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/mutex.h>
19#include <linux/idr.h>
20#include <linux/log2.h>
21#include <linux/pm_runtime.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 NR_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_check_events(struct disk_events *ev,
40 unsigned int *clearing_ptr);
41static void disk_alloc_events(struct gendisk *disk);
42static void disk_add_events(struct gendisk *disk);
43static void disk_del_events(struct gendisk *disk);
44static void disk_release_events(struct gendisk *disk);
45
46/**
47 * disk_get_part - get partition
48 * @disk: disk to look partition from
49 * @partno: partition number
50 *
51 * Look for partition @partno from @disk. If found, increment
52 * reference count and return it.
53 *
54 * CONTEXT:
55 * Don't care.
56 *
57 * RETURNS:
58 * Pointer to the found partition on success, NULL if not found.
59 */
60struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
61{
62 struct hd_struct *part = NULL;
63 struct disk_part_tbl *ptbl;
64
65 if (unlikely(partno < 0))
66 return NULL;
67
68 rcu_read_lock();
69
70 ptbl = rcu_dereference(disk->part_tbl);
71 if (likely(partno < ptbl->len)) {
72 part = rcu_dereference(ptbl->part[partno]);
73 if (part)
74 get_device(part_to_dev(part));
75 }
76
77 rcu_read_unlock();
78
79 return part;
80}
81EXPORT_SYMBOL_GPL(disk_get_part);
82
83/**
84 * disk_part_iter_init - initialize partition iterator
85 * @piter: iterator to initialize
86 * @disk: disk to iterate over
87 * @flags: DISK_PITER_* flags
88 *
89 * Initialize @piter so that it iterates over partitions of @disk.
90 *
91 * CONTEXT:
92 * Don't care.
93 */
94void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
95 unsigned int flags)
96{
97 struct disk_part_tbl *ptbl;
98
99 rcu_read_lock();
100 ptbl = rcu_dereference(disk->part_tbl);
101
102 piter->disk = disk;
103 piter->part = NULL;
104
105 if (flags & DISK_PITER_REVERSE)
106 piter->idx = ptbl->len - 1;
107 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
108 piter->idx = 0;
109 else
110 piter->idx = 1;
111
112 piter->flags = flags;
113
114 rcu_read_unlock();
115}
116EXPORT_SYMBOL_GPL(disk_part_iter_init);
117
118/**
119 * disk_part_iter_next - proceed iterator to the next partition and return it
120 * @piter: iterator of interest
121 *
122 * Proceed @piter to the next partition and return it.
123 *
124 * CONTEXT:
125 * Don't care.
126 */
127struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
128{
129 struct disk_part_tbl *ptbl;
130 int inc, end;
131
132 /* put the last partition */
133 disk_put_part(piter->part);
134 piter->part = NULL;
135
136 /* get part_tbl */
137 rcu_read_lock();
138 ptbl = rcu_dereference(piter->disk->part_tbl);
139
140 /* determine iteration parameters */
141 if (piter->flags & DISK_PITER_REVERSE) {
142 inc = -1;
143 if (piter->flags & (DISK_PITER_INCL_PART0 |
144 DISK_PITER_INCL_EMPTY_PART0))
145 end = -1;
146 else
147 end = 0;
148 } else {
149 inc = 1;
150 end = ptbl->len;
151 }
152
153 /* iterate to the next partition */
154 for (; piter->idx != end; piter->idx += inc) {
155 struct hd_struct *part;
156
157 part = rcu_dereference(ptbl->part[piter->idx]);
158 if (!part)
159 continue;
160 if (!part_nr_sects_read(part) &&
161 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
162 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
163 piter->idx == 0))
164 continue;
165
166 get_device(part_to_dev(part));
167 piter->part = part;
168 piter->idx += inc;
169 break;
170 }
171
172 rcu_read_unlock();
173
174 return piter->part;
175}
176EXPORT_SYMBOL_GPL(disk_part_iter_next);
177
178/**
179 * disk_part_iter_exit - finish up partition iteration
180 * @piter: iter of interest
181 *
182 * Called when iteration is over. Cleans up @piter.
183 *
184 * CONTEXT:
185 * Don't care.
186 */
187void disk_part_iter_exit(struct disk_part_iter *piter)
188{
189 disk_put_part(piter->part);
190 piter->part = NULL;
191}
192EXPORT_SYMBOL_GPL(disk_part_iter_exit);
193
194static inline int sector_in_part(struct hd_struct *part, sector_t sector)
195{
196 return part->start_sect <= sector &&
197 sector < part->start_sect + part_nr_sects_read(part);
198}
199
200/**
201 * disk_map_sector_rcu - map sector to partition
202 * @disk: gendisk of interest
203 * @sector: sector to map
204 *
205 * Find out which partition @sector maps to on @disk. This is
206 * primarily used for stats accounting.
207 *
208 * CONTEXT:
209 * RCU read locked. The returned partition pointer is valid only
210 * while preemption is disabled.
211 *
212 * RETURNS:
213 * Found partition on success, part0 is returned if no partition matches
214 */
215struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
216{
217 struct disk_part_tbl *ptbl;
218 struct hd_struct *part;
219 int i;
220
221 ptbl = rcu_dereference(disk->part_tbl);
222
223 part = rcu_dereference(ptbl->last_lookup);
224 if (part && sector_in_part(part, sector))
225 return part;
226
227 for (i = 1; i < ptbl->len; i++) {
228 part = rcu_dereference(ptbl->part[i]);
229
230 if (part && sector_in_part(part, sector)) {
231 rcu_assign_pointer(ptbl->last_lookup, part);
232 return part;
233 }
234 }
235 return &disk->part0;
236}
237EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
238
239/*
240 * Can be deleted altogether. Later.
241 *
242 */
243static struct blk_major_name {
244 struct blk_major_name *next;
245 int major;
246 char name[16];
247} *major_names[BLKDEV_MAJOR_HASH_SIZE];
248
249/* index in the above - for now: assume no multimajor ranges */
250static inline int major_to_index(unsigned major)
251{
252 return major % BLKDEV_MAJOR_HASH_SIZE;
253}
254
255#ifdef CONFIG_PROC_FS
256void blkdev_show(struct seq_file *seqf, off_t offset)
257{
258 struct blk_major_name *dp;
259
260 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
261 mutex_lock(&block_class_lock);
262 for (dp = major_names[offset]; dp; dp = dp->next)
263 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
264 mutex_unlock(&block_class_lock);
265 }
266}
267#endif /* CONFIG_PROC_FS */
268
269/**
270 * register_blkdev - register a new block device
271 *
272 * @major: the requested major device number [1..255]. If @major=0, try to
273 * allocate any unused major number.
274 * @name: the name of the new block device as a zero terminated string
275 *
276 * The @name must be unique within the system.
277 *
278 * The return value depends on the @major input parameter.
279 * - if a major device number was requested in range [1..255] then the
280 * function returns zero on success, or a negative error code
281 * - if any unused major number was requested with @major=0 parameter
282 * then the return value is the allocated major number in range
283 * [1..255] or a negative error code otherwise
284 */
285int register_blkdev(unsigned int major, const char *name)
286{
287 struct blk_major_name **n, *p;
288 int index, ret = 0;
289
290 mutex_lock(&block_class_lock);
291
292 /* temporary */
293 if (major == 0) {
294 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
295 if (major_names[index] == NULL)
296 break;
297 }
298
299 if (index == 0) {
300 printk("register_blkdev: failed to get major for %s\n",
301 name);
302 ret = -EBUSY;
303 goto out;
304 }
305 major = index;
306 ret = major;
307 }
308
309 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
310 if (p == NULL) {
311 ret = -ENOMEM;
312 goto out;
313 }
314
315 p->major = major;
316 strlcpy(p->name, name, sizeof(p->name));
317 p->next = NULL;
318 index = major_to_index(major);
319
320 for (n = &major_names[index]; *n; n = &(*n)->next) {
321 if ((*n)->major == major)
322 break;
323 }
324 if (!*n)
325 *n = p;
326 else
327 ret = -EBUSY;
328
329 if (ret < 0) {
330 printk("register_blkdev: cannot get major %d for %s\n",
331 major, name);
332 kfree(p);
333 }
334out:
335 mutex_unlock(&block_class_lock);
336 return ret;
337}
338
339EXPORT_SYMBOL(register_blkdev);
340
341void unregister_blkdev(unsigned int major, const char *name)
342{
343 struct blk_major_name **n;
344 struct blk_major_name *p = NULL;
345 int index = major_to_index(major);
346
347 mutex_lock(&block_class_lock);
348 for (n = &major_names[index]; *n; n = &(*n)->next)
349 if ((*n)->major == major)
350 break;
351 if (!*n || strcmp((*n)->name, name)) {
352 WARN_ON(1);
353 } else {
354 p = *n;
355 *n = p->next;
356 }
357 mutex_unlock(&block_class_lock);
358 kfree(p);
359}
360
361EXPORT_SYMBOL(unregister_blkdev);
362
363static struct kobj_map *bdev_map;
364
365/**
366 * blk_mangle_minor - scatter minor numbers apart
367 * @minor: minor number to mangle
368 *
369 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
370 * is enabled. Mangling twice gives the original value.
371 *
372 * RETURNS:
373 * Mangled value.
374 *
375 * CONTEXT:
376 * Don't care.
377 */
378static int blk_mangle_minor(int minor)
379{
380#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
381 int i;
382
383 for (i = 0; i < MINORBITS / 2; i++) {
384 int low = minor & (1 << i);
385 int high = minor & (1 << (MINORBITS - 1 - i));
386 int distance = MINORBITS - 1 - 2 * i;
387
388 minor ^= low | high; /* clear both bits */
389 low <<= distance; /* swap the positions */
390 high >>= distance;
391 minor |= low | high; /* and set */
392 }
393#endif
394 return minor;
395}
396
397/**
398 * blk_alloc_devt - allocate a dev_t for a partition
399 * @part: partition to allocate dev_t for
400 * @devt: out parameter for resulting dev_t
401 *
402 * Allocate a dev_t for block device.
403 *
404 * RETURNS:
405 * 0 on success, allocated dev_t is returned in *@devt. -errno on
406 * failure.
407 *
408 * CONTEXT:
409 * Might sleep.
410 */
411int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
412{
413 struct gendisk *disk = part_to_disk(part);
414 int idx;
415
416 /* in consecutive minor range? */
417 if (part->partno < disk->minors) {
418 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
419 return 0;
420 }
421
422 /* allocate ext devt */
423 mutex_lock(&ext_devt_mutex);
424 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_KERNEL);
425 mutex_unlock(&ext_devt_mutex);
426 if (idx < 0)
427 return idx == -ENOSPC ? -EBUSY : idx;
428
429 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
430 return 0;
431}
432
433/**
434 * blk_free_devt - free a dev_t
435 * @devt: dev_t to free
436 *
437 * Free @devt which was allocated using blk_alloc_devt().
438 *
439 * CONTEXT:
440 * Might sleep.
441 */
442void blk_free_devt(dev_t devt)
443{
444 might_sleep();
445
446 if (devt == MKDEV(0, 0))
447 return;
448
449 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
450 mutex_lock(&ext_devt_mutex);
451 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
452 mutex_unlock(&ext_devt_mutex);
453 }
454}
455
456static char *bdevt_str(dev_t devt, char *buf)
457{
458 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
459 char tbuf[BDEVT_SIZE];
460 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
461 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
462 } else
463 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
464
465 return buf;
466}
467
468/*
469 * Register device numbers dev..(dev+range-1)
470 * range must be nonzero
471 * The hash chain is sorted on range, so that subranges can override.
472 */
473void blk_register_region(dev_t devt, unsigned long range, struct module *module,
474 struct kobject *(*probe)(dev_t, int *, void *),
475 int (*lock)(dev_t, void *), void *data)
476{
477 kobj_map(bdev_map, devt, range, module, probe, lock, data);
478}
479
480EXPORT_SYMBOL(blk_register_region);
481
482void blk_unregister_region(dev_t devt, unsigned long range)
483{
484 kobj_unmap(bdev_map, devt, range);
485}
486
487EXPORT_SYMBOL(blk_unregister_region);
488
489static struct kobject *exact_match(dev_t devt, int *partno, void *data)
490{
491 struct gendisk *p = data;
492
493 return &disk_to_dev(p)->kobj;
494}
495
496static int exact_lock(dev_t devt, void *data)
497{
498 struct gendisk *p = data;
499
500 if (!get_disk(p))
501 return -1;
502 return 0;
503}
504
505static void register_disk(struct gendisk *disk)
506{
507 struct device *ddev = disk_to_dev(disk);
508 struct block_device *bdev;
509 struct disk_part_iter piter;
510 struct hd_struct *part;
511 int err;
512
513 ddev->parent = disk->driverfs_dev;
514
515 dev_set_name(ddev, "%s", disk->disk_name);
516
517 /* delay uevents, until we scanned partition table */
518 dev_set_uevent_suppress(ddev, 1);
519
520 if (device_add(ddev))
521 return;
522 if (!sysfs_deprecated) {
523 err = sysfs_create_link(block_depr, &ddev->kobj,
524 kobject_name(&ddev->kobj));
525 if (err) {
526 device_del(ddev);
527 return;
528 }
529 }
530
531 /*
532 * avoid probable deadlock caused by allocating memory with
533 * GFP_KERNEL in runtime_resume callback of its all ancestor
534 * devices
535 */
536 pm_runtime_set_memalloc_noio(ddev, true);
537
538 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
539 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
540
541 /* No minors to use for partitions */
542 if (!disk_part_scan_enabled(disk))
543 goto exit;
544
545 /* No such device (e.g., media were just removed) */
546 if (!get_capacity(disk))
547 goto exit;
548
549 bdev = bdget_disk(disk, 0);
550 if (!bdev)
551 goto exit;
552
553 bdev->bd_invalidated = 1;
554 err = blkdev_get(bdev, FMODE_READ, NULL);
555 if (err < 0)
556 goto exit;
557 blkdev_put(bdev, FMODE_READ);
558
559exit:
560 /* announce disk after possible partitions are created */
561 dev_set_uevent_suppress(ddev, 0);
562 kobject_uevent(&ddev->kobj, KOBJ_ADD);
563
564 /* announce possible partitions */
565 disk_part_iter_init(&piter, disk, 0);
566 while ((part = disk_part_iter_next(&piter)))
567 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
568 disk_part_iter_exit(&piter);
569}
570
571/**
572 * add_disk - add partitioning information to kernel list
573 * @disk: per-device partitioning information
574 *
575 * This function registers the partitioning information in @disk
576 * with the kernel.
577 *
578 * FIXME: error handling
579 */
580void add_disk(struct gendisk *disk)
581{
582 struct backing_dev_info *bdi;
583 dev_t devt;
584 int retval;
585
586 /* minors == 0 indicates to use ext devt from part0 and should
587 * be accompanied with EXT_DEVT flag. Make sure all
588 * parameters make sense.
