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/*
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/backing-dev.h>
12#include <linux/init.h>
13#include <linux/spinlock.h>
14#include <linux/proc_fs.h>
15#include <linux/seq_file.h>
16#include <linux/slab.h>
17#include <linux/kmod.h>
18#include <linux/kobj_map.h>
19#include <linux/mutex.h>
20#include <linux/idr.h>
21#include <linux/log2.h>
22#include <linux/pm_runtime.h>
23#include <linux/badblocks.h>
24
25#include "blk.h"
26
27static DEFINE_MUTEX(block_class_lock);
28struct kobject *block_depr;
29
30/* for extended dynamic devt allocation, currently only one major is used */
31#define NR_EXT_DEVT (1 << MINORBITS)
32
33/* For extended devt allocation. ext_devt_lock prevents look up
34 * results from going away underneath its user.
35 */
36static DEFINE_SPINLOCK(ext_devt_lock);
37static DEFINE_IDR(ext_devt_idr);
38
39static struct device_type disk_type;
40
41static void disk_check_events(struct disk_events *ev,
42 unsigned int *clearing_ptr);
43static void disk_alloc_events(struct gendisk *disk);
44static void disk_add_events(struct gendisk *disk);
45static void disk_del_events(struct gendisk *disk);
46static void disk_release_events(struct gendisk *disk);
47
48/**
49 * disk_get_part - get partition
50 * @disk: disk to look partition from
51 * @partno: partition number
52 *
53 * Look for partition @partno from @disk. If found, increment
54 * reference count and return it.
55 *
56 * CONTEXT:
57 * Don't care.
58 *
59 * RETURNS:
60 * Pointer to the found partition on success, NULL if not found.
61 */
62struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
63{
64 struct hd_struct *part = NULL;
65 struct disk_part_tbl *ptbl;
66
67 if (unlikely(partno < 0))
68 return NULL;
69
70 rcu_read_lock();
71
72 ptbl = rcu_dereference(disk->part_tbl);
73 if (likely(partno < ptbl->len)) {
74 part = rcu_dereference(ptbl->part[partno]);
75 if (part)
76 get_device(part_to_dev(part));
77 }
78
79 rcu_read_unlock();
80
81 return part;
82}
83EXPORT_SYMBOL_GPL(disk_get_part);
84
85/**
86 * disk_part_iter_init - initialize partition iterator
87 * @piter: iterator to initialize
88 * @disk: disk to iterate over
89 * @flags: DISK_PITER_* flags
90 *
91 * Initialize @piter so that it iterates over partitions of @disk.
92 *
93 * CONTEXT:
94 * Don't care.
95 */
96void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
97 unsigned int flags)
98{
99 struct disk_part_tbl *ptbl;
100
101 rcu_read_lock();
102 ptbl = rcu_dereference(disk->part_tbl);
103
104 piter->disk = disk;
105 piter->part = NULL;
106
107 if (flags & DISK_PITER_REVERSE)
108 piter->idx = ptbl->len - 1;
109 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
110 piter->idx = 0;
111 else
112 piter->idx = 1;
113
114 piter->flags = flags;
115
116 rcu_read_unlock();
117}
118EXPORT_SYMBOL_GPL(disk_part_iter_init);
119
120/**
121 * disk_part_iter_next - proceed iterator to the next partition and return it
122 * @piter: iterator of interest
123 *
124 * Proceed @piter to the next partition and return it.
125 *
126 * CONTEXT:
127 * Don't care.
128 */
129struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
130{
131 struct disk_part_tbl *ptbl;
132 int inc, end;
133
134 /* put the last partition */
135 disk_put_part(piter->part);
136 piter->part = NULL;
137
138 /* get part_tbl */
139 rcu_read_lock();
140 ptbl = rcu_dereference(piter->disk->part_tbl);
141
142 /* determine iteration parameters */
143 if (piter->flags & DISK_PITER_REVERSE) {
144 inc = -1;
145 if (piter->flags & (DISK_PITER_INCL_PART0 |
146 DISK_PITER_INCL_EMPTY_PART0))
147 end = -1;
148 else
149 end = 0;
150 } else {
151 inc = 1;
152 end = ptbl->len;
153 }
154
155 /* iterate to the next partition */
156 for (; piter->idx != end; piter->idx += inc) {
157 struct hd_struct *part;
158
159 part = rcu_dereference(ptbl->part[piter->idx]);
160 if (!part)
161 continue;
162 if (!part_nr_sects_read(part) &&
163 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
164 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
165 piter->idx == 0))
166 continue;
167
168 get_device(part_to_dev(part));
169 piter->part = part;
170 piter->idx += inc;
171 break;
172 }
173
174 rcu_read_unlock();
175
176 return piter->part;
177}
178EXPORT_SYMBOL_GPL(disk_part_iter_next);
179
180/**
181 * disk_part_iter_exit - finish up partition iteration
182 * @piter: iter of interest
183 *
184 * Called when iteration is over. Cleans up @piter.
185 *
186 * CONTEXT:
187 * Don't care.
188 */
189void disk_part_iter_exit(struct disk_part_iter *piter)
190{
191 disk_put_part(piter->part);
192 piter->part = NULL;
193}
194EXPORT_SYMBOL_GPL(disk_part_iter_exit);
195
196static inline int sector_in_part(struct hd_struct *part, sector_t sector)
197{
198 return part->start_sect <= sector &&
199 sector < part->start_sect + part_nr_sects_read(part);
200}
201
202/**
203 * disk_map_sector_rcu - map sector to partition
204 * @disk: gendisk of interest
205 * @sector: sector to map
206 *
207 * Find out which partition @sector maps to on @disk. This is
208 * primarily used for stats accounting.
209 *
210 * CONTEXT:
211 * RCU read locked. The returned partition pointer is valid only
212 * while preemption is disabled.
213 *
214 * RETURNS:
215 * Found partition on success, part0 is returned if no partition matches
216 */
217struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
218{
219 struct disk_part_tbl *ptbl;
220 struct hd_struct *part;
221 int i;
222
223 ptbl = rcu_dereference(disk->part_tbl);
224
225 part = rcu_dereference(ptbl->last_lookup);
226 if (part && sector_in_part(part, sector))
227 return part;
228
229 for (i = 1; i < ptbl->len; i++) {
230 part = rcu_dereference(ptbl->part[i]);
231
232 if (part && sector_in_part(part, sector)) {
233 rcu_assign_pointer(ptbl->last_lookup, part);
234 return part;
235 }
236 }
237 return &disk->part0;
238}
239EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
240
241/*
242 * Can be deleted altogether. Later.
