xref: /linux/block/genhd.c (revision caa97be1a2052f2dfc026c3fe5ef62f620782f24)
1 /*
2  *  gendisk handling
3  */
4 
5 #include <linux/module.h>
6 #include <linux/fs.h>
7 #include <linux/genhd.h>
8 #include <linux/kdev_t.h>
9 #include <linux/kernel.h>
10 #include <linux/blkdev.h>
11 #include <linux/backing-dev.h>
12 #include <linux/init.h>
13 #include <linux/spinlock.h>
14 #include <linux/proc_fs.h>
15 #include <linux/seq_file.h>
16 #include <linux/slab.h>
17 #include <linux/kmod.h>
18 #include <linux/kobj_map.h>
19 #include <linux/mutex.h>
20 #include <linux/idr.h>
21 #include <linux/log2.h>
22 #include <linux/pm_runtime.h>
23 #include <linux/badblocks.h>
24 
25 #include "blk.h"
26 
27 static DEFINE_MUTEX(block_class_lock);
28 struct kobject *block_depr;
29 
30 /* for extended dynamic devt allocation, currently only one major is used */
31 #define NR_EXT_DEVT		(1 << MINORBITS)
32 
33 /* For extended devt allocation.  ext_devt_lock prevents look up
34  * results from going away underneath its user.
35  */
36 static DEFINE_SPINLOCK(ext_devt_lock);
37 static DEFINE_IDR(ext_devt_idr);
38 
39 static const struct device_type disk_type;
40 
41 static void disk_check_events(struct disk_events *ev,
42 			      unsigned int *clearing_ptr);
43 static void disk_alloc_events(struct gendisk *disk);
44 static void disk_add_events(struct gendisk *disk);
45 static void disk_del_events(struct gendisk *disk);
46 static void disk_release_events(struct gendisk *disk);
47 
48 /**
49  * disk_get_part - get partition
50  * @disk: disk to look partition from
51  * @partno: partition number
52  *
53  * Look for partition @partno from @disk.  If found, increment
54  * reference count and return it.
55  *
56  * CONTEXT:
57  * Don't care.
58  *
59  * RETURNS:
60  * Pointer to the found partition on success, NULL if not found.
61  */
62 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
63 {
64 	struct hd_struct *part = NULL;
65 	struct disk_part_tbl *ptbl;
66 
67 	if (unlikely(partno < 0))
68 		return NULL;
69 
70 	rcu_read_lock();
71 
72 	ptbl = rcu_dereference(disk->part_tbl);
73 	if (likely(partno < ptbl->len)) {
74 		part = rcu_dereference(ptbl->part[partno]);
75 		if (part)
76 			get_device(part_to_dev(part));
77 	}
78 
79 	rcu_read_unlock();
80 
81 	return part;
82 }
83 EXPORT_SYMBOL_GPL(disk_get_part);
84 
85 /**
86  * disk_part_iter_init - initialize partition iterator
87  * @piter: iterator to initialize
88  * @disk: disk to iterate over
89  * @flags: DISK_PITER_* flags
90  *
91  * Initialize @piter so that it iterates over partitions of @disk.
92  *
93  * CONTEXT:
94  * Don't care.
95  */
96 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
97 			  unsigned int flags)
98 {
99 	struct disk_part_tbl *ptbl;
100 
101 	rcu_read_lock();
102 	ptbl = rcu_dereference(disk->part_tbl);
103 
104 	piter->disk = disk;
105 	piter->part = NULL;
106 
107 	if (flags & DISK_PITER_REVERSE)
108 		piter->idx = ptbl->len - 1;
109 	else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
110 		piter->idx = 0;
111 	else
112 		piter->idx = 1;
113 
114 	piter->flags = flags;
115 
116 	rcu_read_unlock();
117 }
118 EXPORT_SYMBOL_GPL(disk_part_iter_init);
119 
120 /**
121  * disk_part_iter_next - proceed iterator to the next partition and return it
122  * @piter: iterator of interest
123  *
124  * Proceed @piter to the next partition and return it.
125  *
126  * CONTEXT:
127  * Don't care.