589 */
590 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
591 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
592
593 disk->flags |= GENHD_FL_UP;
594
595 retval = blk_alloc_devt(&disk->part0, &devt);
596 if (retval) {
597 WARN_ON(1);
598 return;
599 }
600 disk_to_dev(disk)->devt = devt;
601
602 /* ->major and ->first_minor aren't supposed to be
603 * dereferenced from here on, but set them just in case.
604 */
605 disk->major = MAJOR(devt);
606 disk->first_minor = MINOR(devt);
607
608 disk_alloc_events(disk);
609
610 /* Register BDI before referencing it from bdev */
611 bdi = &disk->queue->backing_dev_info;
612 bdi_register_dev(bdi, disk_devt(disk));
613
614 blk_register_region(disk_devt(disk), disk->minors, NULL,
615 exact_match, exact_lock, disk);
616 register_disk(disk);
617 blk_register_queue(disk);
618
619 /*
620 * Take an extra ref on queue which will be put on disk_release()
621 * so that it sticks around as long as @disk is there.
622 */
623 WARN_ON_ONCE(!blk_get_queue(disk->queue));
624
625 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
626 "bdi");
627 WARN_ON(retval);
628
629 disk_add_events(disk);
630}
631EXPORT_SYMBOL(add_disk);
632
633void del_gendisk(struct gendisk *disk)
634{
635 struct disk_part_iter piter;
636 struct hd_struct *part;
637
638 disk_del_events(disk);
639
640 /* invalidate stuff */
641 disk_part_iter_init(&piter, disk,
642 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
643 while ((part = disk_part_iter_next(&piter))) {
644 invalidate_partition(disk, part->partno);
645 delete_partition(disk, part->partno);
646 }
647 disk_part_iter_exit(&piter);
648
649 invalidate_partition(disk, 0);
650 set_capacity(disk, 0);
651 disk->flags &= ~GENHD_FL_UP;
652
653 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
654 bdi_unregister(&disk->queue->backing_dev_info);
655 blk_unregister_queue(disk);
656 blk_unregister_region(disk_devt(disk), disk->minors);
657
658 part_stat_set_all(&disk->part0, 0);
659 disk->part0.stamp = 0;
660
661 kobject_put(disk->part0.holder_dir);
662 kobject_put(disk->slave_dir);
663 disk->driverfs_dev = NULL;
664 if (!sysfs_deprecated)
665 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
666 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
667 device_del(disk_to_dev(disk));
668 blk_free_devt(disk_to_dev(disk)->devt);
669}
670EXPORT_SYMBOL(del_gendisk);
671
672/**
673 * get_gendisk - get partitioning information for a given device
674 * @devt: device to get partitioning information for
675 * @partno: returned partition index
676 *
677 * This function gets the structure containing partitioning
678 * information for the given device @devt.
679 */
680struct gendisk *get_gendisk(dev_t devt, int *partno)
681{
682 struct gendisk *disk = NULL;
683
684 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
685 struct kobject *kobj;
686
687 kobj = kobj_lookup(bdev_map, devt, partno);
688 if (kobj)
689 disk = dev_to_disk(kobj_to_dev(kobj));
690 } else {
691 struct hd_struct *part;
692
693 mutex_lock(&ext_devt_mutex);
694 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
695 if (part && get_disk(part_to_disk(part))) {
696 *partno = part->partno;
697 disk = part_to_disk(part);
698 }
699 mutex_unlock(&ext_devt_mutex);
700 }
701
702 return disk;
703}
704EXPORT_SYMBOL(get_gendisk);
705
706/**
707 * bdget_disk - do bdget() by gendisk and partition number
708 * @disk: gendisk of interest
709 * @partno: partition number
710 *
711 * Find partition @partno from @disk, do bdget() on it.
712 *
713 * CONTEXT:
714 * Don't care.
715 *
716 * RETURNS:
717 * Resulting block_device on success, NULL on failure.
718 */
719struct block_device *bdget_disk(struct gendisk *disk, int partno)
720{
721 struct hd_struct *part;
722 struct block_device *bdev = NULL;
723
724 part = disk_get_part(disk, partno);
725 if (part)
726 bdev = bdget(part_devt(part));
727 disk_put_part(part);
728
729 return bdev;
730}
731EXPORT_SYMBOL(bdget_disk);
732
733/*
734 * print a full list of all partitions - intended for places where the root
735 * filesystem can't be mounted and thus to give the victim some idea of what
736 * went wrong
737 */
738void __init printk_all_partitions(void)
739{
740 struct class_dev_iter iter;
741 struct device *dev;
742
743 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
744 while ((dev = class_dev_iter_next(&iter))) {
745 struct gendisk *disk = dev_to_disk(dev);
746 struct disk_part_iter piter;
747 struct hd_struct *part;
748 char name_buf[BDEVNAME_SIZE];
749 char devt_buf[BDEVT_SIZE];
750
751 /*
752 * Don't show empty devices or things that have been
753 * suppressed
754 */
755 if (get_capacity(disk) == 0 ||
756 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
757 continue;
758
759 /*
760 * Note, unlike /proc/partitions, I am showing the
761 * numbers in hex - the same format as the root=
762 * option takes.
763 */
764 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
765 while ((part = disk_part_iter_next(&piter))) {
766 bool is_part0 = part == &disk->part0;
767
768 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
769 bdevt_str(part_devt(part), devt_buf),
770 (unsigned long long)part_nr_sects_read(part) >> 1
771 , disk_name(disk, part->partno, name_buf),
772 part->info ? part->info->uuid : "");
773 if (is_part0) {
774 if (disk->driverfs_dev != NULL &&
775 disk->driverfs_dev->driver != NULL)
776 printk(" driver: %s\n",
777 disk->driverfs_dev->driver->name);
778 else
779 printk(" (driver?)\n");
780 } else
781 printk("\n");
782 }
783 disk_part_iter_exit(&piter);
784 }
785 class_dev_iter_exit(&iter);
786}
787
788#ifdef CONFIG_PROC_FS
789/* iterator */
790static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
791{
792 loff_t skip = *pos;
793 struct class_dev_iter *iter;
794 struct device *dev;
795
796 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
797 if (!iter)
798 return ERR_PTR(-ENOMEM);
799
800 seqf->private = iter;
801 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
802 do {
803 dev = class_dev_iter_next(iter);
804 if (!dev)
805 return NULL;
806 } while (skip--);
807
808 return dev_to_disk(dev);
809}
810
811static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
812{
813 struct device *dev;
814
815 (*pos)++;
816 dev = class_dev_iter_next(seqf->private);
817 if (dev)
818 return dev_to_disk(dev);
819
820 return NULL;
821}
822
823static void disk_seqf_stop(struct seq_file *seqf, void *v)
824{
825 struct class_dev_iter *iter = seqf->private;
826
827 /* stop is called even after start failed :-( */
828 if (iter) {
829 class_dev_iter_exit(iter);
830 kfree(iter);
831 }
832}
833
834static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
835{
836 void *p;
837
838 p = disk_seqf_start(seqf, pos);
839 if (!IS_ERR_OR_NULL(p) && !*pos)
840 seq_puts(seqf, "major minor #blocks name\n\n");
841 return p;
842}
843
844static int show_partition(struct seq_file *seqf, void *v)
845{
846 struct gendisk *sgp = v;
847 struct disk_part_iter piter;
848 struct hd_struct *part;
849 char buf[BDEVNAME_SIZE];
850
851 /* Don't show non-partitionable removeable devices or empty devices */
852 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
853 (sgp->flags & GENHD_FL_REMOVABLE)))
854 return 0;
855 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
856 return 0;
857
858 /* show the full disk and all non-0 size partitions of it */
859 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
860 while ((part = disk_part_iter_next(&piter)))
861 seq_printf(seqf, "%4d %7d %10llu %s\n",
862 MAJOR(part_devt(part)), MINOR(part_devt(part)),
863 (unsigned long long)part_nr_sects_read(part) >> 1,
864 disk_name(sgp, part->partno, buf));
865 disk_part_iter_exit(&piter);
866
867 return 0;
868}
869
870static const struct seq_operations partitions_op = {
871 .start = show_partition_start,
872 .next = disk_seqf_next,
873 .stop = disk_seqf_stop,
874 .show = show_partition
875};
876
877static int partitions_open(struct inode *inode, struct file *file)
878{
879 return seq_open(file, &partitions_op);
880}
881
882static const struct file_operations proc_partitions_operations = {
883 .open = partitions_open,
884 .read = seq_read,
885 .llseek = seq_lseek,
886 .release = seq_release,
887};
888#endif
889
890
891static struct kobject *base_probe(dev_t devt, int *partno, void *data)
892{
893 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
894 /* Make old-style 2.4 aliases work */
895 request_module("block-major-%d", MAJOR(devt));
896 return NULL;
897}
898
899static int __init genhd_device_init(void)
900{
901 int error;
902
903 block_class.dev_kobj = sysfs_dev_block_kobj;
904 error = class_register(&block_class);
905 if (unlikely(error))
906 return error;
907 bdev_map = kobj_map_init(base_probe, &block_class_lock);
908 blk_dev_init();
909
910 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
911
912 /* create top-level block dir */
913 if (!sysfs_deprecated)
914 block_depr = kobject_create_and_add("block", NULL);
915 return 0;
916}
917
918subsys_initcall(genhd_device_init);
919
920static ssize_t disk_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->minors);
926}
927
928static ssize_t disk_ext_range_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", disk_max_parts(disk));
934}
935
936static ssize_t disk_removable_show(struct device *dev,
937 struct device_attribute *attr, char *buf)
938{
939 struct gendisk *disk = dev_to_disk(dev);
940
941 return sprintf(buf, "%d\n",
942 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
943}
944
945static ssize_t disk_ro_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, "%d\n", get_disk_ro(disk) ? 1 : 0);
951}
952
953static ssize_t disk_capability_show(struct device *dev,
954 struct device_attribute *attr, char *buf)
955{
956 struct gendisk *disk = dev_to_disk(dev);
957
958 return sprintf(buf, "%x\n", disk->flags);
959}
960
961static ssize_t disk_alignment_offset_show(struct device *dev,
962 struct device_attribute *attr,
963 char *buf)
964{
965 struct gendisk *disk = dev_to_disk(dev);
966
967 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
968}
969
970static ssize_t disk_discard_alignment_show(struct device *dev,
971 struct device_attribute *attr,
972 char *buf)
973{
974 struct gendisk *disk = dev_to_disk(dev);
975
976 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
977}
978
979static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
980static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
981static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
982static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
983static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
984static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
985static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
986 NULL);
987static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
988static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
989static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
990#ifdef CONFIG_FAIL_MAKE_REQUEST
991static struct device_attribute dev_attr_fail =
992 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
993#endif
994#ifdef CONFIG_FAIL_IO_TIMEOUT
995static struct device_attribute dev_attr_fail_timeout =
996 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
997 part_timeout_store);
998#endif
999
1000static struct attribute *disk_attrs[] = {
1001 &dev_attr_range.attr,
1002 &dev_attr_ext_range.attr,
1003 &dev_attr_removable.attr,
1004 &dev_attr_ro.attr,
1005 &dev_attr_size.attr,
1006 &dev_attr_alignment_offset.attr,
1007 &dev_attr_discard_alignment.attr,
1008 &dev_attr_capability.attr,
1009 &dev_attr_stat.attr,
1010 &dev_attr_inflight.attr,
1011#ifdef CONFIG_FAIL_MAKE_REQUEST
1012 &dev_attr_fail.attr,
1013#endif
1014#ifdef CONFIG_FAIL_IO_TIMEOUT
1015 &dev_attr_fail_timeout.attr,
1016#endif
1017 NULL
1018};
1019
1020static struct attribute_group disk_attr_group = {
1021 .attrs = disk_attrs,
1022};
1023
1024static const struct attribute_group *disk_attr_groups[] = {
1025 &disk_attr_group,
1026 NULL
1027};
1028
1029/**
1030 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1031 * @disk: disk to replace part_tbl for
1032 * @new_ptbl: new part_tbl to install
1033 *
1034 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1035 * original ptbl is freed using RCU callback.
1036 *
1037 * LOCKING:
1038 * Matching bd_mutx locked.
1039 */
1040static void disk_replace_part_tbl(struct gendisk *disk,
1041 struct disk_part_tbl *new_ptbl)
1042{
1043 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1044
1045 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1046
1047 if (old_ptbl) {
1048 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1049 kfree_rcu(old_ptbl, rcu_head);
1050 }
1051}
1052
1053/**
1054 * disk_expand_part_tbl - expand disk->part_tbl
1055 * @disk: disk to expand part_tbl for
1056 * @partno: expand such that this partno can fit in
1057 *
1058 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1059 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1060 *
1061 * LOCKING:
1062 * Matching bd_mutex locked, might sleep.
1063 *
1064 * RETURNS:
1065 * 0 on success, -errno on failure.
1066 */
1067int disk_expand_part_tbl(struct gendisk *disk, int partno)
1068{
1069 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1070 struct disk_part_tbl *new_ptbl;
1071 int len = old_ptbl ? old_ptbl->len : 0;
1072 int target = partno + 1;
1073 size_t size;
1074 int i;
1075
1076 /* disk_max_parts() is zero during initialization, ignore if so */
1077 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1078 return -EINVAL;
1079
1080 if (target <= len)
1081 return 0;
1082
1083 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1084 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1085 if (!new_ptbl)
1086 return -ENOMEM;
1087
1088 new_ptbl->len = target;
1089
1090 for (i = 0; i < len; i++)
1091 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1092
1093 disk_replace_part_tbl(disk, new_ptbl);
1094 return 0;
1095}
1096
1097static void disk_release(struct device *dev)
1098{
1099 struct gendisk *disk = dev_to_disk(dev);
1100
1101 disk_release_events(disk);
1102 kfree(disk->random);
1103 disk_replace_part_tbl(disk, NULL);
1104 free_part_stats(&disk->part0);
1105 free_part_info(&disk->part0);
1106 if (disk->queue)
1107 blk_put_queue(disk->queue);
1108 kfree(disk);
1109}
1110struct class block_class = {
1111 .name = "block",
1112};
1113
1114static char *block_devnode(struct device *dev, umode_t *mode,
1115 kuid_t *uid, kgid_t *gid)
1116{
1117 struct gendisk *disk = dev_to_disk(dev);
1118
1119 if (disk->devnode)
1120 return disk->devnode(disk, mode);
1121 return NULL;
1122}
1123
1124static struct device_type disk_type = {
1125 .name = "disk",
1126 .groups = disk_attr_groups,
1127 .release = disk_release,
1128 .devnode = block_devnode,
1129};
1130
1131#ifdef CONFIG_PROC_FS
1132/*
1133 * aggregate disk stat collector. Uses the same stats that the sysfs
1134 * entries do, above, but makes them available through one seq_file.
1135 *
1136 * The output looks suspiciously like /proc/partitions with a bunch of
1137 * extra fields.
