Loading...
1/*
2 * gendisk handling
3 */
4
5#include <linux/module.h>
6#include <linux/fs.h>
7#include <linux/genhd.h>
8#include <linux/kdev_t.h>
9#include <linux/kernel.h>
10#include <linux/blkdev.h>
11#include <linux/init.h>
12#include <linux/spinlock.h>
13#include <linux/proc_fs.h>
14#include <linux/seq_file.h>
15#include <linux/slab.h>
16#include <linux/kmod.h>
17#include <linux/kobj_map.h>
18#include <linux/buffer_head.h>
19#include <linux/mutex.h>
20#include <linux/idr.h>
21#include <linux/log2.h>
22
23#include "blk.h"
24
25static DEFINE_MUTEX(block_class_lock);
26struct kobject *block_depr;
27
28/* for extended dynamic devt allocation, currently only one major is used */
29#define MAX_EXT_DEVT (1 << MINORBITS)
30
31/* For extended devt allocation. ext_devt_mutex prevents look up
32 * results from going away underneath its user.
33 */
34static DEFINE_MUTEX(ext_devt_mutex);
35static DEFINE_IDR(ext_devt_idr);
36
37static struct device_type disk_type;
38
39static void disk_add_events(struct gendisk *disk);
40static void disk_del_events(struct gendisk *disk);
41static void disk_release_events(struct gendisk *disk);
42
43/**
44 * disk_get_part - get partition
45 * @disk: disk to look partition from
46 * @partno: partition number
47 *
48 * Look for partition @partno from @disk. If found, increment
49 * reference count and return it.
50 *
51 * CONTEXT:
52 * Don't care.
53 *
54 * RETURNS:
55 * Pointer to the found partition on success, NULL if not found.
56 */
57struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
58{
59 struct hd_struct *part = NULL;
60 struct disk_part_tbl *ptbl;
61
62 if (unlikely(partno < 0))
63 return NULL;
64
65 rcu_read_lock();
66
67 ptbl = rcu_dereference(disk->part_tbl);
68 if (likely(partno < ptbl->len)) {
69 part = rcu_dereference(ptbl->part[partno]);
70 if (part)
71 get_device(part_to_dev(part));
72 }
73
74 rcu_read_unlock();
75
76 return part;
77}
78EXPORT_SYMBOL_GPL(disk_get_part);
79
80/**
81 * disk_part_iter_init - initialize partition iterator
82 * @piter: iterator to initialize
83 * @disk: disk to iterate over
84 * @flags: DISK_PITER_* flags
85 *
86 * Initialize @piter so that it iterates over partitions of @disk.
87 *
88 * CONTEXT:
89 * Don't care.
90 */
91void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
92 unsigned int flags)
93{
94 struct disk_part_tbl *ptbl;
95
96 rcu_read_lock();
97 ptbl = rcu_dereference(disk->part_tbl);
98
99 piter->disk = disk;
100 piter->part = NULL;
101
102 if (flags & DISK_PITER_REVERSE)
103 piter->idx = ptbl->len - 1;
104 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
105 piter->idx = 0;
106 else
107 piter->idx = 1;
108
109 piter->flags = flags;
110
111 rcu_read_unlock();
112}
113EXPORT_SYMBOL_GPL(disk_part_iter_init);
114
115/**
116 * disk_part_iter_next - proceed iterator to the next partition and return it
117 * @piter: iterator of interest
118 *
119 * Proceed @piter to the next partition and return it.
120 *
121 * CONTEXT:
122 * Don't care.
123 */
124struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
125{
126 struct disk_part_tbl *ptbl;
127 int inc, end;
128
129 /* put the last partition */
130 disk_put_part(piter->part);
131 piter->part = NULL;
132
133 /* get part_tbl */
134 rcu_read_lock();
135 ptbl = rcu_dereference(piter->disk->part_tbl);
136
137 /* determine iteration parameters */
138 if (piter->flags & DISK_PITER_REVERSE) {
139 inc = -1;
140 if (piter->flags & (DISK_PITER_INCL_PART0 |
141 DISK_PITER_INCL_EMPTY_PART0))
142 end = -1;
143 else
144 end = 0;
145 } else {
146 inc = 1;
147 end = ptbl->len;
148 }
149
150 /* iterate to the next partition */
151 for (; piter->idx != end; piter->idx += inc) {
152 struct hd_struct *part;
153
154 part = rcu_dereference(ptbl->part[piter->idx]);
155 if (!part)
156 continue;
157 if (!part->nr_sects &&
158 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
159 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
160 piter->idx == 0))
161 continue;
162
163 get_device(part_to_dev(part));
164 piter->part = part;
165 piter->idx += inc;
166 break;
167 }
168
169 rcu_read_unlock();
170
171 return piter->part;
172}
173EXPORT_SYMBOL_GPL(disk_part_iter_next);
174
175/**
176 * disk_part_iter_exit - finish up partition iteration
177 * @piter: iter of interest
178 *
179 * Called when iteration is over. Cleans up @piter.
180 *
181 * CONTEXT:
182 * Don't care.
183 */
184void disk_part_iter_exit(struct disk_part_iter *piter)
185{
186 disk_put_part(piter->part);
187 piter->part = NULL;
188}
189EXPORT_SYMBOL_GPL(disk_part_iter_exit);
190
191static inline int sector_in_part(struct hd_struct *part, sector_t sector)
192{
193 return part->start_sect <= sector &&
194 sector < part->start_sect + part->nr_sects;
195}
196
197/**
198 * disk_map_sector_rcu - map sector to partition
199 * @disk: gendisk of interest
200 * @sector: sector to map
201 *
202 * Find out which partition @sector maps to on @disk. This is
203 * primarily used for stats accounting.
204 *
205 * CONTEXT:
206 * RCU read locked. The returned partition pointer is valid only
207 * while preemption is disabled.
208 *
209 * RETURNS:
210 * Found partition on success, part0 is returned if no partition matches
211 */
212struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
213{
214 struct disk_part_tbl *ptbl;
215 struct hd_struct *part;
216 int i;
217
218 ptbl = rcu_dereference(disk->part_tbl);
219
220 part = rcu_dereference(ptbl->last_lookup);
221 if (part && sector_in_part(part, sector))
222 return part;
223
224 for (i = 1; i < ptbl->len; i++) {
225 part = rcu_dereference(ptbl->part[i]);
226
227 if (part && sector_in_part(part, sector)) {
228 rcu_assign_pointer(ptbl->last_lookup, part);
229 return part;
230 }
231 }
232 return &disk->part0;
233}
234EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
235
236/*
237 * Can be deleted altogether. Later.
238 *
239 */
240static struct blk_major_name {
241 struct blk_major_name *next;
242 int major;
243 char name[16];
244} *major_names[BLKDEV_MAJOR_HASH_SIZE];
245
246/* index in the above - for now: assume no multimajor ranges */
247static inline int major_to_index(unsigned major)
248{
249 return major % BLKDEV_MAJOR_HASH_SIZE;
250}
251
252#ifdef CONFIG_PROC_FS
253void blkdev_show(struct seq_file *seqf, off_t offset)
254{
255 struct blk_major_name *dp;
256
257 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
258 mutex_lock(&block_class_lock);
259 for (dp = major_names[offset]; dp; dp = dp->next)
260 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
261 mutex_unlock(&block_class_lock);
262 }
263}
264#endif /* CONFIG_PROC_FS */
265
266/**
267 * register_blkdev - register a new block device
268 *
269 * @major: the requested major device number [1..255]. If @major=0, try to
270 * allocate any unused major number.
271 * @name: the name of the new block device as a zero terminated string
272 *
273 * The @name must be unique within the system.
274 *
275 * The return value depends on the @major input parameter.
276 * - if a major device number was requested in range [1..255] then the
277 * function returns zero on success, or a negative error code
278 * - if any unused major number was requested with @major=0 parameter
279 * then the return value is the allocated major number in range
280 * [1..255] or a negative error code otherwise
281 */
282int register_blkdev(unsigned int major, const char *name)
283{
284 struct blk_major_name **n, *p;
285 int index, ret = 0;
286
287 mutex_lock(&block_class_lock);
288
289 /* temporary */
290 if (major == 0) {
291 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
292 if (major_names[index] == NULL)
293 break;
294 }
295
296 if (index == 0) {
297 printk("register_blkdev: failed to get major for %s\n",
298 name);
299 ret = -EBUSY;
300 goto out;
301 }
302 major = index;
303 ret = major;
304 }
305
306 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
307 if (p == NULL) {
308 ret = -ENOMEM;
309 goto out;
310 }
311
312 p->major = major;
313 strlcpy(p->name, name, sizeof(p->name));
314 p->next = NULL;
315 index = major_to_index(major);
316
317 for (n = &major_names[index]; *n; n = &(*n)->next) {
318 if ((*n)->major == major)
319 break;
320 }
321 if (!*n)
322 *n = p;
323 else
324 ret = -EBUSY;
325
326 if (ret < 0) {
327 printk("register_blkdev: cannot get major %d for %s\n",
328 major, name);
329 kfree(p);
330 }
331out:
332 mutex_unlock(&block_class_lock);
333 return ret;
334}
335
336EXPORT_SYMBOL(register_blkdev);
337
338void unregister_blkdev(unsigned int major, const char *name)
339{
340 struct blk_major_name **n;
341 struct blk_major_name *p = NULL;
342 int index = major_to_index(major);
343
344 mutex_lock(&block_class_lock);
345 for (n = &major_names[index]; *n; n = &(*n)->next)
346 if ((*n)->major == major)
347 break;
348 if (!*n || strcmp((*n)->name, name)) {
349 WARN_ON(1);
350 } else {
351 p = *n;
352 *n = p->next;
353 }
354 mutex_unlock(&block_class_lock);
355 kfree(p);
356}
357
358EXPORT_SYMBOL(unregister_blkdev);
359
360static struct kobj_map *bdev_map;
361
362/**
363 * blk_mangle_minor - scatter minor numbers apart
364 * @minor: minor number to mangle
365 *
366 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
367 * is enabled. Mangling twice gives the original value.
368 *
369 * RETURNS:
370 * Mangled value.
371 *
372 * CONTEXT:
373 * Don't care.
374 */
375static int blk_mangle_minor(int minor)
376{
377#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
378 int i;
379
380 for (i = 0; i < MINORBITS / 2; i++) {
381 int low = minor & (1 << i);
382 int high = minor & (1 << (MINORBITS - 1 - i));
383 int distance = MINORBITS - 1 - 2 * i;
384
385 minor ^= low | high; /* clear both bits */
386 low <<= distance; /* swap the positions */
387 high >>= distance;
388 minor |= low | high; /* and set */
389 }
390#endif
391 return minor;
392}
393
394/**
395 * blk_alloc_devt - allocate a dev_t for a partition
396 * @part: partition to allocate dev_t for
397 * @devt: out parameter for resulting dev_t
398 *
399 * Allocate a dev_t for block device.
400 *
401 * RETURNS:
402 * 0 on success, allocated dev_t is returned in *@devt. -errno on
403 * failure.
404 *
405 * CONTEXT:
406 * Might sleep.
407 */
408int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
409{
410 struct gendisk *disk = part_to_disk(part);
411 int idx, rc;
412
413 /* in consecutive minor range? */
414 if (part->partno < disk->minors) {
415 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
416 return 0;
417 }
418
419 /* allocate ext devt */
420 do {
421 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
422 return -ENOMEM;
423 rc = idr_get_new(&ext_devt_idr, part, &idx);
424 } while (rc == -EAGAIN);
425
426 if (rc)
427 return rc;
428
429 if (idx > MAX_EXT_DEVT) {
430 idr_remove(&ext_devt_idr, idx);
431 return -EBUSY;
432 }
433
434 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
435 return 0;
436}
437
438/**
439 * blk_free_devt - free a dev_t
440 * @devt: dev_t to free
441 *
442 * Free @devt which was allocated using blk_alloc_devt().
443 *
444 * CONTEXT:
445 * Might sleep.
446 */
447void blk_free_devt(dev_t devt)
448{
449 might_sleep();
450
451 if (devt == MKDEV(0, 0))
452 return;
453
454 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
455 mutex_lock(&ext_devt_mutex);
456 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
457 mutex_unlock(&ext_devt_mutex);
458 }
459}
460
461static char *bdevt_str(dev_t devt, char *buf)
462{
463 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
464 char tbuf[BDEVT_SIZE];
465 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
466 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
467 } else
468 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
469
470 return buf;
471}
472
473/*
474 * Register device numbers dev..(dev+range-1)
475 * range must be nonzero
476 * The hash chain is sorted on range, so that subranges can override.