243 *
244 */
245static struct blk_major_name {
246 struct blk_major_name *next;
247 int major;
248 char name[16];
249} *major_names[BLKDEV_MAJOR_HASH_SIZE];
250
251/* index in the above - for now: assume no multimajor ranges */
252static inline int major_to_index(unsigned major)
253{
254 return major % BLKDEV_MAJOR_HASH_SIZE;
255}
256
257#ifdef CONFIG_PROC_FS
258void blkdev_show(struct seq_file *seqf, off_t offset)
259{
260 struct blk_major_name *dp;
261
262 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
263 mutex_lock(&block_class_lock);
264 for (dp = major_names[offset]; dp; dp = dp->next)
265 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
266 mutex_unlock(&block_class_lock);
267 }
268}
269#endif /* CONFIG_PROC_FS */
270
271/**
272 * register_blkdev - register a new block device
273 *
274 * @major: the requested major device number [1..255]. If @major=0, try to
275 * allocate any unused major number.
276 * @name: the name of the new block device as a zero terminated string
277 *
278 * The @name must be unique within the system.
279 *
280 * The return value depends on the @major input parameter.
281 * - if a major device number was requested in range [1..255] then the
282 * function returns zero on success, or a negative error code
283 * - if any unused major number was requested with @major=0 parameter
284 * then the return value is the allocated major number in range
285 * [1..255] or a negative error code otherwise
286 */
287int register_blkdev(unsigned int major, const char *name)
288{
289 struct blk_major_name **n, *p;
290 int index, ret = 0;
291
292 mutex_lock(&block_class_lock);
293
294 /* temporary */
295 if (major == 0) {
296 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
297 if (major_names[index] == NULL)
298 break;
299 }
300
301 if (index == 0) {
302 printk("register_blkdev: failed to get major for %s\n",
303 name);
304 ret = -EBUSY;
305 goto out;
306 }
307 major = index;
308 ret = major;
309 }
310
311 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
312 if (p == NULL) {
313 ret = -ENOMEM;
314 goto out;
315 }
316
317 p->major = major;
318 strlcpy(p->name, name, sizeof(p->name));
319 p->next = NULL;
320 index = major_to_index(major);
321
322 for (n = &major_names[index]; *n; n = &(*n)->next) {
323 if ((*n)->major == major)
324 break;
325 }
326 if (!*n)
327 *n = p;
328 else
329 ret = -EBUSY;
330
331 if (ret < 0) {
332 printk("register_blkdev: cannot get major %d for %s\n",
333 major, name);
334 kfree(p);
335 }
336out:
337 mutex_unlock(&block_class_lock);
338 return ret;
339}
340
341EXPORT_SYMBOL(register_blkdev);
342
343void unregister_blkdev(unsigned int major, const char *name)
344{
345 struct blk_major_name **n;
346 struct blk_major_name *p = NULL;
347 int index = major_to_index(major);
348
349 mutex_lock(&block_class_lock);
350 for (n = &major_names[index]; *n; n = &(*n)->next)
351 if ((*n)->major == major)
352 break;
353 if (!*n || strcmp((*n)->name, name)) {
354 WARN_ON(1);
355 } else {
356 p = *n;
357 *n = p->next;
358 }
359 mutex_unlock(&block_class_lock);
360 kfree(p);
361}
362
363EXPORT_SYMBOL(unregister_blkdev);
364
365static struct kobj_map *bdev_map;
366
367/**
368 * blk_mangle_minor - scatter minor numbers apart
369 * @minor: minor number to mangle
370 *
371 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
372 * is enabled. Mangling twice gives the original value.
373 *
374 * RETURNS:
375 * Mangled value.
376 *
377 * CONTEXT:
378 * Don't care.
379 */
380static int blk_mangle_minor(int minor)
381{
382#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
383 int i;
384
385 for (i = 0; i < MINORBITS / 2; i++) {
386 int low = minor & (1 << i);
387 int high = minor & (1 << (MINORBITS - 1 - i));
388 int distance = MINORBITS - 1 - 2 * i;
389
390 minor ^= low | high; /* clear both bits */
391 low <<= distance; /* swap the positions */
392 high >>= distance;
393 minor |= low | high; /* and set */
394 }
395#endif
396 return minor;
397}
398
399/**
400 * blk_alloc_devt - allocate a dev_t for a partition
401 * @part: partition to allocate dev_t for
402 * @devt: out parameter for resulting dev_t
403 *
404 * Allocate a dev_t for block device.
405 *
406 * RETURNS:
407 * 0 on success, allocated dev_t is returned in *@devt. -errno on
408 * failure.
409 *
410 * CONTEXT:
411 * Might sleep.
412 */
413int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
414{
415 struct gendisk *disk = part_to_disk(part);
416 int idx;
417
418 /* in consecutive minor range? */
419 if (part->partno < disk->minors) {
420 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
421 return 0;
422 }
423
424 /* allocate ext devt */
425 idr_preload(GFP_KERNEL);
426
427 spin_lock_bh(&ext_devt_lock);
428 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
429 spin_unlock_bh(&ext_devt_lock);
430
431 idr_preload_end();
432 if (idx < 0)
433 return idx == -ENOSPC ? -EBUSY : idx;
434
435 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
436 return 0;
437}
438
439/**
440 * blk_free_devt - free a dev_t
441 * @devt: dev_t to free
442 *
443 * Free @devt which was allocated using blk_alloc_devt().
444 *
445 * CONTEXT:
446 * Might sleep.
447 */
448void blk_free_devt(dev_t devt)
449{
450 if (devt == MKDEV(0, 0))
451 return;
452
453 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
454 spin_lock_bh(&ext_devt_lock);
455 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
456 spin_unlock_bh(&ext_devt_lock);
457 }
458}
459
460static char *bdevt_str(dev_t devt, char *buf)
461{
462 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
463 char tbuf[BDEVT_SIZE];
464 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
465 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
466 } else
467 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
468
469 return buf;
470}
471
472/*
473 * Register device numbers dev..(dev+range-1)
474 * range must be nonzero
475 * The hash chain is sorted on range, so that subranges can override.