1138 */
1139static int diskstats_show(struct seq_file *seqf, void *v)
1140{
1141 struct gendisk *gp = v;
1142 struct disk_part_iter piter;
1143 struct hd_struct *hd;
1144 char buf[BDEVNAME_SIZE];
1145 int cpu;
1146
1147 /*
1148 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1149 seq_puts(seqf, "major minor name"
1150 " rio rmerge rsect ruse wio wmerge "
1151 "wsect wuse running use aveq"
1152 "\n\n");
1153 */
1154
1155 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1156 while ((hd = disk_part_iter_next(&piter))) {
1157 cpu = part_stat_lock();
1158 part_round_stats(cpu, hd);
1159 part_stat_unlock();
1160 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1161 "%u %lu %lu %lu %u %u %u %u\n",
1162 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1163 disk_name(gp, hd->partno, buf),
1164 part_stat_read(hd, ios[READ]),
1165 part_stat_read(hd, merges[READ]),
1166 part_stat_read(hd, sectors[READ]),
1167 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1168 part_stat_read(hd, ios[WRITE]),
1169 part_stat_read(hd, merges[WRITE]),
1170 part_stat_read(hd, sectors[WRITE]),
1171 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1172 part_in_flight(hd),
1173 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1174 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1175 );
1176 }
1177 disk_part_iter_exit(&piter);
1178
1179 return 0;
1180}
1181
1182static const struct seq_operations diskstats_op = {
1183 .start = disk_seqf_start,
1184 .next = disk_seqf_next,
1185 .stop = disk_seqf_stop,
1186 .show = diskstats_show
1187};
1188
1189static int diskstats_open(struct inode *inode, struct file *file)
1190{
1191 return seq_open(file, &diskstats_op);
1192}
1193
1194static const struct file_operations proc_diskstats_operations = {
1195 .open = diskstats_open,
1196 .read = seq_read,
1197 .llseek = seq_lseek,
1198 .release = seq_release,
1199};
1200
1201static int __init proc_genhd_init(void)
1202{
1203 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1204 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1205 return 0;
1206}
1207module_init(proc_genhd_init);
1208#endif /* CONFIG_PROC_FS */
1209
1210dev_t blk_lookup_devt(const char *name, int partno)
1211{
1212 dev_t devt = MKDEV(0, 0);
1213 struct class_dev_iter iter;
1214 struct device *dev;
1215
1216 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1217 while ((dev = class_dev_iter_next(&iter))) {
1218 struct gendisk *disk = dev_to_disk(dev);
1219 struct hd_struct *part;
1220
1221 if (strcmp(dev_name(dev), name))
1222 continue;
1223
1224 if (partno < disk->minors) {
1225 /* We need to return the right devno, even
1226 * if the partition doesn't exist yet.
1227 */
1228 devt = MKDEV(MAJOR(dev->devt),
1229 MINOR(dev->devt) + partno);
1230 break;
1231 }
1232 part = disk_get_part(disk, partno);
1233 if (part) {
1234 devt = part_devt(part);
1235 disk_put_part(part);
1236 break;
1237 }
1238 disk_put_part(part);
1239 }
1240 class_dev_iter_exit(&iter);
1241 return devt;
1242}
1243EXPORT_SYMBOL(blk_lookup_devt);
1244
1245struct gendisk *alloc_disk(int minors)
1246{
1247 return alloc_disk_node(minors, NUMA_NO_NODE);
1248}
1249EXPORT_SYMBOL(alloc_disk);
1250
1251struct gendisk *alloc_disk_node(int minors, int node_id)
1252{
1253 struct gendisk *disk;
1254
1255 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1256 if (disk) {
1257 if (!init_part_stats(&disk->part0)) {
1258 kfree(disk);
1259 return NULL;
1260 }
1261 disk->node_id = node_id;
1262 if (disk_expand_part_tbl(disk, 0)) {
1263 free_part_stats(&disk->part0);
1264 kfree(disk);
1265 return NULL;
1266 }
1267 disk->part_tbl->part[0] = &disk->part0;
1268
1269 /*
1270 * set_capacity() and get_capacity() currently don't use
1271 * seqcounter to read/update the part0->nr_sects. Still init
1272 * the counter as we can read the sectors in IO submission
1273 * patch using seqence counters.
1274 *
1275 * TODO: Ideally set_capacity() and get_capacity() should be
1276 * converted to make use of bd_mutex and sequence counters.
1277 */
1278 seqcount_init(&disk->part0.nr_sects_seq);
1279 hd_ref_init(&disk->part0);
1280
1281 disk->minors = minors;
1282 rand_initialize_disk(disk);
1283 disk_to_dev(disk)->class = &block_class;
1284 disk_to_dev(disk)->type = &disk_type;
1285 device_initialize(disk_to_dev(disk));
1286 }
1287 return disk;
1288}
1289EXPORT_SYMBOL(alloc_disk_node);
1290
1291struct kobject *get_disk(struct gendisk *disk)
1292{
1293 struct module *owner;
1294 struct kobject *kobj;
1295
1296 if (!disk->fops)
1297 return NULL;
1298 owner = disk->fops->owner;
1299 if (owner && !try_module_get(owner))
1300 return NULL;
1301 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1302 if (kobj == NULL) {
1303 module_put(owner);
1304 return NULL;
1305 }
1306 return kobj;
1307
1308}
1309
1310EXPORT_SYMBOL(get_disk);
1311
1312void put_disk(struct gendisk *disk)
1313{
1314 if (disk)
1315 kobject_put(&disk_to_dev(disk)->kobj);
1316}
1317
1318EXPORT_SYMBOL(put_disk);
1319
1320static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1321{
1322 char event[] = "DISK_RO=1";
1323 char *envp[] = { event, NULL };
1324
1325 if (!ro)
1326 event[8] = '0';
1327 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1328}
1329
1330void set_device_ro(struct block_device *bdev, int flag)
1331{
1332 bdev->bd_part->policy = flag;
1333}
1334
1335EXPORT_SYMBOL(set_device_ro);
1336
1337void set_disk_ro(struct gendisk *disk, int flag)
1338{
1339 struct disk_part_iter piter;
1340 struct hd_struct *part;
1341
1342 if (disk->part0.policy != flag) {
1343 set_disk_ro_uevent(disk, flag);
1344 disk->part0.policy = flag;
1345 }
1346
1347 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1348 while ((part = disk_part_iter_next(&piter)))
1349 part->policy = flag;
1350 disk_part_iter_exit(&piter);
1351}
1352
1353EXPORT_SYMBOL(set_disk_ro);
1354
1355int bdev_read_only(struct block_device *bdev)
1356{
1357 if (!bdev)
1358 return 0;
1359 return bdev->bd_part->policy;
1360}
1361
1362EXPORT_SYMBOL(bdev_read_only);
1363
1364int invalidate_partition(struct gendisk *disk, int partno)
1365{
1366 int res = 0;
1367 struct block_device *bdev = bdget_disk(disk, partno);
1368 if (bdev) {
1369 fsync_bdev(bdev);
1370 res = __invalidate_device(bdev, true);
1371 bdput(bdev);
1372 }
1373 return res;
1374}
1375
1376EXPORT_SYMBOL(invalidate_partition);
1377
1378/*
1379 * Disk events - monitor disk events like media change and eject request.
1380 */
1381struct disk_events {
1382 struct list_head node; /* all disk_event's */
1383 struct gendisk *disk; /* the associated disk */
1384 spinlock_t lock;
1385
1386 struct mutex block_mutex; /* protects blocking */
1387 int block; /* event blocking depth */
1388 unsigned int pending; /* events already sent out */
1389 unsigned int clearing; /* events being cleared */
1390
1391 long poll_msecs; /* interval, -1 for default */
1392 struct delayed_work dwork;
1393};
1394
1395static const char *disk_events_strs[] = {
1396 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1397 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1398};
1399
1400static char *disk_uevents[] = {
1401 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1402 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1403};
1404
1405/* list of all disk_events */
1406static DEFINE_MUTEX(disk_events_mutex);
1407static LIST_HEAD(disk_events);
1408
1409/* disable in-kernel polling by default */
1410static unsigned long disk_events_dfl_poll_msecs = 0;
1411
1412static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1413{
1414 struct disk_events *ev = disk->ev;
1415 long intv_msecs = 0;
1416
1417 /*
1418 * If device-specific poll interval is set, always use it. If
1419 * the default is being used, poll iff there are events which
1420 * can't be monitored asynchronously.
1421 */
1422 if (ev->poll_msecs >= 0)
1423 intv_msecs = ev->poll_msecs;
1424 else if (disk->events & ~disk->async_events)
1425 intv_msecs = disk_events_dfl_poll_msecs;
1426
1427 return msecs_to_jiffies(intv_msecs);
1428}
1429
1430/**
1431 * disk_block_events - block and flush disk event checking
1432 * @disk: disk to block events for
1433 *
1434 * On return from this function, it is guaranteed that event checking
1435 * isn't in progress and won't happen until unblocked by
1436 * disk_unblock_events(). Events blocking is counted and the actual
1437 * unblocking happens after the matching number of unblocks are done.
1438 *
1439 * Note that this intentionally does not block event checking from
1440 * disk_clear_events().
1441 *
1442 * CONTEXT:
1443 * Might sleep.
1444 */
1445void disk_block_events(struct gendisk *disk)
1446{
1447 struct disk_events *ev = disk->ev;
1448 unsigned long flags;
1449 bool cancel;
1450
1451 if (!ev)
1452 return;
1453
1454 /*
1455 * Outer mutex ensures that the first blocker completes canceling
1456 * the event work before further blockers are allowed to finish.
1457 */
1458 mutex_lock(&ev->block_mutex);
1459
1460 spin_lock_irqsave(&ev->lock, flags);
1461 cancel = !ev->block++;
1462 spin_unlock_irqrestore(&ev->lock, flags);
1463
1464 if (cancel)
1465 cancel_delayed_work_sync(&disk->ev->dwork);
1466
1467 mutex_unlock(&ev->block_mutex);
1468}
1469
1470static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1471{
1472 struct disk_events *ev = disk->ev;
1473 unsigned long intv;
1474 unsigned long flags;
1475
1476 spin_lock_irqsave(&ev->lock, flags);
1477
1478 if (WARN_ON_ONCE(ev->block <= 0))
1479 goto out_unlock;
1480
1481 if (--ev->block)
1482 goto out_unlock;
1483
1484 /*
1485 * Not exactly a latency critical operation, set poll timer
1486 * slack to 25% and kick event check.
1487 */
1488 intv = disk_events_poll_jiffies(disk);
1489 set_timer_slack(&ev->dwork.timer, intv / 4);
1490 if (check_now)
1491 queue_delayed_work(system_freezable_power_efficient_wq,
1492 &ev->dwork, 0);
1493 else if (intv)
1494 queue_delayed_work(system_freezable_power_efficient_wq,
1495 &ev->dwork, intv);
1496out_unlock:
1497 spin_unlock_irqrestore(&ev->lock, flags);
1498}
1499
1500/**
1501 * disk_unblock_events - unblock disk event checking
1502 * @disk: disk to unblock events for
1503 *
1504 * Undo disk_block_events(). When the block count reaches zero, it
1505 * starts events polling if configured.
1506 *
1507 * CONTEXT:
1508 * Don't care. Safe to call from irq context.
1509 */
1510void disk_unblock_events(struct gendisk *disk)
1511{
1512 if (disk->ev)
1513 __disk_unblock_events(disk, false);
1514}
1515
1516/**
1517 * disk_flush_events - schedule immediate event checking and flushing
1518 * @disk: disk to check and flush events for
1519 * @mask: events to flush
1520 *
1521 * Schedule immediate event checking on @disk if not blocked. Events in
1522 * @mask are scheduled to be cleared from the driver. Note that this
1523 * doesn't clear the events from @disk->ev.
1524 *
1525 * CONTEXT:
1526 * If @mask is non-zero must be called with bdev->bd_mutex held.
1527 */
1528void disk_flush_events(struct gendisk *disk, unsigned int mask)
1529{
1530 struct disk_events *ev = disk->ev;
1531
1532 if (!ev)
1533 return;
1534
1535 spin_lock_irq(&ev->lock);
1536 ev->clearing |= mask;
1537 if (!ev->block)
1538 mod_delayed_work(system_freezable_power_efficient_wq,
1539 &ev->dwork, 0);
1540 spin_unlock_irq(&ev->lock);
1541}
1542
1543/**
1544 * disk_clear_events - synchronously check, clear and return pending events
1545 * @disk: disk to fetch and clear events from
1546 * @mask: mask of events to be fetched and clearted
1547 *
1548 * Disk events are synchronously checked and pending events in @mask
1549 * are cleared and returned. This ignores the block count.
1550 *
1551 * CONTEXT:
1552 * Might sleep.
1553 */
1554unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1555{
1556 const struct block_device_operations *bdops = disk->fops;
1557 struct disk_events *ev = disk->ev;
1558 unsigned int pending;
1559 unsigned int clearing = mask;
1560
1561 if (!ev) {
1562 /* for drivers still using the old ->media_changed method */
1563 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1564 bdops->media_changed && bdops->media_changed(disk))
1565 return DISK_EVENT_MEDIA_CHANGE;
1566 return 0;
1567 }
1568
1569 disk_block_events(disk);
1570
1571 /*
1572 * store the union of mask and ev->clearing on the stack so that the
1573 * race with disk_flush_events does not cause ambiguity (ev->clearing
1574 * can still be modified even if events are blocked).
1575 */
1576 spin_lock_irq(&ev->lock);
1577 clearing |= ev->clearing;
1578 ev->clearing = 0;
1579 spin_unlock_irq(&ev->lock);
1580
1581 disk_check_events(ev, &clearing);
1582 /*
1583 * if ev->clearing is not 0, the disk_flush_events got called in the
1584 * middle of this function, so we want to run the workfn without delay.
1585 */
1586 __disk_unblock_events(disk, ev->clearing ? true : false);
1587
1588 /* then, fetch and clear pending events */
1589 spin_lock_irq(&ev->lock);
1590 pending = ev->pending & mask;
1591 ev->pending &= ~mask;
1592 spin_unlock_irq(&ev->lock);
1593 WARN_ON_ONCE(clearing & mask);
1594
1595 return pending;
1596}
1597
1598/*
1599 * Separate this part out so that a different pointer for clearing_ptr can be
1600 * passed in for disk_clear_events.