477 */
478void blk_register_region(dev_t devt, unsigned long range, struct module *module,
479 struct kobject *(*probe)(dev_t, int *, void *),
480 int (*lock)(dev_t, void *), void *data)
481{
482 kobj_map(bdev_map, devt, range, module, probe, lock, data);
483}
484
485EXPORT_SYMBOL(blk_register_region);
486
487void blk_unregister_region(dev_t devt, unsigned long range)
488{
489 kobj_unmap(bdev_map, devt, range);
490}
491
492EXPORT_SYMBOL(blk_unregister_region);
493
494static struct kobject *exact_match(dev_t devt, int *partno, void *data)
495{
496 struct gendisk *p = data;
497
498 return &disk_to_dev(p)->kobj;
499}
500
501static int exact_lock(dev_t devt, void *data)
502{
503 struct gendisk *p = data;
504
505 if (!get_disk(p))
506 return -1;
507 return 0;
508}
509
510void register_disk(struct gendisk *disk)
511{
512 struct device *ddev = disk_to_dev(disk);
513 struct block_device *bdev;
514 struct disk_part_iter piter;
515 struct hd_struct *part;
516 int err;
517
518 ddev->parent = disk->driverfs_dev;
519
520 dev_set_name(ddev, disk->disk_name);
521
522 /* delay uevents, until we scanned partition table */
523 dev_set_uevent_suppress(ddev, 1);
524
525 if (device_add(ddev))
526 return;
527 if (!sysfs_deprecated) {
528 err = sysfs_create_link(block_depr, &ddev->kobj,
529 kobject_name(&ddev->kobj));
530 if (err) {
531 device_del(ddev);
532 return;
533 }
534 }
535 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
536 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
537
538 /* No minors to use for partitions */
539 if (!disk_partitionable(disk))
540 goto exit;
541
542 /* No such device (e.g., media were just removed) */
543 if (!get_capacity(disk))
544 goto exit;
545
546 bdev = bdget_disk(disk, 0);
547 if (!bdev)
548 goto exit;
549
550 bdev->bd_invalidated = 1;
551 err = blkdev_get(bdev, FMODE_READ, NULL);
552 if (err < 0)
553 goto exit;
554 blkdev_put(bdev, FMODE_READ);
555
556exit:
557 /* announce disk after possible partitions are created */
558 dev_set_uevent_suppress(ddev, 0);
559 kobject_uevent(&ddev->kobj, KOBJ_ADD);
560
561 /* announce possible partitions */
562 disk_part_iter_init(&piter, disk, 0);
563 while ((part = disk_part_iter_next(&piter)))
564 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
565 disk_part_iter_exit(&piter);
566}
567
568/**
569 * add_disk - add partitioning information to kernel list
570 * @disk: per-device partitioning information
571 *
572 * This function registers the partitioning information in @disk
573 * with the kernel.
574 *
575 * FIXME: error handling
576 */
577void add_disk(struct gendisk *disk)
578{
579 struct backing_dev_info *bdi;
580 dev_t devt;
581 int retval;
582
583 /* minors == 0 indicates to use ext devt from part0 and should
584 * be accompanied with EXT_DEVT flag. Make sure all
585 * parameters make sense.
586 */
587 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
588 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
589
590 disk->flags |= GENHD_FL_UP;
591
592 retval = blk_alloc_devt(&disk->part0, &devt);
593 if (retval) {
594 WARN_ON(1);
595 return;
596 }
597 disk_to_dev(disk)->devt = devt;
598
599 /* ->major and ->first_minor aren't supposed to be
600 * dereferenced from here on, but set them just in case.
601 */
602 disk->major = MAJOR(devt);
603 disk->first_minor = MINOR(devt);
604
605 /* Register BDI before referencing it from bdev */
606 bdi = &disk->queue->backing_dev_info;
607 bdi_register_dev(bdi, disk_devt(disk));
608
609 blk_register_region(disk_devt(disk), disk->minors, NULL,
610 exact_match, exact_lock, disk);
611 register_disk(disk);
612 blk_register_queue(disk);
613
614 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
615 "bdi");
616 WARN_ON(retval);
617
618 disk_add_events(disk);
619}
620EXPORT_SYMBOL(add_disk);
621
622void del_gendisk(struct gendisk *disk)
623{
624 struct disk_part_iter piter;
625 struct hd_struct *part;
626
627 disk_del_events(disk);
628
629 /* invalidate stuff */
630 disk_part_iter_init(&piter, disk,
631 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
632 while ((part = disk_part_iter_next(&piter))) {
633 invalidate_partition(disk, part->partno);
634 delete_partition(disk, part->partno);
635 }
636 disk_part_iter_exit(&piter);
637
638 invalidate_partition(disk, 0);
639 blk_free_devt(disk_to_dev(disk)->devt);
640 set_capacity(disk, 0);
641 disk->flags &= ~GENHD_FL_UP;
642
643 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
644 bdi_unregister(&disk->queue->backing_dev_info);
645 blk_unregister_queue(disk);
646 blk_unregister_region(disk_devt(disk), disk->minors);
647
648 part_stat_set_all(&disk->part0, 0);
649 disk->part0.stamp = 0;
650
651 kobject_put(disk->part0.holder_dir);
652 kobject_put(disk->slave_dir);
653 disk->driverfs_dev = NULL;
654 if (!sysfs_deprecated)
655 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
656 device_del(disk_to_dev(disk));
657}
658EXPORT_SYMBOL(del_gendisk);
659
660/**
661 * get_gendisk - get partitioning information for a given device
662 * @devt: device to get partitioning information for
663 * @partno: returned partition index
664 *
665 * This function gets the structure containing partitioning
666 * information for the given device @devt.
667 */
668struct gendisk *get_gendisk(dev_t devt, int *partno)
669{
670 struct gendisk *disk = NULL;
671
672 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
673 struct kobject *kobj;
674
675 kobj = kobj_lookup(bdev_map, devt, partno);
676 if (kobj)
677 disk = dev_to_disk(kobj_to_dev(kobj));
678 } else {
679 struct hd_struct *part;
680
681 mutex_lock(&ext_devt_mutex);
682 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
683 if (part && get_disk(part_to_disk(part))) {
684 *partno = part->partno;
685 disk = part_to_disk(part);
686 }
687 mutex_unlock(&ext_devt_mutex);
688 }
689
690 return disk;
691}
692EXPORT_SYMBOL(get_gendisk);
693
694/**
695 * bdget_disk - do bdget() by gendisk and partition number
696 * @disk: gendisk of interest
697 * @partno: partition number
698 *
699 * Find partition @partno from @disk, do bdget() on it.
700 *
701 * CONTEXT:
702 * Don't care.
703 *
704 * RETURNS:
705 * Resulting block_device on success, NULL on failure.
706 */
707struct block_device *bdget_disk(struct gendisk *disk, int partno)
708{
709 struct hd_struct *part;
710 struct block_device *bdev = NULL;
711
712 part = disk_get_part(disk, partno);
713 if (part)
714 bdev = bdget(part_devt(part));
715 disk_put_part(part);
716
717 return bdev;
718}
719EXPORT_SYMBOL(bdget_disk);
720
721/*
722 * print a full list of all partitions - intended for places where the root
723 * filesystem can't be mounted and thus to give the victim some idea of what
724 * went wrong
725 */
726void __init printk_all_partitions(void)
727{
728 struct class_dev_iter iter;
729 struct device *dev;
730
731 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
732 while ((dev = class_dev_iter_next(&iter))) {
733 struct gendisk *disk = dev_to_disk(dev);
734 struct disk_part_iter piter;
735 struct hd_struct *part;
736 char name_buf[BDEVNAME_SIZE];
737 char devt_buf[BDEVT_SIZE];
738 u8 uuid[PARTITION_META_INFO_UUIDLTH * 2 + 1];
739
740 /*
741 * Don't show empty devices or things that have been
742 * suppressed
743 */
744 if (get_capacity(disk) == 0 ||
745 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
746 continue;
747
748 /*
749 * Note, unlike /proc/partitions, I am showing the
750 * numbers in hex - the same format as the root=
751 * option takes.
752 */
753 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
754 while ((part = disk_part_iter_next(&piter))) {
755 bool is_part0 = part == &disk->part0;
756
757 uuid[0] = 0;
758 if (part->info)
759 part_unpack_uuid(part->info->uuid, uuid);
760
761 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
762 bdevt_str(part_devt(part), devt_buf),
763 (unsigned long long)part->nr_sects >> 1,
764 disk_name(disk, part->partno, name_buf), uuid);
765 if (is_part0) {
766 if (disk->driverfs_dev != NULL &&
767 disk->driverfs_dev->driver != NULL)
768 printk(" driver: %s\n",
769 disk->driverfs_dev->driver->name);
770 else
771 printk(" (driver?)\n");
772 } else
773 printk("\n");
774 }
775 disk_part_iter_exit(&piter);
776 }
777 class_dev_iter_exit(&iter);
778}
779
780#ifdef CONFIG_PROC_FS
781/* iterator */
782static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
783{
784 loff_t skip = *pos;
785 struct class_dev_iter *iter;
786 struct device *dev;
787
788 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
789 if (!iter)
790 return ERR_PTR(-ENOMEM);
791
792 seqf->private = iter;
793 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
794 do {
795 dev = class_dev_iter_next(iter);
796 if (!dev)
797 return NULL;
798 } while (skip--);
799
800 return dev_to_disk(dev);
801}
802
803static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
804{
805 struct device *dev;
806
807 (*pos)++;
808 dev = class_dev_iter_next(seqf->private);
809 if (dev)
810 return dev_to_disk(dev);
811
812 return NULL;
813}
814
815static void disk_seqf_stop(struct seq_file *seqf, void *v)
816{
817 struct class_dev_iter *iter = seqf->private;
818
819 /* stop is called even after start failed :-( */
820 if (iter) {
821 class_dev_iter_exit(iter);
822 kfree(iter);
823 }
824}
825
826static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
827{
828 static void *p;
829
830 p = disk_seqf_start(seqf, pos);
831 if (!IS_ERR_OR_NULL(p) && !*pos)
832 seq_puts(seqf, "major minor #blocks name\n\n");
833 return p;
834}
835
836static int show_partition(struct seq_file *seqf, void *v)
837{
838 struct gendisk *sgp = v;
839 struct disk_part_iter piter;
840 struct hd_struct *part;
841 char buf[BDEVNAME_SIZE];
842
843 /* Don't show non-partitionable removeable devices or empty devices */
844 if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
845 (sgp->flags & GENHD_FL_REMOVABLE)))
846 return 0;
847 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
848 return 0;
849
850 /* show the full disk and all non-0 size partitions of it */
851 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
852 while ((part = disk_part_iter_next(&piter)))
853 seq_printf(seqf, "%4d %7d %10llu %s\n",
854 MAJOR(part_devt(part)), MINOR(part_devt(part)),
855 (unsigned long long)part->nr_sects >> 1,
856 disk_name(sgp, part->partno, buf));
857 disk_part_iter_exit(&piter);
858
859 return 0;
860}
861
862static const struct seq_operations partitions_op = {
863 .start = show_partition_start,
864 .next = disk_seqf_next,
865 .stop = disk_seqf_stop,
866 .show = show_partition
867};
868
869static int partitions_open(struct inode *inode, struct file *file)
870{
871 return seq_open(file, &partitions_op);
872}
873
874static const struct file_operations proc_partitions_operations = {
875 .open = partitions_open,
876 .read = seq_read,
877 .llseek = seq_lseek,
878 .release = seq_release,
879};
880#endif
881
882
883static struct kobject *base_probe(dev_t devt, int *partno, void *data)
884{
885 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
886 /* Make old-style 2.4 aliases work */
887 request_module("block-major-%d", MAJOR(devt));
888 return NULL;
889}
890
891static int __init genhd_device_init(void)
892{
893 int error;
894
895 block_class.dev_kobj = sysfs_dev_block_kobj;
896 error = class_register(&block_class);
897 if (unlikely(error))
898 return error;
899 bdev_map = kobj_map_init(base_probe, &block_class_lock);
900 blk_dev_init();
901
902 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
903
904 /* create top-level block dir */
905 if (!sysfs_deprecated)
906 block_depr = kobject_create_and_add("block", NULL);
907 return 0;
908}
909
910subsys_initcall(genhd_device_init);
911
912static ssize_t disk_range_show(struct device *dev,
913 struct device_attribute *attr, char *buf)
914{
915 struct gendisk *disk = dev_to_disk(dev);
916
917 return sprintf(buf, "%d\n", disk->minors);
918}
919
920static ssize_t disk_ext_range_show(struct device *dev,
921 struct device_attribute *attr, char *buf)
922{
923 struct gendisk *disk = dev_to_disk(dev);
924
925 return sprintf(buf, "%d\n", disk_max_parts(disk));
926}
927
928static ssize_t disk_removable_show(struct device *dev,
929 struct device_attribute *attr, char *buf)
930{
931 struct gendisk *disk = dev_to_disk(dev);
932
933 return sprintf(buf, "%d\n",
934 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
935}
936
937static ssize_t disk_ro_show(struct device *dev,
938 struct device_attribute *attr, char *buf)
939{
940 struct gendisk *disk = dev_to_disk(dev);
941
942 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
943}
944
945static ssize_t disk_capability_show(struct device *dev,
946 struct device_attribute *attr, char *buf)
947{
948 struct gendisk *disk = dev_to_disk(dev);
949
950 return sprintf(buf, "%x\n", disk->flags);
951}
952
953static ssize_t disk_alignment_offset_show(struct device *dev,
954 struct device_attribute *attr,
955 char *buf)
956{
957 struct gendisk *disk = dev_to_disk(dev);
958
959 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
960}
961
962static ssize_t disk_discard_alignment_show(struct device *dev,
963 struct device_attribute *attr,
964 char *buf)
965{
966 struct gendisk *disk = dev_to_disk(dev);
967
968 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
969}
970
971static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
972static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
973static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
974static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
975static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
976static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
977static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
978 NULL);
979static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
980static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
981static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
982#ifdef CONFIG_FAIL_MAKE_REQUEST
983static struct device_attribute dev_attr_fail =
984 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
985#endif
986#ifdef CONFIG_FAIL_IO_TIMEOUT
987static struct device_attribute dev_attr_fail_timeout =
988 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
989 part_timeout_store);
990#endif
991
992static struct attribute *disk_attrs[] = {
993 &dev_attr_range.attr,
994 &dev_attr_ext_range.attr,
995 &dev_attr_removable.attr,
996 &dev_attr_ro.attr,
997 &dev_attr_size.attr,
998 &dev_attr_alignment_offset.attr,
999 &dev_attr_discard_alignment.attr,
1000 &dev_attr_capability.attr,
1001 &dev_attr_stat.attr,
1002 &dev_attr_inflight.attr,
1003#ifdef CONFIG_FAIL_MAKE_REQUEST
1004 &dev_attr_fail.attr,
1005#endif
1006#ifdef CONFIG_FAIL_IO_TIMEOUT
1007 &dev_attr_fail_timeout.attr,
1008#endif
1009 NULL
1010};
1011
1012static struct attribute_group disk_attr_group = {
1013 .attrs = disk_attrs,
1014};
1015
1016static const struct attribute_group *disk_attr_groups[] = {
1017 &disk_attr_group,
1018 NULL
1019};
1020
1021/**
1022 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1023 * @disk: disk to replace part_tbl for
1024 * @new_ptbl: new part_tbl to install
1025 *
1026 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1027 * original ptbl is freed using RCU callback.