476 */
477void blk_register_region(dev_t devt, unsigned long range, struct module *module,
478 struct kobject *(*probe)(dev_t, int *, void *),
479 int (*lock)(dev_t, void *), void *data)
480{
481 kobj_map(bdev_map, devt, range, module, probe, lock, data);
482}
483
484EXPORT_SYMBOL(blk_register_region);
485
486void blk_unregister_region(dev_t devt, unsigned long range)
487{
488 kobj_unmap(bdev_map, devt, range);
489}
490
491EXPORT_SYMBOL(blk_unregister_region);
492
493static struct kobject *exact_match(dev_t devt, int *partno, void *data)
494{
495 struct gendisk *p = data;
496
497 return &disk_to_dev(p)->kobj;
498}
499
500static int exact_lock(dev_t devt, void *data)
501{
502 struct gendisk *p = data;
503
504 if (!get_disk(p))
505 return -1;
506 return 0;
507}
508
509static void register_disk(struct device *parent, struct gendisk *disk)
510{
511 struct device *ddev = disk_to_dev(disk);
512 struct block_device *bdev;
513 struct disk_part_iter piter;
514 struct hd_struct *part;
515 int err;
516
517 ddev->parent = parent;
518
519 dev_set_name(ddev, "%s", disk->disk_name);
520
521 /* delay uevents, until we scanned partition table */
522 dev_set_uevent_suppress(ddev, 1);
523
524 if (device_add(ddev))
525 return;
526 if (!sysfs_deprecated) {
527 err = sysfs_create_link(block_depr, &ddev->kobj,
528 kobject_name(&ddev->kobj));
529 if (err) {
530 device_del(ddev);
531 return;
532 }
533 }
534
535 /*
536 * avoid probable deadlock caused by allocating memory with
537 * GFP_KERNEL in runtime_resume callback of its all ancestor
538 * devices
539 */
540 pm_runtime_set_memalloc_noio(ddev, true);
541
542 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
543 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
544
545 /* No minors to use for partitions */
546 if (!disk_part_scan_enabled(disk))
547 goto exit;
548
549 /* No such device (e.g., media were just removed) */
550 if (!get_capacity(disk))
551 goto exit;
552
553 bdev = bdget_disk(disk, 0);
554 if (!bdev)
555 goto exit;
556
557 bdev->bd_invalidated = 1;
558 err = blkdev_get(bdev, FMODE_READ, NULL);
559 if (err < 0)
560 goto exit;
561 blkdev_put(bdev, FMODE_READ);
562
563exit:
564 /* announce disk after possible partitions are created */
565 dev_set_uevent_suppress(ddev, 0);
566 kobject_uevent(&ddev->kobj, KOBJ_ADD);
567
568 /* announce possible partitions */
569 disk_part_iter_init(&piter, disk, 0);
570 while ((part = disk_part_iter_next(&piter)))
571 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
572 disk_part_iter_exit(&piter);
573}
574
575/**
576 * device_add_disk - add partitioning information to kernel list
577 * @parent: parent device for the disk
578 * @disk: per-device partitioning information
579 *
580 * This function registers the partitioning information in @disk
581 * with the kernel.
582 *
583 * FIXME: error handling
584 */
585void device_add_disk(struct device *parent, struct gendisk *disk)
586{
587 struct backing_dev_info *bdi;
588 dev_t devt;
589 int retval;
590
591 /* minors == 0 indicates to use ext devt from part0 and should
592 * be accompanied with EXT_DEVT flag. Make sure all
593 * parameters make sense.
594 */
595 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
596 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
597
598 disk->flags |= GENHD_FL_UP;
599
600 retval = blk_alloc_devt(&disk->part0, &devt);
601 if (retval) {
602 WARN_ON(1);
603 return;
604 }
605 disk_to_dev(disk)->devt = devt;
606
607 /* ->major and ->first_minor aren't supposed to be
608 * dereferenced from here on, but set them just in case.
609 */
610 disk->major = MAJOR(devt);
611 disk->first_minor = MINOR(devt);
612
613 disk_alloc_events(disk);
614
615 /* Register BDI before referencing it from bdev */
616 bdi = &disk->queue->backing_dev_info;
617 bdi_register_owner(bdi, disk_to_dev(disk));
618
619 blk_register_region(disk_devt(disk), disk->minors, NULL,
620 exact_match, exact_lock, disk);
621 register_disk(parent, disk);
622 blk_register_queue(disk);
623
624 /*
625 * Take an extra ref on queue which will be put on disk_release()
626 * so that it sticks around as long as @disk is there.
627 */
628 WARN_ON_ONCE(!blk_get_queue(disk->queue));
629
630 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
631 "bdi");
632 WARN_ON(retval);
633
634 disk_add_events(disk);
635 blk_integrity_add(disk);
636}
637EXPORT_SYMBOL(device_add_disk);
638
639void del_gendisk(struct gendisk *disk)
640{
641 struct disk_part_iter piter;
642 struct hd_struct *part;
643
644 blk_integrity_del(disk);
645 disk_del_events(disk);
646
647 /* invalidate stuff */
648 disk_part_iter_init(&piter, disk,
649 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
650 while ((part = disk_part_iter_next(&piter))) {
651 invalidate_partition(disk, part->partno);
652 delete_partition(disk, part->partno);
653 }
654 disk_part_iter_exit(&piter);
655
656 invalidate_partition(disk, 0);
657 set_capacity(disk, 0);
658 disk->flags &= ~GENHD_FL_UP;
659
660 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
661 blk_unregister_queue(disk);
662 blk_unregister_region(disk_devt(disk), disk->minors);
663
664 part_stat_set_all(&disk->part0, 0);
665 disk->part0.stamp = 0;
666
667 kobject_put(disk->part0.holder_dir);
668 kobject_put(disk->slave_dir);
669 if (!sysfs_deprecated)
670 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
671 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
672 device_del(disk_to_dev(disk));
673}
674EXPORT_SYMBOL(del_gendisk);
675
676/* sysfs access to bad-blocks list. */
677static ssize_t disk_badblocks_show(struct device *dev,
678 struct device_attribute *attr,
679 char *page)
680{
681 struct gendisk *disk = dev_to_disk(dev);
682
683 if (!disk->bb)
684 return sprintf(page, "\n");
685
686 return badblocks_show(disk->bb, page, 0);
687}
688
689static ssize_t disk_badblocks_store(struct device *dev,
690 struct device_attribute *attr,
691 const char *page, size_t len)
692{
693 struct gendisk *disk = dev_to_disk(dev);
694
695 if (!disk->bb)
696 return -ENXIO;
697
698 return badblocks_store(disk->bb, page, len, 0);
699}
700
701/**
702 * get_gendisk - get partitioning information for a given device
703 * @devt: device to get partitioning information for
704 * @partno: returned partition index
705 *
706 * This function gets the structure containing partitioning
707 * information for the given device @devt.
708 */
709struct gendisk *get_gendisk(dev_t devt, int *partno)
710{
711 struct gendisk *disk = NULL;
712
713 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
714 struct kobject *kobj;
715
716 kobj = kobj_lookup(bdev_map, devt, partno);
717 if (kobj)
718 disk = dev_to_disk(kobj_to_dev(kobj));
719 } else {
720 struct hd_struct *part;
721
722 spin_lock_bh(&ext_devt_lock);
723 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
724 if (part && get_disk(part_to_disk(part))) {
725 *partno = part->partno;
726 disk = part_to_disk(part);
727 }
728 spin_unlock_bh(&ext_devt_lock);
729 }
730
731 return disk;
732}
733EXPORT_SYMBOL(get_gendisk);
734
735/**
736 * bdget_disk - do bdget() by gendisk and partition number
737 * @disk: gendisk of interest
738 * @partno: partition number
739 *
740 * Find partition @partno from @disk, do bdget() on it.