1601 */
1602static void disk_events_workfn(struct work_struct *work)
1603{
1604 struct delayed_work *dwork = to_delayed_work(work);
1605 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1606
1607 disk_check_events(ev, &ev->clearing);
1608}
1609
1610static void disk_check_events(struct disk_events *ev,
1611 unsigned int *clearing_ptr)
1612{
1613 struct gendisk *disk = ev->disk;
1614 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1615 unsigned int clearing = *clearing_ptr;
1616 unsigned int events;
1617 unsigned long intv;
1618 int nr_events = 0, i;
1619
1620 /* check events */
1621 events = disk->fops->check_events(disk, clearing);
1622
1623 /* accumulate pending events and schedule next poll if necessary */
1624 spin_lock_irq(&ev->lock);
1625
1626 events &= ~ev->pending;
1627 ev->pending |= events;
1628 *clearing_ptr &= ~clearing;
1629
1630 intv = disk_events_poll_jiffies(disk);
1631 if (!ev->block && intv)
1632 queue_delayed_work(system_freezable_power_efficient_wq,
1633 &ev->dwork, intv);
1634
1635 spin_unlock_irq(&ev->lock);
1636
1637 /*
1638 * Tell userland about new events. Only the events listed in
1639 * @disk->events are reported. Unlisted events are processed the
1640 * same internally but never get reported to userland.
1641 */
1642 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1643 if (events & disk->events & (1 << i))
1644 envp[nr_events++] = disk_uevents[i];
1645
1646 if (nr_events)
1647 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1648}
1649
1650/*
1651 * A disk events enabled device has the following sysfs nodes under
1652 * its /sys/block/X/ directory.
1653 *
1654 * events : list of all supported events
1655 * events_async : list of events which can be detected w/o polling
1656 * events_poll_msecs : polling interval, 0: disable, -1: system default
1657 */
1658static ssize_t __disk_events_show(unsigned int events, char *buf)
1659{
1660 const char *delim = "";
1661 ssize_t pos = 0;
1662 int i;
1663
1664 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1665 if (events & (1 << i)) {
1666 pos += sprintf(buf + pos, "%s%s",
1667 delim, disk_events_strs[i]);
1668 delim = " ";
1669 }
1670 if (pos)
1671 pos += sprintf(buf + pos, "\n");
1672 return pos;
1673}
1674
1675static ssize_t disk_events_show(struct device *dev,
1676 struct device_attribute *attr, char *buf)
1677{
1678 struct gendisk *disk = dev_to_disk(dev);
1679
1680 return __disk_events_show(disk->events, buf);
1681}
1682
1683static ssize_t disk_events_async_show(struct device *dev,
1684 struct device_attribute *attr, char *buf)
1685{
1686 struct gendisk *disk = dev_to_disk(dev);
1687
1688 return __disk_events_show(disk->async_events, buf);
1689}
1690
1691static ssize_t disk_events_poll_msecs_show(struct device *dev,
1692 struct device_attribute *attr,
1693 char *buf)
1694{
1695 struct gendisk *disk = dev_to_disk(dev);
1696
1697 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1698}
1699
1700static ssize_t disk_events_poll_msecs_store(struct device *dev,
1701 struct device_attribute *attr,
1702 const char *buf, size_t count)
1703{
1704 struct gendisk *disk = dev_to_disk(dev);
1705 long intv;
1706
1707 if (!count || !sscanf(buf, "%ld", &intv))
1708 return -EINVAL;
1709
1710 if (intv < 0 && intv != -1)
1711 return -EINVAL;
1712
1713 disk_block_events(disk);
1714 disk->ev->poll_msecs = intv;
1715 __disk_unblock_events(disk, true);
1716
1717 return count;
1718}
1719
1720static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1721static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1722static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1723 disk_events_poll_msecs_show,
1724 disk_events_poll_msecs_store);
1725
1726static const struct attribute *disk_events_attrs[] = {
1727 &dev_attr_events.attr,
1728 &dev_attr_events_async.attr,
1729 &dev_attr_events_poll_msecs.attr,
1730 NULL,
1731};
1732
1733/*
1734 * The default polling interval can be specified by the kernel
1735 * parameter block.events_dfl_poll_msecs which defaults to 0
1736 * (disable). This can also be modified runtime by writing to
1737 * /sys/module/block/events_dfl_poll_msecs.
1738 */
1739static int disk_events_set_dfl_poll_msecs(const char *val,
1740 const struct kernel_param *kp)
1741{
1742 struct disk_events *ev;
1743 int ret;
1744
1745 ret = param_set_ulong(val, kp);
1746 if (ret < 0)
1747 return ret;
1748
1749 mutex_lock(&disk_events_mutex);
1750
1751 list_for_each_entry(ev, &disk_events, node)
1752 disk_flush_events(ev->disk, 0);
1753
1754 mutex_unlock(&disk_events_mutex);
1755
1756 return 0;
1757}
1758
1759static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1760 .set = disk_events_set_dfl_poll_msecs,
1761 .get = param_get_ulong,
1762};
1763
1764#undef MODULE_PARAM_PREFIX
1765#define MODULE_PARAM_PREFIX "block."
1766
1767module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1768 &disk_events_dfl_poll_msecs, 0644);
1769
1770/*
1771 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1772 */
1773static void disk_alloc_events(struct gendisk *disk)
1774{
1775 struct disk_events *ev;
1776
1777 if (!disk->fops->check_events)
1778 return;
1779
1780 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1781 if (!ev) {
1782 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1783 return;
1784 }
1785
1786 INIT_LIST_HEAD(&ev->node);
1787 ev->disk = disk;
1788 spin_lock_init(&ev->lock);
1789 mutex_init(&ev->block_mutex);
1790 ev->block = 1;
1791 ev->poll_msecs = -1;
1792 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1793
1794 disk->ev = ev;
1795}
1796
1797static void disk_add_events(struct gendisk *disk)
1798{
1799 if (!disk->ev)
1800 return;
1801
1802 /* FIXME: error handling */
1803 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1804 pr_warn("%s: failed to create sysfs files for events\n",
1805 disk->disk_name);
1806
1807 mutex_lock(&disk_events_mutex);
1808 list_add_tail(&disk->ev->node, &disk_events);
1809 mutex_unlock(&disk_events_mutex);
1810
1811 /*
1812 * Block count is initialized to 1 and the following initial
1813 * unblock kicks it into action.
1814 */
1815 __disk_unblock_events(disk, true);
1816}
1817
1818static void disk_del_events(struct gendisk *disk)
1819{
1820 if (!disk->ev)
1821 return;
1822
1823 disk_block_events(disk);
1824
1825 mutex_lock(&disk_events_mutex);
1826 list_del_init(&disk->ev->node);
1827 mutex_unlock(&disk_events_mutex);
1828
1829 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1830}
1831
1832static void disk_release_events(struct gendisk *disk)
1833{
1834 /* the block count should be 1 from disk_del_events() */
1835 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1836 kfree(disk->ev);
1837}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * gendisk handling
4 */
5
6#include <linux/module.h>
7#include <linux/ctype.h>
8#include <linux/fs.h>
9#include <linux/genhd.h>
10#include <linux/kdev_t.h>
11#include <linux/kernel.h>
12#include <linux/blkdev.h>
13#include <linux/backing-dev.h>
14#include <linux/init.h>
15#include <linux/spinlock.h>
16#include <linux/proc_fs.h>
17#include <linux/seq_file.h>
18#include <linux/slab.h>
19#include <linux/kmod.h>
20#include <linux/kobj_map.h>
21#include <linux/mutex.h>
22#include <linux/idr.h>
23#include <linux/log2.h>
24#include <linux/pm_runtime.h>
25#include <linux/badblocks.h>
26
27#include "blk.h"
28
29static DEFINE_MUTEX(block_class_lock);
30static struct kobject *block_depr;
31
32/* for extended dynamic devt allocation, currently only one major is used */
33#define NR_EXT_DEVT (1 << MINORBITS)
34
35/* For extended devt allocation. ext_devt_lock prevents look up
36 * results from going away underneath its user.
37 */
38static DEFINE_SPINLOCK(ext_devt_lock);
39static DEFINE_IDR(ext_devt_idr);
40
41static void disk_check_events(struct disk_events *ev,
42 unsigned int *clearing_ptr);
43static void disk_alloc_events(struct gendisk *disk);
44static void disk_add_events(struct gendisk *disk);
45static void disk_del_events(struct gendisk *disk);
46static void disk_release_events(struct gendisk *disk);
47
48/*
49 * Set disk capacity and notify if the size is not currently
50 * zero and will not be set to zero
51 */
52void set_capacity_revalidate_and_notify(struct gendisk *disk, sector_t size,
53 bool revalidate)
54{
55 sector_t capacity = get_capacity(disk);
56
57 set_capacity(disk, size);
58
59 if (revalidate)
60 revalidate_disk(disk);
61
62 if (capacity != size && capacity != 0 && size != 0) {
63 char *envp[] = { "RESIZE=1", NULL };
64
65 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
66 }
67}
68
69EXPORT_SYMBOL_GPL(set_capacity_revalidate_and_notify);
70
71/*
72 * Format the device name of the indicated disk into the supplied buffer and
73 * return a pointer to that same buffer for convenience.
74 */
75char *disk_name(struct gendisk *hd, int partno, char *buf)
76{
77 if (!partno)
78 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
79 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
80 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
81 else
82 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
83
84 return buf;
85}
86
87const char *bdevname(struct block_device *bdev, char *buf)
88{
89 return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
90}
91EXPORT_SYMBOL(bdevname);
92
93static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
94{
95 int cpu;
96
97 memset(stat, 0, sizeof(struct disk_stats));
98 for_each_possible_cpu(cpu) {
99 struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
100 int group;
101
102 for (group = 0; group < NR_STAT_GROUPS; group++) {
103 stat->nsecs[group] += ptr->nsecs[group];
104 stat->sectors[group] += ptr->sectors[group];
105 stat->ios[group] += ptr->ios[group];
106 stat->merges[group] += ptr->merges[group];
107 }
108
109 stat->io_ticks += ptr->io_ticks;
110 }
111}
112
113static unsigned int part_in_flight(struct request_queue *q,
114 struct hd_struct *part)
115{
116 unsigned int inflight = 0;
117 int cpu;
118
119 for_each_possible_cpu(cpu) {
120 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
121 part_stat_local_read_cpu(part, in_flight[1], cpu);
122 }
123 if ((int)inflight < 0)
124 inflight = 0;
125
126 return inflight;
127}
128
129static void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
130 unsigned int inflight[2])
131{
132 int cpu;
133
134 inflight[0] = 0;
135 inflight[1] = 0;
136 for_each_possible_cpu(cpu) {
137 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
138 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
139 }
140 if ((int)inflight[0] < 0)
141 inflight[0] = 0;
142 if ((int)inflight[1] < 0)
143 inflight[1] = 0;
144}
145
146struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
147{
148 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
149
150 if (unlikely(partno < 0 || partno >= ptbl->len))
151 return NULL;
152 return rcu_dereference(ptbl->part[partno]);
153}
154
155/**
156 * disk_get_part - get partition
157 * @disk: disk to look partition from
158 * @partno: partition number
159 *
160 * Look for partition @partno from @disk. If found, increment
161 * reference count and return it.
162 *
163 * CONTEXT:
164 * Don't care.
165 *
166 * RETURNS:
167 * Pointer to the found partition on success, NULL if not found.
168 */
169struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
170{
171 struct hd_struct *part;
172
173 rcu_read_lock();
174 part = __disk_get_part(disk, partno);
175 if (part)
176 get_device(part_to_dev(part));
177 rcu_read_unlock();
178
179 return part;
180}
181
182/**
183 * disk_part_iter_init - initialize partition iterator
184 * @piter: iterator to initialize
185 * @disk: disk to iterate over
186 * @flags: DISK_PITER_* flags
187 *
188 * Initialize @piter so that it iterates over partitions of @disk.
189 *
190 * CONTEXT:
191 * Don't care.
192 */
193void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
194 unsigned int flags)
195{
196 struct disk_part_tbl *ptbl;
197
198 rcu_read_lock();
199 ptbl = rcu_dereference(disk->part_tbl);
200
201 piter->disk = disk;
202 piter->part = NULL;
203
204 if (flags & DISK_PITER_REVERSE)
205 piter->idx = ptbl->len - 1;
206 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
207 piter->idx = 0;
208 else
209 piter->idx = 1;
210
211 piter->flags = flags;
212
213 rcu_read_unlock();
214}
215EXPORT_SYMBOL_GPL(disk_part_iter_init);
216
217/**
218 * disk_part_iter_next - proceed iterator to the next partition and return it
219 * @piter: iterator of interest
220 *
221 * Proceed @piter to the next partition and return it.
222 *
223 * CONTEXT:
224 * Don't care.
225 */
226struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
227{
228 struct disk_part_tbl *ptbl;
229 int inc, end;
230
231 /* put the last partition */
232 disk_put_part(piter->part);
233 piter->part = NULL;
234
235 /* get part_tbl */
236 rcu_read_lock();
237 ptbl = rcu_dereference(piter->disk->part_tbl);
238
239 /* determine iteration parameters */
240 if (piter->flags & DISK_PITER_REVERSE) {
241 inc = -1;
242 if (piter->flags & (DISK_PITER_INCL_PART0 |
243 DISK_PITER_INCL_EMPTY_PART0))
244 end = -1;
245 else
246 end = 0;
247 } else {
248 inc = 1;
249 end = ptbl->len;
250 }
251
252 /* iterate to the next partition */
253 for (; piter->idx != end; piter->idx += inc) {
254 struct hd_struct *part;
255
256 part = rcu_dereference(ptbl->part[piter->idx]);
257 if (!part)
258 continue;
259 if (!part_nr_sects_read(part) &&
260 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
261 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
262 piter->idx == 0))
263 continue;
264
265 get_device(part_to_dev(part));
266 piter->part = part;
267 piter->idx += inc;
268 break;
269 }
270
271 rcu_read_unlock();
272
273 return piter->part;
274}
275EXPORT_SYMBOL_GPL(disk_part_iter_next);
276
277/**
278 * disk_part_iter_exit - finish up partition iteration
279 * @piter: iter of interest
280 *
281 * Called when iteration is over. Cleans up @piter.
282 *
283 * CONTEXT:
284 * Don't care.
285 */
286void disk_part_iter_exit(struct disk_part_iter *piter)
287{
288 disk_put_part(piter->part);
289 piter->part = NULL;
290}
291EXPORT_SYMBOL_GPL(disk_part_iter_exit);
292
293static inline int sector_in_part(struct hd_struct *part, sector_t sector)
294{
295 return part->start_sect <= sector &&
296 sector < part->start_sect + part_nr_sects_read(part);
297}
298
299/**
300 * disk_map_sector_rcu - map sector to partition
301 * @disk: gendisk of interest
302 * @sector: sector to map
303 *
304 * Find out which partition @sector maps to on @disk. This is
305 * primarily used for stats accounting.
306 *
307 * CONTEXT:
308 * RCU read locked. The returned partition pointer is always valid
309 * because its refcount is grabbed except for part0, which lifetime
310 * is same with the disk.
311 *
312 * RETURNS:
313 * Found partition on success, part0 is returned if no partition matches
314 * or the matched partition is being deleted.
