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 * del_gendisk - remove the gendisk
553 * @disk: the struct gendisk to remove
555 * Removes the gendisk and all its associated resources. This deletes the
556 * partitions associated with the gendisk, and unregisters the associated
559 * This is the counter to the respective __device_add_disk() call.
561 * The final removal of the struct gendisk happens when its refcount reaches 0
562 * with put_disk(), which should be called after del_gendisk(), if
563 * __device_add_disk() was used.
565 * Drivers exist which depend on the release of the gendisk to be synchronous,
566 * it should not be deferred.
570 void del_gendisk(struct gendisk *disk)
572 struct request_queue *q = disk->queue;
576 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
579 blk_integrity_del(disk);
580 disk_del_events(disk);
582 mutex_lock(&disk->open_mutex);
583 remove_inode_hash(disk->part0->bd_inode);
584 blk_drop_partitions(disk);
585 mutex_unlock(&disk->open_mutex);
587 fsync_bdev(disk->part0);
588 __invalidate_device(disk->part0, true);
593 set_bit(GD_DEAD, &disk->state);
594 set_capacity(disk, 0);
597 * Prevent new I/O from crossing bio_queue_enter().
599 blk_queue_start_drain(q);
601 if (!(disk->flags & GENHD_FL_HIDDEN)) {
602 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
605 * Unregister bdi before releasing device numbers (as they can
606 * get reused and we'd get clashes in sysfs).
608 bdi_unregister(disk->bdi);
611 blk_unregister_queue(disk);
613 kobject_put(disk->part0->bd_holder_dir);
614 kobject_put(disk->slave_dir);
616 part_stat_set_all(disk->part0, 0);
617 disk->part0->bd_stamp = 0;
618 if (!sysfs_deprecated)
619 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
620 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
621 device_del(disk_to_dev(disk));
623 blk_mq_freeze_queue_wait(q);
627 blk_flush_integrity();
629 * Allow using passthrough request again after the queue is torn down.
631 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
632 __blk_mq_unfreeze_queue(q, true);
635 EXPORT_SYMBOL(del_gendisk);
638 * invalidate_disk - invalidate the disk
639 * @disk: the struct gendisk to invalidate
641 * A helper to invalidates the disk. It will clean the disk's associated
642 * buffer/page caches and reset its internal states so that the disk
643 * can be reused by the drivers.
647 void invalidate_disk(struct gendisk *disk)
649 struct block_device *bdev = disk->part0;
651 invalidate_bdev(bdev);
652 bdev->bd_inode->i_mapping->wb_err = 0;
653 set_capacity(disk, 0);
655 EXPORT_SYMBOL(invalidate_disk);
657 /* sysfs access to bad-blocks list. */
658 static ssize_t disk_badblocks_show(struct device *dev,
659 struct device_attribute *attr,
662 struct gendisk *disk = dev_to_disk(dev);
665 return sprintf(page, "\n");
667 return badblocks_show(disk->bb, page, 0);
670 static ssize_t disk_badblocks_store(struct device *dev,
671 struct device_attribute *attr,
672 const char *page, size_t len)
674 struct gendisk *disk = dev_to_disk(dev);
679 return badblocks_store(disk->bb, page, len, 0);
682 void blk_request_module(dev_t devt)
684 unsigned int major = MAJOR(devt);
685 struct blk_major_name **n;
687 mutex_lock(&major_names_lock);
688 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
689 if ((*n)->major == major && (*n)->probe) {
691 mutex_unlock(&major_names_lock);
695 mutex_unlock(&major_names_lock);
697 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
698 /* Make old-style 2.4 aliases work */
699 request_module("block-major-%d", MAJOR(devt));
703 * print a full list of all partitions - intended for places where the root
704 * filesystem can't be mounted and thus to give the victim some idea of what
707 void __init printk_all_partitions(void)
709 struct class_dev_iter iter;
712 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
713 while ((dev = class_dev_iter_next(&iter))) {
714 struct gendisk *disk = dev_to_disk(dev);
715 struct block_device *part;
716 char devt_buf[BDEVT_SIZE];
720 * Don't show empty devices or things that have been
723 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
727 * Note, unlike /proc/partitions, I am showing the numbers in
728 * hex - the same format as the root= option takes.