128  */
129 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
130 {
131 	struct disk_part_tbl *ptbl;
132 	int inc, end;
133 
134 	/* put the last partition */
135 	disk_put_part(piter->part);
136 	piter->part = NULL;
137 
138 	/* get part_tbl */
139 	rcu_read_lock();
140 	ptbl = rcu_dereference(piter->disk->part_tbl);
141 
142 	/* determine iteration parameters */
143 	if (piter->flags & DISK_PITER_REVERSE) {
144 		inc = -1;
145 		if (piter->flags & (DISK_PITER_INCL_PART0 |
146 				    DISK_PITER_INCL_EMPTY_PART0))
147 			end = -1;
148 		else
149 			end = 0;
150 	} else {
151 		inc = 1;
152 		end = ptbl->len;
153 	}
154 
155 	/* iterate to the next partition */
156 	for (; piter->idx != end; piter->idx += inc) {
157 		struct hd_struct *part;
158 
159 		part = rcu_dereference(ptbl->part[piter->idx]);
160 		if (!part)
161 			continue;
162 		if (!part_nr_sects_read(part) &&
163 		    !(piter->flags & DISK_PITER_INCL_EMPTY) &&
164 		    !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
165 		      piter->idx == 0))
166 			continue;
167 
168 		get_device(part_to_dev(part));
169 		piter->part = part;
170 		piter->idx += inc;
171 		break;
172 	}
173 
174 	rcu_read_unlock();
175 
176 	return piter->part;
177 }
178 EXPORT_SYMBOL_GPL(disk_part_iter_next);
179 
180 /**
181  * disk_part_iter_exit - finish up partition iteration
182  * @piter: iter of interest
183  *
184  * Called when iteration is over.  Cleans up @piter.
185  *
186  * CONTEXT:
187  * Don't care.
188  */
189 void disk_part_iter_exit(struct disk_part_iter *piter)
190 {
191 	disk_put_part(piter->part);
192 	piter->part = NULL;
193 }
194 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
195 
196 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
197 {
198 	return part->start_sect <= sector &&
199 		sector < part->start_sect + part_nr_sects_read(part);
200 }
201 
202 /**
203  * disk_map_sector_rcu - map sector to partition
204  * @disk: gendisk of interest
205  * @sector: sector to map
206  *
207  * Find out which partition @sector maps to on @disk.  This is
208  * primarily used for stats accounting.
209  *
210  * CONTEXT:
211  * RCU read locked.  The returned partition pointer is valid only
212  * while preemption is disabled.
213  *
214  * RETURNS:
215  * Found partition on success, part0 is returned if no partition matches
216  */
217 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
218 {
219 	struct disk_part_tbl *ptbl;
220 	struct hd_struct *part;
221 	int i;
222 
223 	ptbl = rcu_dereference(disk->part_tbl);
224 
225 	part = rcu_dereference(ptbl->last_lookup);
226 	if (part && sector_in_part(part, sector))
227 		return part;
228 
229 	for (i = 1; i < ptbl->len; i++) {
230 		part = rcu_dereference(ptbl->part[i]);
231 
232 		if (part && sector_in_part(part, sector)) {
233 			rcu_assign_pointer(ptbl->last_lookup, part);
234 			return part;
235 		}
236 	}
237 	return &disk->part0;
238 }
239 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
240 
241 /*
242  * Can be deleted altogether. Later.
243  *
244  */
245 #define BLKDEV_MAJOR_HASH_SIZE 255
246 static struct blk_major_name {
247 	struct blk_major_name *next;
248 	int major;
249 	char name[16];
250 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
251 
252 /* index in the above - for now: assume no multimajor ranges */
253 static inline int major_to_index(unsigned major)
254 {
255 	return major % BLKDEV_MAJOR_HASH_SIZE;
256 }
257 
258 #ifdef CONFIG_PROC_FS
259 void blkdev_show(struct seq_file *seqf, off_t offset)
260 {
261 	struct blk_major_name *dp;
262 
263 	mutex_lock(&block_class_lock);
264 	for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
265 		if (dp->major == offset)
266 			seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
267 	mutex_unlock(&block_class_lock);
268 }
269 #endif /* CONFIG_PROC_FS */
270 
271 /**
272  * register_blkdev - register a new block device
273  *
274  * @major: the requested major device number [1..255]. If @major = 0, try to
275  *         allocate any unused major number.