1028 *
1029 * LOCKING:
1030 * Matching bd_mutx locked.
1031 */
1032static void disk_replace_part_tbl(struct gendisk *disk,
1033 struct disk_part_tbl *new_ptbl)
1034{
1035 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1036
1037 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1038
1039 if (old_ptbl) {
1040 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1041 kfree_rcu(old_ptbl, rcu_head);
1042 }
1043}
1044
1045/**
1046 * disk_expand_part_tbl - expand disk->part_tbl
1047 * @disk: disk to expand part_tbl for
1048 * @partno: expand such that this partno can fit in
1049 *
1050 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1051 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1052 *
1053 * LOCKING:
1054 * Matching bd_mutex locked, might sleep.
1055 *
1056 * RETURNS:
1057 * 0 on success, -errno on failure.
1058 */
1059int disk_expand_part_tbl(struct gendisk *disk, int partno)
1060{
1061 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1062 struct disk_part_tbl *new_ptbl;
1063 int len = old_ptbl ? old_ptbl->len : 0;
1064 int target = partno + 1;
1065 size_t size;
1066 int i;
1067
1068 /* disk_max_parts() is zero during initialization, ignore if so */
1069 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1070 return -EINVAL;
1071
1072 if (target <= len)
1073 return 0;
1074
1075 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1076 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1077 if (!new_ptbl)
1078 return -ENOMEM;
1079
1080 new_ptbl->len = target;
1081
1082 for (i = 0; i < len; i++)
1083 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1084
1085 disk_replace_part_tbl(disk, new_ptbl);
1086 return 0;
1087}
1088
1089static void disk_release(struct device *dev)
1090{
1091 struct gendisk *disk = dev_to_disk(dev);
1092
1093 disk_release_events(disk);
1094 kfree(disk->random);
1095 disk_replace_part_tbl(disk, NULL);
1096 free_part_stats(&disk->part0);
1097 free_part_info(&disk->part0);
1098 kfree(disk);
1099}
1100struct class block_class = {
1101 .name = "block",
1102};
1103
1104static char *block_devnode(struct device *dev, mode_t *mode)
1105{
1106 struct gendisk *disk = dev_to_disk(dev);
1107
1108 if (disk->devnode)
1109 return disk->devnode(disk, mode);
1110 return NULL;
1111}
1112
1113static struct device_type disk_type = {
1114 .name = "disk",
1115 .groups = disk_attr_groups,
1116 .release = disk_release,
1117 .devnode = block_devnode,
1118};
1119
1120#ifdef CONFIG_PROC_FS
1121/*
1122 * aggregate disk stat collector. Uses the same stats that the sysfs
1123 * entries do, above, but makes them available through one seq_file.
1124 *
1125 * The output looks suspiciously like /proc/partitions with a bunch of
1126 * extra fields.
1127 */
1128static int diskstats_show(struct seq_file *seqf, void *v)
1129{
1130 struct gendisk *gp = v;
1131 struct disk_part_iter piter;
1132 struct hd_struct *hd;
1133 char buf[BDEVNAME_SIZE];
1134 int cpu;
1135
1136 /*
1137 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1138 seq_puts(seqf, "major minor name"
1139 " rio rmerge rsect ruse wio wmerge "
1140 "wsect wuse running use aveq"
1141 "\n\n");
1142 */
1143
1144 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1145 while ((hd = disk_part_iter_next(&piter))) {
1146 cpu = part_stat_lock();
1147 part_round_stats(cpu, hd);
1148 part_stat_unlock();
1149 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1150 "%u %lu %lu %lu %u %u %u %u\n",
1151 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1152 disk_name(gp, hd->partno, buf),
1153 part_stat_read(hd, ios[READ]),
1154 part_stat_read(hd, merges[READ]),
1155 part_stat_read(hd, sectors[READ]),
1156 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1157 part_stat_read(hd, ios[WRITE]),
1158 part_stat_read(hd, merges[WRITE]),
1159 part_stat_read(hd, sectors[WRITE]),
1160 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1161 part_in_flight(hd),
1162 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1163 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1164 );
1165 }
1166 disk_part_iter_exit(&piter);
1167
1168 return 0;
1169}
1170
1171static const struct seq_operations diskstats_op = {
1172 .start = disk_seqf_start,
1173 .next = disk_seqf_next,
1174 .stop = disk_seqf_stop,
1175 .show = diskstats_show
1176};
1177
1178static int diskstats_open(struct inode *inode, struct file *file)
1179{
1180 return seq_open(file, &diskstats_op);
1181}
1182
1183static const struct file_operations proc_diskstats_operations = {
1184 .open = diskstats_open,
1185 .read = seq_read,
1186 .llseek = seq_lseek,
1187 .release = seq_release,
1188};
1189
1190static int __init proc_genhd_init(void)
1191{
1192 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1193 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1194 return 0;
1195}
1196module_init(proc_genhd_init);
1197#endif /* CONFIG_PROC_FS */
1198
1199dev_t blk_lookup_devt(const char *name, int partno)
1200{
1201 dev_t devt = MKDEV(0, 0);
1202 struct class_dev_iter iter;
1203 struct device *dev;
1204
1205 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1206 while ((dev = class_dev_iter_next(&iter))) {
1207 struct gendisk *disk = dev_to_disk(dev);
1208 struct hd_struct *part;
1209
1210 if (strcmp(dev_name(dev), name))
1211 continue;
1212
1213 if (partno < disk->minors) {
1214 /* We need to return the right devno, even
1215 * if the partition doesn't exist yet.
1216 */
1217 devt = MKDEV(MAJOR(dev->devt),
1218 MINOR(dev->devt) + partno);
1219 break;
1220 }
1221 part = disk_get_part(disk, partno);
1222 if (part) {
1223 devt = part_devt(part);
1224 disk_put_part(part);
1225 break;
1226 }
1227 disk_put_part(part);
1228 }
1229 class_dev_iter_exit(&iter);
1230 return devt;
1231}
1232EXPORT_SYMBOL(blk_lookup_devt);
1233
1234struct gendisk *alloc_disk(int minors)
1235{
1236 return alloc_disk_node(minors, -1);
1237}
1238EXPORT_SYMBOL(alloc_disk);
1239
1240struct gendisk *alloc_disk_node(int minors, int node_id)
1241{
1242 struct gendisk *disk;
1243
1244 disk = kmalloc_node(sizeof(struct gendisk),
1245 GFP_KERNEL | __GFP_ZERO, node_id);
1246 if (disk) {
1247 if (!init_part_stats(&disk->part0)) {
1248 kfree(disk);
1249 return NULL;
1250 }
1251 disk->node_id = node_id;
1252 if (disk_expand_part_tbl(disk, 0)) {
1253 free_part_stats(&disk->part0);
1254 kfree(disk);
1255 return NULL;
1256 }
1257 disk->part_tbl->part[0] = &disk->part0;
1258
1259 hd_ref_init(&disk->part0);
1260
1261 disk->minors = minors;
1262 rand_initialize_disk(disk);
1263 disk_to_dev(disk)->class = &block_class;
1264 disk_to_dev(disk)->type = &disk_type;
1265 device_initialize(disk_to_dev(disk));
1266 }
1267 return disk;
1268}
1269EXPORT_SYMBOL(alloc_disk_node);
1270
1271struct kobject *get_disk(struct gendisk *disk)
1272{
1273 struct module *owner;
1274 struct kobject *kobj;
1275
1276 if (!disk->fops)
1277 return NULL;
1278 owner = disk->fops->owner;
1279 if (owner && !try_module_get(owner))
1280 return NULL;
1281 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1282 if (kobj == NULL) {
1283 module_put(owner);
1284 return NULL;
1285 }
1286 return kobj;
1287
1288}
1289
1290EXPORT_SYMBOL(get_disk);
1291
1292void put_disk(struct gendisk *disk)
1293{
1294 if (disk)
1295 kobject_put(&disk_to_dev(disk)->kobj);
1296}
1297
1298EXPORT_SYMBOL(put_disk);
1299
1300static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1301{
1302 char event[] = "DISK_RO=1";
1303 char *envp[] = { event, NULL };
1304
1305 if (!ro)
1306 event[8] = '0';
1307 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1308}
1309
1310void set_device_ro(struct block_device *bdev, int flag)
1311{
1312 bdev->bd_part->policy = flag;
1313}
1314
1315EXPORT_SYMBOL(set_device_ro);
1316
1317void set_disk_ro(struct gendisk *disk, int flag)
1318{
1319 struct disk_part_iter piter;
1320 struct hd_struct *part;
1321
1322 if (disk->part0.policy != flag) {
1323 set_disk_ro_uevent(disk, flag);
1324 disk->part0.policy = flag;
1325 }
1326
1327 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1328 while ((part = disk_part_iter_next(&piter)))
1329 part->policy = flag;
1330 disk_part_iter_exit(&piter);
1331}
1332
1333EXPORT_SYMBOL(set_disk_ro);
1334
1335int bdev_read_only(struct block_device *bdev)
1336{
1337 if (!bdev)
1338 return 0;
1339 return bdev->bd_part->policy;
1340}
1341
1342EXPORT_SYMBOL(bdev_read_only);
1343
1344int invalidate_partition(struct gendisk *disk, int partno)
1345{
1346 int res = 0;
1347 struct block_device *bdev = bdget_disk(disk, partno);
1348 if (bdev) {
1349 fsync_bdev(bdev);
1350 res = __invalidate_device(bdev, true);
1351 bdput(bdev);
1352 }
1353 return res;
1354}
1355
1356EXPORT_SYMBOL(invalidate_partition);
1357
1358/*
1359 * Disk events - monitor disk events like media change and eject request.
1360 */
1361struct disk_events {
1362 struct list_head node; /* all disk_event's */
1363 struct gendisk *disk; /* the associated disk */
1364 spinlock_t lock;
1365
1366 struct mutex block_mutex; /* protects blocking */
1367 int block; /* event blocking depth */
1368 unsigned int pending; /* events already sent out */
1369 unsigned int clearing; /* events being cleared */
1370
1371 long poll_msecs; /* interval, -1 for default */
1372 struct delayed_work dwork;
1373};
1374
1375static const char *disk_events_strs[] = {
1376 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1377 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1378};
1379
1380static char *disk_uevents[] = {
1381 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1382 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1383};
1384
1385/* list of all disk_events */
1386static DEFINE_MUTEX(disk_events_mutex);
1387static LIST_HEAD(disk_events);
1388
1389/* disable in-kernel polling by default */
1390static unsigned long disk_events_dfl_poll_msecs = 0;
1391
1392static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1393{
1394 struct disk_events *ev = disk->ev;
1395 long intv_msecs = 0;
1396
1397 /*
1398 * If device-specific poll interval is set, always use it. If
1399 * the default is being used, poll iff there are events which
1400 * can't be monitored asynchronously.
1401 */
1402 if (ev->poll_msecs >= 0)
1403 intv_msecs = ev->poll_msecs;
1404 else if (disk->events & ~disk->async_events)
1405 intv_msecs = disk_events_dfl_poll_msecs;
1406
1407 return msecs_to_jiffies(intv_msecs);
1408}
1409
1410/**
1411 * disk_block_events - block and flush disk event checking
1412 * @disk: disk to block events for
1413 *
1414 * On return from this function, it is guaranteed that event checking
1415 * isn't in progress and won't happen until unblocked by
1416 * disk_unblock_events(). Events blocking is counted and the actual
1417 * unblocking happens after the matching number of unblocks are done.
1418 *
1419 * Note that this intentionally does not block event checking from
1420 * disk_clear_events().
1421 *
1422 * CONTEXT:
1423 * Might sleep.
1424 */
1425void disk_block_events(struct gendisk *disk)
1426{
1427 struct disk_events *ev = disk->ev;
1428 unsigned long flags;
1429 bool cancel;
1430
1431 if (!ev)
1432 return;
1433
1434 /*
1435 * Outer mutex ensures that the first blocker completes canceling
1436 * the event work before further blockers are allowed to finish.
1437 */
1438 mutex_lock(&ev->block_mutex);
1439
1440 spin_lock_irqsave(&ev->lock, flags);
1441 cancel = !ev->block++;
1442 spin_unlock_irqrestore(&ev->lock, flags);
1443
1444 if (cancel)
1445 cancel_delayed_work_sync(&disk->ev->dwork);
1446
1447 mutex_unlock(&ev->block_mutex);
1448}
1449
1450static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1451{
1452 struct disk_events *ev = disk->ev;
1453 unsigned long intv;
1454 unsigned long flags;
1455
1456 spin_lock_irqsave(&ev->lock, flags);
1457
1458 if (WARN_ON_ONCE(ev->block <= 0))
1459 goto out_unlock;
1460
1461 if (--ev->block)
1462 goto out_unlock;
1463
1464 /*
1465 * Not exactly a latency critical operation, set poll timer
1466 * slack to 25% and kick event check.
1467 */
1468 intv = disk_events_poll_jiffies(disk);
1469 set_timer_slack(&ev->dwork.timer, intv / 4);
1470 if (check_now)
1471 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1472 else if (intv)
1473 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1474out_unlock:
1475 spin_unlock_irqrestore(&ev->lock, flags);
1476}
1477
1478/**
1479 * disk_unblock_events - unblock disk event checking
1480 * @disk: disk to unblock events for
1481 *
1482 * Undo disk_block_events(). When the block count reaches zero, it
1483 * starts events polling if configured.
1484 *
1485 * CONTEXT:
1486 * Don't care. Safe to call from irq context.
