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);
105 * Format the device name of the indicated block device into the supplied buffer
106 * and return a pointer to that same buffer for convenience.
108 * Note: do not use this in new code, use the %pg specifier to sprintf and
111 const char *bdevname(struct block_device *bdev, char *buf)
113 struct gendisk *hd = bdev->bd_disk;
114 int partno = bdev->bd_partno;
117 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
118 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
119 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
121 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
125 EXPORT_SYMBOL(bdevname);
127 static void part_stat_read_all(struct block_device *part,
128 struct disk_stats *stat)
132 memset(stat, 0, sizeof(struct disk_stats));
133 for_each_possible_cpu(cpu) {
134 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
137 for (group = 0; group < NR_STAT_GROUPS; group++) {
138 stat->nsecs[group] += ptr->nsecs[group];
139 stat->sectors[group] += ptr->sectors[group];
140 stat->ios[group] += ptr->ios[group];
141 stat->merges[group] += ptr->merges[group];
144 stat->io_ticks += ptr->io_ticks;
148 static unsigned int part_in_flight(struct block_device *part)
150 unsigned int inflight = 0;
153 for_each_possible_cpu(cpu) {
154 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
155 part_stat_local_read_cpu(part, in_flight[1], cpu);
157 if ((int)inflight < 0)
163 static void part_in_flight_rw(struct block_device *part,
164 unsigned int inflight[2])
170 for_each_possible_cpu(cpu) {
171 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
172 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
174 if ((int)inflight[0] < 0)
176 if ((int)inflight[1] < 0)
181 * Can be deleted altogether. Later.
184 #define BLKDEV_MAJOR_HASH_SIZE 255
185 static struct blk_major_name {
186 struct blk_major_name *next;
189 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
190 void (*probe)(dev_t devt);
192 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
193 static DEFINE_MUTEX(major_names_lock);
194 static DEFINE_SPINLOCK(major_names_spinlock);
196 /* index in the above - for now: assume no multimajor ranges */
197 static inline int major_to_index(unsigned major)
199 return major % BLKDEV_MAJOR_HASH_SIZE;
202 #ifdef CONFIG_PROC_FS
203 void blkdev_show(struct seq_file *seqf, off_t offset)
205 struct blk_major_name *dp;
207 spin_lock(&major_names_spinlock);
208 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
209 if (dp->major == offset)
210 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
211 spin_unlock(&major_names_spinlock);
213 #endif /* CONFIG_PROC_FS */
216 * __register_blkdev - register a new block device
218 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
219 * @major = 0, try to allocate any unused major number.
220 * @name: the name of the new block device as a zero terminated string
221 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
222 * pre-created device node is accessed. When a probe call uses
223 * add_disk() and it fails the driver must cleanup resources. This
224 * interface may soon be removed.
226 * The @name must be unique within the system.
228 * The return value depends on the @major input parameter:
230 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
231 * then the function returns zero on success, or a negative error code
232 * - if any unused major number was requested with @major = 0 parameter
233 * then the return value is the allocated major number in range
234 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
236 * See Documentation/admin-guide/devices.txt for the list of allocated
239 * Use register_blkdev instead for any new code.