315 */
316struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
317{
318 struct disk_part_tbl *ptbl;
319 struct hd_struct *part;
320 int i;
321
322 rcu_read_lock();
323 ptbl = rcu_dereference(disk->part_tbl);
324
325 part = rcu_dereference(ptbl->last_lookup);
326 if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
327 goto out_unlock;
328
329 for (i = 1; i < ptbl->len; i++) {
330 part = rcu_dereference(ptbl->part[i]);
331
332 if (part && sector_in_part(part, sector)) {
333 /*
334 * only live partition can be cached for lookup,
335 * so use-after-free on cached & deleting partition
336 * can be avoided
337 */
338 if (!hd_struct_try_get(part))
339 break;
340 rcu_assign_pointer(ptbl->last_lookup, part);
341 goto out_unlock;
342 }
343 }
344
345 part = &disk->part0;
346out_unlock:
347 rcu_read_unlock();
348 return part;
349}
350
351/**
352 * disk_has_partitions
353 * @disk: gendisk of interest
354 *
355 * Walk through the partition table and check if valid partition exists.
356 *
357 * CONTEXT:
358 * Don't care.
359 *
360 * RETURNS:
361 * True if the gendisk has at least one valid non-zero size partition.
362 * Otherwise false.
363 */
364bool disk_has_partitions(struct gendisk *disk)
365{
366 struct disk_part_tbl *ptbl;
367 int i;
368 bool ret = false;
369
370 rcu_read_lock();
371 ptbl = rcu_dereference(disk->part_tbl);
372
373 /* Iterate partitions skipping the whole device at index 0 */
374 for (i = 1; i < ptbl->len; i++) {
375 if (rcu_dereference(ptbl->part[i])) {
376 ret = true;
377 break;
378 }
379 }
380
381 rcu_read_unlock();
382
383 return ret;
384}
385EXPORT_SYMBOL_GPL(disk_has_partitions);
386
387/*
388 * Can be deleted altogether. Later.
389 *
390 */
391#define BLKDEV_MAJOR_HASH_SIZE 255
392static struct blk_major_name {
393 struct blk_major_name *next;
394 int major;
395 char name[16];
396} *major_names[BLKDEV_MAJOR_HASH_SIZE];
397
398/* index in the above - for now: assume no multimajor ranges */
399static inline int major_to_index(unsigned major)
400{
401 return major % BLKDEV_MAJOR_HASH_SIZE;
402}
403
404#ifdef CONFIG_PROC_FS
405void blkdev_show(struct seq_file *seqf, off_t offset)
406{
407 struct blk_major_name *dp;
408
409 mutex_lock(&block_class_lock);
410 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
411 if (dp->major == offset)
412 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
413 mutex_unlock(&block_class_lock);
414}
415#endif /* CONFIG_PROC_FS */
416
417/**
418 * register_blkdev - register a new block device
419 *
420 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
421 * @major = 0, try to allocate any unused major number.
422 * @name: the name of the new block device as a zero terminated string
423 *
424 * The @name must be unique within the system.
425 *
426 * The return value depends on the @major input parameter:
427 *
428 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
429 * then the function returns zero on success, or a negative error code
430 * - if any unused major number was requested with @major = 0 parameter
431 * then the return value is the allocated major number in range
432 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
433 *
434 * See Documentation/admin-guide/devices.txt for the list of allocated
435 * major numbers.
436 */
437int register_blkdev(unsigned int major, const char *name)
438{
439 struct blk_major_name **n, *p;
440 int index, ret = 0;
441
442 mutex_lock(&block_class_lock);
443
444 /* temporary */
445 if (major == 0) {
446 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
447 if (major_names[index] == NULL)
448 break;
449 }
450
451 if (index == 0) {
452 printk("%s: failed to get major for %s\n",
453 __func__, name);
454 ret = -EBUSY;
455 goto out;
456 }
457 major = index;
458 ret = major;
459 }
460
461 if (major >= BLKDEV_MAJOR_MAX) {
462 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
463 __func__, major, BLKDEV_MAJOR_MAX-1, name);
464
465 ret = -EINVAL;
466 goto out;
467 }
468
469 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
470 if (p == NULL) {
471 ret = -ENOMEM;
472 goto out;
473 }
474
475 p->major = major;
476 strlcpy(p->name, name, sizeof(p->name));
477 p->next = NULL;
478 index = major_to_index(major);
479
480 for (n = &major_names[index]; *n; n = &(*n)->next) {
481 if ((*n)->major == major)
482 break;
483 }
484 if (!*n)
485 *n = p;
486 else
487 ret = -EBUSY;
488
489 if (ret < 0) {
490 printk("register_blkdev: cannot get major %u for %s\n",
491 major, name);
492 kfree(p);
493 }
494out:
495 mutex_unlock(&block_class_lock);
496 return ret;
497}
498
499EXPORT_SYMBOL(register_blkdev);
500
501void unregister_blkdev(unsigned int major, const char *name)
502{
503 struct blk_major_name **n;
504 struct blk_major_name *p = NULL;
505 int index = major_to_index(major);
506
507 mutex_lock(&block_class_lock);
508 for (n = &major_names[index]; *n; n = &(*n)->next)
509 if ((*n)->major == major)
510 break;
511 if (!*n || strcmp((*n)->name, name)) {
512 WARN_ON(1);
513 } else {
514 p = *n;
515 *n = p->next;
516 }
517 mutex_unlock(&block_class_lock);
518 kfree(p);
519}
520
521EXPORT_SYMBOL(unregister_blkdev);
522
523static struct kobj_map *bdev_map;
524
525/**
526 * blk_mangle_minor - scatter minor numbers apart
527 * @minor: minor number to mangle
528 *
529 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
530 * is enabled. Mangling twice gives the original value.
531 *
532 * RETURNS:
533 * Mangled value.
534 *
535 * CONTEXT:
536 * Don't care.
537 */
538static int blk_mangle_minor(int minor)
539{
540#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
541 int i;
542
543 for (i = 0; i < MINORBITS / 2; i++) {
544 int low = minor & (1 << i);
545 int high = minor & (1 << (MINORBITS - 1 - i));
546 int distance = MINORBITS - 1 - 2 * i;
547
548 minor ^= low | high; /* clear both bits */
549 low <<= distance; /* swap the positions */
550 high >>= distance;
551 minor |= low | high; /* and set */
552 }
553#endif
554 return minor;
555}
556
557/**
558 * blk_alloc_devt - allocate a dev_t for a partition
559 * @part: partition to allocate dev_t for
560 * @devt: out parameter for resulting dev_t
561 *
562 * Allocate a dev_t for block device.
563 *
564 * RETURNS:
565 * 0 on success, allocated dev_t is returned in *@devt. -errno on
566 * failure.
567 *
568 * CONTEXT:
569 * Might sleep.
570 */
571int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
572{
573 struct gendisk *disk = part_to_disk(part);
574 int idx;
575
576 /* in consecutive minor range? */
577 if (part->partno < disk->minors) {
578 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
579 return 0;
580 }
581
582 /* allocate ext devt */
583 idr_preload(GFP_KERNEL);
584
585 spin_lock_bh(&ext_devt_lock);
586 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
587 spin_unlock_bh(&ext_devt_lock);
588
589 idr_preload_end();
590 if (idx < 0)
591 return idx == -ENOSPC ? -EBUSY : idx;
592
593 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
594 return 0;
595}
596
597/**
598 * blk_free_devt - free a dev_t
599 * @devt: dev_t to free
600 *
601 * Free @devt which was allocated using blk_alloc_devt().
602 *
603 * CONTEXT:
604 * Might sleep.
605 */
606void blk_free_devt(dev_t devt)
607{
608 if (devt == MKDEV(0, 0))
609 return;
610
611 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
612 spin_lock_bh(&ext_devt_lock);
613 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
614 spin_unlock_bh(&ext_devt_lock);
615 }
616}
617
618/*
619 * We invalidate devt by assigning NULL pointer for devt in idr.
620 */
621void blk_invalidate_devt(dev_t devt)
622{
623 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
624 spin_lock_bh(&ext_devt_lock);
625 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
626 spin_unlock_bh(&ext_devt_lock);
627 }
628}
629
630static char *bdevt_str(dev_t devt, char *buf)
631{
632 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
633 char tbuf[BDEVT_SIZE];
634 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
635 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
636 } else
637 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
638
639 return buf;
640}
641
642/*
643 * Register device numbers dev..(dev+range-1)
644 * range must be nonzero
645 * The hash chain is sorted on range, so that subranges can override.
646 */
647void blk_register_region(dev_t devt, unsigned long range, struct module *module,
648 struct kobject *(*probe)(dev_t, int *, void *),
649 int (*lock)(dev_t, void *), void *data)
650{
651 kobj_map(bdev_map, devt, range, module, probe, lock, data);
652}
653
654EXPORT_SYMBOL(blk_register_region);
655
656void blk_unregister_region(dev_t devt, unsigned long range)
657{
658 kobj_unmap(bdev_map, devt, range);
659}
660
661EXPORT_SYMBOL(blk_unregister_region);
662
663static struct kobject *exact_match(dev_t devt, int *partno, void *data)
664{
665 struct gendisk *p = data;
666
667 return &disk_to_dev(p)->kobj;
668}
669
670static int exact_lock(dev_t devt, void *data)
671{
672 struct gendisk *p = data;
673
674 if (!get_disk_and_module(p))
675 return -1;
676 return 0;
677}
678
679static void register_disk(struct device *parent, struct gendisk *disk,
680 const struct attribute_group **groups)
681{
682 struct device *ddev = disk_to_dev(disk);
683 struct block_device *bdev;
684 struct disk_part_iter piter;
685 struct hd_struct *part;
686 int err;
687
688 ddev->parent = parent;
689
690 dev_set_name(ddev, "%s", disk->disk_name);
691
692 /* delay uevents, until we scanned partition table */
693 dev_set_uevent_suppress(ddev, 1);
694
695 if (groups) {
696 WARN_ON(ddev->groups);
697 ddev->groups = groups;
698 }
699 if (device_add(ddev))
700 return;
701 if (!sysfs_deprecated) {
702 err = sysfs_create_link(block_depr, &ddev->kobj,
703 kobject_name(&ddev->kobj));
704 if (err) {
705 device_del(ddev);
706 return;
707 }
708 }
709
710 /*
711 * avoid probable deadlock caused by allocating memory with
712 * GFP_KERNEL in runtime_resume callback of its all ancestor
713 * devices
714 */
715 pm_runtime_set_memalloc_noio(ddev, true);
716
717 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
718 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
719
720 if (disk->flags & GENHD_FL_HIDDEN) {
721 dev_set_uevent_suppress(ddev, 0);
722 return;
723 }
724
725 /* No minors to use for partitions */
726 if (!disk_part_scan_enabled(disk))
727 goto exit;
728
729 /* No such device (e.g., media were just removed) */
730 if (!get_capacity(disk))
731 goto exit;
732
733 bdev = bdget_disk(disk, 0);
734 if (!bdev)
735 goto exit;
736
737 bdev->bd_invalidated = 1;
738 err = blkdev_get(bdev, FMODE_READ, NULL);
739 if (err < 0)
740 goto exit;
741 blkdev_put(bdev, FMODE_READ);
742
743exit:
744 /* announce disk after possible partitions are created */
745 dev_set_uevent_suppress(ddev, 0);
746 kobject_uevent(&ddev->kobj, KOBJ_ADD);
747
748 /* announce possible partitions */
749 disk_part_iter_init(&piter, disk, 0);
750 while ((part = disk_part_iter_next(&piter)))
751 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
752 disk_part_iter_exit(&piter);
753
754 if (disk->queue->backing_dev_info->dev) {
755 err = sysfs_create_link(&ddev->kobj,
756 &disk->queue->backing_dev_info->dev->kobj,
757 "bdi");
758 WARN_ON(err);
759 }
760}
761
762/**
763 * __device_add_disk - add disk information to kernel list
764 * @parent: parent device for the disk
765 * @disk: per-device partitioning information
766 * @groups: Additional per-device sysfs groups
767 * @register_queue: register the queue if set to true
768 *
769 * This function registers the partitioning information in @disk
770 * with the kernel.
771 *
772 * FIXME: error handling
773 */
774static void __device_add_disk(struct device *parent, struct gendisk *disk,
775 const struct attribute_group **groups,
776 bool register_queue)
777{
778 dev_t devt;
779 int retval;
780
781 /*
782 * The disk queue should now be all set with enough information about
783 * the device for the elevator code to pick an adequate default
784 * elevator if one is needed, that is, for devices requesting queue
785 * registration.
786 */
787 if (register_queue)
788 elevator_init_mq(disk->queue);
789
790 /* minors == 0 indicates to use ext devt from part0 and should
791 * be accompanied with EXT_DEVT flag. Make sure all
792 * parameters make sense.
793 */
794 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
795 WARN_ON(!disk->minors &&
796 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
797
798 disk->flags |= GENHD_FL_UP;
799
800 retval = blk_alloc_devt(&disk->part0, &devt);
801 if (retval) {
802 WARN_ON(1);
803 return;
804 }
805 disk->major = MAJOR(devt);
806 disk->first_minor = MINOR(devt);
807
808 disk_alloc_events(disk);
809
810 if (disk->flags & GENHD_FL_HIDDEN) {
811 /*
812 * Don't let hidden disks show up in /proc/partitions,
813 * and don't bother scanning for partitions either.
814 */
815 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
816 disk->flags |= GENHD_FL_NO_PART_SCAN;
817 } else {
818 struct backing_dev_info *bdi = disk->queue->backing_dev_info;
819 struct device *dev = disk_to_dev(disk);
820 int ret;
821
822 /* Register BDI before referencing it from bdev */
823 dev->devt = devt;
824 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
825 WARN_ON(ret);
826 bdi_set_owner(bdi, dev);
827 blk_register_region(disk_devt(disk), disk->minors, NULL,
828 exact_match, exact_lock, disk);
829 }
830 register_disk(parent, disk, groups);
831 if (register_queue)
832 blk_register_queue(disk);
833
834 /*
835 * Take an extra ref on queue which will be put on disk_release()
836 * so that it sticks around as long as @disk is there.
837 */
838 WARN_ON_ONCE(!blk_get_queue(disk->queue));
839
840 disk_add_events(disk);
841 blk_integrity_add(disk);
842}
843
844void device_add_disk(struct device *parent, struct gendisk *disk,
845 const struct attribute_group **groups)
846
847{
848 __device_add_disk(parent, disk, groups, true);
849}
850EXPORT_SYMBOL(device_add_disk);
851
852void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
853{
854 __device_add_disk(parent, disk, NULL, false);
855}
856EXPORT_SYMBOL(device_add_disk_no_queue_reg);
857
858static void invalidate_partition(struct gendisk *disk, int partno)
859{
860 struct block_device *bdev;
861
862 bdev = bdget_disk(disk, partno);
863 if (!bdev)
864 return;
865
866 fsync_bdev(bdev);
867 __invalidate_device(bdev, true);
868
869 /*
870 * Unhash the bdev inode for this device so that it gets evicted as soon
871 * as last inode reference is dropped.