1487 */
1488void disk_unblock_events(struct gendisk *disk)
1489{
1490 if (disk->ev)
1491 __disk_unblock_events(disk, false);
1492}
1493
1494/**
1495 * disk_flush_events - schedule immediate event checking and flushing
1496 * @disk: disk to check and flush events for
1497 * @mask: events to flush
1498 *
1499 * Schedule immediate event checking on @disk if not blocked. Events in
1500 * @mask are scheduled to be cleared from the driver. Note that this
1501 * doesn't clear the events from @disk->ev.
1502 *
1503 * CONTEXT:
1504 * If @mask is non-zero must be called with bdev->bd_mutex held.
1505 */
1506void disk_flush_events(struct gendisk *disk, unsigned int mask)
1507{
1508 struct disk_events *ev = disk->ev;
1509
1510 if (!ev)
1511 return;
1512
1513 spin_lock_irq(&ev->lock);
1514 ev->clearing |= mask;
1515 if (!ev->block) {
1516 cancel_delayed_work(&ev->dwork);
1517 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1518 }
1519 spin_unlock_irq(&ev->lock);
1520}
1521
1522/**
1523 * disk_clear_events - synchronously check, clear and return pending events
1524 * @disk: disk to fetch and clear events from
1525 * @mask: mask of events to be fetched and clearted
1526 *
1527 * Disk events are synchronously checked and pending events in @mask
1528 * are cleared and returned. This ignores the block count.
1529 *
1530 * CONTEXT:
1531 * Might sleep.
1532 */
1533unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1534{
1535 const struct block_device_operations *bdops = disk->fops;
1536 struct disk_events *ev = disk->ev;
1537 unsigned int pending;
1538
1539 if (!ev) {
1540 /* for drivers still using the old ->media_changed method */
1541 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1542 bdops->media_changed && bdops->media_changed(disk))
1543 return DISK_EVENT_MEDIA_CHANGE;
1544 return 0;
1545 }
1546
1547 /* tell the workfn about the events being cleared */
1548 spin_lock_irq(&ev->lock);
1549 ev->clearing |= mask;
1550 spin_unlock_irq(&ev->lock);
1551
1552 /* uncondtionally schedule event check and wait for it to finish */
1553 disk_block_events(disk);
1554 queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
1555 flush_delayed_work(&ev->dwork);
1556 __disk_unblock_events(disk, false);
1557
1558 /* then, fetch and clear pending events */
1559 spin_lock_irq(&ev->lock);
1560 WARN_ON_ONCE(ev->clearing & mask); /* cleared by workfn */
1561 pending = ev->pending & mask;
1562 ev->pending &= ~mask;
1563 spin_unlock_irq(&ev->lock);
1564
1565 return pending;
1566}
1567
1568static void disk_events_workfn(struct work_struct *work)
1569{
1570 struct delayed_work *dwork = to_delayed_work(work);
1571 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1572 struct gendisk *disk = ev->disk;
1573 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1574 unsigned int clearing = ev->clearing;
1575 unsigned int events;
1576 unsigned long intv;
1577 int nr_events = 0, i;
1578
1579 /* check events */
1580 events = disk->fops->check_events(disk, clearing);
1581
1582 /* accumulate pending events and schedule next poll if necessary */
1583 spin_lock_irq(&ev->lock);
1584
1585 events &= ~ev->pending;
1586 ev->pending |= events;
1587 ev->clearing &= ~clearing;
1588
1589 intv = disk_events_poll_jiffies(disk);
1590 if (!ev->block && intv)
1591 queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
1592
1593 spin_unlock_irq(&ev->lock);
1594
1595 /*
1596 * Tell userland about new events. Only the events listed in
1597 * @disk->events are reported. Unlisted events are processed the
1598 * same internally but never get reported to userland.
1599 */
1600 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1601 if (events & disk->events & (1 << i))
1602 envp[nr_events++] = disk_uevents[i];
1603
1604 if (nr_events)
1605 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1606}
1607
1608/*
1609 * A disk events enabled device has the following sysfs nodes under
1610 * its /sys/block/X/ directory.
1611 *
1612 * events : list of all supported events
1613 * events_async : list of events which can be detected w/o polling
1614 * events_poll_msecs : polling interval, 0: disable, -1: system default
1615 */
1616static ssize_t __disk_events_show(unsigned int events, char *buf)
1617{
1618 const char *delim = "";
1619 ssize_t pos = 0;
1620 int i;
1621
1622 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1623 if (events & (1 << i)) {
1624 pos += sprintf(buf + pos, "%s%s",
1625 delim, disk_events_strs[i]);
1626 delim = " ";
1627 }
1628 if (pos)
1629 pos += sprintf(buf + pos, "\n");
1630 return pos;
1631}
1632
1633static ssize_t disk_events_show(struct device *dev,
1634 struct device_attribute *attr, char *buf)
1635{
1636 struct gendisk *disk = dev_to_disk(dev);
1637
1638 return __disk_events_show(disk->events, buf);
1639}
1640
1641static ssize_t disk_events_async_show(struct device *dev,
1642 struct device_attribute *attr, char *buf)
1643{
1644 struct gendisk *disk = dev_to_disk(dev);
1645
1646 return __disk_events_show(disk->async_events, buf);
1647}
1648
1649static ssize_t disk_events_poll_msecs_show(struct device *dev,
1650 struct device_attribute *attr,
1651 char *buf)
1652{
1653 struct gendisk *disk = dev_to_disk(dev);
1654
1655 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1656}
1657
1658static ssize_t disk_events_poll_msecs_store(struct device *dev,
1659 struct device_attribute *attr,
1660 const char *buf, size_t count)
1661{
1662 struct gendisk *disk = dev_to_disk(dev);
1663 long intv;
1664
1665 if (!count || !sscanf(buf, "%ld", &intv))
1666 return -EINVAL;
1667
1668 if (intv < 0 && intv != -1)
1669 return -EINVAL;
1670
1671 disk_block_events(disk);
1672 disk->ev->poll_msecs = intv;
1673 __disk_unblock_events(disk, true);
1674
1675 return count;
1676}
1677
1678static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1679static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1680static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1681 disk_events_poll_msecs_show,
1682 disk_events_poll_msecs_store);
1683
1684static const struct attribute *disk_events_attrs[] = {
1685 &dev_attr_events.attr,
1686 &dev_attr_events_async.attr,
1687 &dev_attr_events_poll_msecs.attr,
1688 NULL,
1689};
1690
1691/*
1692 * The default polling interval can be specified by the kernel
1693 * parameter block.events_dfl_poll_msecs which defaults to 0
1694 * (disable). This can also be modified runtime by writing to
1695 * /sys/module/block/events_dfl_poll_msecs.
1696 */
1697static int disk_events_set_dfl_poll_msecs(const char *val,
1698 const struct kernel_param *kp)
1699{
1700 struct disk_events *ev;
1701 int ret;
1702
1703 ret = param_set_ulong(val, kp);
1704 if (ret < 0)
1705 return ret;
1706
1707 mutex_lock(&disk_events_mutex);
1708
1709 list_for_each_entry(ev, &disk_events, node)
1710 disk_flush_events(ev->disk, 0);
1711
1712 mutex_unlock(&disk_events_mutex);
1713
1714 return 0;
1715}
1716
1717static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1718 .set = disk_events_set_dfl_poll_msecs,
1719 .get = param_get_ulong,
1720};
1721
1722#undef MODULE_PARAM_PREFIX
1723#define MODULE_PARAM_PREFIX "block."
1724
1725module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1726 &disk_events_dfl_poll_msecs, 0644);
1727
1728/*
1729 * disk_{add|del|release}_events - initialize and destroy disk_events.
1730 */
1731static void disk_add_events(struct gendisk *disk)
1732{
1733 struct disk_events *ev;
1734
1735 if (!disk->fops->check_events)
1736 return;
1737
1738 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1739 if (!ev) {
1740 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1741 return;
1742 }
1743
1744 if (sysfs_create_files(&disk_to_dev(disk)->kobj,
1745 disk_events_attrs) < 0) {
1746 pr_warn("%s: failed to create sysfs files for events\n",
1747 disk->disk_name);
1748 kfree(ev);
1749 return;
1750 }
1751
1752 disk->ev = ev;
1753
1754 INIT_LIST_HEAD(&ev->node);
1755 ev->disk = disk;
1756 spin_lock_init(&ev->lock);
1757 mutex_init(&ev->block_mutex);
1758 ev->block = 1;
1759 ev->poll_msecs = -1;
1760 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1761
1762 mutex_lock(&disk_events_mutex);
1763 list_add_tail(&ev->node, &disk_events);
1764 mutex_unlock(&disk_events_mutex);
1765
1766 /*
1767 * Block count is initialized to 1 and the following initial
1768 * unblock kicks it into action.
1769 */
1770 __disk_unblock_events(disk, true);
1771}
1772
1773static void disk_del_events(struct gendisk *disk)
1774{
1775 if (!disk->ev)
1776 return;
1777
1778 disk_block_events(disk);
1779
1780 mutex_lock(&disk_events_mutex);
1781 list_del_init(&disk->ev->node);
1782 mutex_unlock(&disk_events_mutex);
1783
1784 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1785}
1786
1787static void disk_release_events(struct gendisk *disk)
1788{
1789 /* the block count should be 1 from disk_del_events() */
1790 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1791 kfree(disk->ev);
1792}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * gendisk handling
4 */
5
6#include <linux/module.h>
7#include <linux/fs.h>
8#include <linux/genhd.h>
9#include <linux/kdev_t.h>
10#include <linux/kernel.h>
11#include <linux/blkdev.h>
12#include <linux/backing-dev.h>
13#include <linux/init.h>
14#include <linux/spinlock.h>
15#include <linux/proc_fs.h>
16#include <linux/seq_file.h>
17#include <linux/slab.h>
18#include <linux/kmod.h>
19#include <linux/kobj_map.h>
20#include <linux/mutex.h>
21#include <linux/idr.h>
22#include <linux/log2.h>
23#include <linux/pm_runtime.h>
24#include <linux/badblocks.h>
25
26#include "blk.h"
27
28static DEFINE_MUTEX(block_class_lock);
29struct kobject *block_depr;
30
31/* for extended dynamic devt allocation, currently only one major is used */
32#define NR_EXT_DEVT (1 << MINORBITS)
33
34/* For extended devt allocation. ext_devt_lock prevents look up
35 * results from going away underneath its user.
36 */
37static DEFINE_SPINLOCK(ext_devt_lock);
38static DEFINE_IDR(ext_devt_idr);
39
40static const struct device_type disk_type;
41
42static void disk_check_events(struct disk_events *ev,
43 unsigned int *clearing_ptr);
44static void disk_alloc_events(struct gendisk *disk);
45static void disk_add_events(struct gendisk *disk);
46static void disk_del_events(struct gendisk *disk);
47static void disk_release_events(struct gendisk *disk);
48
49void part_inc_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
50{
51 if (queue_is_mq(q))
52 return;
53
54 part_stat_local_inc(part, in_flight[rw]);
55 if (part->partno)
56 part_stat_local_inc(&part_to_disk(part)->part0, in_flight[rw]);
57}
58
59void part_dec_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
60{
61 if (queue_is_mq(q))
62 return;
63
64 part_stat_local_dec(part, in_flight[rw]);
65 if (part->partno)
66 part_stat_local_dec(&part_to_disk(part)->part0, in_flight[rw]);
67}
68
69unsigned int part_in_flight(struct request_queue *q, struct hd_struct *part)
70{
71 int cpu;
72 unsigned int inflight;
73
74 if (queue_is_mq(q)) {
75 return blk_mq_in_flight(q, part);
76 }
77
78 inflight = 0;
79 for_each_possible_cpu(cpu) {
80 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
81 part_stat_local_read_cpu(part, in_flight[1], cpu);
82 }
83 if ((int)inflight < 0)
84 inflight = 0;
85
86 return inflight;
87}
88
89void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
90 unsigned int inflight[2])
91{
92 int cpu;
93
94 if (queue_is_mq(q)) {
95 blk_mq_in_flight_rw(q, part, inflight);
96 return;
97 }
98
99 inflight[0] = 0;
100 inflight[1] = 0;
101 for_each_possible_cpu(cpu) {
102 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
103 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
104 }
105 if ((int)inflight[0] < 0)
106 inflight[0] = 0;
107 if ((int)inflight[1] < 0)
108 inflight[1] = 0;
109}
110
111struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
112{
113 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
114
115 if (unlikely(partno < 0 || partno >= ptbl->len))
116 return NULL;
117 return rcu_dereference(ptbl->part[partno]);
118}
119
120/**
121 * disk_get_part - get partition
122 * @disk: disk to look partition from
123 * @partno: partition number
124 *
125 * Look for partition @partno from @disk. If found, increment
126 * reference count and return it.
127 *
128 * CONTEXT:
129 * Don't care.
130 *
131 * RETURNS:
132 * Pointer to the found partition on success, NULL if not found.
133 */
134struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
135{
136 struct hd_struct *part;
137
138 rcu_read_lock();
139 part = __disk_get_part(disk, partno);
140 if (part)
141 get_device(part_to_dev(part));
142 rcu_read_unlock();
143
144 return part;
145}
146EXPORT_SYMBOL_GPL(disk_get_part);
147
148/**
149 * disk_part_iter_init - initialize partition iterator
150 * @piter: iterator to initialize
151 * @disk: disk to iterate over
152 * @flags: DISK_PITER_* flags
153 *
154 * Initialize @piter so that it iterates over partitions of @disk.
155 *
156 * CONTEXT:
157 * Don't care.
158 */
159void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
160 unsigned int flags)
161{
162 struct disk_part_tbl *ptbl;
163
164 rcu_read_lock();
165 ptbl = rcu_dereference(disk->part_tbl);
166
167 piter->disk = disk;
168 piter->part = NULL;
169
170 if (flags & DISK_PITER_REVERSE)
171 piter->idx = ptbl->len - 1;
172 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
173 piter->idx = 0;
174 else
175 piter->idx = 1;
176
177 piter->flags = flags;
178
179 rcu_read_unlock();
180}
181EXPORT_SYMBOL_GPL(disk_part_iter_init);
182
183/**
184 * disk_part_iter_next - proceed iterator to the next partition and return it
185 * @piter: iterator of interest
186 *
187 * Proceed @piter to the next partition and return it.