241 int __register_blkdev(unsigned int major, const char *name,
242 void (*probe)(dev_t devt))
244 struct blk_major_name **n, *p;
247 mutex_lock(&major_names_lock);
251 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
252 if (major_names[index] == NULL)
257 printk("%s: failed to get major for %s\n",
266 if (major >= BLKDEV_MAJOR_MAX) {
267 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
268 __func__, major, BLKDEV_MAJOR_MAX-1, name);
274 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
281 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
284 strlcpy(p->name, name, sizeof(p->name));
286 index = major_to_index(major);
288 spin_lock(&major_names_spinlock);
289 for (n = &major_names[index]; *n; n = &(*n)->next) {
290 if ((*n)->major == major)
297 spin_unlock(&major_names_spinlock);
300 printk("register_blkdev: cannot get major %u for %s\n",
305 mutex_unlock(&major_names_lock);
308 EXPORT_SYMBOL(__register_blkdev);
310 void unregister_blkdev(unsigned int major, const char *name)
312 struct blk_major_name **n;
313 struct blk_major_name *p = NULL;
314 int index = major_to_index(major);
316 mutex_lock(&major_names_lock);
317 spin_lock(&major_names_spinlock);
318 for (n = &major_names[index]; *n; n = &(*n)->next)
319 if ((*n)->major == major)
321 if (!*n || strcmp((*n)->name, name)) {
327 spin_unlock(&major_names_spinlock);
328 mutex_unlock(&major_names_lock);
332 EXPORT_SYMBOL(unregister_blkdev);
334 int blk_alloc_ext_minor(void)
338 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT, GFP_KERNEL);
344 void blk_free_ext_minor(unsigned int minor)
346 ida_free(&ext_devt_ida, minor);
349 static char *bdevt_str(dev_t devt, char *buf)
351 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
352 char tbuf[BDEVT_SIZE];
353 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
354 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
356 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
361 void disk_uevent(struct gendisk *disk, enum kobject_action action)
363 struct block_device *part;
367 xa_for_each(&disk->part_tbl, idx, part) {
368 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
370 if (!kobject_get_unless_zero(&part->bd_device.kobj))
374 kobject_uevent(bdev_kobj(part), action);
375 put_device(&part->bd_device);
380 EXPORT_SYMBOL_GPL(disk_uevent);
382 int disk_scan_partitions(struct gendisk *disk, fmode_t mode)
384 struct block_device *bdev;
386 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
388 if (disk->open_partitions)
391 set_bit(GD_NEED_PART_SCAN, &disk->state);
392 bdev = blkdev_get_by_dev(disk_devt(disk), mode, NULL);
394 return PTR_ERR(bdev);
395 blkdev_put(bdev, mode);
400 * device_add_disk - add disk information to kernel list
401 * @parent: parent device for the disk
402 * @disk: per-device partitioning information
403 * @groups: Additional per-device sysfs groups
405 * This function registers the partitioning information in @disk
408 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
409 const struct attribute_group **groups)
412 struct device *ddev = disk_to_dev(disk);
416 * The disk queue should now be all set with enough information about
417 * the device for the elevator code to pick an adequate default
418 * elevator if one is needed, that is, for devices requesting queue
421 elevator_init_mq(disk->queue);
424 * If the driver provides an explicit major number it also must provide
425 * the number of minors numbers supported, and those will be used to
427 * Otherwise just allocate the device numbers for both the whole device
428 * and all partitions from the extended dev_t space.
431 if (WARN_ON(!disk->minors))
434 if (disk->minors > DISK_MAX_PARTS) {
435 pr_err("block: can't allocate more than %d partitions\n",
437 disk->minors = DISK_MAX_PARTS;
439 if (disk->first_minor + disk->minors > MINORMASK + 1)
442 if (WARN_ON(disk->minors))
445 ret = blk_alloc_ext_minor();
448 disk->major = BLOCK_EXT_MAJOR;
449 disk->first_minor = ret;
452 /* delay uevents, until we scanned partition table */
453 dev_set_uevent_suppress(ddev, 1);
455 ddev->parent = parent;
456 ddev->groups = groups;
457 dev_set_name(ddev, "%s", disk->disk_name);
458 if (!(disk->flags & GENHD_FL_HIDDEN))
459 ddev->devt = MKDEV(disk->major, disk->first_minor);
460 ret = device_add(ddev);
462 goto out_free_ext_minor;
464 ret = disk_alloc_events(disk);
468 if (!sysfs_deprecated) {
469 ret = sysfs_create_link(block_depr, &ddev->kobj,
470 kobject_name(&ddev->kobj));
476 * avoid probable deadlock caused by allocating memory with
477 * GFP_KERNEL in runtime_resume callback of its all ancestor
480 pm_runtime_set_memalloc_noio(ddev, true);
482 ret = blk_integrity_add(disk);
484 goto out_del_block_link;
486 disk->part0->bd_holder_dir =
487 kobject_create_and_add("holders", &ddev->kobj);
488 if (!disk->part0->bd_holder_dir) {
490 goto out_del_integrity;
492 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
493 if (!disk->slave_dir) {
495 goto out_put_holder_dir;
498 ret = bd_register_pending_holders(disk);
500 goto out_put_slave_dir;
502 ret = blk_register_queue(disk);
504 goto out_put_slave_dir;
506 if (!(disk->flags & GENHD_FL_HIDDEN)) {
507 ret = bdi_register(disk->bdi, "%u:%u",
508 disk->major, disk->first_minor);
510 goto out_unregister_queue;
511 bdi_set_owner(disk->bdi, ddev);
512 ret = sysfs_create_link(&ddev->kobj,
513 &disk->bdi->dev->kobj, "bdi");
515 goto out_unregister_bdi;
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 disk_update_readahead(disk);
530 disk_add_events(disk);
531 set_bit(GD_ADDED, &disk->state);
535 if (!(disk->flags & GENHD_FL_HIDDEN))
536 bdi_unregister(disk->bdi);
537 out_unregister_queue:
538 blk_unregister_queue(disk);
540 kobject_put(disk->slave_dir);
542 kobject_put(disk->part0->bd_holder_dir);
544 blk_integrity_del(disk);
546 if (!sysfs_deprecated)
547 sysfs_remove_link(block_depr, dev_name(ddev));
551 if (disk->major == BLOCK_EXT_MAJOR)
552 blk_free_ext_minor(disk->first_minor);
555 EXPORT_SYMBOL(device_add_disk);
558 * del_gendisk - remove the gendisk
559 * @disk: the struct gendisk to remove
561 * Removes the gendisk and all its associated resources. This deletes the
562 * partitions associated with the gendisk, and unregisters the associated
565 * This is the counter to the respective __device_add_disk() call.
