Loading...
1/*
2 * gendisk handling
3 */
4
5#include <linux/module.h>
6#include <linux/fs.h>
7#include <linux/genhd.h>
8#include <linux/kdev_t.h>
9#include <linux/kernel.h>
10#include <linux/blkdev.h>
11#include <linux/init.h>
12#include <linux/spinlock.h>
13#include <linux/proc_fs.h>
14#include <linux/seq_file.h>
15#include <linux/slab.h>
16#include <linux/kmod.h>
17#include <linux/kobj_map.h>
18#include <linux/buffer_head.h>
19#include <linux/mutex.h>
20#include <linux/idr.h>
21#include <linux/log2.h>
22
23#include "blk.h"
24
25static DEFINE_MUTEX(block_class_lock);
26struct kobject *block_depr;
27
28/* for extended dynamic devt allocation, currently only one major is used */
29#define MAX_EXT_DEVT (1 << MINORBITS)
30
31/* For extended devt allocation. ext_devt_mutex prevents look up
32 * results from going away underneath its user.
33 */
34static DEFINE_MUTEX(ext_devt_mutex);
35static DEFINE_IDR(ext_devt_idr);
36
37static struct device_type disk_type;
38
39static void disk_add_events(struct gendisk *disk);
40static void disk_del_events(struct gendisk *disk);
41static void disk_release_events(struct gendisk *disk);
42
43/**
44 * disk_get_part - get partition
45 * @disk: disk to look partition from
46 * @partno: partition number
47 *
48 * Look for partition @partno from @disk. If found, increment
49 * reference count and return it.
50 *
51 * CONTEXT:
52 * Don't care.
53 *
54 * RETURNS:
55 * Pointer to the found partition on success, NULL if not found.
56 */
57struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
58{
59 struct hd_struct *part = NULL;
60 struct disk_part_tbl *ptbl;
61
62 if (unlikely(partno < 0))
63 return NULL;
64
65 rcu_read_lock();
66
67 ptbl = rcu_dereference(disk->part_tbl);
68 if (likely(partno < ptbl->len)) {
69 part = rcu_dereference(ptbl->part[partno]);
70 if (part)
71 get_device(part_to_dev(part));
72 }
73
74 rcu_read_unlock();
75
76 return part;
77}
78EXPORT_SYMBOL_GPL(disk_get_part);
79
80/**
81 * disk_part_iter_init - initialize partition iterator
82 * @piter: iterator to initialize
83 * @disk: disk to iterate over
84 * @flags: DISK_PITER_* flags
85 *
86 * Initialize @piter so that it iterates over partitions of @disk.
87 *
88 * CONTEXT:
89 * Don't care.
90 */
91void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
92 unsigned int flags)
93{
94 struct disk_part_tbl *ptbl;
95
96 rcu_read_lock();
97 ptbl = rcu_dereference(disk->part_tbl);
98
99 piter->disk = disk;
100 piter->part = NULL;
101
102 if (flags & DISK_PITER_REVERSE)
103 piter->idx = ptbl->len - 1;
104 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
105 piter->idx = 0;
106 else
107 piter->idx = 1;
108
109 piter->flags = flags;
110
111 rcu_read_unlock();
112}
113EXPORT_SYMBOL_GPL(disk_part_iter_init);
114
115/**
116 * disk_part_iter_next - proceed iterator to the next partition and return it
117 * @piter: iterator of interest
118 *
119 * Proceed @piter to the next partition and return it.
120 *
121 * CONTEXT:
122 * Don't care.
123 */
124struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
125{
126 struct disk_part_tbl *ptbl;
127 int inc, end;
128
129 /* put the last partition */
130 disk_put_part(piter->part);
131 piter->part = NULL;
132
133 /* get part_tbl */
134 rcu_read_lock();
135 ptbl = rcu_dereference(piter->disk->part_tbl);
136
137 /* determine iteration parameters */
138 if (piter->flags & DISK_PITER_REVERSE) {
139 inc = -1;
140 if (piter->flags & (DISK_PITER_INCL_PART0 |
141 DISK_PITER_INCL_EMPTY_PART0))
142 end = -1;
143 else
144 end = 0;
145 } else {
146 inc = 1;
147 end = ptbl->len;
148 }
149
150 /* iterate to the next partition */
151 for (; piter->idx != end; piter->idx += inc) {
152 struct hd_struct *part;
153
154 part = rcu_dereference(ptbl->part[piter->idx]);
155 if (!part)
156 continue;
157 if (!part->nr_sects &&
158 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
159 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
160 piter->idx == 0))
161 continue;
162
163 get_device(part_to_dev(part));
164 piter->part = part;
165 piter->idx += inc;
166 break;
167 }
168
169 rcu_read_unlock();
170
171 return piter->part;
172}
173EXPORT_SYMBOL_GPL(disk_part_iter_next);
174
175/**
176 * disk_part_iter_exit - finish up partition iteration
177 * @piter: iter of interest
178 *
179 * Called when iteration is over. Cleans up @piter.
180 *
181 * CONTEXT:
182 * Don't care.
183 */
184void disk_part_iter_exit(struct disk_part_iter *piter)
185{
186 disk_put_part(piter->part);
187 piter->part = NULL;
188}
189EXPORT_SYMBOL_GPL(disk_part_iter_exit);
190
191static inline int sector_in_part(struct hd_struct *part, sector_t sector)
192{
193 return part->start_sect <= sector &&
194 sector < part->start_sect + part->nr_sects;
195}
196
197/**
198 * disk_map_sector_rcu - map sector to partition
199 * @disk: gendisk of interest
200 * @sector: sector to map
201 *
202 * Find out which partition @sector maps to on @disk. This is
203 * primarily used for stats accounting.
204 *
205 * CONTEXT:
206 * RCU read locked. The returned partition pointer is valid only
207 * while preemption is disabled.
208 *
209 * RETURNS:
210 * Found partition on success, part0 is returned if no partition matches
211 */
212struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
213{
214 struct disk_part_tbl *ptbl;
215 struct hd_struct *part;
216 int i;
217
218 ptbl = rcu_dereference(disk->part_tbl);
219
220 part = rcu_dereference(ptbl->last_lookup);
221 if (part && sector_in_part(part, sector))
222 return part;
223
224 for (i = 1; i < ptbl->len; i++) {
225 part = rcu_dereference(ptbl->part[i]);
226
227 if (part && sector_in_part(part, sector)) {
228 rcu_assign_pointer(ptbl->last_lookup, part);
229 return part;
230 }
231 }
232 return &disk->part0;
233}
234EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
235
236/*
237 * Can be deleted altogether. Later.
238 *
239 */
240static struct blk_major_name {
241 struct blk_major_name *next;
242 int major;
243 char name[16];
244} *major_names[BLKDEV_MAJOR_HASH_SIZE];
245
246/* index in the above - for now: assume no multimajor ranges */
247static inline int major_to_index(unsigned major)
248{
249 return major % BLKDEV_MAJOR_HASH_SIZE;
250}
251
252#ifdef CONFIG_PROC_FS
253void blkdev_show(struct seq_file *seqf, off_t offset)
254{
255 struct blk_major_name *dp;
256
257 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
258 mutex_lock(&block_class_lock);
259 for (dp = major_names[offset]; dp; dp = dp->next)
260 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
261 mutex_unlock(&block_class_lock);
262 }
263}
264#endif /* CONFIG_PROC_FS */
265
266/**
267 * register_blkdev - register a new block device
268 *
269 * @major: the requested major device number [1..255]. If @major=0, try to
270 * allocate any unused major number.
271 * @name: the name of the new block device as a zero terminated string
272 *
273 * The @name must be unique within the system.
274 *
275 * The return value depends on the @major input parameter.
276 * - if a major device number was requested in range [1..255] then the
277 * function returns zero on success, or a negative error code
278 * - if any unused major number was requested with @major=0 parameter
279 * then the return value is the allocated major number in range
280 * [1..255] or a negative error code otherwise
281 */
282int register_blkdev(unsigned int major, const char *name)
283{
284 struct blk_major_name **n, *p;
285 int index, ret = 0;
286
287 mutex_lock(&block_class_lock);
288
289 /* temporary */
290 if (major == 0) {
291 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
292 if (major_names[index] == NULL)
293 break;
294 }
295
296 if (index == 0) {
297 printk("register_blkdev: failed to get major for %s\n",
298 name);
299 ret = -EBUSY;
300 goto out;
301 }
302 major = index;
303 ret = major;
304 }
305
306 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
307 if (p == NULL) {
308 ret = -ENOMEM;
309 goto out;
310 }
311
312 p->major = major;
313 strlcpy(p->name, name, sizeof(p->name));
314 p->next = NULL;
315 index = major_to_index(major);
316
317 for (n = &major_names[index]; *n; n = &(*n)->next) {
318 if ((*n)->major == major)
319 break;
320 }
321 if (!*n)
322 *n = p;
323 else
324 ret = -EBUSY;
325
326 if (ret < 0) {
327 printk("register_blkdev: cannot get major %d for %s\n",
328 major, name);
329 kfree(p);
330 }
331out:
332 mutex_unlock(&block_class_lock);
333 return ret;
334}
335
336EXPORT_SYMBOL(register_blkdev);
337
338void unregister_blkdev(unsigned int major, const char *name)
339{
340 struct blk_major_name **n;
341 struct blk_major_name *p = NULL;
342 int index = major_to_index(major);
343
344 mutex_lock(&block_class_lock);
345 for (n = &major_names[index]; *n; n = &(*n)->next)
346 if ((*n)->major == major)
347 break;
348 if (!*n || strcmp((*n)->name, name)) {
349 WARN_ON(1);
350 } else {
351 p = *n;
352 *n = p->next;
353 }
354 mutex_unlock(&block_class_lock);
355 kfree(p);
356}
357
358EXPORT_SYMBOL(unregister_blkdev);
359
360static struct kobj_map *bdev_map;
361
362/**
363 * blk_mangle_minor - scatter minor numbers apart
364 * @minor: minor number to mangle
365 *
366 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
367 * is enabled. Mangling twice gives the original value.
368 *
369 * RETURNS:
370 * Mangled value.
371 *
372 * CONTEXT:
373 * Don't care.
374 */
375static int blk_mangle_minor(int minor)
376{
377#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
378 int i;
379
380 for (i = 0; i < MINORBITS / 2; i++) {
381 int low = minor & (1 << i);
382 int high = minor & (1 << (MINORBITS - 1 - i));
383 int distance = MINORBITS - 1 - 2 * i;
384
385 minor ^= low | high; /* clear both bits */
386 low <<= distance; /* swap the positions */
387 high >>= distance;
388 minor |= low | high; /* and set */
389 }
390#endif
391 return minor;
392}
393
394/**
395 * blk_alloc_devt - allocate a dev_t for a partition
396 * @part: partition to allocate dev_t for
397 * @devt: out parameter for resulting dev_t
398 *
399 * Allocate a dev_t for block device.
400 *
401 * RETURNS:
402 * 0 on success, allocated dev_t is returned in *@devt. -errno on
403 * failure.
404 *
405 * CONTEXT:
406 * Might sleep.
407 */
408int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
409{
410 struct gendisk *disk = part_to_disk(part);
411 int idx, rc;
412
413 /* in consecutive minor range? */
414 if (part->partno < disk->minors) {
415 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
416 return 0;
417 }
418
419 /* allocate ext devt */
420 do {
421 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
422 return -ENOMEM;
423 rc = idr_get_new(&ext_devt_idr, part, &idx);
424 } while (rc == -EAGAIN);
425
426 if (rc)
427 return rc;
428
429 if (idx > MAX_EXT_DEVT) {
430 idr_remove(&ext_devt_idr, idx);
431 return -EBUSY;
432 }
433
434 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
435 return 0;
436}
437
438/**
439 * blk_free_devt - free a dev_t
440 * @devt: dev_t to free
441 *
442 * Free @devt which was allocated using blk_alloc_devt().
