_disk_ro - set a gendisk read-only
1414 * @disk: gendisk to operate on
1415 * @read_only: %true to set the disk read-only, %false set the disk read/write
1417 * This function is used to indicate whether a given disk device should have its
1418 * read-only flag set. set_disk_ro() is typically used by device drivers to
1419 * indicate whether the underlying physical device is write-protected.
1421 void set_disk_ro(struct gendisk *disk, bool read_only)
1424 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1427 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1430 set_disk_ro_uevent(disk, read_only);
1432 EXPORT_SYMBOL(set_disk_ro);
1434 void inc_diskseq(struct gendisk *disk)
1436 disk->diskseq = atomic64_inc_return(&diskseq);