188 *
189 * CONTEXT:
190 * Don't care.
191 */
192struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
193{
194 struct disk_part_tbl *ptbl;
195 int inc, end;
196
197 /* put the last partition */
198 disk_put_part(piter->part);
199 piter->part = NULL;
200
201 /* get part_tbl */
202 rcu_read_lock();
203 ptbl = rcu_dereference(piter->disk->part_tbl);
204
205 /* determine iteration parameters */
206 if (piter->flags & DISK_PITER_REVERSE) {
207 inc = -1;
208 if (piter->flags & (DISK_PITER_INCL_PART0 |
209 DISK_PITER_INCL_EMPTY_PART0))
210 end = -1;
211 else
212 end = 0;
213 } else {
214 inc = 1;
215 end = ptbl->len;
216 }
217
218 /* iterate to the next partition */
219 for (; piter->idx != end; piter->idx += inc) {
220 struct hd_struct *part;
221
222 part = rcu_dereference(ptbl->part[piter->idx]);
223 if (!part)
224 continue;
225 if (!part_nr_sects_read(part) &&
226 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
227 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
228 piter->idx == 0))
229 continue;
230
231 get_device(part_to_dev(part));
232 piter->part = part;
233 piter->idx += inc;
234 break;
235 }
236
237 rcu_read_unlock();
238
239 return piter->part;
240}
241EXPORT_SYMBOL_GPL(disk_part_iter_next);
242
243/**
244 * disk_part_iter_exit - finish up partition iteration
245 * @piter: iter of interest
246 *
247 * Called when iteration is over. Cleans up @piter.
248 *
249 * CONTEXT:
250 * Don't care.
251 */
252void disk_part_iter_exit(struct disk_part_iter *piter)
253{
254 disk_put_part(piter->part);
255 piter->part = NULL;
256}
257EXPORT_SYMBOL_GPL(disk_part_iter_exit);
258
259static inline int sector_in_part(struct hd_struct *part, sector_t sector)
260{
261 return part->start_sect <= sector &&
262 sector < part->start_sect + part_nr_sects_read(part);
263}
264
265/**
266 * disk_map_sector_rcu - map sector to partition
267 * @disk: gendisk of interest
268 * @sector: sector to map
269 *
270 * Find out which partition @sector maps to on @disk. This is
271 * primarily used for stats accounting.
272 *
273 * CONTEXT:
274 * RCU read locked. The returned partition pointer is valid only
275 * while preemption is disabled.
276 *
277 * RETURNS:
278 * Found partition on success, part0 is returned if no partition matches
279 */
280struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
281{
282 struct disk_part_tbl *ptbl;
283 struct hd_struct *part;
284 int i;
285
286 ptbl = rcu_dereference(disk->part_tbl);
287
288 part = rcu_dereference(ptbl->last_lookup);
289 if (part && sector_in_part(part, sector))
290 return part;
291
292 for (i = 1; i < ptbl->len; i++) {
293 part = rcu_dereference(ptbl->part[i]);
294
295 if (part && sector_in_part(part, sector)) {
296 rcu_assign_pointer(ptbl->last_lookup, part);
297 return part;
298 }
299 }
300 return &disk->part0;
301}
302EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
303
304/*
305 * Can be deleted altogether. Later.
306 *
307 */
308#define BLKDEV_MAJOR_HASH_SIZE 255
309static struct blk_major_name {
310 struct blk_major_name *next;
311 int major;
312 char name[16];
313} *major_names[BLKDEV_MAJOR_HASH_SIZE];
314
315/* index in the above - for now: assume no multimajor ranges */
316static inline int major_to_index(unsigned major)
317{
318 return major % BLKDEV_MAJOR_HASH_SIZE;
319}
320
321#ifdef CONFIG_PROC_FS
322void blkdev_show(struct seq_file *seqf, off_t offset)
323{
324 struct blk_major_name *dp;
325
326 mutex_lock(&block_class_lock);
327 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
328 if (dp->major == offset)
329 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
330 mutex_unlock(&block_class_lock);
331}
332#endif /* CONFIG_PROC_FS */
333
334/**
335 * register_blkdev - register a new block device
336 *
337 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
338 * @major = 0, try to allocate any unused major number.
339 * @name: the name of the new block device as a zero terminated string
340 *
341 * The @name must be unique within the system.
342 *
343 * The return value depends on the @major input parameter:
344 *
345 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
346 * then the function returns zero on success, or a negative error code
347 * - if any unused major number was requested with @major = 0 parameter
348 * then the return value is the allocated major number in range
349 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
350 *
351 * See Documentation/admin-guide/devices.txt for the list of allocated
352 * major numbers.
353 */
354int register_blkdev(unsigned int major, const char *name)
355{
356 struct blk_major_name **n, *p;
357 int index, ret = 0;
358
359 mutex_lock(&block_class_lock);
360
361 /* temporary */
362 if (major == 0) {
363 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
364 if (major_names[index] == NULL)
365 break;
366 }
367
368 if (index == 0) {
369 printk("%s: failed to get major for %s\n",
370 __func__, name);
371 ret = -EBUSY;
372 goto out;
373 }
374 major = index;
375 ret = major;
376 }
377
378 if (major >= BLKDEV_MAJOR_MAX) {
379 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
380 __func__, major, BLKDEV_MAJOR_MAX-1, name);
381
382 ret = -EINVAL;
383 goto out;
384 }
385
386 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
387 if (p == NULL) {
388 ret = -ENOMEM;
389 goto out;
390 }
391
392 p->major = major;
393 strlcpy(p->name, name, sizeof(p->name));
394 p->next = NULL;
395 index = major_to_index(major);
396
397 for (n = &major_names[index]; *n; n = &(*n)->next) {
398 if ((*n)->major == major)
399 break;
400 }
401 if (!*n)
402 *n = p;
403 else
404 ret = -EBUSY;
405
406 if (ret < 0) {
407 printk("register_blkdev: cannot get major %u for %s\n",
408 major, name);
409 kfree(p);
410 }
411out:
412 mutex_unlock(&block_class_lock);
413 return ret;
414}
415
416EXPORT_SYMBOL(register_blkdev);
417
418void unregister_blkdev(unsigned int major, const char *name)
419{
420 struct blk_major_name **n;
421 struct blk_major_name *p = NULL;
422 int index = major_to_index(major);
423
424 mutex_lock(&block_class_lock);
425 for (n = &major_names[index]; *n; n = &(*n)->next)
426 if ((*n)->major == major)
427 break;
428 if (!*n || strcmp((*n)->name, name)) {
429 WARN_ON(1);
430 } else {
431 p = *n;
432 *n = p->next;
433 }
434 mutex_unlock(&block_class_lock);
435 kfree(p);
436}
437
438EXPORT_SYMBOL(unregister_blkdev);
439
440static struct kobj_map *bdev_map;
441
442/**
443 * blk_mangle_minor - scatter minor numbers apart
444 * @minor: minor number to mangle
445 *
446 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
447 * is enabled. Mangling twice gives the original value.
448 *
449 * RETURNS:
450 * Mangled value.
451 *
452 * CONTEXT:
453 * Don't care.
454 */
455static int blk_mangle_minor(int minor)
456{
457#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
458 int i;
459
460 for (i = 0; i < MINORBITS / 2; i++) {
461 int low = minor & (1 << i);
462 int high = minor & (1 << (MINORBITS - 1 - i));
463 int distance = MINORBITS - 1 - 2 * i;
464
465 minor ^= low | high; /* clear both bits */
466 low <<= distance; /* swap the positions */
467 high >>= distance;
468 minor |= low | high; /* and set */
469 }
470#endif
471 return minor;
472}
473
474/**
475 * blk_alloc_devt - allocate a dev_t for a partition
476 * @part: partition to allocate dev_t for
477 * @devt: out parameter for resulting dev_t
478 *
479 * Allocate a dev_t for block device.
480 *
481 * RETURNS:
482 * 0 on success, allocated dev_t is returned in *@devt. -errno on
483 * failure.
484 *
485 * CONTEXT:
486 * Might sleep.
487 */
488int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
489{
490 struct gendisk *disk = part_to_disk(part);
491 int idx;
492
493 /* in consecutive minor range? */
494 if (part->partno < disk->minors) {
495 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
496 return 0;
497 }
498
499 /* allocate ext devt */
500 idr_preload(GFP_KERNEL);
501
502 spin_lock_bh(&ext_devt_lock);
503 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
504 spin_unlock_bh(&ext_devt_lock);
505
506 idr_preload_end();
507 if (idx < 0)
508 return idx == -ENOSPC ? -EBUSY : idx;
509
510 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
511 return 0;
512}
513
514/**
515 * blk_free_devt - free a dev_t
516 * @devt: dev_t to free
517 *
518 * Free @devt which was allocated using blk_alloc_devt().
519 *
520 * CONTEXT:
521 * Might sleep.
522 */
523void blk_free_devt(dev_t devt)
524{
525 if (devt == MKDEV(0, 0))
526 return;
527
528 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
529 spin_lock_bh(&ext_devt_lock);
530 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
531 spin_unlock_bh(&ext_devt_lock);
532 }
533}
534
535/*
536 * We invalidate devt by assigning NULL pointer for devt in idr.
537 */
538void blk_invalidate_devt(dev_t devt)
539{
540 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
541 spin_lock_bh(&ext_devt_lock);
542 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
543 spin_unlock_bh(&ext_devt_lock);
544 }
545}
546
547static char *bdevt_str(dev_t devt, char *buf)
548{
549 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
550 char tbuf[BDEVT_SIZE];
551 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
552 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
553 } else
554 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
555
556 return buf;
557}
558
559/*
560 * Register device numbers dev..(dev+range-1)
561 * range must be nonzero
562 * The hash chain is sorted on range, so that subranges can override.
563 */
564void blk_register_region(dev_t devt, unsigned long range, struct module *module,
565 struct kobject *(*probe)(dev_t, int *, void *),
566 int (*lock)(dev_t, void *), void *data)
567{
568 kobj_map(bdev_map, devt, range, module, probe, lock, data);
569}
570
571EXPORT_SYMBOL(blk_register_region);
572
573void blk_unregister_region(dev_t devt, unsigned long range)
574{
575 kobj_unmap(bdev_map, devt, range);
576}
577
578EXPORT_SYMBOL(blk_unregister_region);
579
580static struct kobject *exact_match(dev_t devt, int *partno, void *data)
581{
582 struct gendisk *p = data;
583
584 return &disk_to_dev(p)->kobj;
585}
586
587static int exact_lock(dev_t devt, void *data)
588{
589 struct gendisk *p = data;
590
591 if (!get_disk_and_module(p))
592 return -1;
593 return 0;
594}
595
596static void register_disk(struct device *parent, struct gendisk *disk,
597 const struct attribute_group **groups)
598{
599 struct device *ddev = disk_to_dev(disk);
600 struct block_device *bdev;
601 struct disk_part_iter piter;
602 struct hd_struct *part;
603 int err;
604
605 ddev->parent = parent;
606
607 dev_set_name(ddev, "%s", disk->disk_name);
608
609 /* delay uevents, until we scanned partition table */
610 dev_set_uevent_suppress(ddev, 1);
611
612 if (groups) {
613 WARN_ON(ddev->groups);
614 ddev->groups = groups;
615 }
616 if (device_add(ddev))
617 return;
618 if (!sysfs_deprecated) {
619 err = sysfs_create_link(block_depr, &ddev->kobj,
620 kobject_name(&ddev->kobj));
621 if (err) {
622 device_del(ddev);
623 return;
624 }
625 }
626
627 /*
628 * avoid probable deadlock caused by allocating memory with
629 * GFP_KERNEL in runtime_resume callback of its all ancestor
630 * devices
631 */
632 pm_runtime_set_memalloc_noio(ddev, true);
633
634 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
635 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
636
637 if (disk->flags & GENHD_FL_HIDDEN) {
638 dev_set_uevent_suppress(ddev, 0);
639 return;
640 }
641
642 /* No minors to use for partitions */
643 if (!disk_part_scan_enabled(disk))
644 goto exit;
645
646 /* No such device (e.g., media were just removed) */
647 if (!get_capacity(disk))
648 goto exit;
649
650 bdev = bdget_disk(disk, 0);
651 if (!bdev)
652 goto exit;
653
654 bdev->bd_invalidated = 1;
655 err = blkdev_get(bdev, FMODE_READ, NULL);
656 if (err < 0)
657 goto exit;
658 blkdev_put(bdev, FMODE_READ);
659
660exit:
661 /* announce disk after possible partitions are created */
662 dev_set_uevent_suppress(ddev, 0);
663 kobject_uevent(&ddev->kobj, KOBJ_ADD);
664
665 /* announce possible partitions */
666 disk_part_iter_init(&piter, disk, 0);
667 while ((part = disk_part_iter_next(&piter)))
668 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
669 disk_part_iter_exit(&piter);
670
671 if (disk->queue->backing_dev_info->dev) {
672 err = sysfs_create_link(&ddev->kobj,
673 &disk->queue->backing_dev_info->dev->kobj,
674 "bdi");
675 WARN_ON(err);
676 }
677}
678
679/**
680 * __device_add_disk - add disk information to kernel list
681 * @parent: parent device for the disk
682 * @disk: per-device partitioning information
683 * @groups: Additional per-device sysfs groups
684 * @register_queue: register the queue if set to true
685 *
686 * This function registers the partitioning information in @disk
687 * with the kernel.
688 *
689 * FIXME: error handling
690 */
691static void __device_add_disk(struct device *parent, struct gendisk *disk,
692 const struct attribute_group **groups,
693 bool register_queue)
694{
695 dev_t devt;
696 int retval;
697
698 /*
699 * The disk queue should now be all set with enough information about
700 * the device for the elevator code to pick an adequate default
701 * elevator if one is needed, that is, for devices requesting queue
702 * registration.