741 *
742 * CONTEXT:
743 * Don't care.
744 *
745 * RETURNS:
746 * Resulting block_device on success, NULL on failure.
747 */
748struct block_device *bdget_disk(struct gendisk *disk, int partno)
749{
750 struct hd_struct *part;
751 struct block_device *bdev = NULL;
752
753 part = disk_get_part(disk, partno);
754 if (part)
755 bdev = bdget(part_devt(part));
756 disk_put_part(part);
757
758 return bdev;
759}
760EXPORT_SYMBOL(bdget_disk);
761
762/*
763 * print a full list of all partitions - intended for places where the root
764 * filesystem can't be mounted and thus to give the victim some idea of what
765 * went wrong
766 */
767void __init printk_all_partitions(void)
768{
769 struct class_dev_iter iter;
770 struct device *dev;
771
772 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
773 while ((dev = class_dev_iter_next(&iter))) {
774 struct gendisk *disk = dev_to_disk(dev);
775 struct disk_part_iter piter;
776 struct hd_struct *part;
777 char name_buf[BDEVNAME_SIZE];
778 char devt_buf[BDEVT_SIZE];
779
780 /*
781 * Don't show empty devices or things that have been
782 * suppressed
783 */
784 if (get_capacity(disk) == 0 ||
785 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
786 continue;
787
788 /*
789 * Note, unlike /proc/partitions, I am showing the
790 * numbers in hex - the same format as the root=
791 * option takes.
792 */
793 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
794 while ((part = disk_part_iter_next(&piter))) {
795 bool is_part0 = part == &disk->part0;
796
797 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
798 bdevt_str(part_devt(part), devt_buf),
799 (unsigned long long)part_nr_sects_read(part) >> 1
800 , disk_name(disk, part->partno, name_buf),
801 part->info ? part->info->uuid : "");
802 if (is_part0) {
803 if (dev->parent && dev->parent->driver)
804 printk(" driver: %s\n",
805 dev->parent->driver->name);
806 else
807 printk(" (driver?)\n");
808 } else
809 printk("\n");
810 }
811 disk_part_iter_exit(&piter);
812 }
813 class_dev_iter_exit(&iter);
814}
815
816#ifdef CONFIG_PROC_FS
817/* iterator */
818static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
819{
820 loff_t skip = *pos;
821 struct class_dev_iter *iter;
822 struct device *dev;
823
824 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
825 if (!iter)
826 return ERR_PTR(-ENOMEM);
827
828 seqf->private = iter;
829 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
830 do {
831 dev = class_dev_iter_next(iter);
832 if (!dev)
833 return NULL;
834 } while (skip--);
835
836 return dev_to_disk(dev);
837}
838
839static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
840{
841 struct device *dev;
842
843 (*pos)++;
844 dev = class_dev_iter_next(seqf->private);
845 if (dev)
846 return dev_to_disk(dev);
847
848 return NULL;
849}
850
851static void disk_seqf_stop(struct seq_file *seqf, void *v)
852{
853 struct class_dev_iter *iter = seqf->private;
854
855 /* stop is called even after start failed :-( */
856 if (iter) {
857 class_dev_iter_exit(iter);
858 kfree(iter);
859 seqf->private = NULL;
860 }
861}
862
863static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
864{
865 void *p;
866
867 p = disk_seqf_start(seqf, pos);
868 if (!IS_ERR_OR_NULL(p) && !*pos)
869 seq_puts(seqf, "major minor #blocks name\n\n");
870 return p;
871}
872
873static int show_partition(struct seq_file *seqf, void *v)
874{
875 struct gendisk *sgp = v;
876 struct disk_part_iter piter;
877 struct hd_struct *part;
878 char buf[BDEVNAME_SIZE];
879
880 /* Don't show non-partitionable removeable devices or empty devices */
881 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
882 (sgp->flags & GENHD_FL_REMOVABLE)))
883 return 0;
884 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
885 return 0;
886
887 /* show the full disk and all non-0 size partitions of it */
888 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
889 while ((part = disk_part_iter_next(&piter)))
890 seq_printf(seqf, "%4d %7d %10llu %s\n",
891 MAJOR(part_devt(part)), MINOR(part_devt(part)),
892 (unsigned long long)part_nr_sects_read(part) >> 1,
893 disk_name(sgp, part->partno, buf));
894 disk_part_iter_exit(&piter);
895
896 return 0;
897}
898
899static const struct seq_operations partitions_op = {
900 .start = show_partition_start,
901 .next = disk_seqf_next,
902 .stop = disk_seqf_stop,
903 .show = show_partition
904};
905
906static int partitions_open(struct inode *inode, struct file *file)
907{
908 return seq_open(file, &partitions_op);
909}
910
911static const struct file_operations proc_partitions_operations = {
912 .open = partitions_open,
913 .read = seq_read,
914 .llseek = seq_lseek,
915 .release = seq_release,
916};
917#endif
918
919
920static struct kobject *base_probe(dev_t devt, int *partno, void *data)
921{
922 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
923 /* Make old-style 2.4 aliases work */
924 request_module("block-major-%d", MAJOR(devt));
925 return NULL;
926}
927
928static int __init genhd_device_init(void)
929{
930 int error;
931
932 block_class.dev_kobj = sysfs_dev_block_kobj;
933 error = class_register(&block_class);
934 if (unlikely(error))
935 return error;
936 bdev_map = kobj_map_init(base_probe, &block_class_lock);
937 blk_dev_init();
938
939 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
940
941 /* create top-level block dir */
942 if (!sysfs_deprecated)
943 block_depr = kobject_create_and_add("block", NULL);
944 return 0;
945}
946
947subsys_initcall(genhd_device_init);
948
949static ssize_t disk_range_show(struct device *dev,
950 struct device_attribute *attr, char *buf)
951{
952 struct gendisk *disk = dev_to_disk(dev);
953
954 return sprintf(buf, "%d\n", disk->minors);
955}
956
957static ssize_t disk_ext_range_show(struct device *dev,
958 struct device_attribute *attr, char *buf)
959{
960 struct gendisk *disk = dev_to_disk(dev);
961
962 return sprintf(buf, "%d\n", disk_max_parts(disk));
963}
964
965static ssize_t disk_removable_show(struct device *dev,
966 struct device_attribute *attr, char *buf)
967{
968 struct gendisk *disk = dev_to_disk(dev);
969
970 return sprintf(buf, "%d\n",