443 *
444 * CONTEXT:
445 * Might sleep.
446 */
447void blk_free_devt(dev_t devt)
448{
449 might_sleep();
450
451 if (devt == MKDEV(0, 0))
452 return;
453
454 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
455 mutex_lock(&ext_devt_mutex);
456 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
457 mutex_unlock(&ext_devt_mutex);
458 }
459}
460
461static char *bdevt_str(dev_t devt, char *buf)
462{
463 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
464 char tbuf[BDEVT_SIZE];
465 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
466 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
467 } else
468 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
469
470 return buf;
471}
472
473/*
474 * Register device numbers dev..(dev+range-1)
475 * range must be nonzero
476 * The hash chain is sorted on range, so that subranges can override.
477 */
478void blk_register_region(dev_t devt, unsigned long range, struct module *module,
479 struct kobject *(*probe)(dev_t, int *, void *),
480 int (*lock)(dev_t, void *), void *data)
481{
482 kobj_map(bdev_map, devt, range, module, probe, lock, data);
483}
484
485EXPORT_SYMBOL(blk_register_region);
486
487void blk_unregister_region(dev_t devt, unsigned long range)
488{
489 kobj_unmap(bdev_map, devt, range);
490}
491
492EXPORT_SYMBOL(blk_unregister_region);
493
494static struct kobject *exact_match(dev_t devt, int *partno, void *data)
495{
496 struct gendisk *p = data;
497
498 return &disk_to_dev(p)->kobj;
499}
500
501static int exact_lock(dev_t devt, void *data)
502{
503 struct gendisk *p = data;
504
505 if (!get_disk(p))
506 return -1;
507 return 0;
508}
509
510void register_disk(struct gendisk *disk)
511{
512 struct device *ddev = disk_to_dev(disk);
513 struct block_device *bdev;
514 struct disk_part_iter piter;
515 struct hd_struct *part;
516 int err;
517
518 ddev->parent = disk->driverfs_dev;
519
520 dev_set_name(ddev, disk->disk_name);
521
522 /* delay uevents, until we scanned partition table */
523 dev_set_uevent_suppress(ddev, 1);
524
525 if (device_add(ddev))
526 return;
527 if (!sysfs_deprecated) {
528 err = sysfs_create_link(block_depr, &ddev->kobj,
529 kobject_name(&ddev->kobj));
530 if (err) {
531 device_del(ddev);
532 return;
533 }
534 }
535 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
536 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
537
538 /* No minors to use for partitions */
539 if (!disk_partitionable(disk))
540 goto exit;
541
542 /* No such device (e.g., media were just removed) */
543 if (!get_capacity(disk))
544 goto exit;
545
546 bdev = bdget_disk(disk, 0);
547 if (!bdev)
548 goto exit;
549
550 bdev->bd_invalidated = 1;
551 err = blkdev_get(bdev, FMODE_READ, NULL);
552 if (err < 0)
553 goto exit;
554 blkdev_put(bdev, FMODE_READ);
555
556exit:
557 /* announce disk after possible partitions are created */
558 dev_set_uevent_suppress(ddev, 0);
559 kobject_uevent(&ddev->kobj, KOBJ_ADD);
560
561 /* announce possible partitions */
562 disk_part_iter_init(&piter, disk, 0);
563 while ((part = disk_part_iter_next(&piter)))
564 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
565 disk_part_iter_exit(&piter);
566}
567
568/**
569 * add_disk - add partitioning information to kernel list
570 * @disk: per-device partitioning information
571 *
572 * This function registers the partitioning information in @disk
573 * with the kernel.
574 *
575 * FIXME: error handling
576 */
577void add_disk(struct gendisk *disk)
578{
579 struct backing_dev_info *bdi;
580 dev_t devt;
581 int retval;
582
583 /* minors == 0 indicates to use ext devt from part0 and should
584 * be accompanied with EXT_DEVT flag. Make sure all
585 * parameters make sense.
586 */
587 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
588 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
589
590 disk->flags |= GENHD_FL_UP;
591
592 retval = blk_alloc_devt(&disk->part0, &devt);
593 if (retval) {
594 WARN_ON(1);
595 return;
596 }
597 disk_to_dev(disk)->devt = devt;
598
599 /* ->major and ->first_minor aren't supposed to be
600 * dereferenced from here on, but set them just in case.
601 */
602 disk->major = MAJOR(devt);
603 disk->first_minor = MINOR(devt);
604
605 /* Register BDI before referencing it from bdev */
606 bdi = &disk->queue->backing_dev_info;
607 bdi_register_dev(bdi, disk_devt(disk));
608
609 blk_register_region(disk_devt(disk), disk->minors, NULL,
610 exact_match, exact_lock, disk);
611 register_disk(disk);
612 blk_register_queue(disk);
613
614 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
615 "bdi");
616 WARN_ON(retval);
617
618 disk_add_events(disk);
619}
620EXPORT_SYMBOL(add_disk);
621
622void del_gendisk(struct gendisk *disk)
623{
624 struct disk_part_iter piter;
625 struct hd_struct *part;
626
627 disk_del_events(disk);
628
629 /* invalidate stuff */
630 disk_part_iter_init(&piter, disk,
631 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
632 while ((part = disk_part_iter_next(&piter))) {
633 invalidate_partition(disk, part->partno);
634 delete_partition(disk, part->partno);
635 }
636 disk_part_iter_exit(&piter);
637
638 invalidate_partition(disk, 0);
639 blk_free_devt(disk_to_dev(disk)->devt);
640 set_capacity(disk, 0);
641 disk->flags &= ~GENHD_FL_UP;
642
643 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
644 bdi_unregister(&disk->queue->backing_dev_info);
645 blk_unregister_queue(disk);
646 blk_unregister_region(disk_devt(disk), disk->minors);
647
648 part_stat_set_all(&disk->part0, 0);
649 disk->part0.stamp = 0;
650
651 kobject_put(disk->part0.holder_dir);
652 kobject_put(disk->slave_dir);
653 disk->driverfs_dev = NULL;
654 if (!sysfs_deprecated)
655 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
656 device_del(disk_to_dev(disk));
657}
658EXPORT_SYMBOL(del_gendisk);
659
660/**
661 * get_gendisk - get partitioning information for a given device
662 * @devt: device to get partitioning information for
663 * @partno: returned partition index
664 *
665 * This function gets the structure containing partitioning
666 * information for the given device @devt.
667 */
668struct gendisk *get_gendisk(dev_t devt, int *partno)
669{
670 struct gendisk *disk = NULL;
671
672 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
673 struct kobject *kobj;
674
675 kobj = kobj_lookup(bdev_map, devt, partno);
676 if (kobj)
677 disk = dev_to_disk(kobj_to_dev(kobj));
678 } else {
679 struct hd_struct *part;
680
681 mutex_lock(&ext_devt_mutex);
682 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
683 if (part && get_disk(part_to_disk(part))) {
684 *partno = part->partno;
685 disk = part_to_disk(part);
686 }
687 mutex_unlock(&ext_devt_mutex);
688 }
689
690 return disk;
691}
692EXPORT_SYMBOL(get_gendisk);
693
694/**
695 * bdget_disk - do bdget() by gendisk and partition number
696 * @disk: gendisk of interest
697 * @partno: partition number
698 *
699 * Find partition @partno from @disk, do bdget() on it.
700 *
701 * CONTEXT:
702 * Don't care.
703 *
704 * RETURNS:
705 * Resulting block_device on success, NULL on failure.
706 */
707struct block_device *bdget_disk(struct gendisk *disk, int partno)
708{
709 struct hd_struct *part;
710 struct block_device *bdev = NULL;
711
712 part = disk_get_part(disk, partno);
713 if (part)
714 bdev = bdget(part_devt(part));
715 disk_put_part(part);
716
717 return bdev;
718}
719EXPORT_SYMBOL(bdget_disk);
720
721/*
722 * print a full list of all partitions - intended for places where the root
723 * filesystem can't be mounted and thus to give the victim some idea of what
724 * went wrong
725 */
726void __init printk_all_partitions(void)
727{
728 struct class_dev_iter iter;
729 struct device *dev;
730
731 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
732 while ((dev = class_dev_iter_next(&iter))) {
733 struct gendisk *disk = dev_to_disk(dev);
734 struct disk_part_iter piter;
735 struct hd_struct *part;
736 char name_buf[BDEVNAME_SIZE];
737 char devt_buf[BDEVT_SIZE];
738 u8 uuid[PARTITION_META_INFO_UUIDLTH * 2 + 1];
739
740 /*
741 * Don't show empty devices or things that have been
742 * suppressed
743 */
744 if (get_capacity(disk) == 0 ||
745 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
746 continue;
747
748 /*
749 * Note, unlike /proc/partitions, I am showing the
750 * numbers in hex - the same format as the root=
751 * option takes.