703 */
704 if (register_queue)
705 elevator_init_mq(disk->queue);
706
707 /* minors == 0 indicates to use ext devt from part0 and should
708 * be accompanied with EXT_DEVT flag. Make sure all
709 * parameters make sense.
710 */
711 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
712 WARN_ON(!disk->minors &&
713 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
714
715 disk->flags |= GENHD_FL_UP;
716
717 retval = blk_alloc_devt(&disk->part0, &devt);
718 if (retval) {
719 WARN_ON(1);
720 return;
721 }
722 disk->major = MAJOR(devt);
723 disk->first_minor = MINOR(devt);
724
725 disk_alloc_events(disk);
726
727 if (disk->flags & GENHD_FL_HIDDEN) {
728 /*
729 * Don't let hidden disks show up in /proc/partitions,
730 * and don't bother scanning for partitions either.
731 */
732 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
733 disk->flags |= GENHD_FL_NO_PART_SCAN;
734 } else {
735 int ret;
736
737 /* Register BDI before referencing it from bdev */
738 disk_to_dev(disk)->devt = devt;
739 ret = bdi_register_owner(disk->queue->backing_dev_info,
740 disk_to_dev(disk));
741 WARN_ON(ret);
742 blk_register_region(disk_devt(disk), disk->minors, NULL,
743 exact_match, exact_lock, disk);
744 }
745 register_disk(parent, disk, groups);
746 if (register_queue)
747 blk_register_queue(disk);
748
749 /*
750 * Take an extra ref on queue which will be put on disk_release()
751 * so that it sticks around as long as @disk is there.
752 */
753 WARN_ON_ONCE(!blk_get_queue(disk->queue));
754
755 disk_add_events(disk);
756 blk_integrity_add(disk);
757}
758
759void device_add_disk(struct device *parent, struct gendisk *disk,
760 const struct attribute_group **groups)
761
762{
763 __device_add_disk(parent, disk, groups, true);
764}
765EXPORT_SYMBOL(device_add_disk);
766
767void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
768{
769 __device_add_disk(parent, disk, NULL, false);
770}
771EXPORT_SYMBOL(device_add_disk_no_queue_reg);
772
773void del_gendisk(struct gendisk *disk)
774{
775 struct disk_part_iter piter;
776 struct hd_struct *part;
777
778 blk_integrity_del(disk);
779 disk_del_events(disk);
780
781 /*
782 * Block lookups of the disk until all bdevs are unhashed and the
783 * disk is marked as dead (GENHD_FL_UP cleared).
784 */
785 down_write(&disk->lookup_sem);
786 /* invalidate stuff */
787 disk_part_iter_init(&piter, disk,
788 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
789 while ((part = disk_part_iter_next(&piter))) {
790 invalidate_partition(disk, part->partno);
791 bdev_unhash_inode(part_devt(part));
792 delete_partition(disk, part->partno);
793 }
794 disk_part_iter_exit(&piter);
795
796 invalidate_partition(disk, 0);
797 bdev_unhash_inode(disk_devt(disk));
798 set_capacity(disk, 0);
799 disk->flags &= ~GENHD_FL_UP;
800 up_write(&disk->lookup_sem);
801
802 if (!(disk->flags & GENHD_FL_HIDDEN))
803 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
804 if (disk->queue) {
805 /*
806 * Unregister bdi before releasing device numbers (as they can
807 * get reused and we'd get clashes in sysfs).
808 */
809 if (!(disk->flags & GENHD_FL_HIDDEN))
810 bdi_unregister(disk->queue->backing_dev_info);
811 blk_unregister_queue(disk);
812 } else {
813 WARN_ON(1);
814 }
815
816 if (!(disk->flags & GENHD_FL_HIDDEN))
817 blk_unregister_region(disk_devt(disk), disk->minors);
818 /*
819 * Remove gendisk pointer from idr so that it cannot be looked up
820 * while RCU period before freeing gendisk is running to prevent
821 * use-after-free issues. Note that the device number stays
822 * "in-use" until we really free the gendisk.
823 */
824 blk_invalidate_devt(disk_devt(disk));
825
826 kobject_put(disk->part0.holder_dir);
827 kobject_put(disk->slave_dir);
828
829 part_stat_set_all(&disk->part0, 0);
830 disk->part0.stamp = 0;
831 if (!sysfs_deprecated)
832 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
833 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
834 device_del(disk_to_dev(disk));
835}
836EXPORT_SYMBOL(del_gendisk);
837
838/* sysfs access to bad-blocks list. */
839static ssize_t disk_badblocks_show(struct device *dev,
840 struct device_attribute *attr,
841 char *page)
842{
843 struct gendisk *disk = dev_to_disk(dev);
844
845 if (!disk->bb)
846 return sprintf(page, "\n");
847
848 return badblocks_show(disk->bb, page, 0);
849}
850
851static ssize_t disk_badblocks_store(struct device *dev,
852 struct device_attribute *attr,
853 const char *page, size_t len)
854{
855 struct gendisk *disk = dev_to_disk(dev);
856
857 if (!disk->bb)
858 return -ENXIO;
859
860 return badblocks_store(disk->bb, page, len, 0);
861}
862
863/**
864 * get_gendisk - get partitioning information for a given device
865 * @devt: device to get partitioning information for
866 * @partno: returned partition index
867 *
868 * This function gets the structure containing partitioning
869 * information for the given device @devt.
870 */
871struct gendisk *get_gendisk(dev_t devt, int *partno)
872{
873 struct gendisk *disk = NULL;
874
875 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
876 struct kobject *kobj;
877
878 kobj = kobj_lookup(bdev_map, devt, partno);
879 if (kobj)
880 disk = dev_to_disk(kobj_to_dev(kobj));
881 } else {
882 struct hd_struct *part;
883
884 spin_lock_bh(&ext_devt_lock);
885 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
886 if (part && get_disk_and_module(part_to_disk(part))) {
887 *partno = part->partno;
888 disk = part_to_disk(part);
889 }
890 spin_unlock_bh(&ext_devt_lock);
891 }
892
893 if (!disk)
894 return NULL;
895
896 /*
897 * Synchronize with del_gendisk() to not return disk that is being
898 * destroyed.
899 */
900 down_read(&disk->lookup_sem);
901 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
902 !(disk->flags & GENHD_FL_UP))) {
903 up_read(&disk->lookup_sem);
904 put_disk_and_module(disk);
905 disk = NULL;
906 } else {
907 up_read(&disk->lookup_sem);
908 }
909 return disk;
910}
911EXPORT_SYMBOL(get_gendisk);
912
913/**
914 * bdget_disk - do bdget() by gendisk and partition number
915 * @disk: gendisk of interest
916 * @partno: partition number
917 *
918 * Find partition @partno from @disk, do bdget() on it.
919 *
920 * CONTEXT:
921 * Don't care.
922 *
923 * RETURNS:
924 * Resulting block_device on success, NULL on failure.
925 */
926struct block_device *bdget_disk(struct gendisk *disk, int partno)
927{
928 struct hd_struct *part;
929 struct block_device *bdev = NULL;
930
931 part = disk_get_part(disk, partno);
932 if (part)
933 bdev = bdget(part_devt(part));
934 disk_put_part(part);
935
936 return bdev;
937}
938EXPORT_SYMBOL(bdget_disk);
939
940/*
941 * print a full list of all partitions - intended for places where the root
942 * filesystem can't be mounted and thus to give the victim some idea of what
943 * went wrong
944 */
945void __init printk_all_partitions(void)
946{
947 struct class_dev_iter iter;
948 struct device *dev;
949
950 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
951 while ((dev = class_dev_iter_next(&iter))) {
952 struct gendisk *disk = dev_to_disk(dev);
953 struct disk_part_iter piter;
954 struct hd_struct *part;
955 char name_buf[BDEVNAME_SIZE];
956 char devt_buf[BDEVT_SIZE];
957
958 /*
959 * Don't show empty devices or things that have been
960 * suppressed
961 */
962 if (get_capacity(disk) == 0 ||
963 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
964 continue;
965
966 /*
967 * Note, unlike /proc/partitions, I am showing the
968 * numbers in hex - the same format as the root=
969 * option takes.
970 */
971 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
972 while ((part = disk_part_iter_next(&piter))) {
973 bool is_part0 = part == &disk->part0;
974
975 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
976 bdevt_str(part_devt(part), devt_buf),
977 (unsigned long long)part_nr_sects_read(part) >> 1
978 , disk_name(disk, part->partno, name_buf),
979 part->info ? part->info->uuid : "");
980 if (is_part0) {
981 if (dev->parent && dev->parent->driver)
982 printk(" driver: %s\n",
983 dev->parent->driver->name);
984 else
985 printk(" (driver?)\n");
986 } else
987 printk("\n");
988 }
989 disk_part_iter_exit(&piter);
990 }
991 class_dev_iter_exit(&iter);
992}
993
994#ifdef CONFIG_PROC_FS
995/* iterator */
996static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
997{
998 loff_t skip = *pos;
999 struct class_dev_iter *iter;
1000 struct device *dev;
1001
1002 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1003 if (!iter)
1004 return ERR_PTR(-ENOMEM);
1005
1006 seqf->private = iter;
1007 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1008 do {
1009 dev = class_dev_iter_next(iter);
1010 if (!dev)
1011 return NULL;
1012 } while (skip--);
1013
1014 return dev_to_disk(dev);
1015}
1016
1017static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1018{
1019 struct device *dev;
1020
1021 (*pos)++;
1022 dev = class_dev_iter_next(seqf->private);
1023 if (dev)
1024 return dev_to_disk(dev);
1025
1026 return NULL;
1027}
1028
1029static void disk_seqf_stop(struct seq_file *seqf, void *v)
1030{
1031 struct class_dev_iter *iter = seqf->private;
1032
1033 /* stop is called even after start failed :-( */
1034 if (iter) {
1035 class_dev_iter_exit(iter);
1036 kfree(iter);
1037 seqf->private = NULL;
1038 }
1039}
1040
1041static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1042{
1043 void *p;
1044
1045 p = disk_seqf_start(seqf, pos);
1046 if (!IS_ERR_OR_NULL(p) && !*pos)
1047 seq_puts(seqf, "major minor #blocks name\n\n");
1048 return p;
1049}
1050
1051static int show_partition(struct seq_file *seqf, void *v)
1052{
1053 struct gendisk *sgp = v;
1054 struct disk_part_iter piter;
1055 struct hd_struct *part;
1056 char buf[BDEVNAME_SIZE];
1057
1058 /* Don't show non-partitionable removeable devices or empty devices */
1059 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1060 (sgp->flags & GENHD_FL_REMOVABLE)))
1061 return 0;
1062 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1063 return 0;
1064
1065 /* show the full disk and all non-0 size partitions of it */
1066 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1067 while ((part = disk_part_iter_next(&piter)))
1068 seq_printf(seqf, "%4d %7d %10llu %s\n",
1069 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1070 (unsigned long long)part_nr_sects_read(part) >> 1,
1071 disk_name(sgp, part->partno, buf));
1072 disk_part_iter_exit(&piter);
1073
1074 return 0;
1075}
1076
1077static const struct seq_operations partitions_op = {
1078 .start = show_partition_start,
1079 .next = disk_seqf_next,
1080 .stop = disk_seqf_stop,
1081 .show = show_partition
1082};
1083#endif
1084
1085
1086static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1087{
1088 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1089 /* Make old-style 2.4 aliases work */
1090 request_module("block-major-%d", MAJOR(devt));
1091 return NULL;
1092}
1093
1094static int __init genhd_device_init(void)
1095{
1096 int error;
1097
1098 block_class.dev_kobj = sysfs_dev_block_kobj;
1099 error = class_register(&block_class);
1100 if (unlikely(error))
1101 return error;
1102 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1103 blk_dev_init();
1104
1105 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1106
1107 /* create top-level block dir */
1108 if (!sysfs_deprecated)
1109 block_depr = kobject_create_and_add("block", NULL);
1110 return 0;
1111}
1112
1113subsys_initcall(genhd_device_init);
1114
1115static ssize_t disk_range_show(struct device *dev,
1116 struct device_attribute *attr, char *buf)
1117{
1118 struct gendisk *disk = dev_to_disk(dev);
1119
1120 return sprintf(buf, "%d\n", disk->minors);
1121}
1122
1123static ssize_t disk_ext_range_show(struct device *dev,
1124 struct device_attribute *attr, char *buf)
1125{
1126 struct gendisk *disk = dev_to_disk(dev);
1127
1128 return sprintf(buf, "%d\n", disk_max_parts(disk));
1129}
1130
1131static ssize_t disk_removable_show(struct device *dev,
1132 struct device_attribute *attr, char *buf)
1133{
1134 struct gendisk *disk = dev_to_disk(dev);
1135
1136 return sprintf(buf, "%d\n",
1137 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1138}
1139
1140static ssize_t disk_hidden_show(struct device *dev,
1141 struct device_attribute *attr, char *buf)
1142{
1143 struct gendisk *disk = dev_to_disk(dev);
1144
1145 return sprintf(buf, "%d\n",
1146 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1147}
1148
1149static ssize_t disk_ro_show(struct device *dev,
1150 struct device_attribute *attr, char *buf)
1151{
1152 struct gendisk *disk = dev_to_disk(dev);
1153
1154 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1155}
1156
1157static ssize_t disk_capability_show(struct device *dev,
1158 struct device_attribute *attr, char *buf)
1159{
1160 struct gendisk *disk = dev_to_disk(dev);
1161
1162 return sprintf(buf, "%x\n", disk->flags);
1163}
1164
1165static ssize_t disk_alignment_offset_show(struct device *dev,
1166 struct device_attribute *attr,
1167 char *buf)
1168{
1169 struct gendisk *disk = dev_to_disk(dev);
1170
1171 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1172}
1173
1174static ssize_t disk_discard_alignment_show(struct device *dev,
1175 struct device_attribute *attr,
1176 char *buf)
1177{
1178 struct gendisk *disk = dev_to_disk(dev);
1179
1180 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1181}
1182
1183static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1184static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1185static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1186static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1187static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1188static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1189static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1190static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1191static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1192static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1193static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1194static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1195#ifdef CONFIG_FAIL_MAKE_REQUEST
1196static struct device_attribute dev_attr_fail =
1197 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1198#endif
1199#ifdef CONFIG_FAIL_IO_TIMEOUT
1200static struct device_attribute dev_attr_fail_timeout =
1201 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1202#endif
1203
1204static struct attribute *disk_attrs[] = {
1205 &dev_attr_range.attr,
1206 &dev_attr_ext_range.attr,
1207 &dev_attr_removable.attr,
1208 &dev_attr_hidden.attr,
1209 &dev_attr_ro.attr,
1210 &dev_attr_size.attr,
1211 &dev_attr_alignment_offset.attr,
1212 &dev_attr_discard_alignment.attr,
1213 &dev_attr_capability.attr,
1214 &dev_attr_stat.attr,
1215 &dev_attr_inflight.attr,
1216 &dev_attr_badblocks.attr,
1217#ifdef CONFIG_FAIL_MAKE_REQUEST
1218 &dev_attr_fail.attr,
1219#endif
1220#ifdef CONFIG_FAIL_IO_TIMEOUT
1221 &dev_attr_fail_timeout.attr,
1222#endif
1223 NULL
1224};
1225
1226static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1227{
1228 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1229 struct gendisk *disk = dev_to_disk(dev);
1230
1231 if (a == &dev_attr_badblocks.attr && !disk->bb)
1232 return 0;
1233 return a->mode;
1234}
1235
1236static struct attribute_group disk_attr_group = {
1237 .attrs = disk_attrs,
1238 .is_visible = disk_visible,
1239};
1240
1241static const struct attribute_group *disk_attr_groups[] = {
1242 &disk_attr_group,
1243 NULL
1244};
1245
1246/**
1247 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1248 * @disk: disk to replace part_tbl for
1249 * @new_ptbl: new part_tbl to install
1250 *
1251 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1252 * original ptbl is freed using RCU callback.