971 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
972}
973
974static ssize_t disk_ro_show(struct device *dev,
975 struct device_attribute *attr, char *buf)
976{
977 struct gendisk *disk = dev_to_disk(dev);
978
979 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
980}
981
982static ssize_t disk_capability_show(struct device *dev,
983 struct device_attribute *attr, char *buf)
984{
985 struct gendisk *disk = dev_to_disk(dev);
986
987 return sprintf(buf, "%x\n", disk->flags);
988}
989
990static ssize_t disk_alignment_offset_show(struct device *dev,
991 struct device_attribute *attr,
992 char *buf)
993{
994 struct gendisk *disk = dev_to_disk(dev);
995
996 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
997}
998
999static ssize_t disk_discard_alignment_show(struct device *dev,
1000 struct device_attribute *attr,
1001 char *buf)
1002{
1003 struct gendisk *disk = dev_to_disk(dev);
1004
1005 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1006}
1007
1008static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
1009static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
1010static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
1011static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
1012static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
1013static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
1014static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
1015 NULL);
1016static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
1017static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
1018static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
1019static DEVICE_ATTR(badblocks, S_IRUGO | S_IWUSR, disk_badblocks_show,
1020 disk_badblocks_store);
1021#ifdef CONFIG_FAIL_MAKE_REQUEST
1022static struct device_attribute dev_attr_fail =
1023 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
1024#endif
1025#ifdef CONFIG_FAIL_IO_TIMEOUT
1026static struct device_attribute dev_attr_fail_timeout =
1027 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
1028 part_timeout_store);
1029#endif
1030
1031static struct attribute *disk_attrs[] = {
1032 &dev_attr_range.attr,
1033 &dev_attr_ext_range.attr,
1034 &dev_attr_removable.attr,
1035 &dev_attr_ro.attr,
1036 &dev_attr_size.attr,
1037 &dev_attr_alignment_offset.attr,
1038 &dev_attr_discard_alignment.attr,
1039 &dev_attr_capability.attr,
1040 &dev_attr_stat.attr,
1041 &dev_attr_inflight.attr,
1042 &dev_attr_badblocks.attr,
1043#ifdef CONFIG_FAIL_MAKE_REQUEST
1044 &dev_attr_fail.attr,
1045#endif
1046#ifdef CONFIG_FAIL_IO_TIMEOUT
1047 &dev_attr_fail_timeout.attr,
1048#endif
1049 NULL
1050};
1051
1052static struct attribute_group disk_attr_group = {
1053 .attrs = disk_attrs,
1054};
1055
1056static const struct attribute_group *disk_attr_groups[] = {
1057 &disk_attr_group,
1058 NULL
1059};
1060
1061/**
1062 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1063 * @disk: disk to replace part_tbl for
1064 * @new_ptbl: new part_tbl to install
1065 *
1066 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1067 * original ptbl is freed using RCU callback.
1068 *
1069 * LOCKING:
1070 * Matching bd_mutx locked.
1071 */
1072static void disk_replace_part_tbl(struct gendisk *disk,
1073 struct disk_part_tbl *new_ptbl)
1074{
1075 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1076
1077 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1078
1079 if (old_ptbl) {
1080 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1081 kfree_rcu(old_ptbl, rcu_head);
1082 }
1083}
1084
1085/**
1086 * disk_expand_part_tbl - expand disk->part_tbl
1087 * @disk: disk to expand part_tbl for
1088 * @partno: expand such that this partno can fit in
1089 *
1090 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1091 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1092 *
1093 * LOCKING:
1094 * Matching bd_mutex locked, might sleep.
1095 *
1096 * RETURNS:
1097 * 0 on success, -errno on failure.
1098 */
1099int disk_expand_part_tbl(struct gendisk *disk, int partno)
1100{
1101 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1102 struct disk_part_tbl *new_ptbl;
1103 int len = old_ptbl ? old_ptbl->len : 0;
1104 int i, target;
1105 size_t size;
1106
1107 /*
1108 * check for int overflow, since we can get here from blkpg_ioctl()
1109 * with a user passed 'partno'.
1110 */
1111 target = partno + 1;
1112 if (target < 0)
1113 return -EINVAL;
1114
1115 /* disk_max_parts() is zero during initialization, ignore if so */
1116 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1117 return -EINVAL;
1118
1119 if (target <= len)
1120 return 0;
1121
1122 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1123 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1124 if (!new_ptbl)
1125 return -ENOMEM;
1126
1127 new_ptbl->len = target;
1128
1129 for (i = 0; i < len; i++)
1130 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1131
1132 disk_replace_part_tbl(disk, new_ptbl);
1133 return 0;
1134}
1135
1136static void disk_release(struct device *dev)
1137{
1138 struct gendisk *disk = dev_to_disk(dev);
1139
1140 blk_free_devt(dev->devt);
1141 disk_release_events(disk);
1142 kfree(disk->random);
1143 disk_replace_part_tbl(disk, NULL);
1144 hd_free_part(&disk->part0);
1145 if (disk->queue)
1146 blk_put_queue(disk->queue);
1147 kfree(disk);
1148}
1149struct class block_class = {
1150 .name = "block",
1151};
1152
1153static char *block_devnode(struct device *dev, umode_t *mode,
1154 kuid_t *uid, kgid_t *gid)
1155{
1156 struct gendisk *disk = dev_to_disk(dev);
1157
1158 if (disk->devnode)
1159 return disk->devnode(disk, mode);
1160 return NULL;
1161}
1162
1163static struct device_type disk_type = {
1164 .name = "disk",
1165 .groups = disk_attr_groups,
1166 .release = disk_release,
1167 .devnode = block_devnode,
1168};
1169
1170#ifdef CONFIG_PROC_FS
1171/*
1172 * aggregate disk stat collector. Uses the same stats that the sysfs
1173 * entries do, above, but makes them available through one seq_file.
1174 *
1175 * The output looks suspiciously like /proc/partitions with a bunch of
1176 * extra fields.