872 */
873 remove_inode_hash(bdev->bd_inode);
874 bdput(bdev);
875}
876
877/**
878 * del_gendisk - remove the gendisk
879 * @disk: the struct gendisk to remove
880 *
881 * Removes the gendisk and all its associated resources. This deletes the
882 * partitions associated with the gendisk, and unregisters the associated
883 * request_queue.
884 *
885 * This is the counter to the respective __device_add_disk() call.
886 *
887 * The final removal of the struct gendisk happens when its refcount reaches 0
888 * with put_disk(), which should be called after del_gendisk(), if
889 * __device_add_disk() was used.
890 *
891 * Drivers exist which depend on the release of the gendisk to be synchronous,
892 * it should not be deferred.
893 *
894 * Context: can sleep
895 */
896void del_gendisk(struct gendisk *disk)
897{
898 struct disk_part_iter piter;
899 struct hd_struct *part;
900
901 might_sleep();
902
903 blk_integrity_del(disk);
904 disk_del_events(disk);
905
906 /*
907 * Block lookups of the disk until all bdevs are unhashed and the
908 * disk is marked as dead (GENHD_FL_UP cleared).
909 */
910 down_write(&disk->lookup_sem);
911 /* invalidate stuff */
912 disk_part_iter_init(&piter, disk,
913 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
914 while ((part = disk_part_iter_next(&piter))) {
915 invalidate_partition(disk, part->partno);
916 delete_partition(disk, part);
917 }
918 disk_part_iter_exit(&piter);
919
920 invalidate_partition(disk, 0);
921 set_capacity(disk, 0);
922 disk->flags &= ~GENHD_FL_UP;
923 up_write(&disk->lookup_sem);
924
925 if (!(disk->flags & GENHD_FL_HIDDEN))
926 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
927 if (disk->queue) {
928 /*
929 * Unregister bdi before releasing device numbers (as they can
930 * get reused and we'd get clashes in sysfs).
931 */
932 if (!(disk->flags & GENHD_FL_HIDDEN))
933 bdi_unregister(disk->queue->backing_dev_info);
934 blk_unregister_queue(disk);
935 } else {
936 WARN_ON(1);
937 }
938
939 if (!(disk->flags & GENHD_FL_HIDDEN))
940 blk_unregister_region(disk_devt(disk), disk->minors);
941 /*
942 * Remove gendisk pointer from idr so that it cannot be looked up
943 * while RCU period before freeing gendisk is running to prevent
944 * use-after-free issues. Note that the device number stays
945 * "in-use" until we really free the gendisk.
946 */
947 blk_invalidate_devt(disk_devt(disk));
948
949 kobject_put(disk->part0.holder_dir);
950 kobject_put(disk->slave_dir);
951
952 part_stat_set_all(&disk->part0, 0);
953 disk->part0.stamp = 0;
954 if (!sysfs_deprecated)
955 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
956 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
957 device_del(disk_to_dev(disk));
958}
959EXPORT_SYMBOL(del_gendisk);
960
961/* sysfs access to bad-blocks list. */
962static ssize_t disk_badblocks_show(struct device *dev,
963 struct device_attribute *attr,
964 char *page)
965{
966 struct gendisk *disk = dev_to_disk(dev);
967
968 if (!disk->bb)
969 return sprintf(page, "\n");
970
971 return badblocks_show(disk->bb, page, 0);
972}
973
974static ssize_t disk_badblocks_store(struct device *dev,
975 struct device_attribute *attr,
976 const char *page, size_t len)
977{
978 struct gendisk *disk = dev_to_disk(dev);
979
980 if (!disk->bb)
981 return -ENXIO;
982
983 return badblocks_store(disk->bb, page, len, 0);
984}
985
986/**
987 * get_gendisk - get partitioning information for a given device
988 * @devt: device to get partitioning information for
989 * @partno: returned partition index
990 *
991 * This function gets the structure containing partitioning
992 * information for the given device @devt.
993 *
994 * Context: can sleep
995 */
996struct gendisk *get_gendisk(dev_t devt, int *partno)
997{
998 struct gendisk *disk = NULL;
999
1000 might_sleep();
1001
1002 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
1003 struct kobject *kobj;
1004
1005 kobj = kobj_lookup(bdev_map, devt, partno);
1006 if (kobj)
1007 disk = dev_to_disk(kobj_to_dev(kobj));
1008 } else {
1009 struct hd_struct *part;
1010
1011 spin_lock_bh(&ext_devt_lock);
1012 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
1013 if (part && get_disk_and_module(part_to_disk(part))) {
1014 *partno = part->partno;
1015 disk = part_to_disk(part);
1016 }
1017 spin_unlock_bh(&ext_devt_lock);
1018 }
1019
1020 if (!disk)
1021 return NULL;
1022
1023 /*
1024 * Synchronize with del_gendisk() to not return disk that is being
1025 * destroyed.
1026 */
1027 down_read(&disk->lookup_sem);
1028 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1029 !(disk->flags & GENHD_FL_UP))) {
1030 up_read(&disk->lookup_sem);
1031 put_disk_and_module(disk);
1032 disk = NULL;
1033 } else {
1034 up_read(&disk->lookup_sem);
1035 }
1036 return disk;
1037}
1038
1039/**
1040 * bdget_disk - do bdget() by gendisk and partition number
1041 * @disk: gendisk of interest
1042 * @partno: partition number
1043 *
1044 * Find partition @partno from @disk, do bdget() on it.
1045 *
1046 * CONTEXT:
1047 * Don't care.
1048 *
1049 * RETURNS:
1050 * Resulting block_device on success, NULL on failure.
1051 */
1052struct block_device *bdget_disk(struct gendisk *disk, int partno)
1053{
1054 struct hd_struct *part;
1055 struct block_device *bdev = NULL;
1056
1057 part = disk_get_part(disk, partno);
1058 if (part)
1059 bdev = bdget(part_devt(part));
1060 disk_put_part(part);
1061
1062 return bdev;
1063}
1064EXPORT_SYMBOL(bdget_disk);
1065
1066/*
1067 * print a full list of all partitions - intended for places where the root
1068 * filesystem can't be mounted and thus to give the victim some idea of what
1069 * went wrong
1070 */
1071void __init printk_all_partitions(void)
1072{
1073 struct class_dev_iter iter;
1074 struct device *dev;
1075
1076 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1077 while ((dev = class_dev_iter_next(&iter))) {
1078 struct gendisk *disk = dev_to_disk(dev);
1079 struct disk_part_iter piter;
1080 struct hd_struct *part;
1081 char name_buf[BDEVNAME_SIZE];
1082 char devt_buf[BDEVT_SIZE];
1083
1084 /*
1085 * Don't show empty devices or things that have been
1086 * suppressed
1087 */
1088 if (get_capacity(disk) == 0 ||
1089 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1090 continue;
1091
1092 /*
1093 * Note, unlike /proc/partitions, I am showing the
1094 * numbers in hex - the same format as the root=
1095 * option takes.
1096 */
1097 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1098 while ((part = disk_part_iter_next(&piter))) {
1099 bool is_part0 = part == &disk->part0;
1100
1101 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1102 bdevt_str(part_devt(part), devt_buf),
1103 (unsigned long long)part_nr_sects_read(part) >> 1
1104 , disk_name(disk, part->partno, name_buf),
1105 part->info ? part->info->uuid : "");
1106 if (is_part0) {
1107 if (dev->parent && dev->parent->driver)
1108 printk(" driver: %s\n",
1109 dev->parent->driver->name);
1110 else
1111 printk(" (driver?)\n");
1112 } else
1113 printk("\n");
1114 }
1115 disk_part_iter_exit(&piter);
1116 }
1117 class_dev_iter_exit(&iter);
1118}
1119
1120#ifdef CONFIG_PROC_FS
1121/* iterator */
1122static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1123{
1124 loff_t skip = *pos;
1125 struct class_dev_iter *iter;
1126 struct device *dev;
1127
1128 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1129 if (!iter)
1130 return ERR_PTR(-ENOMEM);
1131
1132 seqf->private = iter;
1133 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1134 do {
1135 dev = class_dev_iter_next(iter);
1136 if (!dev)
1137 return NULL;
1138 } while (skip--);
1139
1140 return dev_to_disk(dev);
1141}
1142
1143static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1144{
1145 struct device *dev;
1146
1147 (*pos)++;
1148 dev = class_dev_iter_next(seqf->private);
1149 if (dev)
1150 return dev_to_disk(dev);
1151
1152 return NULL;
1153}
1154
1155static void disk_seqf_stop(struct seq_file *seqf, void *v)
1156{
1157 struct class_dev_iter *iter = seqf->private;
1158
1159 /* stop is called even after start failed :-( */
1160 if (iter) {
1161 class_dev_iter_exit(iter);
1162 kfree(iter);
1163 seqf->private = NULL;
1164 }
1165}
1166
1167static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1168{
1169 void *p;
1170
1171 p = disk_seqf_start(seqf, pos);
1172 if (!IS_ERR_OR_NULL(p) && !*pos)
1173 seq_puts(seqf, "major minor #blocks name\n\n");
1174 return p;
1175}
1176
1177static int show_partition(struct seq_file *seqf, void *v)
1178{
1179 struct gendisk *sgp = v;
1180 struct disk_part_iter piter;
1181 struct hd_struct *part;
1182 char buf[BDEVNAME_SIZE];
1183
1184 /* Don't show non-partitionable removeable devices or empty devices */
1185 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1186 (sgp->flags & GENHD_FL_REMOVABLE)))
1187 return 0;
1188 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1189 return 0;
1190
1191 /* show the full disk and all non-0 size partitions of it */
1192 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1193 while ((part = disk_part_iter_next(&piter)))
1194 seq_printf(seqf, "%4d %7d %10llu %s\n",
1195 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1196 (unsigned long long)part_nr_sects_read(part) >> 1,
1197 disk_name(sgp, part->partno, buf));
1198 disk_part_iter_exit(&piter);
1199
1200 return 0;
1201}
1202
1203static const struct seq_operations partitions_op = {
1204 .start = show_partition_start,
1205 .next = disk_seqf_next,
1206 .stop = disk_seqf_stop,
1207 .show = show_partition
1208};
1209#endif
1210
1211
1212static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1213{
1214 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1215 /* Make old-style 2.4 aliases work */
1216 request_module("block-major-%d", MAJOR(devt));
1217 return NULL;
1218}
1219
1220static int __init genhd_device_init(void)
1221{
1222 int error;
1223
1224 block_class.dev_kobj = sysfs_dev_block_kobj;
1225 error = class_register(&block_class);
1226 if (unlikely(error))
1227 return error;
1228 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1229 blk_dev_init();
1230
1231 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1232
1233 /* create top-level block dir */
1234 if (!sysfs_deprecated)
1235 block_depr = kobject_create_and_add("block", NULL);
1236 return 0;
1237}
1238
1239subsys_initcall(genhd_device_init);
1240
1241static ssize_t disk_range_show(struct device *dev,
1242 struct device_attribute *attr, char *buf)
1243{
1244 struct gendisk *disk = dev_to_disk(dev);
1245
1246 return sprintf(buf, "%d\n", disk->minors);
1247}
1248
1249static ssize_t disk_ext_range_show(struct device *dev,
1250 struct device_attribute *attr, char *buf)
1251{
1252 struct gendisk *disk = dev_to_disk(dev);
1253
1254 return sprintf(buf, "%d\n", disk_max_parts(disk));
1255}
1256
1257static ssize_t disk_removable_show(struct device *dev,
1258 struct device_attribute *attr, char *buf)
1259{
1260 struct gendisk *disk = dev_to_disk(dev);
1261
1262 return sprintf(buf, "%d\n",
1263 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1264}
1265
1266static ssize_t disk_hidden_show(struct device *dev,
1267 struct device_attribute *attr, char *buf)
1268{
1269 struct gendisk *disk = dev_to_disk(dev);
1270
1271 return sprintf(buf, "%d\n",
1272 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1273}
1274
1275static ssize_t disk_ro_show(struct device *dev,
1276 struct device_attribute *attr, char *buf)
1277{
1278 struct gendisk *disk = dev_to_disk(dev);
1279
1280 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1281}
1282
1283ssize_t part_size_show(struct device *dev,
1284 struct device_attribute *attr, char *buf)
1285{
1286 struct hd_struct *p = dev_to_part(dev);
1287
1288 return sprintf(buf, "%llu\n",
1289 (unsigned long long)part_nr_sects_read(p));
1290}
1291
1292ssize_t part_stat_show(struct device *dev,
1293 struct device_attribute *attr, char *buf)
1294{
1295 struct hd_struct *p = dev_to_part(dev);
1296 struct request_queue *q = part_to_disk(p)->queue;
1297 struct disk_stats stat;
1298 unsigned int inflight;
1299
1300 part_stat_read_all(p, &stat);
1301 if (queue_is_mq(q))
1302 inflight = blk_mq_in_flight(q, p);
1303 else
1304 inflight = part_in_flight(q, p);
1305
1306 return sprintf(buf,
1307 "%8lu %8lu %8llu %8u "
1308 "%8lu %8lu %8llu %8u "
1309 "%8u %8u %8u "
1310 "%8lu %8lu %8llu %8u "
1311 "%8lu %8u"
1312 "\n",
1313 stat.ios[STAT_READ],
1314 stat.merges[STAT_READ],
1315 (unsigned long long)stat.sectors[STAT_READ],
1316 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1317 stat.ios[STAT_WRITE],
1318 stat.merges[STAT_WRITE],
1319 (unsigned long long)stat.sectors[STAT_WRITE],
1320 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1321 inflight,
1322 jiffies_to_msecs(stat.io_ticks),
1323 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1324 stat.nsecs[STAT_WRITE] +
1325 stat.nsecs[STAT_DISCARD] +
1326 stat.nsecs[STAT_FLUSH],
1327 NSEC_PER_MSEC),
1328 stat.ios[STAT_DISCARD],
1329 stat.merges[STAT_DISCARD],
1330 (unsigned long long)stat.sectors[STAT_DISCARD],
1331 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1332 stat.ios[STAT_FLUSH],
1333 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1334}
1335
1336ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1337 char *buf)
1338{
1339 struct hd_struct *p = dev_to_part(dev);
1340 struct request_queue *q = part_to_disk(p)->queue;
1341 unsigned int inflight[2];
1342
1343 if (queue_is_mq(q))
1344 blk_mq_in_flight_rw(q, p, inflight);
1345 else
1346 part_in_flight_rw(q, p, inflight);
1347
1348 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1349}
1350
1351static ssize_t disk_capability_show(struct device *dev,
1352 struct device_attribute *attr, char *buf)
1353{
1354 struct gendisk *disk = dev_to_disk(dev);
1355
1356 return sprintf(buf, "%x\n", disk->flags);
1357}
1358
1359static ssize_t disk_alignment_offset_show(struct device *dev,
1360 struct device_attribute *attr,
1361 char *buf)
1362{
1363 struct gendisk *disk = dev_to_disk(dev);
1364
1365 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1366}
1367
1368static ssize_t disk_discard_alignment_show(struct device *dev,
1369 struct device_attribute *attr,
1370 char *buf)
1371{
1372 struct gendisk *disk = dev_to_disk(dev);
1373
1374 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1375}
1376
1377static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1378static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1379static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1380static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1381static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1382static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1383static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1384static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1385static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1386static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1387static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1388static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1389
1390#ifdef CONFIG_FAIL_MAKE_REQUEST
1391ssize_t part_fail_show(struct device *dev,
1392 struct device_attribute *attr, char *buf)
1393{
1394 struct hd_struct *p = dev_to_part(dev);
1395
1396 return sprintf(buf, "%d\n", p->make_it_fail);
1397}
1398
1399ssize_t part_fail_store(struct device *dev,
1400 struct device_attribute *attr,
1401 const char *buf, size_t count)
1402{
1403 struct hd_struct *p = dev_to_part(dev);
1404 int i;
1405
1406 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1407 p->make_it_fail = (i == 0) ? 0 : 1;
1408
1409 return count;
1410}
1411
1412static struct device_attribute dev_attr_fail =
1413 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1414#endif /* CONFIG_FAIL_MAKE_REQUEST */
1415
1416#ifdef CONFIG_FAIL_IO_TIMEOUT
1417static struct device_attribute dev_attr_fail_timeout =
1418 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1419#endif
1420
1421static struct attribute *disk_attrs[] = {
1422 &dev_attr_range.attr,
1423 &dev_attr_ext_range.attr,
1424 &dev_attr_removable.attr,
1425 &dev_attr_hidden.attr,
1426 &dev_attr_ro.attr,
1427 &dev_attr_size.attr,
1428 &dev_attr_alignment_offset.attr,
1429 &dev_attr_discard_alignment.attr,
1430 &dev_attr_capability.attr,
1431 &dev_attr_stat.attr,
1432 &dev_attr_inflight.attr,
1433 &dev_attr_badblocks.attr,
1434#ifdef CONFIG_FAIL_MAKE_REQUEST
1435 &dev_attr_fail.attr,
1436#endif
1437#ifdef CONFIG_FAIL_IO_TIMEOUT
1438 &dev_attr_fail_timeout.attr,
1439#endif
1440 NULL
1441};
1442
1443static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1444{
1445 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1446 struct gendisk *disk = dev_to_disk(dev);
1447
1448 if (a == &dev_attr_badblocks.attr && !disk->bb)
1449 return 0;
1450 return a->mode;
1451}
1452
1453static struct attribute_group disk_attr_group = {
1454 .attrs = disk_attrs,
1455 .is_visible = disk_visible,
1456};
1457
1458static const struct attribute_group *disk_attr_groups[] = {
1459 &disk_attr_group,
1460 NULL
1461};
1462
1463/**
1464 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1465 * @disk: disk to replace part_tbl for
1466 * @new_ptbl: new part_tbl to install
1467 *
1468 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1469 * original ptbl is freed using RCU callback.