731 xa_for_each(&disk->part_tbl, idx, part) {
732 if (!bdev_nr_sectors(part))
734 printk("%s%s %10llu %pg %s",
735 bdev_is_partition(part) ? " " : "",
736 bdevt_str(part->bd_dev, devt_buf),
737 bdev_nr_sectors(part) >> 1, part,
739 part->bd_meta_info->uuid : "");
740 if (bdev_is_partition(part))
742 else if (dev->parent && dev->parent->driver)
743 printk(" driver: %s\n",
744 dev->parent->driver->name);
746 printk(" (driver?)\n");
750 class_dev_iter_exit(&iter);
753 #ifdef CONFIG_PROC_FS
755 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
758 struct class_dev_iter *iter;
761 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
763 return ERR_PTR(-ENOMEM);
765 seqf->private = iter;
766 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
768 dev = class_dev_iter_next(iter);
773 return dev_to_disk(dev);
776 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
781 dev = class_dev_iter_next(seqf->private);
783 return dev_to_disk(dev);
788 static void disk_seqf_stop(struct seq_file *seqf, void *v)
790 struct class_dev_iter *iter = seqf->private;
792 /* stop is called even after start failed :-( */
794 class_dev_iter_exit(iter);
796 seqf->private = NULL;
800 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
804 p = disk_seqf_start(seqf, pos);
805 if (!IS_ERR_OR_NULL(p) && !*pos)
806 seq_puts(seqf, "major minor #blocks name\n\n");
810 static int show_partition(struct seq_file *seqf, void *v)
812 struct gendisk *sgp = v;
813 struct block_device *part;
816 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
820 xa_for_each(&sgp->part_tbl, idx, part) {
821 if (!bdev_nr_sectors(part))
823 seq_printf(seqf, "%4d %7d %10llu %pg\n",
824 MAJOR(part->bd_dev), MINOR(part->bd_dev),
825 bdev_nr_sectors(part) >> 1, part);
831 static const struct seq_operations partitions_op = {
832 .start = show_partition_start,
833 .next = disk_seqf_next,
834 .stop = disk_seqf_stop,
835 .show = show_partition
839 static int __init genhd_device_init(void)
843 block_class.dev_kobj = sysfs_dev_block_kobj;
844 error = class_register(&block_class);
849 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
851 /* create top-level block dir */
852 if (!sysfs_deprecated)
853 block_depr = kobject_create_and_add("block", NULL);
857 subsys_initcall(genhd_device_init);
859 static ssize_t disk_range_show(struct device *dev,
860 struct device_attribute *attr, char *buf)
862 struct gendisk *disk = dev_to_disk(dev);
864 return sprintf(buf, "%d\n", disk->minors);
867 static ssize_t disk_ext_range_show(struct device *dev,
868 struct device_attribute *attr, char *buf)
870 struct gendisk *disk = dev_to_disk(dev);
872 return sprintf(buf, "%d\n",
873 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
876 static ssize_t disk_removable_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_REMOVABLE ? 1 : 0));
885 static ssize_t disk_hidden_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_HIDDEN ? 1 : 0));
894 static ssize_t disk_ro_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", get_disk_ro(disk) ? 1 : 0);
902 ssize_t part_size_show(struct device *dev,
903 struct device_attribute *attr, char *buf)
905 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
908 ssize_t part_stat_show(struct device *dev,
909 struct device_attribute *attr, char *buf)
911 struct block_device *bdev = dev_to_bdev(dev);
912 struct request_queue *q = bdev_get_queue(bdev);
913 struct disk_stats stat;
914 unsigned int inflight;
916 part_stat_read_all(bdev, &stat);
918 inflight = blk_mq_in_flight(q, bdev);
920 inflight = part_in_flight(bdev);
923 "%8lu %8lu %8llu %8u "
924 "%8lu %8lu %8llu %8u "
926 "%8lu %8lu %8llu %8u "
930 stat.merges[STAT_READ],
931 (unsigned long long)stat.sectors[STAT_READ],
932 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
933 stat.ios[STAT_WRITE],
934 stat.merges[STAT_WRITE],
935 (unsigned long long)stat.sectors[STAT_WRITE],
936 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
938 jiffies_to_msecs(stat.io_ticks),
939 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
940 stat.nsecs[STAT_WRITE] +
941 stat.nsecs[STAT_DISCARD] +
942 stat.nsecs[STAT_FLUSH],