1253 *
1254 * LOCKING:
1255 * Matching bd_mutex locked or the caller is the only user of @disk.
1256 */
1257static void disk_replace_part_tbl(struct gendisk *disk,
1258 struct disk_part_tbl *new_ptbl)
1259{
1260 struct disk_part_tbl *old_ptbl =
1261 rcu_dereference_protected(disk->part_tbl, 1);
1262
1263 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1264
1265 if (old_ptbl) {
1266 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1267 kfree_rcu(old_ptbl, rcu_head);
1268 }
1269}
1270
1271/**
1272 * disk_expand_part_tbl - expand disk->part_tbl
1273 * @disk: disk to expand part_tbl for
1274 * @partno: expand such that this partno can fit in
1275 *
1276 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1277 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1278 *
1279 * LOCKING:
1280 * Matching bd_mutex locked or the caller is the only user of @disk.
1281 * Might sleep.
1282 *
1283 * RETURNS:
1284 * 0 on success, -errno on failure.
1285 */
1286int disk_expand_part_tbl(struct gendisk *disk, int partno)
1287{
1288 struct disk_part_tbl *old_ptbl =
1289 rcu_dereference_protected(disk->part_tbl, 1);
1290 struct disk_part_tbl *new_ptbl;
1291 int len = old_ptbl ? old_ptbl->len : 0;
1292 int i, target;
1293
1294 /*
1295 * check for int overflow, since we can get here from blkpg_ioctl()
1296 * with a user passed 'partno'.
1297 */
1298 target = partno + 1;
1299 if (target < 0)
1300 return -EINVAL;
1301
1302 /* disk_max_parts() is zero during initialization, ignore if so */
1303 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1304 return -EINVAL;
1305
1306 if (target <= len)
1307 return 0;
1308
1309 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1310 disk->node_id);
1311 if (!new_ptbl)
1312 return -ENOMEM;
1313
1314 new_ptbl->len = target;
1315
1316 for (i = 0; i < len; i++)
1317 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1318
1319 disk_replace_part_tbl(disk, new_ptbl);
1320 return 0;
1321}
1322
1323static void disk_release(struct device *dev)
1324{
1325 struct gendisk *disk = dev_to_disk(dev);
1326
1327 blk_free_devt(dev->devt);
1328 disk_release_events(disk);
1329 kfree(disk->random);
1330 disk_replace_part_tbl(disk, NULL);
1331 hd_free_part(&disk->part0);
1332 if (disk->queue)
1333 blk_put_queue(disk->queue);
1334 kfree(disk);
1335}
1336struct class block_class = {
1337 .name = "block",
1338};
1339
1340static char *block_devnode(struct device *dev, umode_t *mode,
1341 kuid_t *uid, kgid_t *gid)
1342{
1343 struct gendisk *disk = dev_to_disk(dev);
1344
1345 if (disk->devnode)
1346 return disk->devnode(disk, mode);
1347 return NULL;
1348}
1349
1350static const struct device_type disk_type = {
1351 .name = "disk",
1352 .groups = disk_attr_groups,
1353 .release = disk_release,
1354 .devnode = block_devnode,
1355};
1356
1357#ifdef CONFIG_PROC_FS
1358/*
1359 * aggregate disk stat collector. Uses the same stats that the sysfs
1360 * entries do, above, but makes them available through one seq_file.
1361 *
1362 * The output looks suspiciously like /proc/partitions with a bunch of
1363 * extra fields.
1364 */
1365static int diskstats_show(struct seq_file *seqf, void *v)
1366{
1367 struct gendisk *gp = v;
1368 struct disk_part_iter piter;
1369 struct hd_struct *hd;
1370 char buf[BDEVNAME_SIZE];
1371 unsigned int inflight;
1372
1373 /*
1374 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1375 seq_puts(seqf, "major minor name"
1376 " rio rmerge rsect ruse wio wmerge "
1377 "wsect wuse running use aveq"
1378 "\n\n");
1379 */
1380
1381 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1382 while ((hd = disk_part_iter_next(&piter))) {
1383 inflight = part_in_flight(gp->queue, hd);
1384 seq_printf(seqf, "%4d %7d %s "
1385 "%lu %lu %lu %u "
1386 "%lu %lu %lu %u "
1387 "%u %u %u "
1388 "%lu %lu %lu %u\n",
1389 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1390 disk_name(gp, hd->partno, buf),
1391 part_stat_read(hd, ios[STAT_READ]),
1392 part_stat_read(hd, merges[STAT_READ]),
1393 part_stat_read(hd, sectors[STAT_READ]),
1394 (unsigned int)part_stat_read_msecs(hd, STAT_READ),
1395 part_stat_read(hd, ios[STAT_WRITE]),
1396 part_stat_read(hd, merges[STAT_WRITE]),
1397 part_stat_read(hd, sectors[STAT_WRITE]),
1398 (unsigned int)part_stat_read_msecs(hd, STAT_WRITE),
1399 inflight,
1400 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1401 jiffies_to_msecs(part_stat_read(hd, time_in_queue)),
1402 part_stat_read(hd, ios[STAT_DISCARD]),
1403 part_stat_read(hd, merges[STAT_DISCARD]),
1404 part_stat_read(hd, sectors[STAT_DISCARD]),
1405 (unsigned int)part_stat_read_msecs(hd, STAT_DISCARD)
1406 );
1407 }
1408 disk_part_iter_exit(&piter);
1409
1410 return 0;
1411}
1412
1413static const struct seq_operations diskstats_op = {
1414 .start = disk_seqf_start,
1415 .next = disk_seqf_next,
1416 .stop = disk_seqf_stop,
1417 .show = diskstats_show
1418};
1419
1420static int __init proc_genhd_init(void)
1421{
1422 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1423 proc_create_seq("partitions", 0, NULL, &partitions_op);
1424 return 0;
1425}
1426module_init(proc_genhd_init);
1427#endif /* CONFIG_PROC_FS */
1428
1429dev_t blk_lookup_devt(const char *name, int partno)
1430{
1431 dev_t devt = MKDEV(0, 0);
1432 struct class_dev_iter iter;
1433 struct device *dev;
1434
1435 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1436 while ((dev = class_dev_iter_next(&iter))) {
1437 struct gendisk *disk = dev_to_disk(dev);
1438 struct hd_struct *part;
1439
1440 if (strcmp(dev_name(dev), name))
1441 continue;
1442
1443 if (partno < disk->minors) {
1444 /* We need to return the right devno, even
1445 * if the partition doesn't exist yet.
1446 */
1447 devt = MKDEV(MAJOR(dev->devt),
1448 MINOR(dev->devt) + partno);
1449 break;
1450 }
1451 part = disk_get_part(disk, partno);
1452 if (part) {
1453 devt = part_devt(part);
1454 disk_put_part(part);
1455 break;
1456 }
1457 disk_put_part(part);
1458 }
1459 class_dev_iter_exit(&iter);
1460 return devt;
1461}
1462EXPORT_SYMBOL(blk_lookup_devt);
1463
1464struct gendisk *__alloc_disk_node(int minors, int node_id)
1465{
1466 struct gendisk *disk;
1467 struct disk_part_tbl *ptbl;
1468
1469 if (minors > DISK_MAX_PARTS) {
1470 printk(KERN_ERR
1471 "block: can't allocate more than %d partitions\n",
1472 DISK_MAX_PARTS);
1473 minors = DISK_MAX_PARTS;
1474 }
1475
1476 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1477 if (disk) {
1478 if (!init_part_stats(&disk->part0)) {
1479 kfree(disk);
1480 return NULL;
1481 }
1482 init_rwsem(&disk->lookup_sem);
1483 disk->node_id = node_id;
1484 if (disk_expand_part_tbl(disk, 0)) {
1485 free_part_stats(&disk->part0);
1486 kfree(disk);
1487 return NULL;
1488 }
1489 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1490 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1491
1492 /*
1493 * set_capacity() and get_capacity() currently don't use
1494 * seqcounter to read/update the part0->nr_sects. Still init
1495 * the counter as we can read the sectors in IO submission
1496 * patch using seqence counters.
1497 *
1498 * TODO: Ideally set_capacity() and get_capacity() should be
1499 * converted to make use of bd_mutex and sequence counters.
1500 */
1501 seqcount_init(&disk->part0.nr_sects_seq);
1502 if (hd_ref_init(&disk->part0)) {
1503 hd_free_part(&disk->part0);
1504 kfree(disk);
1505 return NULL;
1506 }
1507
1508 disk->minors = minors;
1509 rand_initialize_disk(disk);
1510 disk_to_dev(disk)->class = &block_class;
1511 disk_to_dev(disk)->type = &disk_type;
1512 device_initialize(disk_to_dev(disk));
1513 }
1514 return disk;
1515}
1516EXPORT_SYMBOL(__alloc_disk_node);
1517
1518struct kobject *get_disk_and_module(struct gendisk *disk)
1519{
1520 struct module *owner;
1521 struct kobject *kobj;
1522
1523 if (!disk->fops)
1524 return NULL;
1525 owner = disk->fops->owner;
1526 if (owner && !try_module_get(owner))
1527 return NULL;
1528 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1529 if (kobj == NULL) {
1530 module_put(owner);
1531 return NULL;
1532 }
1533 return kobj;
1534
1535}
1536EXPORT_SYMBOL(get_disk_and_module);
1537
1538void put_disk(struct gendisk *disk)
1539{
1540 if (disk)
1541 kobject_put(&disk_to_dev(disk)->kobj);
1542}
1543EXPORT_SYMBOL(put_disk);
1544
1545/*
1546 * This is a counterpart of get_disk_and_module() and thus also of
1547 * get_gendisk().
1548 */
1549void put_disk_and_module(struct gendisk *disk)
1550{
1551 if (disk) {
1552 struct module *owner = disk->fops->owner;
1553
1554 put_disk(disk);
1555 module_put(owner);
1556 }
1557}
1558EXPORT_SYMBOL(put_disk_and_module);
1559
1560static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1561{
1562 char event[] = "DISK_RO=1";
1563 char *envp[] = { event, NULL };
1564
1565 if (!ro)
1566 event[8] = '0';
1567 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1568}
1569
1570void set_device_ro(struct block_device *bdev, int flag)
1571{
1572 bdev->bd_part->policy = flag;
1573}
1574
1575EXPORT_SYMBOL(set_device_ro);
1576
1577void set_disk_ro(struct gendisk *disk, int flag)
1578{
1579 struct disk_part_iter piter;
1580 struct hd_struct *part;
1581
1582 if (disk->part0.policy != flag) {
1583 set_disk_ro_uevent(disk, flag);
1584 disk->part0.policy = flag;
1585 }
1586
1587 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1588 while ((part = disk_part_iter_next(&piter)))
1589 part->policy = flag;
1590 disk_part_iter_exit(&piter);
1591}
1592
1593EXPORT_SYMBOL(set_disk_ro);
1594
1595int bdev_read_only(struct block_device *bdev)
1596{
1597 if (!bdev)
1598 return 0;
1599 return bdev->bd_part->policy;
1600}
1601
1602EXPORT_SYMBOL(bdev_read_only);
1603
1604int invalidate_partition(struct gendisk *disk, int partno)
1605{
1606 int res = 0;
1607 struct block_device *bdev = bdget_disk(disk, partno);
1608 if (bdev) {
1609 fsync_bdev(bdev);
1610 res = __invalidate_device(bdev, true);
1611 bdput(bdev);
1612 }
1613 return res;
1614}
1615
1616EXPORT_SYMBOL(invalidate_partition);
1617
1618/*
1619 * Disk events - monitor disk events like media change and eject request.