1177 */
1178static int diskstats_show(struct seq_file *seqf, void *v)
1179{
1180 struct gendisk *gp = v;
1181 struct disk_part_iter piter;
1182 struct hd_struct *hd;
1183 char buf[BDEVNAME_SIZE];
1184 int cpu;
1185
1186 /*
1187 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1188 seq_puts(seqf, "major minor name"
1189 " rio rmerge rsect ruse wio wmerge "
1190 "wsect wuse running use aveq"
1191 "\n\n");
1192 */
1193
1194 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1195 while ((hd = disk_part_iter_next(&piter))) {
1196 cpu = part_stat_lock();
1197 part_round_stats(cpu, hd);
1198 part_stat_unlock();
1199 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1200 "%u %lu %lu %lu %u %u %u %u\n",
1201 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1202 disk_name(gp, hd->partno, buf),
1203 part_stat_read(hd, ios[READ]),
1204 part_stat_read(hd, merges[READ]),
1205 part_stat_read(hd, sectors[READ]),
1206 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1207 part_stat_read(hd, ios[WRITE]),
1208 part_stat_read(hd, merges[WRITE]),
1209 part_stat_read(hd, sectors[WRITE]),
1210 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1211 part_in_flight(hd),
1212 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1213 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1214 );
1215 }
1216 disk_part_iter_exit(&piter);
1217
1218 return 0;
1219}
1220
1221static const struct seq_operations diskstats_op = {
1222 .start = disk_seqf_start,
1223 .next = disk_seqf_next,
1224 .stop = disk_seqf_stop,
1225 .show = diskstats_show
1226};
1227
1228static int diskstats_open(struct inode *inode, struct file *file)
1229{
1230 return seq_open(file, &diskstats_op);
1231}
1232
1233static const struct file_operations proc_diskstats_operations = {
1234 .open = diskstats_open,
1235 .read = seq_read,
1236 .llseek = seq_lseek,
1237 .release = seq_release,
1238};
1239
1240static int __init proc_genhd_init(void)
1241{
1242 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1243 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1244 return 0;
1245}
1246module_init(proc_genhd_init);
1247#endif /* CONFIG_PROC_FS */
1248
1249dev_t blk_lookup_devt(const char *name, int partno)
1250{
1251 dev_t devt = MKDEV(0, 0);
1252 struct class_dev_iter iter;
1253 struct device *dev;
1254
1255 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1256 while ((dev = class_dev_iter_next(&iter))) {
1257 struct gendisk *disk = dev_to_disk(dev);
1258 struct hd_struct *part;
1259
1260 if (strcmp(dev_name(dev), name))
1261 continue;
1262
1263 if (partno < disk->minors) {
1264 /* We need to return the right devno, even
1265 * if the partition doesn't exist yet.
1266 */
1267 devt = MKDEV(MAJOR(dev->devt),
1268 MINOR(dev->devt) + partno);
1269 break;
1270 }
1271 part = disk_get_part(disk, partno);
1272 if (part) {
1273 devt = part_devt(part);
1274 disk_put_part(part);
1275 break;
1276 }
1277 disk_put_part(part);
1278 }
1279 class_dev_iter_exit(&iter);
1280 return devt;
1281}
1282EXPORT_SYMBOL(blk_lookup_devt);
1283
1284struct gendisk *alloc_disk(int minors)
1285{
1286 return alloc_disk_node(minors, NUMA_NO_NODE);
1287}
1288EXPORT_SYMBOL(alloc_disk);
1289
1290struct gendisk *alloc_disk_node(int minors, int node_id)
1291{
1292 struct gendisk *disk;
1293
1294 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1295 if (disk) {
1296 if (!init_part_stats(&disk->part0)) {
1297 kfree(disk);
1298 return NULL;
1299 }
1300 disk->node_id = node_id;
1301 if (disk_expand_part_tbl(disk, 0)) {
1302 free_part_stats(&disk->part0);
1303 kfree(disk);
1304 return NULL;
1305 }
1306 disk->part_tbl->part[0] = &disk->part0;
1307
1308 /*
1309 * set_capacity() and get_capacity() currently don't use
1310 * seqcounter to read/update the part0->nr_sects. Still init
1311 * the counter as we can read the sectors in IO submission
1312 * patch using seqence counters.
1313 *
1314 * TODO: Ideally set_capacity() and get_capacity() should be
1315 * converted to make use of bd_mutex and sequence counters.
1316 */
1317 seqcount_init(&disk->part0.nr_sects_seq);
1318 if (hd_ref_init(&disk->part0)) {
1319 hd_free_part(&disk->part0);
1320 kfree(disk);
1321 return NULL;
1322 }
1323
1324 disk->minors = minors;
1325 rand_initialize_disk(disk);
1326 disk_to_dev(disk)->class = &block_class;
1327 disk_to_dev(disk)->type = &disk_type;
1328 device_initialize(disk_to_dev(disk));
1329 }
1330 return disk;
1331}
1332EXPORT_SYMBOL(alloc_disk_node);
1333
1334struct kobject *get_disk(struct gendisk *disk)
1335{
1336 struct module *owner;
1337 struct kobject *kobj;
1338
1339 if (!disk->fops)
1340 return NULL;
1341 owner = disk->fops->owner;
1342 if (owner && !try_module_get(owner))
1343 return NULL;
1344 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1345 if (kobj == NULL) {
1346 module_put(owner);
1347 return NULL;
1348 }
1349 return kobj;
1350
1351}
1352
1353EXPORT_SYMBOL(get_disk);
1354
1355void put_disk(struct gendisk *disk)
1356{
1357 if (disk)
1358 kobject_put(&disk_to_dev(disk)->kobj);
1359}
1360
1361EXPORT_SYMBOL(put_disk);
1362
1363static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1364{
1365 char event[] = "DISK_RO=1";
1366 char *envp[] = { event, NULL };
1367
1368 if (!ro)
1369 event[8] = '0';
1370 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1371}
1372
1373void set_device_ro(struct block_device *bdev, int flag)
1374{
1375 bdev->bd_part->policy = flag;
1376}
1377
1378EXPORT_SYMBOL(set_device_ro);
1379
1380void set_disk_ro(struct gendisk *disk, int flag)
1381{
1382 struct disk_part_iter piter;
1383 struct hd_struct *part;
1384
1385 if (disk->part0.policy != flag) {
1386 set_disk_ro_uevent(disk, flag);
1387 disk->part0.policy = flag;
1388 }
1389
1390 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1391 while ((part = disk_part_iter_next(&piter)))
1392 part->policy = flag;
1393 disk_part_iter_exit(&piter);
1394}
1395
1396EXPORT_SYMBOL(set_disk_ro);
1397
1398int bdev_read_only(struct block_device *bdev)
1399{
1400 if (!bdev)
1401 return 0;
1402 return bdev->bd_part->policy;
1403}
1404
1405EXPORT_SYMBOL(bdev_read_only);
1406
1407int invalidate_partition(struct gendisk *disk, int partno)
1408{
1409 int res = 0;
1410 struct block_device *bdev = bdget_disk(disk, partno);
1411 if (bdev) {
1412 fsync_bdev(bdev);
1413 res = __invalidate_device(bdev, true);
1414 bdput(bdev);
1415 }
1416 return res;
1417}
1418
1419EXPORT_SYMBOL(invalidate_partition);
1420
1421/*
1422 * Disk events - monitor disk events like media change and eject request.