752 */
753 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
754 while ((part = disk_part_iter_next(&piter))) {
755 bool is_part0 = part == &disk->part0;
756
757 uuid[0] = 0;
758 if (part->info)
759 part_unpack_uuid(part->info->uuid, uuid);
760
761 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
762 bdevt_str(part_devt(part), devt_buf),
763 (unsigned long long)part->nr_sects >> 1,
764 disk_name(disk, part->partno, name_buf), uuid);
765 if (is_part0) {
766 if (disk->driverfs_dev != NULL &&
767 disk->driverfs_dev->driver != NULL)
768 printk(" driver: %s\n",
769 disk->driverfs_dev->driver->name);
770 else
771 printk(" (driver?)\n");
772 } else
773 printk("\n");
774 }
775 disk_part_iter_exit(&piter);
776 }
777 class_dev_iter_exit(&iter);
778}
779
780#ifdef CONFIG_PROC_FS
781/* iterator */
782static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
783{
784 loff_t skip = *pos;
785 struct class_dev_iter *iter;
786 struct device *dev;
787
788 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
789 if (!iter)
790 return ERR_PTR(-ENOMEM);
791
792 seqf->private = iter;
793 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
794 do {
795 dev = class_dev_iter_next(iter);
796 if (!dev)
797 return NULL;
798 } while (skip--);
799
800 return dev_to_disk(dev);
801}
802
803static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
804{
805 struct device *dev;
806
807 (*pos)++;
808 dev = class_dev_iter_next(seqf->private);
809 if (dev)
810 return dev_to_disk(dev);
811
812 return NULL;
813}
814
815static void disk_seqf_stop(struct seq_file *seqf, void *v)
816{
817 struct class_dev_iter *iter = seqf->private;
818
819 /* stop is called even after start failed :-( */
820 if (iter) {
821 class_dev_iter_exit(iter);
822 kfree(iter);
823 }
824}
825
826static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
827{
828 static void *p;
829
830 p = disk_seqf_start(seqf, pos);
831 if (!IS_ERR_OR_NULL(p) && !*pos)
832 seq_puts(seqf, "major minor #blocks name\n\n");
833 return p;
834}
835
836static int show_partition(struct seq_file *seqf, void *v)
837{
838 struct gendisk *sgp = v;
839 struct disk_part_iter piter;
840 struct hd_struct *part;
841 char buf[BDEVNAME_SIZE];
842
843 /* Don't show non-partitionable removeable devices or empty devices */
844 if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
845 (sgp->flags & GENHD_FL_REMOVABLE)))
846 return 0;
847 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
848 return 0;
849
850 /* show the full disk and all non-0 size partitions of it */
851 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
852 while ((part = disk_part_iter_next(&piter)))
853 seq_printf(seqf, "%4d %7d %10llu %s\n",
854 MAJOR(part_devt(part)), MINOR(part_devt(part)),
855 (unsigned long long)part->nr_sects >> 1,
856 disk_name(sgp, part->partno, buf));
857 disk_part_iter_exit(&piter);
858
859 return 0;
860}
861
862static const struct seq_operations partitions_op = {
863 .start = show_partition_start,
864 .next = disk_seqf_next,
865 .stop = disk_seqf_stop,
866 .show = show_partition
867};
868
869static int partitions_open(struct inode *inode, struct file *file)
870{
871 return seq_open(file, &partitions_op);
872}
873
874static const struct file_operations proc_partitions_operations = {
875 .open = partitions_open,
876 .read = seq_read,
877 .llseek = seq_lseek,
878 .release = seq_release,
879};
880#endif
881
882
883static struct kobject *base_probe(dev_t devt, int *partno, void *data)
884{
885 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
886 /* Make old-style 2.4 aliases work */
887 request_module("block-major-%d", MAJOR(devt));
888 return NULL;
889}
890
891static int __init genhd_device_init(void)
892{
893 int error;
894
895 block_class.dev_kobj = sysfs_dev_block_kobj;
896 error = class_register(&block_class);
897 if (unlikely(error))
898 return error;
899 bdev_map = kobj_map_init(base_probe, &block_class_lock);
900 blk_dev_init();
901
902 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
903
904 /* create top-level block dir */
905 if (!sysfs_deprecated)
906 block_depr = kobject_create_and_add("block", NULL);
907 return 0;
908}
909
910subsys_initcall(genhd_device_init);
911
912static ssize_t disk_range_show(struct device *dev,
913 struct device_attribute *attr, char *buf)
914{
915 struct gendisk *disk = dev_to_disk(dev);
916
917 return sprintf(buf, "%d\n", disk->minors);
918}
919
920static ssize_t disk_ext_range_show(struct device *dev,
921 struct device_attribute *attr, char *buf)
922{
923 struct gendisk *disk = dev_to_disk(dev);
924
925 return sprintf(buf, "%d\n", disk_max_parts(disk));
926}
927
928static ssize_t disk_removable_show(struct device *dev,
929 struct device_attribute *attr, char *buf)
930{
931 struct gendisk *disk = dev_to_disk(dev);
932
933 return sprintf(buf, "%d\n",
934 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
935}
936
937static ssize_t disk_ro_show(struct device *dev,
938 struct device_attribute *attr, char *buf)
939{
940 struct gendisk *disk = dev_to_disk(dev);
941
942 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
943}
944
945static ssize_t disk_capability_show(struct device *dev,
946 struct device_attribute *attr, char *buf)
947{
948 struct gendisk *disk = dev_to_disk(dev);
949
950 return sprintf(buf, "%x\n", disk->flags);
951}
952
953static ssize_t disk_alignment_offset_show(struct device *dev,
954 struct device_attribute *attr,
955 char *buf)
956{
957 struct gendisk *disk = dev_to_disk(dev);
958
959 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
960}
961
962static ssize_t disk_discard_alignment_show(struct device *dev,
963 struct device_attribute *attr,
964 char *buf)
965{
966 struct gendisk *disk = dev_to_disk(dev);
967
968 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
969}
970
971static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
972static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
973static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
974static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
975static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
976static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
977static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
978 NULL);
979static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
980static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
981static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
982#ifdef CONFIG_FAIL_MAKE_REQUEST
983static struct device_attribute dev_attr_fail =
984 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
985#endif
986#ifdef CONFIG_FAIL_IO_TIMEOUT
987static struct device_attribute dev_attr_fail_timeout =
988 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
989 part_timeout_store);
990#endif
991
992static struct attribute *disk_attrs[] = {
993 &dev_attr_range.attr,
994 &dev_attr_ext_range.attr,
995 &dev_attr_removable.attr,
996 &dev_attr_ro.attr,
997 &dev_attr_size.attr,
998 &dev_attr_alignment_offset.attr,
999 &dev_attr_discard_alignment.attr,
1000 &dev_attr_capability.attr,
1001 &dev_attr_stat.attr,
1002 &dev_attr_inflight.attr,
1003#ifdef CONFIG_FAIL_MAKE_REQUEST
1004 &dev_attr_fail.attr,
1005#endif
1006#ifdef CONFIG_FAIL_IO_TIMEOUT
1007 &dev_attr_fail_timeout.attr,
1008#endif
1009 NULL
1010};
1011
1012static struct attribute_group disk_attr_group = {
1013 .attrs = disk_attrs,
1014};
1015
1016static const struct attribute_group *disk_attr_groups[] = {
1017 &disk_attr_group,
1018 NULL
1019};
1020
1021/**
1022 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1023 * @disk: disk to replace part_tbl for
1024 * @new_ptbl: new part_tbl to install
1025 *
1026 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1027 * original ptbl is freed using RCU callback.
1028 *
1029 * LOCKING:
1030 * Matching bd_mutx locked.
1031 */
1032static void disk_replace_part_tbl(struct gendisk *disk,
1033 struct disk_part_tbl *new_ptbl)
1034{
1035 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1036
1037 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1038
1039 if (old_ptbl) {
1040 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1041 kfree_rcu(old_ptbl, rcu_head);
1042 }
1043}
1044
1045/**
1046 * disk_expand_part_tbl - expand disk->part_tbl
1047 * @disk: disk to expand part_tbl for
1048 * @partno: expand such that this partno can fit in
1049 *
1050 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1051 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1052 *
1053 * LOCKING:
1054 * Matching bd_mutex locked, might sleep.
1055 *
1056 * RETURNS:
1057 * 0 on success, -errno on failure.
1058 */
1059int disk_expand_part_tbl(struct gendisk *disk, int partno)
1060{
1061 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1062 struct disk_part_tbl *new_ptbl;
1063 int len = old_ptbl ? old_ptbl->len : 0;
1064 int target = partno + 1;
1065 size_t size;
1066 int i;
1067
1068 /* disk_max_parts() is zero during initialization, ignore if so */
1069 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1070 return -EINVAL;
1071
1072 if (target <= len)
1073 return 0;
1074
1075 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1076 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1077 if (!new_ptbl)
1078 return -ENOMEM;
1079
1080 new_ptbl->len = target;
1081
1082 for (i = 0; i < len; i++)
1083 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1084
1085 disk_replace_part_tbl(disk, new_ptbl);
1086 return 0;
1087}
1088
1089static void disk_release(struct device *dev)
1090{
1091 struct gendisk *disk = dev_to_disk(dev);
1092
1093 disk_release_events(disk);
1094 kfree(disk->random);
1095 disk_replace_part_tbl(disk, NULL);
1096 free_part_stats(&disk->part0);
1097 free_part_info(&disk->part0);
1098 kfree(disk);
1099}
1100struct class block_class = {
1101 .name = "block",
1102};
1103
1104static char *block_devnode(struct device *dev, mode_t *mode)
1105{
1106 struct gendisk *disk = dev_to_disk(dev);
1107
1108 if (disk->devnode)
1109 return disk->devnode(disk, mode);
1110 return NULL;
1111}
1112
1113static struct device_type disk_type = {
1114 .name = "disk",
1115 .groups = disk_attr_groups,
1116 .release = disk_release,
1117 .devnode = block_devnode,
1118};
1119
1120#ifdef CONFIG_PROC_FS
1121/*
1122 * aggregate disk stat collector. Uses the same stats that the sysfs
1123 * entries do, above, but makes them available through one seq_file.
1124 *
1125 * The output looks suspiciously like /proc/partitions with a bunch of
1126 * extra fields.
1127 */
1128static int diskstats_show(struct seq_file *seqf, void *v)
1129{
1130 struct gendisk *gp = v;
1131 struct disk_part_iter piter;
1132 struct hd_struct *hd;
1133 char buf[BDEVNAME_SIZE];
1134 int cpu;
1135
1136 /*
1137 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1138 seq_puts(seqf, "major minor name"
1139 " rio rmerge rsect ruse wio wmerge "
1140 "wsect wuse running use aveq"
1141 "\n\n");
1142 */
1143
1144 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1145 while ((hd = disk_part_iter_next(&piter))) {
1146 cpu = part_stat_lock();
1147 part_round_stats(cpu, hd);
1148 part_stat_unlock();
1149 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1150 "%u %lu %lu %lu %u %u %u %u\n",
1151 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1152 disk_name(gp, hd->partno, buf),
1153 part_stat_read(hd, ios[READ]),
1154 part_stat_read(hd, merges[READ]),
1155 part_stat_read(hd, sectors[READ]),
1156 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1157 part_stat_read(hd, ios[WRITE]),
1158 part_stat_read(hd, merges[WRITE]),
1159 part_stat_read(hd, sectors[WRITE]),
1160 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1161 part_in_flight(hd),
1162 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1163 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1164 );
1165 }
1166 disk_part_iter_exit(&piter);
1167
1168 return 0;
1169}
1170
1171static const struct seq_operations diskstats_op = {
1172 .start = disk_seqf_start,
1173 .next = disk_seqf_next,
1174 .stop = disk_seqf_stop,
1175 .show = diskstats_show
1176};
1177
1178static int diskstats_open(struct inode *inode, struct file *file)
1179{
1180 return seq_open(file, &diskstats_op);
1181}
1182
1183static const struct file_operations proc_diskstats_operations = {
1184 .open = diskstats_open,
1185 .read = seq_read,
1186 .llseek = seq_lseek,
1187 .release = seq_release,
1188};
1189
1190static int __init proc_genhd_init(void)
1191{
1192 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1193 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1194 return 0;
1195}
1196module_init(proc_genhd_init);
1197#endif /* CONFIG_PROC_FS */
1198
1199dev_t blk_lookup_devt(const char *name, int partno)
1200{
1201 dev_t devt = MKDEV(0, 0);
1202 struct class_dev_iter iter;
1203 struct device *dev;
1204
1205 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1206 while ((dev = class_dev_iter_next(&iter))) {
1207 struct gendisk *disk = dev_to_disk(dev);
1208 struct hd_struct *part;
1209
1210 if (strcmp(dev_name(dev), name))
1211 continue;
1212
1213 if (partno < disk->minors) {
1214 /* We need to return the right devno, even
1215 * if the partition doesn't exist yet.