276  * @name: the name of the new block device as a zero terminated string
277  *
278  * The @name must be unique within the system.
279  *
280  * The return value depends on the @major input parameter:
281  *
282  *  - if a major device number was requested in range [1..255] then the
283  *    function returns zero on success, or a negative error code
284  *  - if any unused major number was requested with @major = 0 parameter
285  *    then the return value is the allocated major number in range
286  *    [1..255] or a negative error code otherwise
287  */
288 int register_blkdev(unsigned int major, const char *name)
289 {
290 	struct blk_major_name **n, *p;
291 	int index, ret = 0;
292 
293 	mutex_lock(&block_class_lock);
294 
295 	/* temporary */
296 	if (major == 0) {
297 		for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
298 			if (major_names[index] == NULL)
299 				break;
300 		}
301 
302 		if (index == 0) {
303 			printk("register_blkdev: failed to get major for %s\n",
304 			       name);
305 			ret = -EBUSY;
306 			goto out;
307 		}
308 		major = index;
309 		ret = major;
310 	}
311 
312 	if (major >= BLKDEV_MAJOR_MAX) {
313 		pr_err("register_blkdev: major requested (%d) is greater than the maximum (%d) for %s\n",
314 		       major, BLKDEV_MAJOR_MAX, name);
315 
316 		ret = -EINVAL;
317 		goto out;
318 	}
319 
320 	p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
321 	if (p == NULL) {
322 		ret = -ENOMEM;
323 		goto out;
324 	}
325 
326 	p->major = major;
327 	strlcpy(p->name, name, sizeof(p->name));
328 	p->next = NULL;
329 	index = major_to_index(major);
330 
331 	for (n = &major_names[index]; *n; n = &(*n)->next) {
332 		if ((*n)->major == major)
333 			break;
334 	}
335 	if (!*n)
336 		*n = p;
337 	else
338 		ret = -EBUSY;
339 
340 	if (ret < 0) {
341 		printk("register_blkdev: cannot get major %d for %s\n",
342 		       major, name);
343 		kfree(p);
344 	}
345 out:
346 	mutex_unlock(&block_class_lock);
347 	return ret;
348 }
349 
350 EXPORT_SYMBOL(register_blkdev);
351 
352 void unregister_blkdev(unsigned int major, const char *name)
353 {
354 	struct blk_major_name **n;
355 	struct blk_major_name *p = NULL;
356 	int index = major_to_index(major);
357 
358 	mutex_lock(&block_class_lock);
359 	for (n = &major_names[index]; *n; n = &(*n)->next)
360 		if ((*n)->major == major)
361 			break;
362 	if (!*n || strcmp((*n)->name, name)) {
363 		WARN_ON(1);
364 	} else {
365 		p = *n;
366 		*n = p->next;
367 	}
368 	mutex_unlock(&block_class_lock);
369 	kfree(p);
370 }
371 
372 EXPORT_SYMBOL(unregister_blkdev);
373 
374 static struct kobj_map *bdev_map;
375 
376 /**
377  * blk_mangle_minor - scatter minor numbers apart
378  * @minor: minor number to mangle
379  *
380  * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
381  * is enabled.  Mangling twice gives the original value.
382  *
383  * RETURNS:
384  * Mangled value.
385  *
386  * CONTEXT:
387  * Don't care.
388  */
389 static int blk_mangle_minor(int minor)
390 {
391 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
392 	int i;
393 
394 	for (i = 0; i < MINORBITS / 2; i++) {
395 		int low = minor & (1 << i);
396 		int high = minor & (1 << (MINORBITS - 1 - i));
397 		int distance = MINORBITS - 1 - 2 * i;
398 
399 		minor ^= low | high;	/* clear both bits */
400 		low <<= distance;	/* swap the positions */
401 		high >>= distance;
402 		minor |= low | high;	/* and set */
403 	}
404 #endif
405 	return minor;
406 }
407 
408 /**
409  * blk_alloc_devt - allocate a dev_t for a partition
410  * @part: partition to allocate dev_t for
411  * @devt: out parameter for resulting dev_t
412  *
413  * Allocate a dev_t for block device.
414  *
415  * RETURNS:
416  * 0 on success, allocated dev_t is returned in *@devt.  -errno on
417  * failure.