1423 */
1424struct disk_events {
1425 struct list_head node; /* all disk_event's */
1426 struct gendisk *disk; /* the associated disk */
1427 spinlock_t lock;
1428
1429 struct mutex block_mutex; /* protects blocking */
1430 int block; /* event blocking depth */
1431 unsigned int pending; /* events already sent out */
1432 unsigned int clearing; /* events being cleared */
1433
1434 long poll_msecs; /* interval, -1 for default */
1435 struct delayed_work dwork;
1436};
1437
1438static const char *disk_events_strs[] = {
1439 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1440 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1441};
1442
1443static char *disk_uevents[] = {
1444 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1445 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1446};
1447
1448/* list of all disk_events */
1449static DEFINE_MUTEX(disk_events_mutex);
1450static LIST_HEAD(disk_events);
1451
1452/* disable in-kernel polling by default */
1453static unsigned long disk_events_dfl_poll_msecs;
1454
1455static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1456{
1457 struct disk_events *ev = disk->ev;
1458 long intv_msecs = 0;
1459
1460 /*
1461 * If device-specific poll interval is set, always use it. If
1462 * the default is being used, poll iff there are events which
1463 * can't be monitored asynchronously.
1464 */
1465 if (ev->poll_msecs >= 0)
1466 intv_msecs = ev->poll_msecs;
1467 else if (disk->events & ~disk->async_events)
1468 intv_msecs = disk_events_dfl_poll_msecs;
1469
1470 return msecs_to_jiffies(intv_msecs);
1471}
1472
1473/**
1474 * disk_block_events - block and flush disk event checking
1475 * @disk: disk to block events for
1476 *
1477 * On return from this function, it is guaranteed that event checking
1478 * isn't in progress and won't happen until unblocked by
1479 * disk_unblock_events(). Events blocking is counted and the actual
1480 * unblocking happens after the matching number of unblocks are done.
1481 *
1482 * Note that this intentionally does not block event checking from
1483 * disk_clear_events().
1484 *
1485 * CONTEXT:
1486 * Might sleep.
1487 */
1488void disk_block_events(struct gendisk *disk)
1489{
1490 struct disk_events *ev = disk->ev;
1491 unsigned long flags;
1492 bool cancel;
1493
1494 if (!ev)
1495 return;
1496
1497 /*
1498 * Outer mutex ensures that the first blocker completes canceling
1499 * the event work before further blockers are allowed to finish.
1500 */
1501 mutex_lock(&ev->block_mutex);
1502
1503 spin_lock_irqsave(&ev->lock, flags);
1504 cancel = !ev->block++;
1505 spin_unlock_irqrestore(&ev->lock, flags);
1506
1507 if (cancel)
1508 cancel_delayed_work_sync(&disk->ev->dwork);
1509
1510 mutex_unlock(&ev->block_mutex);
1511}
1512
1513static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1514{
1515 struct disk_events *ev = disk->ev;
1516 unsigned long intv;
1517 unsigned long flags;
1518
1519 spin_lock_irqsave(&ev->lock, flags);
1520
1521 if (WARN_ON_ONCE(ev->block <= 0))
1522 goto out_unlock;
1523
1524 if (--ev->block)
1525 goto out_unlock;
1526
1527 intv = disk_events_poll_jiffies(disk);
1528 if (check_now)
1529 queue_delayed_work(system_freezable_power_efficient_wq,
1530 &ev->dwork, 0);
1531 else if (intv)
1532 queue_delayed_work(system_freezable_power_efficient_wq,
1533 &ev->dwork, intv);
1534out_unlock:
1535 spin_unlock_irqrestore(&ev->lock, flags);
1536}
1537
1538/**
1539 * disk_unblock_events - unblock disk event checking
1540 * @disk: disk to unblock events for
1541 *
1542 * Undo disk_block_events(). When the block count reaches zero, it
1543 * starts events polling if configured.
1544 *
1545 * CONTEXT:
1546 * Don't care. Safe to call from irq context.
1547 */
1548void disk_unblock_events(struct gendisk *disk)
1549{
1550 if (disk->ev)
1551 __disk_unblock_events(disk, false);
1552}
1553
1554/**
1555 * disk_flush_events - schedule immediate event checking and flushing
1556 * @disk: disk to check and flush events for
1557 * @mask: events to flush
1558 *
1559 * Schedule immediate event checking on @disk if not blocked. Events in
1560 * @mask are scheduled to be cleared from the driver. Note that this
1561 * doesn't clear the events from @disk->ev.
1562 *
1563 * CONTEXT:
1564 * If @mask is non-zero must be called with bdev->bd_mutex held.
1565 */
1566void disk_flush_events(struct gendisk *disk, unsigned int mask)
1567{
1568 struct disk_events *ev = disk->ev;
1569
1570 if (!ev)
1571 return;
1572
1573 spin_lock_irq(&ev->lock);
1574 ev->clearing |= mask;
1575 if (!ev->block)
1576 mod_delayed_work(system_freezable_power_efficient_wq,
1577 &ev->dwork, 0);
1578 spin_unlock_irq(&ev->lock);
1579}
1580
1581/**
1582 * disk_clear_events - synchronously check, clear and return pending events
1583 * @disk: disk to fetch and clear events from
1584 * @mask: mask of events to be fetched and cleared
1585 *
1586 * Disk events are synchronously checked and pending events in @mask
1587 * are cleared and returned. This ignores the block count.
1588 *
1589 * CONTEXT:
1590 * Might sleep.
1591 */
1592unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1593{
1594 const struct block_device_operations *bdops = disk->fops;
1595 struct disk_events *ev = disk->ev;
1596 unsigned int pending;
1597 unsigned int clearing = mask;
1598
1599 if (!ev) {
1600 /* for drivers still using the old ->media_changed method */
1601 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1602 bdops->media_changed && bdops->media_changed(disk))
1603 return DISK_EVENT_MEDIA_CHANGE;
1604 return 0;
1605 }
1606
1607 disk_block_events(disk);
1608
1609 /*
1610 * store the union of mask and ev->clearing on the stack so that the
1611 * race with disk_flush_events does not cause ambiguity (ev->clearing
1612 * can still be modified even if events are blocked).
1613 */
1614 spin_lock_irq(&ev->lock);
1615 clearing |= ev->clearing;
1616 ev->clearing = 0;
1617 spin_unlock_irq(&ev->lock);
1618
1619 disk_check_events(ev, &clearing);
1620 /*
1621 * if ev->clearing is not 0, the disk_flush_events got called in the
1622 * middle of this function, so we want to run the workfn without delay.
1623 */
1624 __disk_unblock_events(disk, ev->clearing ? true : false);
1625
1626 /* then, fetch and clear pending events */
1627 spin_lock_irq(&ev->lock);
1628 pending = ev->pending & mask;
1629 ev->pending &= ~mask;
1630 spin_unlock_irq(&ev->lock);
1631 WARN_ON_ONCE(clearing & mask);
1632
1633 return pending;
1634}
1635
1636/*
1637 * Separate this part out so that a different pointer for clearing_ptr can be
1638 * passed in for disk_clear_events.