1216 */
1217 devt = MKDEV(MAJOR(dev->devt),
1218 MINOR(dev->devt) + partno);
1219 break;
1220 }
1221 part = disk_get_part(disk, partno);
1222 if (part) {
1223 devt = part_devt(part);
1224 disk_put_part(part);
1225 break;
1226 }
1227 disk_put_part(part);
1228 }
1229 class_dev_iter_exit(&iter);
1230 return devt;
1231}
1232EXPORT_SYMBOL(blk_lookup_devt);
1233
1234struct gendisk *alloc_disk(int minors)
1235{
1236 return alloc_disk_node(minors, -1);
1237}
1238EXPORT_SYMBOL(alloc_disk);
1239
1240struct gendisk *alloc_disk_node(int minors, int node_id)
1241{
1242 struct gendisk *disk;
1243
1244 disk = kmalloc_node(sizeof(struct gendisk),
1245 GFP_KERNEL | __GFP_ZERO, node_id);
1246 if (disk) {
1247 if (!init_part_stats(&disk->part0)) {
1248 kfree(disk);
1249 return NULL;
1250 }
1251 disk->node_id = node_id;
1252 if (disk_expand_part_tbl(disk, 0)) {
1253 free_part_stats(&disk->part0);
1254 kfree(disk);
1255 return NULL;
1256 }
1257 disk->part_tbl->part[0] = &disk->part0;
1258
1259 hd_ref_init(&disk->part0);
1260
1261 disk->minors = minors;
1262 rand_initialize_disk(disk);
1263 disk_to_dev(disk)->class = &block_class;
1264 disk_to_dev(disk)->type = &disk_type;
1265 device_initialize(disk_to_dev(disk));
1266 }
1267 return disk;
1268}
1269EXPORT_SYMBOL(alloc_disk_node);
1270
1271struct kobject *get_disk(struct gendisk *disk)
1272{
1273 struct module *owner;
1274 struct kobject *kobj;
1275
1276 if (!disk->fops)
1277 return NULL;
1278 owner = disk->fops->owner;
1279 if (owner && !try_module_get(owner))
1280 return NULL;
1281 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1282 if (kobj == NULL) {
1283 module_put(owner);
1284 return NULL;
1285 }
1286 return kobj;
1287
1288}
1289
1290EXPORT_SYMBOL(get_disk);
1291
1292void put_disk(struct gendisk *disk)
1293{
1294 if (disk)
1295 kobject_put(&disk_to_dev(disk)->kobj);
1296}
1297
1298EXPORT_SYMBOL(put_disk);
1299
1300static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1301{
1302 char event[] = "DISK_RO=1";
1303 char *envp[] = { event, NULL };
1304
1305 if (!ro)
1306 event[8] = '0';
1307 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1308}
1309
1310void set_device_ro(struct block_device *bdev, int flag)
1311{
1312 bdev->bd_part->policy = flag;
1313}
1314
1315EXPORT_SYMBOL(set_device_ro);
1316
1317void set_disk_ro(struct gendisk *disk, int flag)
1318{
1319 struct disk_part_iter piter;
1320 struct hd_struct *part;
1321
1322 if (disk->part0.policy != flag) {
1323 set_disk_ro_uevent(disk, flag);
1324 disk->part0.policy = flag;
1325 }
1326
1327 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1328 while ((part = disk_part_iter_next(&piter)))
1329 part->policy = flag;
1330 disk_part_iter_exit(&piter);
1331}
1332
1333EXPORT_SYMBOL(set_disk_ro);
1334
1335int bdev_read_only(struct block_device *bdev)
1336{
1337 if (!bdev)
1338 return 0;
1339 return bdev->bd_part->policy;
1340}
1341
1342EXPORT_SYMBOL(bdev_read_only);
1343
1344int invalidate_partition(struct gendisk *disk, int partno)
1345{
1346 int res = 0;
1347 struct block_device *bdev = bdget_disk(disk, partno);
1348 if (bdev) {
1349 fsync_bdev(bdev);
1350 res = __invalidate_device(bdev, true);
1351 bdput(bdev);
1352 }
1353 return res;
1354}
1355
1356EXPORT_SYMBOL(invalidate_partition);
1357
1358/*
1359 * Disk events - monitor disk events like media change and eject request.
1360 */
1361struct disk_events {
1362 struct list_head node; /* all disk_event's */
1363 struct gendisk *disk; /* the associated disk */
1364 spinlock_t lock;
1365
1366 struct mutex block_mutex; /* protects blocking */
1367 int block; /* event blocking depth */
1368 unsigned int pending; /* events already sent out */
1369 unsigned int clearing; /* events being cleared */
1370
1371 long poll_msecs; /* interval, -1 for default */
1372 struct delayed_work dwork;
1373};
1374
1375static const char *disk_events_strs[] = {
1376 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1377 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1378};
1379
1380static char *disk_uevents[] = {
1381 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1382 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1383};
1384
1385/* list of all disk_events */
1386static DEFINE_MUTEX(disk_events_mutex);
1387static LIST_HEAD(disk_events);
1388
1389/* disable in-kernel polling by default */
1390static unsigned long disk_events_dfl_poll_msecs = 0;
1391
1392static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1393{
1394 struct disk_events *ev = disk->ev;
1395 long intv_msecs = 0;
1396
1397 /*
1398 * If device-specific poll interval is set, always use it. If
1399 * the default is being used, poll iff there are events which
1400 * can't be monitored asynchronously.
1401 */
1402 if (ev->poll_msecs >= 0)
1403 intv_msecs = ev->poll_msecs;
1404 else if (disk->events & ~disk->async_events)
1405 intv_msecs = disk_events_dfl_poll_msecs;
1406
1407 return msecs_to_jiffies(intv_msecs);
1408}
1409
1410/**
1411 * disk_block_events - block and flush disk event checking
1412 * @disk: disk to block events for
1413 *
1414 * On return from this function, it is guaranteed that event checking
1415 * isn't in progress and won't happen until unblocked by
1416 * disk_unblock_events(). Events blocking is counted and the actual
1417 * unblocking happens after the matching number of unblocks are done.
1418 *
1419 * Note that this intentionally does not block event checking from
1420 * disk_clear_events().
1421 *
1422 * CONTEXT:
1423 * Might sleep.
1424 */
1425void disk_block_events(struct gendisk *disk)
1426{
1427 struct disk_events *ev = disk->ev;
1428 unsigned long flags;
1429 bool cancel;
1430
1431 if (!ev)
1432 return;
1433
1434 /*
1435 * Outer mutex ensures that the first blocker completes canceling
1436 * the event work before further blockers are allowed to finish.
1437 */
1438 mutex_lock(&ev->block_mutex);
1439
1440 spin_lock_irqsave(&ev->lock, flags);
1441 cancel = !ev->block++;
1442 spin_unlock_irqrestore(&ev->lock, flags);
1443
1444 if (cancel)
1445 cancel_delayed_work_sync(&disk->ev->dwork);
1446
1447 mutex_unlock(&ev->block_mutex);
1448}
1449
1450static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1451{
1452 struct disk_events *ev = disk->ev;
1453 unsigned long intv;
1454 unsigned long flags;
1455
1456 spin_lock_irqsave(&ev->lock, flags);
1457
1458 if (WARN_ON_ONCE(ev->block <= 0))
1459 goto out_unlock;
1460
1461 if (--ev->block)
1462 goto out_unlock;
1463
1464 /*
1465 * Not exactly a latency critical operation, set poll timer
1466 * slack to 25% and kick event check.
1467 */
1468 intv = disk_events_poll_jiffies(disk);
1469 set_timer_slack(&ev->dwork.timer, intv / 4);
1470 if (check_now)
1471 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1472 else if (intv)
1473 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1474out_unlock:
1475 spin_unlock_irqrestore(&ev->lock, flags);
1476}
1477
1478/**
1479 * disk_unblock_events - unblock disk event checking
1480 * @disk: disk to unblock events for
1481 *
1482 * Undo disk_block_events(). When the block count reaches zero, it
1483 * starts events polling if configured.
1484 *
1485 * CONTEXT:
1486 * Don't care. Safe to call from irq context.
1487 */
1488void disk_unblock_events(struct gendisk *disk)
1489{
1490 if (disk->ev)
1491 __disk_unblock_events(disk, false);
1492}
1493
1494/**
1495 * disk_flush_events - schedule immediate event checking and flushing
1496 * @disk: disk to check and flush events for
1497 * @mask: events to flush
1498 *
1499 * Schedule immediate event checking on @disk if not blocked. Events in
1500 * @mask are scheduled to be cleared from the driver. Note that this
1501 * doesn't clear the events from @disk->ev.
1502 *
1503 * CONTEXT:
1504 * If @mask is non-zero must be called with bdev->bd_mutex held.
1505 */
1506void disk_flush_events(struct gendisk *disk, unsigned int mask)
1507{
1508 struct disk_events *ev = disk->ev;
1509
1510 if (!ev)
1511 return;
1512
1513 spin_lock_irq(&ev->lock);
1514 ev->clearing |= mask;
1515 if (!ev->block) {
1516 cancel_delayed_work(&ev->dwork);
1517 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1518 }
1519 spin_unlock_irq(&ev->lock);
1520}
1521
1522/**
1523 * disk_clear_events - synchronously check, clear and return pending events
1524 * @disk: disk to fetch and clear events from
1525 * @mask: mask of events to be fetched and clearted
1526 *
1527 * Disk events are synchronously checked and pending events in @mask
1528 * are cleared and returned. This ignores the block count.
1529 *
1530 * CONTEXT:
1531 * Might sleep.
1532 */
1533unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1534{
1535 const struct block_device_operations *bdops = disk->fops;
1536 struct disk_events *ev = disk->ev;
1537 unsigned int pending;
1538
1539 if (!ev) {
1540 /* for drivers still using the old ->media_changed method */
1541 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1542 bdops->media_changed && bdops->media_changed(disk))
1543 return DISK_EVENT_MEDIA_CHANGE;
1544 return 0;
1545 }
1546
1547 /* tell the workfn about the events being cleared */
1548 spin_lock_irq(&ev->lock);
1549 ev->clearing |= mask;
1550 spin_unlock_irq(&ev->lock);
1551
1552 /* uncondtionally schedule event check and wait for it to finish */
1553 disk_block_events(disk);
1554 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1555 flush_delayed_work(&ev->dwork);
1556 __disk_unblock_events(disk, false);
1557
1558 /* then, fetch and clear pending events */
1559 spin_lock_irq(&ev->lock);
1560 WARN_ON_ONCE(ev->clearing & mask); /* cleared by workfn */
1561 pending = ev->pending & mask;
1562 ev->pending &= ~mask;
1563 spin_unlock_irq(&ev->lock);
1564
1565 return pending;
1566}
1567
1568static void disk_events_workfn(struct work_struct *work)
1569{
1570 struct delayed_work *dwork = to_delayed_work(work);
1571 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1572 struct gendisk *disk = ev->disk;
1573 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1574 unsigned int clearing = ev->clearing;
1575 unsigned int events;
1576 unsigned long intv;
1577 int nr_events = 0, i;
1578
1579 /* check events */
1580 events = disk->fops->check_events(disk, clearing);
1581
1582 /* accumulate pending events and schedule next poll if necessary */
1583 spin_lock_irq(&ev->lock);
1584
1585 events &= ~ev->pending;
1586 ev->pending |= events;
1587 ev->clearing &= ~clearing;
1588
1589 intv = disk_events_poll_jiffies(disk);
1590 if (!ev->block && intv)
1591 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1592
1593 spin_unlock_irq(&ev->lock);
1594
1595 /*
1596 * Tell userland about new events. Only the events listed in
1597 * @disk->events are reported. Unlisted events are processed the
1598 * same internally but never get reported to userland.
1599 */
1600 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1601 if (events & disk->events & (1 << i))
1602 envp[nr_events++] = disk_uevents[i];
1603
1604 if (nr_events)
1605 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1606}
1607
1608/*
1609 * A disk events enabled device has the following sysfs nodes under
1610 * its /sys/block/X/ directory.