567 * The final removal of the struct gendisk happens when its refcount reaches 0
568 * with put_disk(), which should be called after del_gendisk(), if
569 * __device_add_disk() was used.
571 * Drivers exist which depend on the release of the gendisk to be synchronous,
572 * it should not be deferred.
576 void del_gendisk(struct gendisk *disk)
578 struct request_queue *q = disk->queue;
582 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
585 blk_integrity_del(disk);
586 disk_del_events(disk);
588 mutex_lock(&disk->open_mutex);
589 remove_inode_hash(disk->part0->bd_inode);
590 blk_drop_partitions(disk);
591 mutex_unlock(&disk->open_mutex);
593 fsync_bdev(disk->part0);
594 __invalidate_device(disk->part0, true);
599 set_bit(GD_DEAD, &disk->state);
600 set_capacity(disk, 0);
603 * Prevent new I/O from crossing bio_queue_enter().
605 blk_queue_start_drain(q);
607 if (!(disk->flags & GENHD_FL_HIDDEN)) {
608 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
611 * Unregister bdi before releasing device numbers (as they can
612 * get reused and we'd get clashes in sysfs).
614 bdi_unregister(disk->bdi);
617 blk_unregister_queue(disk);
619 kobject_put(disk->part0->bd_holder_dir);
620 kobject_put(disk->slave_dir);
622 part_stat_set_all(disk->part0, 0);
623 disk->part0->bd_stamp = 0;
624 if (!sysfs_deprecated)
625 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
626 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
627 device_del(disk_to_dev(disk));
629 blk_mq_freeze_queue_wait(q);
631 blk_throtl_cancel_bios(disk->queue);
634 blk_flush_integrity();
636 * Allow using passthrough request again after the queue is torn down.
638 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
639 __blk_mq_unfreeze_queue(q, true);
642 EXPORT_SYMBOL(del_gendisk);
645 * invalidate_disk - invalidate the disk
646 * @disk: the struct gendisk to invalidate
648 * A helper to invalidates the disk. It will clean the disk's associated
649 * buffer/page caches and reset its internal states so that the disk
650 * can be reused by the drivers.
654 void invalidate_disk(struct gendisk *disk)
656 struct block_device *bdev = disk->part0;
658 invalidate_bdev(bdev);
659 bdev->bd_inode->i_mapping->wb_err = 0;
660 set_capacity(disk, 0);
662 EXPORT_SYMBOL(invalidate_disk);
664 /* sysfs access to bad-blocks list. */
665 static ssize_t disk_badblocks_show(struct device *dev,
666 struct device_attribute *attr,
669 struct gendisk *disk = dev_to_disk(dev);
672 return sprintf(page, "\n");
674 return badblocks_show(disk->bb, page, 0);
677 static ssize_t disk_badblocks_store(struct device *dev,
678 struct device_attribute *attr,
679 const char *page, size_t len)
681 struct gendisk *disk = dev_to_disk(dev);
686 return badblocks_store(disk->bb, page, len, 0);
689 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
690 void blk_request_module(dev_t devt)
692 unsigned int major = MAJOR(devt);
693 struct blk_major_name **n;
695 mutex_lock(&major_names_lock);
696 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
697 if ((*n)->major == major && (*n)->probe) {
699 mutex_unlock(&major_names_lock);
703 mutex_unlock(&major_names_lock);
705 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
706 /* Make old-style 2.4 aliases work */
707 request_module("block-major-%d", MAJOR(devt));
709 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
712 * print a full list of all partitions - intended for places where the root
713 * filesystem can't be mounted and thus to give the victim some idea of what
716 void __init printk_all_partitions(void)
718 struct class_dev_iter iter;
721 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
722 while ((dev = class_dev_iter_next(&iter))) {
723 struct gendisk *disk = dev_to_disk(dev);
724 struct block_device *part;
725 char devt_buf[BDEVT_SIZE];
729 * Don't show empty devices or things that have been
732 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
736 * Note, unlike /proc/partitions, I am showing the numbers in
737 * hex - the same format as the root= option takes.