1470 *
1471 * LOCKING:
1472 * Matching bd_mutex locked or the caller is the only user of @disk.
1473 */
1474static void disk_replace_part_tbl(struct gendisk *disk,
1475 struct disk_part_tbl *new_ptbl)
1476{
1477 struct disk_part_tbl *old_ptbl =
1478 rcu_dereference_protected(disk->part_tbl, 1);
1479
1480 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1481
1482 if (old_ptbl) {
1483 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1484 kfree_rcu(old_ptbl, rcu_head);
1485 }
1486}
1487
1488/**
1489 * disk_expand_part_tbl - expand disk->part_tbl
1490 * @disk: disk to expand part_tbl for
1491 * @partno: expand such that this partno can fit in
1492 *
1493 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1494 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1495 *
1496 * LOCKING:
1497 * Matching bd_mutex locked or the caller is the only user of @disk.
1498 * Might sleep.
1499 *
1500 * RETURNS:
1501 * 0 on success, -errno on failure.
1502 */
1503int disk_expand_part_tbl(struct gendisk *disk, int partno)
1504{
1505 struct disk_part_tbl *old_ptbl =
1506 rcu_dereference_protected(disk->part_tbl, 1);
1507 struct disk_part_tbl *new_ptbl;
1508 int len = old_ptbl ? old_ptbl->len : 0;
1509 int i, target;
1510
1511 /*
1512 * check for int overflow, since we can get here from blkpg_ioctl()
1513 * with a user passed 'partno'.
1514 */
1515 target = partno + 1;
1516 if (target < 0)
1517 return -EINVAL;
1518
1519 /* disk_max_parts() is zero during initialization, ignore if so */
1520 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1521 return -EINVAL;
1522
1523 if (target <= len)
1524 return 0;
1525
1526 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1527 disk->node_id);
1528 if (!new_ptbl)
1529 return -ENOMEM;
1530
1531 new_ptbl->len = target;
1532
1533 for (i = 0; i < len; i++)
1534 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1535
1536 disk_replace_part_tbl(disk, new_ptbl);
1537 return 0;
1538}
1539
1540/**
1541 * disk_release - releases all allocated resources of the gendisk
1542 * @dev: the device representing this disk
1543 *
1544 * This function releases all allocated resources of the gendisk.
1545 *
1546 * The struct gendisk refcount is incremented with get_gendisk() or
1547 * get_disk_and_module(), and its refcount is decremented with
1548 * put_disk_and_module() or put_disk(). Once the refcount reaches 0 this
1549 * function is called.
1550 *
1551 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1552 * assigned. Since the request_queue sits on top of the gendisk for these
1553 * drivers we also call blk_put_queue() for them, and we expect the
1554 * request_queue refcount to reach 0 at this point, and so the request_queue
1555 * will also be freed prior to the disk.
1556 *
1557 * Context: can sleep
1558 */
1559static void disk_release(struct device *dev)
1560{
1561 struct gendisk *disk = dev_to_disk(dev);
1562
1563 might_sleep();
1564
1565 blk_free_devt(dev->devt);
1566 disk_release_events(disk);
1567 kfree(disk->random);
1568 disk_replace_part_tbl(disk, NULL);
1569 hd_free_part(&disk->part0);
1570 if (disk->queue)
1571 blk_put_queue(disk->queue);
1572 kfree(disk);
1573}
1574struct class block_class = {
1575 .name = "block",
1576};
1577
1578static char *block_devnode(struct device *dev, umode_t *mode,
1579 kuid_t *uid, kgid_t *gid)
1580{
1581 struct gendisk *disk = dev_to_disk(dev);
1582
1583 if (disk->fops->devnode)
1584 return disk->fops->devnode(disk, mode);
1585 return NULL;
1586}
1587
1588const struct device_type disk_type = {
1589 .name = "disk",
1590 .groups = disk_attr_groups,
1591 .release = disk_release,
1592 .devnode = block_devnode,
1593};
1594
1595#ifdef CONFIG_PROC_FS
1596/*
1597 * aggregate disk stat collector. Uses the same stats that the sysfs
1598 * entries do, above, but makes them available through one seq_file.
1599 *
1600 * The output looks suspiciously like /proc/partitions with a bunch of
1601 * extra fields.
1602 */
1603static int diskstats_show(struct seq_file *seqf, void *v)
1604{
1605 struct gendisk *gp = v;
1606 struct disk_part_iter piter;
1607 struct hd_struct *hd;
1608 char buf[BDEVNAME_SIZE];
1609 unsigned int inflight;
1610 struct disk_stats stat;
1611
1612 /*
1613 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1614 seq_puts(seqf, "major minor name"
1615 " rio rmerge rsect ruse wio wmerge "
1616 "wsect wuse running use aveq"
1617 "\n\n");
1618 */
1619
1620 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1621 while ((hd = disk_part_iter_next(&piter))) {
1622 part_stat_read_all(hd, &stat);
1623 if (queue_is_mq(gp->queue))
1624 inflight = blk_mq_in_flight(gp->queue, hd);
1625 else
1626 inflight = part_in_flight(gp->queue, hd);
1627
1628 seq_printf(seqf, "%4d %7d %s "
1629 "%lu %lu %lu %u "
1630 "%lu %lu %lu %u "
1631 "%u %u %u "
1632 "%lu %lu %lu %u "
1633 "%lu %u"
1634 "\n",
1635 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1636 disk_name(gp, hd->partno, buf),
1637 stat.ios[STAT_READ],
1638 stat.merges[STAT_READ],
1639 stat.sectors[STAT_READ],
1640 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1641 NSEC_PER_MSEC),
1642 stat.ios[STAT_WRITE],
1643 stat.merges[STAT_WRITE],
1644 stat.sectors[STAT_WRITE],
1645 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1646 NSEC_PER_MSEC),
1647 inflight,
1648 jiffies_to_msecs(stat.io_ticks),
1649 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1650 stat.nsecs[STAT_WRITE] +
1651 stat.nsecs[STAT_DISCARD] +
1652 stat.nsecs[STAT_FLUSH],
1653 NSEC_PER_MSEC),
1654 stat.ios[STAT_DISCARD],
1655 stat.merges[STAT_DISCARD],
1656 stat.sectors[STAT_DISCARD],
1657 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1658 NSEC_PER_MSEC),
1659 stat.ios[STAT_FLUSH],
1660 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1661 NSEC_PER_MSEC)
1662 );
1663 }
1664 disk_part_iter_exit(&piter);
1665
1666 return 0;
1667}
1668
1669static const struct seq_operations diskstats_op = {
1670 .start = disk_seqf_start,
1671 .next = disk_seqf_next,
1672 .stop = disk_seqf_stop,
1673 .show = diskstats_show
1674};
1675
1676static int __init proc_genhd_init(void)
1677{
1678 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1679 proc_create_seq("partitions", 0, NULL, &partitions_op);
1680 return 0;
1681}
1682module_init(proc_genhd_init);
1683#endif /* CONFIG_PROC_FS */
1684
1685dev_t blk_lookup_devt(const char *name, int partno)
1686{
1687 dev_t devt = MKDEV(0, 0);
1688 struct class_dev_iter iter;
1689 struct device *dev;
1690
1691 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1692 while ((dev = class_dev_iter_next(&iter))) {
1693 struct gendisk *disk = dev_to_disk(dev);
1694 struct hd_struct *part;
1695
1696 if (strcmp(dev_name(dev), name))
1697 continue;
1698
1699 if (partno < disk->minors) {
1700 /* We need to return the right devno, even
1701 * if the partition doesn't exist yet.
1702 */
1703 devt = MKDEV(MAJOR(dev->devt),
1704 MINOR(dev->devt) + partno);
1705 break;
1706 }
1707 part = disk_get_part(disk, partno);
1708 if (part) {
1709 devt = part_devt(part);
1710 disk_put_part(part);
1711 break;
1712 }
1713 disk_put_part(part);
1714 }
1715 class_dev_iter_exit(&iter);
1716 return devt;
1717}
1718
1719struct gendisk *__alloc_disk_node(int minors, int node_id)
1720{
1721 struct gendisk *disk;
1722 struct disk_part_tbl *ptbl;
1723
1724 if (minors > DISK_MAX_PARTS) {
1725 printk(KERN_ERR
1726 "block: can't allocate more than %d partitions\n",
1727 DISK_MAX_PARTS);
1728 minors = DISK_MAX_PARTS;
1729 }
1730
1731 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1732 if (disk) {
1733 disk->part0.dkstats = alloc_percpu(struct disk_stats);
1734 if (!disk->part0.dkstats) {
1735 kfree(disk);
1736 return NULL;
1737 }
1738 init_rwsem(&disk->lookup_sem);
1739 disk->node_id = node_id;
1740 if (disk_expand_part_tbl(disk, 0)) {
1741 free_percpu(disk->part0.dkstats);
1742 kfree(disk);
1743 return NULL;
1744 }
1745 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1746 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1747
1748 /*
1749 * set_capacity() and get_capacity() currently don't use
1750 * seqcounter to read/update the part0->nr_sects. Still init
1751 * the counter as we can read the sectors in IO submission
1752 * patch using seqence counters.
1753 *
1754 * TODO: Ideally set_capacity() and get_capacity() should be
1755 * converted to make use of bd_mutex and sequence counters.
1756 */
1757 hd_sects_seq_init(&disk->part0);
1758 if (hd_ref_init(&disk->part0)) {
1759 hd_free_part(&disk->part0);
1760 kfree(disk);
1761 return NULL;
1762 }
1763
1764 disk->minors = minors;
1765 rand_initialize_disk(disk);
1766 disk_to_dev(disk)->class = &block_class;
1767 disk_to_dev(disk)->type = &disk_type;
1768 device_initialize(disk_to_dev(disk));
1769 }
1770 return disk;
1771}
1772EXPORT_SYMBOL(__alloc_disk_node);
1773
1774/**
1775 * get_disk_and_module - increments the gendisk and gendisk fops module refcount
1776 * @disk: the struct gendisk to increment the refcount for
1777 *
1778 * This increments the refcount for the struct gendisk, and the gendisk's
1779 * fops module owner.
1780 *
1781 * Context: Any context.
1782 */
1783struct kobject *get_disk_and_module(struct gendisk *disk)
1784{
1785 struct module *owner;
1786 struct kobject *kobj;
1787
1788 if (!disk->fops)
1789 return NULL;
1790 owner = disk->fops->owner;
1791 if (owner && !try_module_get(owner))
1792 return NULL;
1793 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1794 if (kobj == NULL) {
1795 module_put(owner);
1796 return NULL;
1797 }
1798 return kobj;
1799
1800}
1801EXPORT_SYMBOL(get_disk_and_module);
1802
1803/**
1804 * put_disk - decrements the gendisk refcount
1805 * @disk: the struct gendisk to decrement the refcount for
1806 *
1807 * This decrements the refcount for the struct gendisk. When this reaches 0
1808 * we'll have disk_release() called.
1809 *
1810 * Context: Any context, but the last reference must not be dropped from
1811 * atomic context.
1812 */
1813void put_disk(struct gendisk *disk)
1814{
1815 if (disk)
1816 kobject_put(&disk_to_dev(disk)->kobj);
1817}
1818EXPORT_SYMBOL(put_disk);
1819
1820/**
1821 * put_disk_and_module - decrements the module and gendisk refcount
1822 * @disk: the struct gendisk to decrement the refcount for
1823 *
1824 * This is a counterpart of get_disk_and_module() and thus also of
1825 * get_gendisk().
1826 *
1827 * Context: Any context, but the last reference must not be dropped from
1828 * atomic context.