1611 *
1612 * events : list of all supported events
1613 * events_async : list of events which can be detected w/o polling
1614 * events_poll_msecs : polling interval, 0: disable, -1: system default
1615 */
1616static ssize_t __disk_events_show(unsigned int events, char *buf)
1617{
1618 const char *delim = "";
1619 ssize_t pos = 0;
1620 int i;
1621
1622 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1623 if (events & (1 << i)) {
1624 pos += sprintf(buf + pos, "%s%s",
1625 delim, disk_events_strs[i]);
1626 delim = " ";
1627 }
1628 if (pos)
1629 pos += sprintf(buf + pos, "\n");
1630 return pos;
1631}
1632
1633static ssize_t disk_events_show(struct device *dev,
1634 struct device_attribute *attr, char *buf)
1635{
1636 struct gendisk *disk = dev_to_disk(dev);
1637
1638 return __disk_events_show(disk->events, buf);
1639}
1640
1641static ssize_t disk_events_async_show(struct device *dev,
1642 struct device_attribute *attr, char *buf)
1643{
1644 struct gendisk *disk = dev_to_disk(dev);
1645
1646 return __disk_events_show(disk->async_events, buf);
1647}
1648
1649static ssize_t disk_events_poll_msecs_show(struct device *dev,
1650 struct device_attribute *attr,
1651 char *buf)
1652{
1653 struct gendisk *disk = dev_to_disk(dev);
1654
1655 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1656}
1657
1658static ssize_t disk_events_poll_msecs_store(struct device *dev,
1659 struct device_attribute *attr,
1660 const char *buf, size_t count)
1661{
1662 struct gendisk *disk = dev_to_disk(dev);
1663 long intv;
1664
1665 if (!count || !sscanf(buf, "%ld", &intv))
1666 return -EINVAL;
1667
1668 if (intv < 0 && intv != -1)
1669 return -EINVAL;
1670
1671 disk_block_events(disk);
1672 disk->ev->poll_msecs = intv;
1673 __disk_unblock_events(disk, true);
1674
1675 return count;
1676}
1677
1678static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1679static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1680static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1681 disk_events_poll_msecs_show,
1682 disk_events_poll_msecs_store);
1683
1684static const struct attribute *disk_events_attrs[] = {
1685 &dev_attr_events.attr,
1686 &dev_attr_events_async.attr,
1687 &dev_attr_events_poll_msecs.attr,
1688 NULL,
1689};
1690
1691/*
1692 * The default polling interval can be specified by the kernel
1693 * parameter block.events_dfl_poll_msecs which defaults to 0
1694 * (disable). This can also be modified runtime by writing to
1695 * /sys/module/block/events_dfl_poll_msecs.
1696 */
1697static int disk_events_set_dfl_poll_msecs(const char *val,
1698 const struct kernel_param *kp)
1699{
1700 struct disk_events *ev;
1701 int ret;
1702
1703 ret = param_set_ulong(val, kp);
1704 if (ret < 0)
1705 return ret;
1706
1707 mutex_lock(&disk_events_mutex);
1708
1709 list_for_each_entry(ev, &disk_events, node)
1710 disk_flush_events(ev->disk, 0);
1711
1712 mutex_unlock(&disk_events_mutex);
1713
1714 return 0;
1715}
1716
1717static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1718 .set = disk_events_set_dfl_poll_msecs,
1719 .get = param_get_ulong,
1720};
1721
1722#undef MODULE_PARAM_PREFIX
1723#define MODULE_PARAM_PREFIX "block."
1724
1725module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1726 &disk_events_dfl_poll_msecs, 0644);
1727
1728/*
1729 * disk_{add|del|release}_events - initialize and destroy disk_events.
1730 */
1731static void disk_add_events(struct gendisk *disk)
1732{
1733 struct disk_events *ev;
1734
1735 if (!disk->fops->check_events)
1736 return;
1737
1738 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1739 if (!ev) {
1740 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1741 return;
1742 }
1743
1744 if (sysfs_create_files(&disk_to_dev(disk)->kobj,
1745 disk_events_attrs) < 0) {
1746 pr_warn("%s: failed to create sysfs files for events\n",
1747 disk->disk_name);
1748 kfree(ev);
1749 return;
1750 }
1751
1752 disk->ev = ev;
1753
1754 INIT_LIST_HEAD(&ev->node);
1755 ev->disk = disk;
1756 spin_lock_init(&ev->lock);
1757 mutex_init(&ev->block_mutex);
1758 ev->block = 1;
1759 ev->poll_msecs = -1;
1760 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1761
1762 mutex_lock(&disk_events_mutex);
1763 list_add_tail(&ev->node, &disk_events);
1764 mutex_unlock(&disk_events_mutex);
1765
1766 /*
1767 * Block count is initialized to 1 and the following initial
1768 * unblock kicks it into action.
1769 */
1770 __disk_unblock_events(disk, true);
1771}
1772
1773static void disk_del_events(struct gendisk *disk)
1774{
1775 if (!disk->ev)
1776 return;
1777
1778 disk_block_events(disk);
1779
1780 mutex_lock(&disk_events_mutex);
1781 list_del_init(&disk->ev->node);
1782 mutex_unlock(&disk_events_mutex);
1783
1784 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1785}
1786
1787static void disk_release_events(struct gendisk *disk)
1788{
1789 /* the block count should be 1 from disk_del_events() */
1790 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1791 kfree(disk->ev);
1792}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * gendisk handling
4 *
5 * Portions Copyright (C) 2020 Christoph Hellwig
6 */
7
8#include <linux/module.h>
9#include <linux/ctype.h>
10#include <linux/fs.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 <linux/blktrace_api.h>
29
30#include "blk-throttle.h"
31#include "blk.h"
32#include "blk-mq-sched.h"
33#include "blk-rq-qos.h"
34#include "blk-cgroup.h"
35
36static struct kobject *block_depr;
37
38/*
39 * Unique, monotonically increasing sequential number associated with block
40 * devices instances (i.e. incremented each time a device is attached).
41 * Associating uevents with block devices in userspace is difficult and racy:
42 * the uevent netlink socket is lossy, and on slow and overloaded systems has
43 * a very high latency.
44 * Block devices do not have exclusive owners in userspace, any process can set
45 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
46 * can be reused again and again).
47 * A userspace process setting up a block device and watching for its events
48 * cannot thus reliably tell whether an event relates to the device it just set
49 * up or another earlier instance with the same name.
50 * This sequential number allows userspace processes to solve this problem, and
51 * uniquely associate an uevent to the lifetime to a device.
52 */
53static atomic64_t diskseq;
54
55/* for extended dynamic devt allocation, currently only one major is used */
56#define NR_EXT_DEVT (1 << MINORBITS)
57static DEFINE_IDA(ext_devt_ida);
58
59void set_capacity(struct gendisk *disk, sector_t sectors)
60{
61 bdev_set_nr_sectors(disk->part0, sectors);
62}
63EXPORT_SYMBOL(set_capacity);
64
65/*
66 * Set disk capacity and notify if the size is not currently zero and will not
67 * be set to zero. Returns true if a uevent was sent, otherwise false.
68 */
69bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
70{
71 sector_t capacity = get_capacity(disk);
72 char *envp[] = { "RESIZE=1", NULL };
73
74 set_capacity(disk, size);
75
76 /*
77 * Only print a message and send a uevent if the gendisk is user visible
78 * and alive. This avoids spamming the log and udev when setting the
79 * initial capacity during probing.
80 */
81 if (size == capacity ||
82 !disk_live(disk) ||
83 (disk->flags & GENHD_FL_HIDDEN))
84 return false;
85
86 pr_info("%s: detected capacity change from %lld to %lld\n",
87 disk->disk_name, capacity, size);
88
89 /*
90 * Historically we did not send a uevent for changes to/from an empty
91 * device.
92 */
93 if (!capacity || !size)
94 return false;
95 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
96 return true;
97}
98EXPORT_SYMBOL_GPL(set_capacity_and_notify);
99
100static void part_stat_read_all(struct block_device *part,
101 struct disk_stats *stat)
102{
103 int cpu;
104
105 memset(stat, 0, sizeof(struct disk_stats));
106 for_each_possible_cpu(cpu) {
107 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
108 int group;
109
110 for (group = 0; group < NR_STAT_GROUPS; group++) {
111 stat->nsecs[group] += ptr->nsecs[group];
112 stat->sectors[group] += ptr->sectors[group];
113 stat->ios[group] += ptr->ios[group];
114 stat->merges[group] += ptr->merges[group];
115 }
116
117 stat->io_ticks += ptr->io_ticks;
118 }
119}
120
121static unsigned int part_in_flight(struct block_device *part)
122{
123 unsigned int inflight = 0;
124 int cpu;
125
126 for_each_possible_cpu(cpu) {
127 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
128 part_stat_local_read_cpu(part, in_flight[1], cpu);
129 }
130 if ((int)inflight < 0)
131 inflight = 0;
132
133 return inflight;
134}
135
136static void part_in_flight_rw(struct block_device *part,
137 unsigned int inflight[2])
138{
139 int cpu;
140
141 inflight[0] = 0;
142 inflight[1] = 0;
143 for_each_possible_cpu(cpu) {
144 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
145 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
146 }
147 if ((int)inflight[0] < 0)
148 inflight[0] = 0;
149 if ((int)inflight[1] < 0)
150 inflight[1] = 0;
151}
152
153/*
154 * Can be deleted altogether. Later.
155 *
156 */
157#define BLKDEV_MAJOR_HASH_SIZE 255
158static struct blk_major_name {
159 struct blk_major_name *next;
160 int major;
161 char name[16];
162#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
163 void (*probe)(dev_t devt);
164#endif
165} *major_names[BLKDEV_MAJOR_HASH_SIZE];
166static DEFINE_MUTEX(major_names_lock);
167static DEFINE_SPINLOCK(major_names_spinlock);
168
169/* index in the above - for now: assume no multimajor ranges */
170static inline int major_to_index(unsigned major)
171{
172 return major % BLKDEV_MAJOR_HASH_SIZE;
173}
174
175#ifdef CONFIG_PROC_FS
176void blkdev_show(struct seq_file *seqf, off_t offset)
177{
178 struct blk_major_name *dp;
179
180 spin_lock(&major_names_spinlock);
181 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
182 if (dp->major == offset)
183 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
184 spin_unlock(&major_names_spinlock);
185}
186#endif /* CONFIG_PROC_FS */
187
188/**
189 * __register_blkdev - register a new block device
190 *
191 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
192 * @major = 0, try to allocate any unused major number.
193 * @name: the name of the new block device as a zero terminated string
194 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
195 * pre-created device node is accessed. When a probe call uses
196 * add_disk() and it fails the driver must cleanup resources. This
197 * interface may soon be removed.
198 *
199 * The @name must be unique within the system.
200 *
201 * The return value depends on the @major input parameter:
202 *
203 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
204 * then the function returns zero on success, or a negative error code
205 * - if any unused major number was requested with @major = 0 parameter
206 * then the return value is the allocated major number in range
207 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
208 *
209 * See Documentation/admin-guide/devices.txt for the list of allocated
210 * major numbers.
211 *
212 * Use register_blkdev instead for any new code.