740 xa_for_each(&disk->part_tbl, idx, part) {
741 if (!bdev_nr_sectors(part))
743 printk("%s%s %10llu %pg %s",
744 bdev_is_partition(part) ? " " : "",
745 bdevt_str(part->bd_dev, devt_buf),
746 bdev_nr_sectors(part) >> 1, part,
748 part->bd_meta_info->uuid : "");
749 if (bdev_is_partition(part))
751 else if (dev->parent && dev->parent->driver)
752 printk(" driver: %s\n",
753 dev->parent->driver->name);
755 printk(" (driver?)\n");
759 class_dev_iter_exit(&iter);
762 #ifdef CONFIG_PROC_FS
764 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
767 struct class_dev_iter *iter;
770 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
772 return ERR_PTR(-ENOMEM);
774 seqf->private = iter;
775 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
777 dev = class_dev_iter_next(iter);
782 return dev_to_disk(dev);
785 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
790 dev = class_dev_iter_next(seqf->private);
792 return dev_to_disk(dev);
797 static void disk_seqf_stop(struct seq_file *seqf, void *v)
799 struct class_dev_iter *iter = seqf->private;
801 /* stop is called even after start failed :-( */
803 class_dev_iter_exit(iter);
805 seqf->private = NULL;
809 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
813 p = disk_seqf_start(seqf, pos);
814 if (!IS_ERR_OR_NULL(p) && !*pos)
815 seq_puts(seqf, "major minor #blocks name\n\n");
819 static int show_partition(struct seq_file *seqf, void *v)
821 struct gendisk *sgp = v;
822 struct block_device *part;
825 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
829 xa_for_each(&sgp->part_tbl, idx, part) {
830 if (!bdev_nr_sectors(part))
832 seq_printf(seqf, "%4d %7d %10llu %pg\n",
833 MAJOR(part->bd_dev), MINOR(part->bd_dev),
834 bdev_nr_sectors(part) >> 1, part);
840 static const struct seq_operations partitions_op = {
841 .start = show_partition_start,
842 .next = disk_seqf_next,
843 .stop = disk_seqf_stop,
844 .show = show_partition
848 static int __init genhd_device_init(void)
852 block_class.dev_kobj = sysfs_dev_block_kobj;
853 error = class_register(&block_class);
858 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
860 /* create top-level block dir */
861 if (!sysfs_deprecated)
862 block_depr = kobject_create_and_add("block", NULL);
866 subsys_initcall(genhd_device_init);
868 static ssize_t disk_range_show(struct device *dev,
869 struct device_attribute *attr, char *buf)
871 struct gendisk *disk = dev_to_disk(dev);
873 return sprintf(buf, "%d\n", disk->minors);
876 static ssize_t disk_ext_range_show(struct device *dev,
877 struct device_attribute *attr, char *buf)
879 struct gendisk *disk = dev_to_disk(dev);
881 return sprintf(buf, "%d\n",
882 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
885 static ssize_t disk_removable_show(struct device *dev,
886 struct device_attribute *attr, char *buf)
888 struct gendisk *disk = dev_to_disk(dev);
890 return sprintf(buf, "%d\n",
891 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
894 static ssize_t disk_hidden_show(struct device *dev,
895 struct device_attribute *attr, char *buf)
897 struct gendisk *disk = dev_to_disk(dev);
899 return sprintf(buf, "%d\n",
900 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
903 static ssize_t disk_ro_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", get_disk_ro(disk) ? 1 : 0);
911 ssize_t part_size_show(struct device *dev,
912 struct device_attribute *attr, char *buf)
914 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
917 ssize_t part_stat_show(struct device *dev,
918 struct device_attribute *attr, char *buf)
920 struct block_device *bdev = dev_to_bdev(dev);
921 struct request_queue *q = bdev_get_queue(bdev);
922 struct disk_stats stat;
923 unsigned int inflight;
926 inflight = blk_mq_in_flight(q, bdev);
928 inflight = part_in_flight(bdev);
932 update_io_ticks(bdev, jiffies, true);
935 part_stat_read_all(bdev, &stat);
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[] = {
1103 static void disk_release_mq(struct request_queue *q)
1105 blk_mq_cancel_work_sync(q);
1108 * There can't be any non non-passthrough bios in flight here, but
1109 * requests stay around longer, including passthrough ones so we
1110 * still need to freeze the queue here.