1639 */
1640static void disk_events_workfn(struct work_struct *work)
1641{
1642 struct delayed_work *dwork = to_delayed_work(work);
1643 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1644
1645 disk_check_events(ev, &ev->clearing);
1646}
1647
1648static void disk_check_events(struct disk_events *ev,
1649 unsigned int *clearing_ptr)
1650{
1651 struct gendisk *disk = ev->disk;
1652 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1653 unsigned int clearing = *clearing_ptr;
1654 unsigned int events;
1655 unsigned long intv;
1656 int nr_events = 0, i;
1657
1658 /* check events */
1659 events = disk->fops->check_events(disk, clearing);
1660
1661 /* accumulate pending events and schedule next poll if necessary */
1662 spin_lock_irq(&ev->lock);
1663
1664 events &= ~ev->pending;
1665 ev->pending |= events;
1666 *clearing_ptr &= ~clearing;
1667
1668 intv = disk_events_poll_jiffies(disk);
1669 if (!ev->block && intv)
1670 queue_delayed_work(system_freezable_power_efficient_wq,
1671 &ev->dwork, intv);
1672
1673 spin_unlock_irq(&ev->lock);
1674
1675 /*
1676 * Tell userland about new events. Only the events listed in
1677 * @disk->events are reported. Unlisted events are processed the
1678 * same internally but never get reported to userland.
1679 */
1680 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1681 if (events & disk->events & (1 << i))
1682 envp[nr_events++] = disk_uevents[i];
1683
1684 if (nr_events)
1685 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1686}
1687
1688/*
1689 * A disk events enabled device has the following sysfs nodes under
1690 * its /sys/block/X/ directory.
1691 *
1692 * events : list of all supported events
1693 * events_async : list of events which can be detected w/o polling
1694 * events_poll_msecs : polling interval, 0: disable, -1: system default
1695 */
1696static ssize_t __disk_events_show(unsigned int events, char *buf)
1697{
1698 const char *delim = "";
1699 ssize_t pos = 0;
1700 int i;
1701
1702 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1703 if (events & (1 << i)) {
1704 pos += sprintf(buf + pos, "%s%s",
1705 delim, disk_events_strs[i]);
1706 delim = " ";
1707 }
1708 if (pos)
1709 pos += sprintf(buf + pos, "\n");
1710 return pos;
1711}
1712
1713static ssize_t disk_events_show(struct device *dev,
1714 struct device_attribute *attr, char *buf)
1715{
1716 struct gendisk *disk = dev_to_disk(dev);
1717
1718 return __disk_events_show(disk->events, buf);
1719}
1720
1721static ssize_t disk_events_async_show(struct device *dev,
1722 struct device_attribute *attr, char *buf)
1723{
1724 struct gendisk *disk = dev_to_disk(dev);
1725
1726 return __disk_events_show(disk->async_events, buf);
1727}
1728
1729static ssize_t disk_events_poll_msecs_show(struct device *dev,
1730 struct device_attribute *attr,
1731 char *buf)
1732{
1733 struct gendisk *disk = dev_to_disk(dev);
1734
1735 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1736}
1737
1738static ssize_t disk_events_poll_msecs_store(struct device *dev,
1739 struct device_attribute *attr,
1740 const char *buf, size_t count)
1741{
1742 struct gendisk *disk = dev_to_disk(dev);
1743 long intv;
1744
1745 if (!count || !sscanf(buf, "%ld", &intv))
1746 return -EINVAL;
1747
1748 if (intv < 0 && intv != -1)
1749 return -EINVAL;
1750
1751 disk_block_events(disk);
1752 disk->ev->poll_msecs = intv;
1753 __disk_unblock_events(disk, true);
1754
1755 return count;
1756}
1757
1758static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1759static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1760static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1761 disk_events_poll_msecs_show,
1762 disk_events_poll_msecs_store);
1763
1764static const struct attribute *disk_events_attrs[] = {
1765 &dev_attr_events.attr,
1766 &dev_attr_events_async.attr,
1767 &dev_attr_events_poll_msecs.attr,
1768 NULL,
1769};
1770
1771/*
1772 * The default polling interval can be specified by the kernel
1773 * parameter block.events_dfl_poll_msecs which defaults to 0
1774 * (disable). This can also be modified runtime by writing to
1775 * /sys/module/block/events_dfl_poll_msecs.
1776 */
1777static int disk_events_set_dfl_poll_msecs(const char *val,
1778 const struct kernel_param *kp)
1779{
1780 struct disk_events *ev;
1781 int ret;
1782
1783 ret = param_set_ulong(val, kp);
1784 if (ret < 0)
1785 return ret;
1786
1787 mutex_lock(&disk_events_mutex);
1788
1789 list_for_each_entry(ev, &disk_events, node)
1790 disk_flush_events(ev->disk, 0);
1791
1792 mutex_unlock(&disk_events_mutex);
1793
1794 return 0;
1795}
1796
1797static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1798 .set = disk_events_set_dfl_poll_msecs,
1799 .get = param_get_ulong,
1800};
1801
1802#undef MODULE_PARAM_PREFIX
1803#define MODULE_PARAM_PREFIX "block."
1804
1805module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1806 &disk_events_dfl_poll_msecs, 0644);
1807
1808/*
1809 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1810 */
1811static void disk_alloc_events(struct gendisk *disk)
1812{
1813 struct disk_events *ev;
1814
1815 if (!disk->fops->check_events)
1816 return;
1817
1818 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1819 if (!ev) {
1820 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1821 return;
1822 }
1823
1824 INIT_LIST_HEAD(&ev->node);
1825 ev->disk = disk;
1826 spin_lock_init(&ev->lock);
1827 mutex_init(&ev->block_mutex);
1828 ev->block = 1;
1829 ev->poll_msecs = -1;
1830 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1831
1832 disk->ev = ev;
1833}
1834
1835static void disk_add_events(struct gendisk *disk)
1836{
1837 if (!disk->ev)
1838 return;
1839
1840 /* FIXME: error handling */
1841 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1842 pr_warn("%s: failed to create sysfs files for events\n",
1843 disk->disk_name);
1844
1845 mutex_lock(&disk_events_mutex);
1846 list_add_tail(&disk->ev->node, &disk_events);
1847 mutex_unlock(&disk_events_mutex);
1848
1849 /*
1850 * Block count is initialized to 1 and the following initial
1851 * unblock kicks it into action.
1852 */
1853 __disk_unblock_events(disk, true);
1854}
1855
1856static void disk_del_events(struct gendisk *disk)
1857{
1858 if (!disk->ev)
1859 return;
1860
1861 disk_block_events(disk);
1862
1863 mutex_lock(&disk_events_mutex);
1864 list_del_init(&disk->ev->node);
1865 mutex_unlock(&disk_events_mutex);
1866
1867 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1868}
1869
1870static void disk_release_events(struct gendisk *disk)
1871{
1872 /* the block count should be 1 from disk_del_events() */
1873 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1874 kfree(disk->ev);
1875}