213 */
214int __register_blkdev(unsigned int major, const char *name,
215 void (*probe)(dev_t devt))
216{
217 struct blk_major_name **n, *p;
218 int index, ret = 0;
219
220 mutex_lock(&major_names_lock);
221
222 /* temporary */
223 if (major == 0) {
224 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
225 if (major_names[index] == NULL)
226 break;
227 }
228
229 if (index == 0) {
230 printk("%s: failed to get major for %s\n",
231 __func__, name);
232 ret = -EBUSY;
233 goto out;
234 }
235 major = index;
236 ret = major;
237 }
238
239 if (major >= BLKDEV_MAJOR_MAX) {
240 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
241 __func__, major, BLKDEV_MAJOR_MAX-1, name);
242
243 ret = -EINVAL;
244 goto out;
245 }
246
247 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
248 if (p == NULL) {
249 ret = -ENOMEM;
250 goto out;
251 }
252
253 p->major = major;
254#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
255 p->probe = probe;
256#endif
257 strscpy(p->name, name, sizeof(p->name));
258 p->next = NULL;
259 index = major_to_index(major);
260
261 spin_lock(&major_names_spinlock);
262 for (n = &major_names[index]; *n; n = &(*n)->next) {
263 if ((*n)->major == major)
264 break;
265 }
266 if (!*n)
267 *n = p;
268 else
269 ret = -EBUSY;
270 spin_unlock(&major_names_spinlock);
271
272 if (ret < 0) {
273 printk("register_blkdev: cannot get major %u for %s\n",
274 major, name);
275 kfree(p);
276 }
277out:
278 mutex_unlock(&major_names_lock);
279 return ret;
280}
281EXPORT_SYMBOL(__register_blkdev);
282
283void unregister_blkdev(unsigned int major, const char *name)
284{
285 struct blk_major_name **n;
286 struct blk_major_name *p = NULL;
287 int index = major_to_index(major);
288
289 mutex_lock(&major_names_lock);
290 spin_lock(&major_names_spinlock);
291 for (n = &major_names[index]; *n; n = &(*n)->next)
292 if ((*n)->major == major)
293 break;
294 if (!*n || strcmp((*n)->name, name)) {
295 WARN_ON(1);
296 } else {
297 p = *n;
298 *n = p->next;
299 }
300 spin_unlock(&major_names_spinlock);
301 mutex_unlock(&major_names_lock);
302 kfree(p);
303}
304
305EXPORT_SYMBOL(unregister_blkdev);
306
307int blk_alloc_ext_minor(void)
308{
309 int idx;
310
311 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
312 if (idx == -ENOSPC)
313 return -EBUSY;
314 return idx;
315}
316
317void blk_free_ext_minor(unsigned int minor)
318{
319 ida_free(&ext_devt_ida, minor);
320}
321
322void disk_uevent(struct gendisk *disk, enum kobject_action action)
323{
324 struct block_device *part;
325 unsigned long idx;
326
327 rcu_read_lock();
328 xa_for_each(&disk->part_tbl, idx, part) {
329 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
330 continue;
331 if (!kobject_get_unless_zero(&part->bd_device.kobj))
332 continue;
333
334 rcu_read_unlock();
335 kobject_uevent(bdev_kobj(part), action);
336 put_device(&part->bd_device);
337 rcu_read_lock();
338 }
339 rcu_read_unlock();
340}
341EXPORT_SYMBOL_GPL(disk_uevent);
342
343int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode)
344{
345 struct bdev_handle *handle;
346 int ret = 0;
347
348 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
349 return -EINVAL;
350 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
351 return -EINVAL;
352 if (disk->open_partitions)
353 return -EBUSY;
354
355 /*
356 * If the device is opened exclusively by current thread already, it's
357 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
358 * synchronize with other exclusive openers and other partition
359 * scanners.
360 */
361 if (!(mode & BLK_OPEN_EXCL)) {
362 ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions,
363 NULL);
364 if (ret)
365 return ret;
366 }
367
368 set_bit(GD_NEED_PART_SCAN, &disk->state);
369 handle = bdev_open_by_dev(disk_devt(disk), mode & ~BLK_OPEN_EXCL, NULL,
370 NULL);
371 if (IS_ERR(handle))
372 ret = PTR_ERR(handle);
373 else
374 bdev_release(handle);
375
376 /*
377 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
378 * and this will cause that re-assemble partitioned raid device will
379 * creat partition for underlying disk.
380 */
381 clear_bit(GD_NEED_PART_SCAN, &disk->state);
382 if (!(mode & BLK_OPEN_EXCL))
383 bd_abort_claiming(disk->part0, disk_scan_partitions);
384 return ret;
385}
386
387/**
388 * device_add_disk - add disk information to kernel list
389 * @parent: parent device for the disk
390 * @disk: per-device partitioning information
391 * @groups: Additional per-device sysfs groups
392 *
393 * This function registers the partitioning information in @disk
394 * with the kernel.
395 */
396int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
397 const struct attribute_group **groups)
398
399{
400 struct device *ddev = disk_to_dev(disk);
401 int ret;
402
403 /* Only makes sense for bio-based to set ->poll_bio */
404 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
405 return -EINVAL;
406
407 /*
408 * The disk queue should now be all set with enough information about
409 * the device for the elevator code to pick an adequate default
410 * elevator if one is needed, that is, for devices requesting queue
411 * registration.
412 */
413 elevator_init_mq(disk->queue);
414
415 /* Mark bdev as having a submit_bio, if needed */
416 disk->part0->bd_has_submit_bio = disk->fops->submit_bio != NULL;
417
418 /*
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
421 * setup the gendisk.
422 * Otherwise just allocate the device numbers for both the whole device
423 * and all partitions from the extended dev_t space.
424 */
425 ret = -EINVAL;
426 if (disk->major) {
427 if (WARN_ON(!disk->minors))
428 goto out_exit_elevator;
429
430 if (disk->minors > DISK_MAX_PARTS) {
431 pr_err("block: can't allocate more than %d partitions\n",
432 DISK_MAX_PARTS);
433 disk->minors = DISK_MAX_PARTS;
434 }
435 if (disk->first_minor > MINORMASK ||
436 disk->minors > MINORMASK + 1 ||
437 disk->first_minor + disk->minors > MINORMASK + 1)
438 goto out_exit_elevator;
439 } else {
440 if (WARN_ON(disk->minors))
441 goto out_exit_elevator;
442
443 ret = blk_alloc_ext_minor();
444 if (ret < 0)
445 goto out_exit_elevator;
446 disk->major = BLOCK_EXT_MAJOR;
447 disk->first_minor = ret;
448 }
449
450 /* delay uevents, until we scanned partition table */
451 dev_set_uevent_suppress(ddev, 1);
452
453 ddev->parent = parent;
454 ddev->groups = groups;
455 dev_set_name(ddev, "%s", disk->disk_name);
456 if (!(disk->flags & GENHD_FL_HIDDEN))
457 ddev->devt = MKDEV(disk->major, disk->first_minor);
458 ret = device_add(ddev);
459 if (ret)
460 goto out_free_ext_minor;
461
462 ret = disk_alloc_events(disk);
463 if (ret)
464 goto out_device_del;
465
466 ret = sysfs_create_link(block_depr, &ddev->kobj,
467 kobject_name(&ddev->kobj));
468 if (ret)
469 goto out_device_del;
470
471 /*
472 * avoid probable deadlock caused by allocating memory with
473 * GFP_KERNEL in runtime_resume callback of its all ancestor
474 * devices
475 */
476 pm_runtime_set_memalloc_noio(ddev, true);
477
478 disk->part0->bd_holder_dir =
479 kobject_create_and_add("holders", &ddev->kobj);
480 if (!disk->part0->bd_holder_dir) {
481 ret = -ENOMEM;
482 goto out_del_block_link;
483 }
484 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
485 if (!disk->slave_dir) {
486 ret = -ENOMEM;
487 goto out_put_holder_dir;
488 }
489
490 ret = blk_register_queue(disk);
491 if (ret)
492 goto out_put_slave_dir;
493
494 if (!(disk->flags & GENHD_FL_HIDDEN)) {
495 ret = bdi_register(disk->bdi, "%u:%u",
496 disk->major, disk->first_minor);
497 if (ret)
498 goto out_unregister_queue;
499 bdi_set_owner(disk->bdi, ddev);
500 ret = sysfs_create_link(&ddev->kobj,
501 &disk->bdi->dev->kobj, "bdi");
502 if (ret)
503 goto out_unregister_bdi;
504
505 /* Make sure the first partition scan will be proceed */
506 if (get_capacity(disk) && !(disk->flags & GENHD_FL_NO_PART) &&
507 !test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
508 set_bit(GD_NEED_PART_SCAN, &disk->state);
509
510 bdev_add(disk->part0, ddev->devt);
511 if (get_capacity(disk))
512 disk_scan_partitions(disk, BLK_OPEN_READ);
513
514 /*
515 * Announce the disk and partitions after all partitions are
516 * created. (for hidden disks uevents remain suppressed forever)
517 */
518 dev_set_uevent_suppress(ddev, 0);
519 disk_uevent(disk, KOBJ_ADD);
520 } else {
521 /*
522 * Even if the block_device for a hidden gendisk is not
523 * registered, it needs to have a valid bd_dev so that the
524 * freeing of the dynamic major works.
525 */
526 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
527 }
528
529 disk_update_readahead(disk);
530 disk_add_events(disk);
531 set_bit(GD_ADDED, &disk->state);
532 return 0;
533
534out_unregister_bdi:
535 if (!(disk->flags & GENHD_FL_HIDDEN))
536 bdi_unregister(disk->bdi);
537out_unregister_queue:
538 blk_unregister_queue(disk);
539 rq_qos_exit(disk->queue);
540out_put_slave_dir:
541 kobject_put(disk->slave_dir);
542 disk->slave_dir = NULL;
543out_put_holder_dir:
544 kobject_put(disk->part0->bd_holder_dir);
545out_del_block_link:
546 sysfs_remove_link(block_depr, dev_name(ddev));
547 pm_runtime_set_memalloc_noio(ddev, false);
548out_device_del:
549 device_del(ddev);
550out_free_ext_minor:
551 if (disk->major == BLOCK_EXT_MAJOR)
552 blk_free_ext_minor(disk->first_minor);
553out_exit_elevator:
554 if (disk->queue->elevator)
555 elevator_exit(disk->queue);
556 return ret;
557}
558EXPORT_SYMBOL(device_add_disk);
559
560static void blk_report_disk_dead(struct gendisk *disk, bool surprise)
561{
562 struct block_device *bdev;
563 unsigned long idx;
564
565 /*
566 * On surprise disk removal, bdev_mark_dead() may call into file
567 * systems below. Make it clear that we're expecting to not hold
568 * disk->open_mutex.
569 */
570 lockdep_assert_not_held(&disk->open_mutex);
571
572 rcu_read_lock();
573 xa_for_each(&disk->part_tbl, idx, bdev) {
574 if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
575 continue;
576 rcu_read_unlock();
577
578 bdev_mark_dead(bdev, surprise);
579
580 put_device(&bdev->bd_device);
581 rcu_read_lock();
582 }
583 rcu_read_unlock();
584}
585
586static void __blk_mark_disk_dead(struct gendisk *disk)
587{
588 /*
589 * Fail any new I/O.
590 */
591 if (test_and_set_bit(GD_DEAD, &disk->state))
592 return;
593
594 if (test_bit(GD_OWNS_QUEUE, &disk->state))
595 blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);
596
597 /*
598 * Stop buffered writers from dirtying pages that can't be written out.
599 */
600 set_capacity(disk, 0);
601
602 /*
603 * Prevent new I/O from crossing bio_queue_enter().
604 */
605 blk_queue_start_drain(disk->queue);
606}
607
608/**
609 * blk_mark_disk_dead - mark a disk as dead
610 * @disk: disk to mark as dead
611 *
612 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
613 * to this disk.
614 */
615void blk_mark_disk_dead(struct gendisk *disk)
616{
617 __blk_mark_disk_dead(disk);
618 blk_report_disk_dead(disk, true);
619}
620EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
621
622/**
623 * del_gendisk - remove the gendisk
624 * @disk: the struct gendisk to remove
625 *
626 * Removes the gendisk and all its associated resources. This deletes the
627 * partitions associated with the gendisk, and unregisters the associated
628 * request_queue.
629 *
630 * This is the counter to the respective __device_add_disk() call.
631 *
632 * The final removal of the struct gendisk happens when its refcount reaches 0
633 * with put_disk(), which should be called after del_gendisk(), if
634 * __device_add_disk() was used.
635 *
636 * Drivers exist which depend on the release of the gendisk to be synchronous,
637 * it should not be deferred.
638 *
639 * Context: can sleep
640 */
641void del_gendisk(struct gendisk *disk)
642{
643 struct request_queue *q = disk->queue;
644 struct block_device *part;
645 unsigned long idx;
646
647 might_sleep();
648
649 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
650 return;
651
652 disk_del_events(disk);
653
654 /*
655 * Prevent new openers by unlinked the bdev inode.
656 */
657 mutex_lock(&disk->open_mutex);
658 xa_for_each(&disk->part_tbl, idx, part)
659 remove_inode_hash(part->bd_inode);
660 mutex_unlock(&disk->open_mutex);
661
662 /*
663 * Tell the file system to write back all dirty data and shut down if
664 * it hasn't been notified earlier.
665 */
666 if (!test_bit(GD_DEAD, &disk->state))
667 blk_report_disk_dead(disk, false);
668 __blk_mark_disk_dead(disk);
669
670 /*
671 * Drop all partitions now that the disk is marked dead.