1112 blk_mq_freeze_queue(q);
1115 * Since the I/O scheduler exit code may access cgroup information,
1116 * perform I/O scheduler exit before disassociating from the block
1117 * cgroup controller.
1120 mutex_lock(&q->sysfs_lock);
1122 mutex_unlock(&q->sysfs_lock);
1125 __blk_mq_unfreeze_queue(q, true);
1129 * disk_release - releases all allocated resources of the gendisk
1130 * @dev: the device representing this disk
1132 * This function releases all allocated resources of the gendisk.
1134 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1135 * assigned. Since the request_queue sits on top of the gendisk for these
1136 * drivers we also call blk_put_queue() for them, and we expect the
1137 * request_queue refcount to reach 0 at this point, and so the request_queue
1138 * will also be freed prior to the disk.
1140 * Context: can sleep
1142 static void disk_release(struct device *dev)
1144 struct gendisk *disk = dev_to_disk(dev);
1147 WARN_ON_ONCE(disk_live(disk));
1149 if (queue_is_mq(disk->queue))
1150 disk_release_mq(disk->queue);
1152 blkcg_exit_queue(disk->queue);
1154 disk_release_events(disk);
1155 kfree(disk->random);
1156 xa_destroy(&disk->part_tbl);
1158 disk->queue->disk = NULL;
1159 blk_put_queue(disk->queue);
1161 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1162 disk->fops->free_disk(disk);
1164 iput(disk->part0->bd_inode); /* frees the disk */
1167 static int block_uevent(struct device *dev, struct kobj_uevent_env *env)
1169 struct gendisk *disk = dev_to_disk(dev);
1171 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1174 struct class block_class = {
1176 .dev_uevent = block_uevent,
1179 static char *block_devnode(struct device *dev, umode_t *mode,
1180 kuid_t *uid, kgid_t *gid)
1182 struct gendisk *disk = dev_to_disk(dev);
1184 if (disk->fops->devnode)
1185 return disk->fops->devnode(disk, mode);
1189 const struct device_type disk_type = {
1191 .groups = disk_attr_groups,
1192 .release = disk_release,
1193 .devnode = block_devnode,
1196 #ifdef CONFIG_PROC_FS
1198 * aggregate disk stat collector. Uses the same stats that the sysfs
1199 * entries do, above, but makes them available through one seq_file.
1201 * The output looks suspiciously like /proc/partitions with a bunch of
1204 static int diskstats_show(struct seq_file *seqf, void *v)
1206 struct gendisk *gp = v;
1207 struct block_device *hd;
1208 unsigned int inflight;
1209 struct disk_stats stat;
1213 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1214 seq_puts(seqf, "major minor name"
1215 " rio rmerge rsect ruse wio wmerge "
1216 "wsect wuse running use aveq"
1221 xa_for_each(&gp->part_tbl, idx, hd) {
1222 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1224 if (queue_is_mq(gp->queue))
1225 inflight = blk_mq_in_flight(gp->queue, hd);
1227 inflight = part_in_flight(hd);
1231 update_io_ticks(hd, jiffies, true);
1234 part_stat_read_all(hd, &stat);
1235 seq_printf(seqf, "%4d %7d %pg "
1242 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1243 stat.ios[STAT_READ],
1244 stat.merges[STAT_READ],
1245 stat.sectors[STAT_READ],
1246 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1248 stat.ios[STAT_WRITE],
1249 stat.merges[STAT_WRITE],
1250 stat.sectors[STAT_WRITE],
1251 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1254 jiffies_to_msecs(stat.io_ticks),
1255 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1256 stat.nsecs[STAT_WRITE] +
1257 stat.nsecs[STAT_DISCARD] +
1258 stat.nsecs[STAT_FLUSH],
1260 stat.ios[STAT_DISCARD],
1261 stat.merges[STAT_DISCARD],
1262 stat.sectors[STAT_DISCARD],
1263 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1265 stat.ios[STAT_FLUSH],
1266 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1275 static const struct seq_operations diskstats_op = {
1276 .start = disk_seqf_start,
1277 .next = disk_seqf_next,
1278 .stop = disk_seqf_stop,
1279 .show = diskstats_show
1282 static int __init proc_genhd_init(void)
1284 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1285 proc_create_seq("partitions", 0, NULL, &partitions_op);
1288 module_init(proc_genhd_init);
1289 #endif /* CONFIG_PROC_FS */
1291 dev_t part_devt(struct gendisk *disk, u8 partno)
1293 struct block_device *part;
1297 part = xa_load(&disk->part_tbl, partno);
1299 devt = part->bd_dev;
1305 dev_t blk_lookup_devt(const char *name, int partno)
1307 dev_t devt = MKDEV(0, 0);
1308 struct class_dev_iter iter;
1311 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1312 while ((dev = class_dev_iter_next(&iter))) {
1313 struct gendisk *disk = dev_to_disk(dev);
1315 if (strcmp(dev_name(dev), name))
1318 if (partno < disk->minors) {
1319 /* We need to return the right devno, even
1320 * if the partition doesn't exist yet.