944 stat.ios[STAT_DISCARD],
945 stat.merges[STAT_DISCARD],
946 (unsigned long long)stat.sectors[STAT_DISCARD],
947 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
948 stat.ios[STAT_FLUSH],
949 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
952 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
955 struct block_device *bdev = dev_to_bdev(dev);
956 struct request_queue *q = bdev_get_queue(bdev);
957 unsigned int inflight[2];
960 blk_mq_in_flight_rw(q, bdev, inflight);
962 part_in_flight_rw(bdev, inflight);
964 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
967 static ssize_t disk_capability_show(struct device *dev,
968 struct device_attribute *attr, char *buf)
970 struct gendisk *disk = dev_to_disk(dev);
972 return sprintf(buf, "%x\n", disk->flags);
975 static ssize_t disk_alignment_offset_show(struct device *dev,
976 struct device_attribute *attr,
979 struct gendisk *disk = dev_to_disk(dev);
981 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
984 static ssize_t disk_discard_alignment_show(struct device *dev,
985 struct device_attribute *attr,
988 struct gendisk *disk = dev_to_disk(dev);
990 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
993 static ssize_t diskseq_show(struct device *dev,
994 struct device_attribute *attr, char *buf)
996 struct gendisk *disk = dev_to_disk(dev);
998 return sprintf(buf, "%llu\n", disk->diskseq);
1001 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1002 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1003 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1004 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1005 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1006 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1007 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1008 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1009 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1010 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1011 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1012 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1013 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1015 #ifdef CONFIG_FAIL_MAKE_REQUEST
1016 ssize_t part_fail_show(struct device *dev,
1017 struct device_attribute *attr, char *buf)
1019 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1022 ssize_t part_fail_store(struct device *dev,
1023 struct device_attribute *attr,
1024 const char *buf, size_t count)
1028 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1029 dev_to_bdev(dev)->bd_make_it_fail = i;
1034 static struct device_attribute dev_attr_fail =
1035 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1036 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1038 #ifdef CONFIG_FAIL_IO_TIMEOUT
1039 static struct device_attribute dev_attr_fail_timeout =
1040 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1043 static struct attribute *disk_attrs[] = {
1044 &dev_attr_range.attr,
1045 &dev_attr_ext_range.attr,
1046 &dev_attr_removable.attr,
1047 &dev_attr_hidden.attr,
1049 &dev_attr_size.attr,
1050 &dev_attr_alignment_offset.attr,
1051 &dev_attr_discard_alignment.attr,
1052 &dev_attr_capability.attr,
1053 &dev_attr_stat.attr,
1054 &dev_attr_inflight.attr,
1055 &dev_attr_badblocks.attr,
1056 &dev_attr_events.attr,
1057 &dev_attr_events_async.attr,
1058 &dev_attr_events_poll_msecs.attr,
1059 &dev_attr_diskseq.attr,
1060 #ifdef CONFIG_FAIL_MAKE_REQUEST
1061 &dev_attr_fail.attr,
1063 #ifdef CONFIG_FAIL_IO_TIMEOUT
1064 &dev_attr_fail_timeout.attr,
1069 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1071 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1072 struct gendisk *disk = dev_to_disk(dev);
1074 if (a == &dev_attr_badblocks.attr && !disk->bb)
1079 static struct attribute_group disk_attr_group = {
1080 .attrs = disk_attrs,
1081 .is_visible = disk_visible,
1084 static const struct attribute_group *disk_attr_groups[] = {
1090 * disk_release - releases all allocated resources of the gendisk
1091 * @dev: the device representing this disk
1093 * This function releases all allocated resources of the gendisk.