1829 */
1830void put_disk_and_module(struct gendisk *disk)
1831{
1832 if (disk) {
1833 struct module *owner = disk->fops->owner;
1834
1835 put_disk(disk);
1836 module_put(owner);
1837 }
1838}
1839EXPORT_SYMBOL(put_disk_and_module);
1840
1841static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1842{
1843 char event[] = "DISK_RO=1";
1844 char *envp[] = { event, NULL };
1845
1846 if (!ro)
1847 event[8] = '0';
1848 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1849}
1850
1851void set_device_ro(struct block_device *bdev, int flag)
1852{
1853 bdev->bd_part->policy = flag;
1854}
1855
1856EXPORT_SYMBOL(set_device_ro);
1857
1858void set_disk_ro(struct gendisk *disk, int flag)
1859{
1860 struct disk_part_iter piter;
1861 struct hd_struct *part;
1862
1863 if (disk->part0.policy != flag) {
1864 set_disk_ro_uevent(disk, flag);
1865 disk->part0.policy = flag;
1866 }
1867
1868 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1869 while ((part = disk_part_iter_next(&piter)))
1870 part->policy = flag;
1871 disk_part_iter_exit(&piter);
1872}
1873
1874EXPORT_SYMBOL(set_disk_ro);
1875
1876int bdev_read_only(struct block_device *bdev)
1877{
1878 if (!bdev)
1879 return 0;
1880 return bdev->bd_part->policy;
1881}
1882
1883EXPORT_SYMBOL(bdev_read_only);
1884
1885/*
1886 * Disk events - monitor disk events like media change and eject request.
1887 */
1888struct disk_events {
1889 struct list_head node; /* all disk_event's */
1890 struct gendisk *disk; /* the associated disk */
1891 spinlock_t lock;
1892
1893 struct mutex block_mutex; /* protects blocking */
1894 int block; /* event blocking depth */
1895 unsigned int pending; /* events already sent out */
1896 unsigned int clearing; /* events being cleared */
1897
1898 long poll_msecs; /* interval, -1 for default */
1899 struct delayed_work dwork;
1900};
1901
1902static const char *disk_events_strs[] = {
1903 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1904 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1905};
1906
1907static char *disk_uevents[] = {
1908 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1909 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1910};
1911
1912/* list of all disk_events */
1913static DEFINE_MUTEX(disk_events_mutex);
1914static LIST_HEAD(disk_events);
1915
1916/* disable in-kernel polling by default */
1917static unsigned long disk_events_dfl_poll_msecs;
1918
1919static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1920{
1921 struct disk_events *ev = disk->ev;
1922 long intv_msecs = 0;
1923
1924 /*
1925 * If device-specific poll interval is set, always use it. If
1926 * the default is being used, poll if the POLL flag is set.
1927 */
1928 if (ev->poll_msecs >= 0)
1929 intv_msecs = ev->poll_msecs;
1930 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1931 intv_msecs = disk_events_dfl_poll_msecs;
1932
1933 return msecs_to_jiffies(intv_msecs);
1934}
1935
1936/**
1937 * disk_block_events - block and flush disk event checking
1938 * @disk: disk to block events for
1939 *
1940 * On return from this function, it is guaranteed that event checking
1941 * isn't in progress and won't happen until unblocked by
1942 * disk_unblock_events(). Events blocking is counted and the actual
1943 * unblocking happens after the matching number of unblocks are done.
1944 *
1945 * Note that this intentionally does not block event checking from
1946 * disk_clear_events().
1947 *
1948 * CONTEXT:
1949 * Might sleep.
1950 */
1951void disk_block_events(struct gendisk *disk)
1952{
1953 struct disk_events *ev = disk->ev;
1954 unsigned long flags;
1955 bool cancel;
1956
1957 if (!ev)
1958 return;
1959
1960 /*
1961 * Outer mutex ensures that the first blocker completes canceling
1962 * the event work before further blockers are allowed to finish.
1963 */
1964 mutex_lock(&ev->block_mutex);
1965
1966 spin_lock_irqsave(&ev->lock, flags);
1967 cancel = !ev->block++;
1968 spin_unlock_irqrestore(&ev->lock, flags);
1969
1970 if (cancel)
1971 cancel_delayed_work_sync(&disk->ev->dwork);
1972
1973 mutex_unlock(&ev->block_mutex);
1974}
1975
1976static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1977{
1978 struct disk_events *ev = disk->ev;
1979 unsigned long intv;
1980 unsigned long flags;
1981
1982 spin_lock_irqsave(&ev->lock, flags);
1983
1984 if (WARN_ON_ONCE(ev->block <= 0))
1985 goto out_unlock;
1986
1987 if (--ev->block)
1988 goto out_unlock;
1989
1990 intv = disk_events_poll_jiffies(disk);
1991 if (check_now)
1992 queue_delayed_work(system_freezable_power_efficient_wq,
1993 &ev->dwork, 0);
1994 else if (intv)
1995 queue_delayed_work(system_freezable_power_efficient_wq,
1996 &ev->dwork, intv);
1997out_unlock:
1998 spin_unlock_irqrestore(&ev->lock, flags);
1999}
2000
2001/**
2002 * disk_unblock_events - unblock disk event checking
2003 * @disk: disk to unblock events for
2004 *
2005 * Undo disk_block_events(). When the block count reaches zero, it
2006 * starts events polling if configured.
2007 *
2008 * CONTEXT:
2009 * Don't care. Safe to call from irq context.
2010 */
2011void disk_unblock_events(struct gendisk *disk)
2012{
2013 if (disk->ev)
2014 __disk_unblock_events(disk, false);
2015}
2016
2017/**
2018 * disk_flush_events - schedule immediate event checking and flushing
2019 * @disk: disk to check and flush events for
2020 * @mask: events to flush
2021 *
2022 * Schedule immediate event checking on @disk if not blocked. Events in
2023 * @mask are scheduled to be cleared from the driver. Note that this
2024 * doesn't clear the events from @disk->ev.
2025 *
2026 * CONTEXT:
2027 * If @mask is non-zero must be called with bdev->bd_mutex held.
2028 */
2029void disk_flush_events(struct gendisk *disk, unsigned int mask)
2030{
2031 struct disk_events *ev = disk->ev;
2032
2033 if (!ev)
2034 return;
2035
2036 spin_lock_irq(&ev->lock);
2037 ev->clearing |= mask;
2038 if (!ev->block)
2039 mod_delayed_work(system_freezable_power_efficient_wq,
2040 &ev->dwork, 0);
2041 spin_unlock_irq(&ev->lock);
2042}
2043
2044/**
2045 * disk_clear_events - synchronously check, clear and return pending events
2046 * @disk: disk to fetch and clear events from
2047 * @mask: mask of events to be fetched and cleared
2048 *
2049 * Disk events are synchronously checked and pending events in @mask
2050 * are cleared and returned. This ignores the block count.
2051 *
2052 * CONTEXT:
2053 * Might sleep.
2054 */
2055unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
2056{
2057 struct disk_events *ev = disk->ev;
2058 unsigned int pending;
2059 unsigned int clearing = mask;
2060
2061 if (!ev)
2062 return 0;
2063
2064 disk_block_events(disk);
2065
2066 /*
2067 * store the union of mask and ev->clearing on the stack so that the
2068 * race with disk_flush_events does not cause ambiguity (ev->clearing
2069 * can still be modified even if events are blocked).
2070 */
2071 spin_lock_irq(&ev->lock);
2072 clearing |= ev->clearing;
2073 ev->clearing = 0;
2074 spin_unlock_irq(&ev->lock);
2075
2076 disk_check_events(ev, &clearing);
2077 /*
2078 * if ev->clearing is not 0, the disk_flush_events got called in the
2079 * middle of this function, so we want to run the workfn without delay.
2080 */
2081 __disk_unblock_events(disk, ev->clearing ? true : false);
2082
2083 /* then, fetch and clear pending events */
2084 spin_lock_irq(&ev->lock);
2085 pending = ev->pending & mask;
2086 ev->pending &= ~mask;
2087 spin_unlock_irq(&ev->lock);
2088 WARN_ON_ONCE(clearing & mask);
2089
2090 return pending;
2091}
2092
2093/*
2094 * Separate this part out so that a different pointer for clearing_ptr can be
2095 * passed in for disk_clear_events.
2096 */
2097static void disk_events_workfn(struct work_struct *work)
2098{
2099 struct delayed_work *dwork = to_delayed_work(work);
2100 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2101
2102 disk_check_events(ev, &ev->clearing);
2103}
2104
2105static void disk_check_events(struct disk_events *ev,
2106 unsigned int *clearing_ptr)
2107{
2108 struct gendisk *disk = ev->disk;
2109 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2110 unsigned int clearing = *clearing_ptr;
2111 unsigned int events;
2112 unsigned long intv;
2113 int nr_events = 0, i;
2114
2115 /* check events */
2116 events = disk->fops->check_events(disk, clearing);
2117
2118 /* accumulate pending events and schedule next poll if necessary */
2119 spin_lock_irq(&ev->lock);
2120
2121 events &= ~ev->pending;
2122 ev->pending |= events;
2123 *clearing_ptr &= ~clearing;
2124
2125 intv = disk_events_poll_jiffies(disk);
2126 if (!ev->block && intv)
2127 queue_delayed_work(system_freezable_power_efficient_wq,
2128 &ev->dwork, intv);
2129
2130 spin_unlock_irq(&ev->lock);
2131
2132 /*
2133 * Tell userland about new events. Only the events listed in
2134 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2135 * is set. Otherwise, events are processed internally but never
2136 * get reported to userland.
2137 */
2138 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2139 if ((events & disk->events & (1 << i)) &&
2140 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2141 envp[nr_events++] = disk_uevents[i];
2142
2143 if (nr_events)
2144 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2145}
2146
2147/*
2148 * A disk events enabled device has the following sysfs nodes under
2149 * its /sys/block/X/ directory.
2150 *
2151 * events : list of all supported events
2152 * events_async : list of events which can be detected w/o polling
2153 * (always empty, only for backwards compatibility)
2154 * events_poll_msecs : polling interval, 0: disable, -1: system default
2155 */
2156static ssize_t __disk_events_show(unsigned int events, char *buf)
2157{
2158 const char *delim = "";
2159 ssize_t pos = 0;
2160 int i;
2161
2162 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2163 if (events & (1 << i)) {
2164 pos += sprintf(buf + pos, "%s%s",
2165 delim, disk_events_strs[i]);
2166 delim = " ";
2167 }
2168 if (pos)
2169 pos += sprintf(buf + pos, "\n");
2170 return pos;
2171}
2172
2173static ssize_t disk_events_show(struct device *dev,
2174 struct device_attribute *attr, char *buf)
2175{
2176 struct gendisk *disk = dev_to_disk(dev);
2177
2178 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2179 return 0;
2180
2181 return __disk_events_show(disk->events, buf);
2182}
2183
2184static ssize_t disk_events_async_show(struct device *dev,
2185 struct device_attribute *attr, char *buf)
2186{
2187 return 0;
2188}
2189
2190static ssize_t disk_events_poll_msecs_show(struct device *dev,
2191 struct device_attribute *attr,
2192 char *buf)
2193{
2194 struct gendisk *disk = dev_to_disk(dev);
2195
2196 if (!disk->ev)
2197 return sprintf(buf, "-1\n");
2198
2199 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2200}
2201
2202static ssize_t disk_events_poll_msecs_store(struct device *dev,
2203 struct device_attribute *attr,
2204 const char *buf, size_t count)
2205{
2206 struct gendisk *disk = dev_to_disk(dev);
2207 long intv;
2208
2209 if (!count || !sscanf(buf, "%ld", &intv))
2210 return -EINVAL;
2211
2212 if (intv < 0 && intv != -1)
2213 return -EINVAL;
2214
2215 if (!disk->ev)
2216 return -ENODEV;
2217
2218 disk_block_events(disk);
2219 disk->ev->poll_msecs = intv;
2220 __disk_unblock_events(disk, true);
2221
2222 return count;
2223}
2224
2225static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2226static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2227static const DEVICE_ATTR(events_poll_msecs, 0644,
2228 disk_events_poll_msecs_show,
2229 disk_events_poll_msecs_store);
2230
2231static const struct attribute *disk_events_attrs[] = {
2232 &dev_attr_events.attr,
2233 &dev_attr_events_async.attr,
2234 &dev_attr_events_poll_msecs.attr,
2235 NULL,
2236};
2237
2238/*
2239 * The default polling interval can be specified by the kernel
2240 * parameter block.events_dfl_poll_msecs which defaults to 0
2241 * (disable). This can also be modified runtime by writing to
2242 * /sys/module/block/parameters/events_dfl_poll_msecs.
2243 */
2244static int disk_events_set_dfl_poll_msecs(const char *val,
2245 const struct kernel_param *kp)
2246{
2247 struct disk_events *ev;
2248 int ret;
2249
2250 ret = param_set_ulong(val, kp);
2251 if (ret < 0)
2252 return ret;
2253
2254 mutex_lock(&disk_events_mutex);
2255
2256 list_for_each_entry(ev, &disk_events, node)
2257 disk_flush_events(ev->disk, 0);
2258
2259 mutex_unlock(&disk_events_mutex);
2260
2261 return 0;
2262}
2263
2264static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2265 .set = disk_events_set_dfl_poll_msecs,
2266 .get = param_get_ulong,
2267};
2268
2269#undef MODULE_PARAM_PREFIX
2270#define MODULE_PARAM_PREFIX "block."
2271
2272module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2273 &disk_events_dfl_poll_msecs, 0644);
2274
2275/*
2276 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2277 */
2278static void disk_alloc_events(struct gendisk *disk)
2279{
2280 struct disk_events *ev;
2281
2282 if (!disk->fops->check_events || !disk->events)
2283 return;
2284
2285 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2286 if (!ev) {
2287 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2288 return;
2289 }
2290
2291 INIT_LIST_HEAD(&ev->node);
2292 ev->disk = disk;
2293 spin_lock_init(&ev->lock);
2294 mutex_init(&ev->block_mutex);
2295 ev->block = 1;
2296 ev->poll_msecs = -1;
2297 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2298
2299 disk->ev = ev;
2300}
2301
2302static void disk_add_events(struct gendisk *disk)
2303{
2304 /* FIXME: error handling */
2305 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2306 pr_warn("%s: failed to create sysfs files for events\n",
2307 disk->disk_name);
2308
2309 if (!disk->ev)
2310 return;
2311
2312 mutex_lock(&disk_events_mutex);
2313 list_add_tail(&disk->ev->node, &disk_events);
2314 mutex_unlock(&disk_events_mutex);
2315
2316 /*
2317 * Block count is initialized to 1 and the following initial
2318 * unblock kicks it into action.
2319 */
2320 __disk_unblock_events(disk, true);
2321}
2322
2323static void disk_del_events(struct gendisk *disk)
2324{
2325 if (disk->ev) {
2326 disk_block_events(disk);
2327
2328 mutex_lock(&disk_events_mutex);
2329 list_del_init(&disk->ev->node);
2330 mutex_unlock(&disk_events_mutex);
2331 }
2332
2333 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2334}
2335
2336static void disk_release_events(struct gendisk *disk)
2337{
2338 /* the block count should be 1 from disk_del_events() */
2339 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
2340 kfree(disk->ev);
2341}