672 */
673 mutex_lock(&disk->open_mutex);
674 xa_for_each_start(&disk->part_tbl, idx, part, 1)
675 drop_partition(part);
676 mutex_unlock(&disk->open_mutex);
677
678 if (!(disk->flags & GENHD_FL_HIDDEN)) {
679 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
680
681 /*
682 * Unregister bdi before releasing device numbers (as they can
683 * get reused and we'd get clashes in sysfs).
684 */
685 bdi_unregister(disk->bdi);
686 }
687
688 blk_unregister_queue(disk);
689
690 kobject_put(disk->part0->bd_holder_dir);
691 kobject_put(disk->slave_dir);
692 disk->slave_dir = NULL;
693
694 part_stat_set_all(disk->part0, 0);
695 disk->part0->bd_stamp = 0;
696 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
697 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
698 device_del(disk_to_dev(disk));
699
700 blk_mq_freeze_queue_wait(q);
701
702 blk_throtl_cancel_bios(disk);
703
704 blk_sync_queue(q);
705 blk_flush_integrity();
706
707 if (queue_is_mq(q))
708 blk_mq_cancel_work_sync(q);
709
710 blk_mq_quiesce_queue(q);
711 if (q->elevator) {
712 mutex_lock(&q->sysfs_lock);
713 elevator_exit(q);
714 mutex_unlock(&q->sysfs_lock);
715 }
716 rq_qos_exit(q);
717 blk_mq_unquiesce_queue(q);
718
719 /*
720 * If the disk does not own the queue, allow using passthrough requests
721 * again. Else leave the queue frozen to fail all I/O.
722 */
723 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
724 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
725 __blk_mq_unfreeze_queue(q, true);
726 } else {
727 if (queue_is_mq(q))
728 blk_mq_exit_queue(q);
729 }
730}
731EXPORT_SYMBOL(del_gendisk);
732
733/**
734 * invalidate_disk - invalidate the disk
735 * @disk: the struct gendisk to invalidate
736 *
737 * A helper to invalidates the disk. It will clean the disk's associated
738 * buffer/page caches and reset its internal states so that the disk
739 * can be reused by the drivers.
740 *
741 * Context: can sleep
742 */
743void invalidate_disk(struct gendisk *disk)
744{
745 struct block_device *bdev = disk->part0;
746
747 invalidate_bdev(bdev);
748 bdev->bd_inode->i_mapping->wb_err = 0;
749 set_capacity(disk, 0);
750}
751EXPORT_SYMBOL(invalidate_disk);
752
753/* sysfs access to bad-blocks list. */
754static ssize_t disk_badblocks_show(struct device *dev,
755 struct device_attribute *attr,
756 char *page)
757{
758 struct gendisk *disk = dev_to_disk(dev);
759
760 if (!disk->bb)
761 return sprintf(page, "\n");
762
763 return badblocks_show(disk->bb, page, 0);
764}
765
766static ssize_t disk_badblocks_store(struct device *dev,
767 struct device_attribute *attr,
768 const char *page, size_t len)
769{
770 struct gendisk *disk = dev_to_disk(dev);
771
772 if (!disk->bb)
773 return -ENXIO;
774
775 return badblocks_store(disk->bb, page, len, 0);
776}
777
778#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
779void blk_request_module(dev_t devt)
780{
781 unsigned int major = MAJOR(devt);
782 struct blk_major_name **n;
783
784 mutex_lock(&major_names_lock);
785 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
786 if ((*n)->major == major && (*n)->probe) {
787 (*n)->probe(devt);
788 mutex_unlock(&major_names_lock);
789 return;
790 }
791 }
792 mutex_unlock(&major_names_lock);
793
794 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
795 /* Make old-style 2.4 aliases work */
796 request_module("block-major-%d", MAJOR(devt));
797}
798#endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
799
800#ifdef CONFIG_PROC_FS
801/* iterator */
802static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
803{
804 loff_t skip = *pos;
805 struct class_dev_iter *iter;
806 struct device *dev;
807
808 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
809 if (!iter)
810 return ERR_PTR(-ENOMEM);
811
812 seqf->private = iter;
813 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
814 do {
815 dev = class_dev_iter_next(iter);
816 if (!dev)
817 return NULL;
818 } while (skip--);
819
820 return dev_to_disk(dev);
821}
822
823static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
824{
825 struct device *dev;
826
827 (*pos)++;
828 dev = class_dev_iter_next(seqf->private);
829 if (dev)
830 return dev_to_disk(dev);
831
832 return NULL;
833}
834
835static void disk_seqf_stop(struct seq_file *seqf, void *v)
836{
837 struct class_dev_iter *iter = seqf->private;
838
839 /* stop is called even after start failed :-( */
840 if (iter) {
841 class_dev_iter_exit(iter);
842 kfree(iter);
843 seqf->private = NULL;
844 }
845}
846
847static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
848{
849 void *p;
850
851 p = disk_seqf_start(seqf, pos);
852 if (!IS_ERR_OR_NULL(p) && !*pos)
853 seq_puts(seqf, "major minor #blocks name\n\n");
854 return p;
855}
856
857static int show_partition(struct seq_file *seqf, void *v)
858{
859 struct gendisk *sgp = v;
860 struct block_device *part;
861 unsigned long idx;
862
863 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
864 return 0;
865
866 rcu_read_lock();
867 xa_for_each(&sgp->part_tbl, idx, part) {
868 if (!bdev_nr_sectors(part))
869 continue;
870 seq_printf(seqf, "%4d %7d %10llu %pg\n",
871 MAJOR(part->bd_dev), MINOR(part->bd_dev),
872 bdev_nr_sectors(part) >> 1, part);
873 }
874 rcu_read_unlock();
875 return 0;
876}
877
878static const struct seq_operations partitions_op = {
879 .start = show_partition_start,
880 .next = disk_seqf_next,
881 .stop = disk_seqf_stop,
882 .show = show_partition
883};
884#endif
885
886static int __init genhd_device_init(void)
887{
888 int error;
889
890 error = class_register(&block_class);
891 if (unlikely(error))
892 return error;
893 blk_dev_init();
894
895 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
896
897 /* create top-level block dir */
898 block_depr = kobject_create_and_add("block", NULL);
899 return 0;
900}
901
902subsys_initcall(genhd_device_init);
903
904static ssize_t disk_range_show(struct device *dev,
905 struct device_attribute *attr, char *buf)
906{
907 struct gendisk *disk = dev_to_disk(dev);
908
909 return sprintf(buf, "%d\n", disk->minors);
910}
911
912static ssize_t disk_ext_range_show(struct device *dev,
913 struct device_attribute *attr, char *buf)
914{
915 struct gendisk *disk = dev_to_disk(dev);
916
917 return sprintf(buf, "%d\n",
918 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
919}
920
921static ssize_t disk_removable_show(struct device *dev,
922 struct device_attribute *attr, char *buf)
923{
924 struct gendisk *disk = dev_to_disk(dev);
925
926 return sprintf(buf, "%d\n",
927 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
928}
929
930static ssize_t disk_hidden_show(struct device *dev,
931 struct device_attribute *attr, char *buf)
932{
933 struct gendisk *disk = dev_to_disk(dev);
934
935 return sprintf(buf, "%d\n",
936 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
937}
938
939static ssize_t disk_ro_show(struct device *dev,
940 struct device_attribute *attr, char *buf)
941{
942 struct gendisk *disk = dev_to_disk(dev);
943
944 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
945}
946
947ssize_t part_size_show(struct device *dev,
948 struct device_attribute *attr, char *buf)
949{
950 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
951}
952
953ssize_t part_stat_show(struct device *dev,
954 struct device_attribute *attr, char *buf)
955{
956 struct block_device *bdev = dev_to_bdev(dev);
957 struct request_queue *q = bdev_get_queue(bdev);
958 struct disk_stats stat;
959 unsigned int inflight;
960
961 if (queue_is_mq(q))
962 inflight = blk_mq_in_flight(q, bdev);
963 else
964 inflight = part_in_flight(bdev);
965
966 if (inflight) {
967 part_stat_lock();
968 update_io_ticks(bdev, jiffies, true);
969 part_stat_unlock();
970 }
971 part_stat_read_all(bdev, &stat);
972 return sprintf(buf,
973 "%8lu %8lu %8llu %8u "
974 "%8lu %8lu %8llu %8u "
975 "%8u %8u %8u "
976 "%8lu %8lu %8llu %8u "
977 "%8lu %8u"
978 "\n",
979 stat.ios[STAT_READ],
980 stat.merges[STAT_READ],
981 (unsigned long long)stat.sectors[STAT_READ],
982 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
983 stat.ios[STAT_WRITE],
984 stat.merges[STAT_WRITE],
985 (unsigned long long)stat.sectors[STAT_WRITE],
986 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
987 inflight,
988 jiffies_to_msecs(stat.io_ticks),
989 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
990 stat.nsecs[STAT_WRITE] +
991 stat.nsecs[STAT_DISCARD] +
992 stat.nsecs[STAT_FLUSH],
993 NSEC_PER_MSEC),
994 stat.ios[STAT_DISCARD],
995 stat.merges[STAT_DISCARD],
996 (unsigned long long)stat.sectors[STAT_DISCARD],
997 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
998 stat.ios[STAT_FLUSH],
999 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1000}
1001
1002ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1003 char *buf)
1004{
1005 struct block_device *bdev = dev_to_bdev(dev);
1006 struct request_queue *q = bdev_get_queue(bdev);
1007 unsigned int inflight[2];
1008
1009 if (queue_is_mq(q))
1010 blk_mq_in_flight_rw(q, bdev, inflight);
1011 else
1012 part_in_flight_rw(bdev, inflight);
1013
1014 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1015}
1016
1017static ssize_t disk_capability_show(struct device *dev,
1018 struct device_attribute *attr, char *buf)
1019{
1020 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1021 return sprintf(buf, "0\n");
1022}
1023
1024static ssize_t disk_alignment_offset_show(struct device *dev,
1025 struct device_attribute *attr,
1026 char *buf)
1027{
1028 struct gendisk *disk = dev_to_disk(dev);
1029
1030 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1031}
1032
1033static ssize_t disk_discard_alignment_show(struct device *dev,
1034 struct device_attribute *attr,
1035 char *buf)
1036{
1037 struct gendisk *disk = dev_to_disk(dev);
1038
1039 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1040}
1041
1042static ssize_t diskseq_show(struct device *dev,
1043 struct device_attribute *attr, char *buf)
1044{
1045 struct gendisk *disk = dev_to_disk(dev);
1046
1047 return sprintf(buf, "%llu\n", disk->diskseq);
1048}
1049
1050static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1051static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1052static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1053static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1054static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1055static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1056static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1057static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1058static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1059static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1060static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1061static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1062static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1063
1064#ifdef CONFIG_FAIL_MAKE_REQUEST
1065ssize_t part_fail_show(struct device *dev,
1066 struct device_attribute *attr, char *buf)
1067{
1068 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1069}
1070
1071ssize_t part_fail_store(struct device *dev,
1072 struct device_attribute *attr,
1073 const char *buf, size_t count)
1074{
1075 int i;
1076
1077 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1078 dev_to_bdev(dev)->bd_make_it_fail = i;
1079
1080 return count;
1081}
1082
1083static struct device_attribute dev_attr_fail =
1084 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1085#endif /* CONFIG_FAIL_MAKE_REQUEST */
1086
1087#ifdef CONFIG_FAIL_IO_TIMEOUT
1088static struct device_attribute dev_attr_fail_timeout =
1089 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1090#endif
1091
1092static struct attribute *disk_attrs[] = {
1093 &dev_attr_range.attr,
1094 &dev_attr_ext_range.attr,
1095 &dev_attr_removable.attr,
1096 &dev_attr_hidden.attr,
1097 &dev_attr_ro.attr,
1098 &dev_attr_size.attr,
1099 &dev_attr_alignment_offset.attr,
1100 &dev_attr_discard_alignment.attr,
1101 &dev_attr_capability.attr,
1102 &dev_attr_stat.attr,
1103 &dev_attr_inflight.attr,
1104 &dev_attr_badblocks.attr,
1105 &dev_attr_events.attr,
1106 &dev_attr_events_async.attr,
1107 &dev_attr_events_poll_msecs.attr,
1108 &dev_attr_diskseq.attr,
1109#ifdef CONFIG_FAIL_MAKE_REQUEST
1110 &dev_attr_fail.attr,
1111#endif
1112#ifdef CONFIG_FAIL_IO_TIMEOUT
1113 &dev_attr_fail_timeout.attr,
1114#endif
1115 NULL
1116};
1117
1118static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1119{
1120 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1121 struct gendisk *disk = dev_to_disk(dev);
1122
1123 if (a == &dev_attr_badblocks.attr && !disk->bb)
1124 return 0;
1125 return a->mode;
1126}
1127
1128static struct attribute_group disk_attr_group = {
1129 .attrs = disk_attrs,
1130 .is_visible = disk_visible,
1131};
1132
1133static const struct attribute_group *disk_attr_groups[] = {
1134 &disk_attr_group,
1135#ifdef CONFIG_BLK_DEV_IO_TRACE
1136 &blk_trace_attr_group,
1137#endif
1138#ifdef CONFIG_BLK_DEV_INTEGRITY
1139 &blk_integrity_attr_group,
1140#endif
1141 NULL
1142};
1143
1144/**
1145 * disk_release - releases all allocated resources of the gendisk
1146 * @dev: the device representing this disk
1147 *
1148 * This function releases all allocated resources of the gendisk.