1322 devt = MKDEV(MAJOR(dev->devt),
1323 MINOR(dev->devt) + partno);
1325 devt = part_devt(disk, partno);
1330 class_dev_iter_exit(&iter);
1334 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1335 struct lock_class_key *lkclass)
1337 struct gendisk *disk;
1339 if (!blk_get_queue(q))
1342 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1346 disk->bdi = bdi_alloc(node_id);
1350 /* bdev_alloc() might need the queue, set before the first call */
1353 disk->part0 = bdev_alloc(disk, 0);
1357 disk->node_id = node_id;
1358 mutex_init(&disk->open_mutex);
1359 xa_init(&disk->part_tbl);
1360 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1361 goto out_destroy_part_tbl;
1363 if (blkcg_init_queue(q))
1364 goto out_erase_part0;
1366 rand_initialize_disk(disk);
1367 disk_to_dev(disk)->class = &block_class;
1368 disk_to_dev(disk)->type = &disk_type;
1369 device_initialize(disk_to_dev(disk));
1372 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1373 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1374 INIT_LIST_HEAD(&disk->slave_bdevs);
1379 xa_erase(&disk->part_tbl, 0);
1380 out_destroy_part_tbl:
1381 xa_destroy(&disk->part_tbl);
1382 disk->part0->bd_disk = NULL;
1383 iput(disk->part0->bd_inode);
1392 EXPORT_SYMBOL(__alloc_disk_node);
1394 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1396 struct request_queue *q;
1397 struct gendisk *disk;
1399 q = blk_alloc_queue(node, false);
1403 disk = __alloc_disk_node(q, node, lkclass);
1405 blk_cleanup_queue(q);
1410 EXPORT_SYMBOL(__blk_alloc_disk);
1413 * put_disk - decrements the gendisk refcount
1414 * @disk: the struct gendisk to decrement the refcount for
1416 * This decrements the refcount for the struct gendisk. When this reaches 0
1417 * we'll have disk_release() called.
1419 * Context: Any context, but the last reference must not be dropped from
1422 void put_disk(struct gendisk *disk)
1425 put_device(disk_to_dev(disk));
1427 EXPORT_SYMBOL(put_disk);
1430 * blk_cleanup_disk - shutdown a gendisk allocated by blk_alloc_disk
1431 * @disk: gendisk to shutdown
1433 * Mark the queue hanging off @disk DYING, drain all pending requests, then mark
1434 * the queue DEAD, destroy and put it and the gendisk structure.
1436 * Context: can sleep
1438 void blk_cleanup_disk(struct gendisk *disk)
1440 blk_cleanup_queue(disk->queue);
1443 EXPORT_SYMBOL(blk_cleanup_disk);
1445 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1447 char event[] = "DISK_RO=1";
1448 char *envp[] = { event, NULL };
1452 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1456 * set_disk_ro - set a gendisk read-only
1457 * @disk: gendisk to operate on
1458 * @read_only: %true to set the disk read-only, %false set the disk read/write
1460 * This function is used to indicate whether a given disk device should have its
1461 * read-only flag set. set_disk_ro() is typically used by device drivers to
1462 * indicate whether the underlying physical device is write-protected.
1464 void set_disk_ro(struct gendisk *disk, bool read_only)
1467 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1470 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1473 set_disk_ro_uevent(disk, read_only);
1475 EXPORT_SYMBOL(set_disk_ro);
1477 void inc_diskseq(struct gendisk *disk)
1479 disk->diskseq = atomic64_inc_return(&diskseq);