1095 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1096 * assigned. Since the request_queue sits on top of the gendisk for these
1097 * drivers we also call blk_put_queue() for them, and we expect the
1098 * request_queue refcount to reach 0 at this point, and so the request_queue
1099 * will also be freed prior to the disk.
1101 * Context: can sleep
1103 static void disk_release(struct device *dev)
1105 struct gendisk *disk = dev_to_disk(dev);
1108 WARN_ON_ONCE(disk_live(disk));
1110 blk_mq_cancel_work_sync(disk->queue);
1112 disk_release_events(disk);
1113 kfree(disk->random);
1114 xa_destroy(&disk->part_tbl);
1115 disk->queue->disk = NULL;
1116 blk_put_queue(disk->queue);
1117 iput(disk->part0->bd_inode); /* frees the disk */
1120 static int block_uevent(struct device *dev, struct kobj_uevent_env *env)
1122 struct gendisk *disk = dev_to_disk(dev);
1124 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1127 struct class block_class = {
1129 .dev_uevent = block_uevent,
1132 static char *block_devnode(struct device *dev, umode_t *mode,
1133 kuid_t *uid, kgid_t *gid)
1135 struct gendisk *disk = dev_to_disk(dev);
1137 if (disk->fops->devnode)
1138 return disk->fops->devnode(disk, mode);
1142 const struct device_type disk_type = {
1144 .groups = disk_attr_groups,
1145 .release = disk_release,
1146 .devnode = block_devnode,
1149 #ifdef CONFIG_PROC_FS
1151 * aggregate disk stat collector. Uses the same stats that the sysfs
1152 * entries do, above, but makes them available through one seq_file.
1154 * The output looks suspiciously like /proc/partitions with a bunch of
1157 static int diskstats_show(struct seq_file *seqf, void *v)
1159 struct gendisk *gp = v;
1160 struct block_device *hd;
1161 unsigned int inflight;
1162 struct disk_stats stat;
1166 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1167 seq_puts(seqf, "major minor name"
1168 " rio rmerge rsect ruse wio wmerge "
1169 "wsect wuse running use aveq"
1174 xa_for_each(&gp->part_tbl, idx, hd) {
1175 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1177 part_stat_read_all(hd, &stat);
1178 if (queue_is_mq(gp->queue))
1179 inflight = blk_mq_in_flight(gp->queue, hd);
1181 inflight = part_in_flight(hd);
1183 seq_printf(seqf, "%4d %7d %pg "
1190 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1191 stat.ios[STAT_READ],
1192 stat.merges[STAT_READ],
1193 stat.sectors[STAT_READ],
1194 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1196 stat.ios[STAT_WRITE],
1197 stat.merges[STAT_WRITE],
1198 stat.sectors[STAT_WRITE],
1199 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1202 jiffies_to_msecs(stat.io_ticks),
1203 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1204 stat.nsecs[STAT_WRITE] +
1205 stat.nsecs[STAT_DISCARD] +
1206 stat.nsecs[STAT_FLUSH],
1208 stat.ios[STAT_DISCARD],
1209 stat.merges[STAT_DISCARD],
1210 stat.sectors[STAT_DISCARD],
1211 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1213 stat.ios[STAT_FLUSH],
1214 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1223 static const struct seq_operations diskstats_op = {
1224 .start = disk_seqf_start,
1225 .next = disk_seqf_next,
1226 .stop = disk_seqf_stop,
1227 .show = diskstats_show
1230 static int __init proc_genhd_init(void)
1232 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1233 proc_create_seq("partitions", 0, NULL, &partitions_op);
1236 module_init(proc_genhd_init);
1237 #endif /* CONFIG_PROC_FS */
1239 dev_t part_devt(struct gendisk *disk, u8 partno)
1241 struct block_device *part;
1245 part = xa_load(&disk->part_tbl, partno);
1247 devt = part->bd_dev;
1253 dev_t blk_lookup_devt(const char *name, int partno)
1255 dev_t devt = MKDEV(0, 0);
1256 struct class_dev_iter iter;
1259 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1260 while ((dev = class_dev_iter_next(&iter))) {
1261 struct gendisk *disk = dev_to_disk(dev);
1263 if (strcmp(dev_name(dev), name))
1266 if (partno < disk->minors) {
1267 /* We need to return the right devno, even
1268 * if the partition doesn't exist yet.