1149 *
1150 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1151 * assigned. Since the request_queue sits on top of the gendisk for these
1152 * drivers we also call blk_put_queue() for them, and we expect the
1153 * request_queue refcount to reach 0 at this point, and so the request_queue
1154 * will also be freed prior to the disk.
1155 *
1156 * Context: can sleep
1157 */
1158static void disk_release(struct device *dev)
1159{
1160 struct gendisk *disk = dev_to_disk(dev);
1161
1162 might_sleep();
1163 WARN_ON_ONCE(disk_live(disk));
1164
1165 blk_trace_remove(disk->queue);
1166
1167 /*
1168 * To undo the all initialization from blk_mq_init_allocated_queue in
1169 * case of a probe failure where add_disk is never called we have to
1170 * call blk_mq_exit_queue here. We can't do this for the more common
1171 * teardown case (yet) as the tagset can be gone by the time the disk
1172 * is released once it was added.
1173 */
1174 if (queue_is_mq(disk->queue) &&
1175 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1176 !test_bit(GD_ADDED, &disk->state))
1177 blk_mq_exit_queue(disk->queue);
1178
1179 blkcg_exit_disk(disk);
1180
1181 bioset_exit(&disk->bio_split);
1182
1183 disk_release_events(disk);
1184 kfree(disk->random);
1185 disk_free_zone_bitmaps(disk);
1186 xa_destroy(&disk->part_tbl);
1187
1188 disk->queue->disk = NULL;
1189 blk_put_queue(disk->queue);
1190
1191 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1192 disk->fops->free_disk(disk);
1193
1194 iput(disk->part0->bd_inode); /* frees the disk */
1195}
1196
1197static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1198{
1199 const struct gendisk *disk = dev_to_disk(dev);
1200
1201 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1202}
1203
1204struct class block_class = {
1205 .name = "block",
1206 .dev_uevent = block_uevent,
1207};
1208
1209static char *block_devnode(const struct device *dev, umode_t *mode,
1210 kuid_t *uid, kgid_t *gid)
1211{
1212 struct gendisk *disk = dev_to_disk(dev);
1213
1214 if (disk->fops->devnode)
1215 return disk->fops->devnode(disk, mode);
1216 return NULL;
1217}
1218
1219const struct device_type disk_type = {
1220 .name = "disk",
1221 .groups = disk_attr_groups,
1222 .release = disk_release,
1223 .devnode = block_devnode,
1224};
1225
1226#ifdef CONFIG_PROC_FS
1227/*
1228 * aggregate disk stat collector. Uses the same stats that the sysfs
1229 * entries do, above, but makes them available through one seq_file.
1230 *
1231 * The output looks suspiciously like /proc/partitions with a bunch of
1232 * extra fields.
1233 */
1234static int diskstats_show(struct seq_file *seqf, void *v)
1235{
1236 struct gendisk *gp = v;
1237 struct block_device *hd;
1238 unsigned int inflight;
1239 struct disk_stats stat;
1240 unsigned long idx;
1241
1242 /*
1243 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1244 seq_puts(seqf, "major minor name"
1245 " rio rmerge rsect ruse wio wmerge "
1246 "wsect wuse running use aveq"
1247 "\n\n");
1248 */
1249
1250 rcu_read_lock();
1251 xa_for_each(&gp->part_tbl, idx, hd) {
1252 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1253 continue;
1254 if (queue_is_mq(gp->queue))
1255 inflight = blk_mq_in_flight(gp->queue, hd);
1256 else
1257 inflight = part_in_flight(hd);
1258
1259 if (inflight) {
1260 part_stat_lock();
1261 update_io_ticks(hd, jiffies, true);
1262 part_stat_unlock();
1263 }
1264 part_stat_read_all(hd, &stat);
1265 seq_printf(seqf, "%4d %7d %pg "
1266 "%lu %lu %lu %u "
1267 "%lu %lu %lu %u "
1268 "%u %u %u "
1269 "%lu %lu %lu %u "
1270 "%lu %u"
1271 "\n",
1272 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1273 stat.ios[STAT_READ],
1274 stat.merges[STAT_READ],
1275 stat.sectors[STAT_READ],
1276 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1277 NSEC_PER_MSEC),
1278 stat.ios[STAT_WRITE],
1279 stat.merges[STAT_WRITE],
1280 stat.sectors[STAT_WRITE],
1281 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1282 NSEC_PER_MSEC),
1283 inflight,
1284 jiffies_to_msecs(stat.io_ticks),
1285 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1286 stat.nsecs[STAT_WRITE] +
1287 stat.nsecs[STAT_DISCARD] +
1288 stat.nsecs[STAT_FLUSH],
1289 NSEC_PER_MSEC),
1290 stat.ios[STAT_DISCARD],
1291 stat.merges[STAT_DISCARD],
1292 stat.sectors[STAT_DISCARD],
1293 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1294 NSEC_PER_MSEC),
1295 stat.ios[STAT_FLUSH],
1296 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1297 NSEC_PER_MSEC)
1298 );
1299 }
1300 rcu_read_unlock();
1301
1302 return 0;
1303}
1304
1305static const struct seq_operations diskstats_op = {
1306 .start = disk_seqf_start,
1307 .next = disk_seqf_next,
1308 .stop = disk_seqf_stop,
1309 .show = diskstats_show
1310};
1311
1312static int __init proc_genhd_init(void)
1313{
1314 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1315 proc_create_seq("partitions", 0, NULL, &partitions_op);
1316 return 0;
1317}
1318module_init(proc_genhd_init);
1319#endif /* CONFIG_PROC_FS */
1320
1321dev_t part_devt(struct gendisk *disk, u8 partno)
1322{
1323 struct block_device *part;
1324 dev_t devt = 0;
1325
1326 rcu_read_lock();
1327 part = xa_load(&disk->part_tbl, partno);
1328 if (part)
1329 devt = part->bd_dev;
1330 rcu_read_unlock();
1331
1332 return devt;
1333}
1334
1335struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1336 struct lock_class_key *lkclass)
1337{
1338 struct gendisk *disk;
1339
1340 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1341 if (!disk)
1342 return NULL;
1343
1344 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1345 goto out_free_disk;
1346
1347 disk->bdi = bdi_alloc(node_id);
1348 if (!disk->bdi)
1349 goto out_free_bioset;
1350
1351 /* bdev_alloc() might need the queue, set before the first call */
1352 disk->queue = q;
1353
1354 disk->part0 = bdev_alloc(disk, 0);
1355 if (!disk->part0)
1356 goto out_free_bdi;
1357
1358 disk->node_id = node_id;
1359 mutex_init(&disk->open_mutex);
1360 xa_init(&disk->part_tbl);
1361 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1362 goto out_destroy_part_tbl;
1363
1364 if (blkcg_init_disk(disk))
1365 goto out_erase_part0;
1366
1367 rand_initialize_disk(disk);
1368 disk_to_dev(disk)->class = &block_class;
1369 disk_to_dev(disk)->type = &disk_type;
1370 device_initialize(disk_to_dev(disk));
1371 inc_diskseq(disk);
1372 q->disk = disk;
1373 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1374#ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1375 INIT_LIST_HEAD(&disk->slave_bdevs);
1376#endif
1377 return disk;
1378
1379out_erase_part0:
1380 xa_erase(&disk->part_tbl, 0);
1381out_destroy_part_tbl:
1382 xa_destroy(&disk->part_tbl);
1383 disk->part0->bd_disk = NULL;
1384 iput(disk->part0->bd_inode);
1385out_free_bdi:
1386 bdi_put(disk->bdi);
1387out_free_bioset:
1388 bioset_exit(&disk->bio_split);
1389out_free_disk:
1390 kfree(disk);
1391 return NULL;
1392}
1393
1394struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1395{
1396 struct request_queue *q;
1397 struct gendisk *disk;
1398
1399 q = blk_alloc_queue(node);
1400 if (!q)
1401 return NULL;
1402
1403 disk = __alloc_disk_node(q, node, lkclass);
1404 if (!disk) {
1405 blk_put_queue(q);
1406 return NULL;
1407 }
1408 set_bit(GD_OWNS_QUEUE, &disk->state);
1409 return disk;
1410}
1411EXPORT_SYMBOL(__blk_alloc_disk);
1412
1413/**
1414 * put_disk - decrements the gendisk refcount
1415 * @disk: the struct gendisk to decrement the refcount for
1416 *
1417 * This decrements the refcount for the struct gendisk. When this reaches 0
1418 * we'll have disk_release() called.
1419 *
1420 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1421 * when handling probe errors (that is before add_disk() is called).
1422 *
1423 * Context: Any context, but the last reference must not be dropped from
1424 * atomic context.
1425 */
1426void put_disk(struct gendisk *disk)
1427{
1428 if (disk)
1429 put_device(disk_to_dev(disk));
1430}
1431EXPORT_SYMBOL(put_disk);
1432
1433static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1434{
1435 char event[] = "DISK_RO=1";
1436 char *envp[] = { event, NULL };
1437
1438 if (!ro)
1439 event[8] = '0';
1440 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1441}
1442
1443/**
1444 * set_disk_ro - set a gendisk read-only
1445 * @disk: gendisk to operate on
1446 * @read_only: %true to set the disk read-only, %false set the disk read/write
1447 *
1448 * This function is used to indicate whether a given disk device should have its
1449 * read-only flag set. set_disk_ro() is typically used by device drivers to
1450 * indicate whether the underlying physical device is write-protected.
1451 */
1452void set_disk_ro(struct gendisk *disk, bool read_only)
1453{
1454 if (read_only) {
1455 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1456 return;
1457 } else {
1458 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1459 return;
1460 }
1461 set_disk_ro_uevent(disk, read_only);
1462}
1463EXPORT_SYMBOL(set_disk_ro);
1464
1465void inc_diskseq(struct gendisk *disk)
1466{
1467 disk->diskseq = atomic64_inc_return(&diskseq);
1468}