1620 */
1621struct disk_events {
1622 struct list_head node; /* all disk_event's */
1623 struct gendisk *disk; /* the associated disk */
1624 spinlock_t lock;
1625
1626 struct mutex block_mutex; /* protects blocking */
1627 int block; /* event blocking depth */
1628 unsigned int pending; /* events already sent out */
1629 unsigned int clearing; /* events being cleared */
1630
1631 long poll_msecs; /* interval, -1 for default */
1632 struct delayed_work dwork;
1633};
1634
1635static const char *disk_events_strs[] = {
1636 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1637 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1638};
1639
1640static char *disk_uevents[] = {
1641 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1642 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1643};
1644
1645/* list of all disk_events */
1646static DEFINE_MUTEX(disk_events_mutex);
1647static LIST_HEAD(disk_events);
1648
1649/* disable in-kernel polling by default */
1650static unsigned long disk_events_dfl_poll_msecs;
1651
1652static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1653{
1654 struct disk_events *ev = disk->ev;
1655 long intv_msecs = 0;
1656
1657 /*
1658 * If device-specific poll interval is set, always use it. If
1659 * the default is being used, poll if the POLL flag is set.
1660 */
1661 if (ev->poll_msecs >= 0)
1662 intv_msecs = ev->poll_msecs;
1663 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1664 intv_msecs = disk_events_dfl_poll_msecs;
1665
1666 return msecs_to_jiffies(intv_msecs);
1667}
1668
1669/**
1670 * disk_block_events - block and flush disk event checking
1671 * @disk: disk to block events for
1672 *
1673 * On return from this function, it is guaranteed that event checking
1674 * isn't in progress and won't happen until unblocked by
1675 * disk_unblock_events(). Events blocking is counted and the actual
1676 * unblocking happens after the matching number of unblocks are done.
1677 *
1678 * Note that this intentionally does not block event checking from
1679 * disk_clear_events().
1680 *
1681 * CONTEXT:
1682 * Might sleep.
1683 */
1684void disk_block_events(struct gendisk *disk)
1685{
1686 struct disk_events *ev = disk->ev;
1687 unsigned long flags;
1688 bool cancel;
1689
1690 if (!ev)
1691 return;
1692
1693 /*
1694 * Outer mutex ensures that the first blocker completes canceling
1695 * the event work before further blockers are allowed to finish.
1696 */
1697 mutex_lock(&ev->block_mutex);
1698
1699 spin_lock_irqsave(&ev->lock, flags);
1700 cancel = !ev->block++;
1701 spin_unlock_irqrestore(&ev->lock, flags);
1702
1703 if (cancel)
1704 cancel_delayed_work_sync(&disk->ev->dwork);
1705
1706 mutex_unlock(&ev->block_mutex);
1707}
1708
1709static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1710{
1711 struct disk_events *ev = disk->ev;
1712 unsigned long intv;
1713 unsigned long flags;
1714
1715 spin_lock_irqsave(&ev->lock, flags);
1716
1717 if (WARN_ON_ONCE(ev->block <= 0))
1718 goto out_unlock;
1719
1720 if (--ev->block)
1721 goto out_unlock;
1722
1723 intv = disk_events_poll_jiffies(disk);
1724 if (check_now)
1725 queue_delayed_work(system_freezable_power_efficient_wq,
1726 &ev->dwork, 0);
1727 else if (intv)
1728 queue_delayed_work(system_freezable_power_efficient_wq,
1729 &ev->dwork, intv);
1730out_unlock:
1731 spin_unlock_irqrestore(&ev->lock, flags);
1732}
1733
1734/**
1735 * disk_unblock_events - unblock disk event checking
1736 * @disk: disk to unblock events for
1737 *
1738 * Undo disk_block_events(). When the block count reaches zero, it
1739 * starts events polling if configured.
1740 *
1741 * CONTEXT:
1742 * Don't care. Safe to call from irq context.
1743 */
1744void disk_unblock_events(struct gendisk *disk)
1745{
1746 if (disk->ev)
1747 __disk_unblock_events(disk, false);
1748}
1749
1750/**
1751 * disk_flush_events - schedule immediate event checking and flushing
1752 * @disk: disk to check and flush events for
1753 * @mask: events to flush
1754 *
1755 * Schedule immediate event checking on @disk if not blocked. Events in
1756 * @mask are scheduled to be cleared from the driver. Note that this
1757 * doesn't clear the events from @disk->ev.
1758 *
1759 * CONTEXT:
1760 * If @mask is non-zero must be called with bdev->bd_mutex held.
1761 */
1762void disk_flush_events(struct gendisk *disk, unsigned int mask)
1763{
1764 struct disk_events *ev = disk->ev;
1765
1766 if (!ev)
1767 return;
1768
1769 spin_lock_irq(&ev->lock);
1770 ev->clearing |= mask;
1771 if (!ev->block)
1772 mod_delayed_work(system_freezable_power_efficient_wq,
1773 &ev->dwork, 0);
1774 spin_unlock_irq(&ev->lock);
1775}
1776
1777/**
1778 * disk_clear_events - synchronously check, clear and return pending events
1779 * @disk: disk to fetch and clear events from
1780 * @mask: mask of events to be fetched and cleared
1781 *
1782 * Disk events are synchronously checked and pending events in @mask
1783 * are cleared and returned. This ignores the block count.
1784 *
1785 * CONTEXT:
1786 * Might sleep.
1787 */
1788unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1789{
1790 const struct block_device_operations *bdops = disk->fops;
1791 struct disk_events *ev = disk->ev;
1792 unsigned int pending;
1793 unsigned int clearing = mask;
1794
1795 if (!ev) {
1796 /* for drivers still using the old ->media_changed method */
1797 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1798 bdops->media_changed && bdops->media_changed(disk))
1799 return DISK_EVENT_MEDIA_CHANGE;
1800 return 0;
1801 }
1802
1803 disk_block_events(disk);
1804
1805 /*
1806 * store the union of mask and ev->clearing on the stack so that the
1807 * race with disk_flush_events does not cause ambiguity (ev->clearing
1808 * can still be modified even if events are blocked).
1809 */
1810 spin_lock_irq(&ev->lock);
1811 clearing |= ev->clearing;
1812 ev->clearing = 0;
1813 spin_unlock_irq(&ev->lock);
1814
1815 disk_check_events(ev, &clearing);
1816 /*
1817 * if ev->clearing is not 0, the disk_flush_events got called in the
1818 * middle of this function, so we want to run the workfn without delay.
1819 */
1820 __disk_unblock_events(disk, ev->clearing ? true : false);
1821
1822 /* then, fetch and clear pending events */
1823 spin_lock_irq(&ev->lock);
1824 pending = ev->pending & mask;
1825 ev->pending &= ~mask;
1826 spin_unlock_irq(&ev->lock);
1827 WARN_ON_ONCE(clearing & mask);
1828
1829 return pending;
1830}
1831
1832/*
1833 * Separate this part out so that a different pointer for clearing_ptr can be
1834 * passed in for disk_clear_events.
1835 */
1836static void disk_events_workfn(struct work_struct *work)
1837{
1838 struct delayed_work *dwork = to_delayed_work(work);
1839 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1840
1841 disk_check_events(ev, &ev->clearing);
1842}
1843
1844static void disk_check_events(struct disk_events *ev,
1845 unsigned int *clearing_ptr)
1846{
1847 struct gendisk *disk = ev->disk;
1848 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1849 unsigned int clearing = *clearing_ptr;
1850 unsigned int events;
1851 unsigned long intv;
1852 int nr_events = 0, i;
1853
1854 /* check events */
1855 events = disk->fops->check_events(disk, clearing);
1856
1857 /* accumulate pending events and schedule next poll if necessary */
1858 spin_lock_irq(&ev->lock);
1859
1860 events &= ~ev->pending;
1861 ev->pending |= events;
1862 *clearing_ptr &= ~clearing;
1863
1864 intv = disk_events_poll_jiffies(disk);
1865 if (!ev->block && intv)
1866 queue_delayed_work(system_freezable_power_efficient_wq,
1867 &ev->dwork, intv);
1868
1869 spin_unlock_irq(&ev->lock);
1870
1871 /*
1872 * Tell userland about new events. Only the events listed in
1873 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
1874 * is set. Otherwise, events are processed internally but never
1875 * get reported to userland.
1876 */
1877 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1878 if ((events & disk->events & (1 << i)) &&
1879 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
1880 envp[nr_events++] = disk_uevents[i];
1881
1882 if (nr_events)
1883 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1884}
1885
1886/*
1887 * A disk events enabled device has the following sysfs nodes under
1888 * its /sys/block/X/ directory.
1889 *
1890 * events : list of all supported events
1891 * events_async : list of events which can be detected w/o polling
1892 * (always empty, only for backwards compatibility)
1893 * events_poll_msecs : polling interval, 0: disable, -1: system default
1894 */
1895static ssize_t __disk_events_show(unsigned int events, char *buf)
1896{
1897 const char *delim = "";
1898 ssize_t pos = 0;
1899 int i;
1900
1901 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1902 if (events & (1 << i)) {
1903 pos += sprintf(buf + pos, "%s%s",
1904 delim, disk_events_strs[i]);
1905 delim = " ";
1906 }
1907 if (pos)
1908 pos += sprintf(buf + pos, "\n");
1909 return pos;
1910}
1911
1912static ssize_t disk_events_show(struct device *dev,
1913 struct device_attribute *attr, char *buf)
1914{
1915 struct gendisk *disk = dev_to_disk(dev);
1916
1917 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
1918 return 0;
1919
1920 return __disk_events_show(disk->events, buf);
1921}
1922
1923static ssize_t disk_events_async_show(struct device *dev,
1924 struct device_attribute *attr, char *buf)
1925{
1926 return 0;
1927}
1928
1929static ssize_t disk_events_poll_msecs_show(struct device *dev,
1930 struct device_attribute *attr,
1931 char *buf)
1932{
1933 struct gendisk *disk = dev_to_disk(dev);
1934
1935 if (!disk->ev)
1936 return sprintf(buf, "-1\n");
1937
1938 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1939}
1940
1941static ssize_t disk_events_poll_msecs_store(struct device *dev,
1942 struct device_attribute *attr,
1943 const char *buf, size_t count)
1944{
1945 struct gendisk *disk = dev_to_disk(dev);
1946 long intv;
1947
1948 if (!count || !sscanf(buf, "%ld", &intv))
1949 return -EINVAL;
1950
1951 if (intv < 0 && intv != -1)
1952 return -EINVAL;
1953
1954 if (!disk->ev)
1955 return -ENODEV;
1956
1957 disk_block_events(disk);
1958 disk->ev->poll_msecs = intv;
1959 __disk_unblock_events(disk, true);
1960
1961 return count;
1962}
1963
1964static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
1965static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
1966static const DEVICE_ATTR(events_poll_msecs, 0644,
1967 disk_events_poll_msecs_show,
1968 disk_events_poll_msecs_store);
1969
1970static const struct attribute *disk_events_attrs[] = {
1971 &dev_attr_events.attr,
1972 &dev_attr_events_async.attr,
1973 &dev_attr_events_poll_msecs.attr,
1974 NULL,
1975};
1976
1977/*
1978 * The default polling interval can be specified by the kernel
1979 * parameter block.events_dfl_poll_msecs which defaults to 0
1980 * (disable). This can also be modified runtime by writing to
1981 * /sys/module/block/parameters/events_dfl_poll_msecs.
1982 */
1983static int disk_events_set_dfl_poll_msecs(const char *val,
1984 const struct kernel_param *kp)
1985{
1986 struct disk_events *ev;
1987 int ret;
1988
1989 ret = param_set_ulong(val, kp);
1990 if (ret < 0)
1991 return ret;
1992
1993 mutex_lock(&disk_events_mutex);
1994
1995 list_for_each_entry(ev, &disk_events, node)
1996 disk_flush_events(ev->disk, 0);
1997
1998 mutex_unlock(&disk_events_mutex);
1999
2000 return 0;
2001}
2002
2003static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2004 .set = disk_events_set_dfl_poll_msecs,
2005 .get = param_get_ulong,
2006};
2007
2008#undef MODULE_PARAM_PREFIX
2009#define MODULE_PARAM_PREFIX "block."
2010
2011module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2012 &disk_events_dfl_poll_msecs, 0644);
2013
2014/*
2015 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2016 */
2017static void disk_alloc_events(struct gendisk *disk)
2018{
2019 struct disk_events *ev;
2020
2021 if (!disk->fops->check_events || !disk->events)
2022 return;
2023
2024 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2025 if (!ev) {
2026 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2027 return;
2028 }
2029
2030 INIT_LIST_HEAD(&ev->node);
2031 ev->disk = disk;
2032 spin_lock_init(&ev->lock);
2033 mutex_init(&ev->block_mutex);
2034 ev->block = 1;
2035 ev->poll_msecs = -1;
2036 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2037
2038 disk->ev = ev;
2039}
2040
2041static void disk_add_events(struct gendisk *disk)
2042{
2043 /* FIXME: error handling */
2044 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2045 pr_warn("%s: failed to create sysfs files for events\n",
2046 disk->disk_name);
2047
2048 if (!disk->ev)
2049 return;
2050
2051 mutex_lock(&disk_events_mutex);
2052 list_add_tail(&disk->ev->node, &disk_events);
2053 mutex_unlock(&disk_events_mutex);
2054
2055 /*
2056 * Block count is initialized to 1 and the following initial
2057 * unblock kicks it into action.
2058 */
2059 __disk_unblock_events(disk, true);
2060}
2061
2062static void disk_del_events(struct gendisk *disk)
2063{
2064 if (disk->ev) {
2065 disk_block_events(disk);
2066
2067 mutex_lock(&disk_events_mutex);
2068 list_del_init(&disk->ev->node);
2069 mutex_unlock(&disk_events_mutex);
2070 }
2071
2072 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2073}
2074
2075static void disk_release_events(struct gendisk *disk)
2076{
2077 /* the block count should be 1 from disk_del_events() */
2078 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
2079 kfree(disk->ev);
2080}