1270 devt = MKDEV(MAJOR(dev->devt),
1271 MINOR(dev->devt) + partno);
1273 devt = part_devt(disk, partno);
1278 class_dev_iter_exit(&iter);
1282 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1283 struct lock_class_key *lkclass)
1285 struct gendisk *disk;
1287 if (!blk_get_queue(q))
1290 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1294 disk->bdi = bdi_alloc(node_id);
1298 /* bdev_alloc() might need the queue, set before the first call */
1301 disk->part0 = bdev_alloc(disk, 0);
1305 disk->node_id = node_id;
1306 mutex_init(&disk->open_mutex);
1307 xa_init(&disk->part_tbl);
1308 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1309 goto out_destroy_part_tbl;
1311 rand_initialize_disk(disk);
1312 disk_to_dev(disk)->class = &block_class;
1313 disk_to_dev(disk)->type = &disk_type;
1314 device_initialize(disk_to_dev(disk));
1317 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1318 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1319 INIT_LIST_HEAD(&disk->slave_bdevs);
1323 out_destroy_part_tbl:
1324 xa_destroy(&disk->part_tbl);
1325 disk->part0->bd_disk = NULL;
1326 iput(disk->part0->bd_inode);
1335 EXPORT_SYMBOL(__alloc_disk_node);
1337 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1339 struct request_queue *q;
1340 struct gendisk *disk;
1342 q = blk_alloc_queue(node, false);
1346 disk = __alloc_disk_node(q, node, lkclass);
1348 blk_cleanup_queue(q);
1353 EXPORT_SYMBOL(__blk_alloc_disk);
1356 * put_disk - decrements the gendisk refcount
1357 * @disk: the struct gendisk to decrement the refcount for
1359 * This decrements the refcount for the struct gendisk. When this reaches 0
1360 * we'll have disk_release() called.
1362 * Context: Any context, but the last reference must not be dropped from
1365 void put_disk(struct gendisk *disk)
1368 put_device(disk_to_dev(disk));
1370 EXPORT_SYMBOL(put_disk);
1373 * blk_cleanup_disk - shutdown a gendisk allocated by blk_alloc_disk
1374 * @disk: gendisk to shutdown
1376 * Mark the queue hanging off @disk DYING, drain all pending requests, then mark
1377 * the queue DEAD, destroy and put it and the gendisk structure.
1379 * Context: can sleep
1381 void blk_cleanup_disk(struct gendisk *disk)
1383 blk_cleanup_queue(disk->queue);
1386 EXPORT_SYMBOL(blk_cleanup_disk);
1388 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1390 char event[] = "DISK_RO=1";
1391 char *envp[] = { event, NULL };
1395 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1399 * set_disk_ro - set a gendisk read-only
1400 * @disk: gendisk to operate on
1401 * @read_only: %true to set the disk read-only, %false set the disk read/write
1403 * This function is used to indicate whether a given disk device should have its
1404 * read-only flag set. set_disk_ro() is typically used by device drivers to
1405 * indicate whether the underlying physical device is write-protected.
1407 void set_disk_ro(struct gendisk *disk, bool read_only)
1410 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1413 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1416 set_disk_ro_uevent(disk, read_only);
1418 EXPORT_SYMBOL(set_disk_ro);
1420 void inc_diskseq(struct gendisk *disk)
1422 disk->diskseq = atomic64_inc_return(&diskseq);