418  *
419  * CONTEXT:
420  * Might sleep.
421  */
422 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
423 {
424 	struct gendisk *disk = part_to_disk(part);
425 	int idx;
426 
427 	/* in consecutive minor range? */
428 	if (part->partno < disk->minors) {
429 		*devt = MKDEV(disk->major, disk->first_minor + part->partno);
430 		return 0;
431 	}
432 
433 	/* allocate ext devt */
434 	idr_preload(GFP_KERNEL);
435 
436 	spin_lock_bh(&ext_devt_lock);
437 	idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
438 	spin_unlock_bh(&ext_devt_lock);
439 
440 	idr_preload_end();
441 	if (idx < 0)
442 		return idx == -ENOSPC ? -EBUSY : idx;
443 
444 	*devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
445 	return 0;
446 }
447 
448 /**
449  * blk_free_devt - free a dev_t
450  * @devt: dev_t to free
451  *
452  * Free @devt which was allocated using blk_alloc_devt().
453  *
454  * CONTEXT:
455  * Might sleep.
456  */
457 void blk_free_devt(dev_t devt)
458 {
459 	if (devt == MKDEV(0, 0))
460 		return;
461 
462 	if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
463 		spin_lock_bh(&ext_devt_lock);
464 		idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
465 		spin_unlock_bh(&ext_devt_lock);
466 	}
467 }
468 
469 static char *bdevt_str(dev_t devt, char *buf)
470 {
471 	if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
472 		char tbuf[BDEVT_SIZE];
473 		snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
474 		snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
475 	} else
476 		snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
477 
478 	return buf;
479 }
480 
481 /*
482  * Register device numbers dev..(dev+range-1)
483  * range must be nonzero
484  * The hash chain is sorted on range, so that subranges can override.
485  */
486 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
487 			 struct kobject *(*probe)(dev_t, int *, void *),
488 			 int (*lock)(dev_t, void *), void *data)
489 {
490 	kobj_map(bdev_map, devt, range, module, probe, lock, data);
491 }
492 
493 EXPORT_SYMBOL(blk_register_region);
494 
495 void blk_unregister_region(dev_t devt, unsigned long range)
496 {
497 	kobj_unmap(bdev_map, devt, range);
498 }
499 
500 EXPORT_SYMBOL(blk_unregister_region);
501 
502 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
503 {
504 	struct gendisk *p = data;
505 
506 	return &disk_to_dev(p)->kobj;
507 }
508 
509 static int exact_lock(dev_t devt, void *data)
510 {
511 	struct gendisk *p = data;
512 
513 	if (!get_disk(p))
514 		return -1;
515 	return 0;
516 }
517 
518 static void register_disk(struct device *parent, struct gendisk *disk)
519 {
520 	struct device *ddev = disk_to_dev(disk);
521 	struct block_device *bdev;
522 	struct disk_part_iter piter;
523 	struct hd_struct *part;
524 	int err;
525 
526 	ddev->parent = parent;
527 
528 	dev_set_name(ddev, "%s", disk->disk_name);
529 
530 	/* delay uevents, until we scanned partition table */
531 	dev_set_uevent_suppress(ddev, 1);
532 
533 	if (device_add(ddev))
534 		return;
535 	if (!sysfs_deprecated) {
536 		err = sysfs_create_link(block_depr, &ddev->kobj,
537 					kobject_name(&ddev->kobj));
538 		if (err) {
539 			device_del(ddev);
540 			return;
541 		}
542 	}
543 
544 	/*
545 	 * avoid probable deadlock caused by allocating memory with
546 	 * GFP_KERNEL in runtime_resume callback of its all ancestor
547 	 * devices
548 	 */
549 	pm_runtime_set_memalloc_noio(ddev, true);
550 
551 	disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
552 	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
553 
554 	/* No minors to use for partitions */
555 	if (!disk_part_scan_enabled(disk))
556 		goto exit;
557 
558 	/* No such device (e.g., media were just removed) */
559 	if (!get_capacity(disk))
560 		goto exit;
561 
562 	bdev = bdget_disk(disk, 0);
563 	if (!bdev)
564 		goto exit;
565 
566 	bdev->bd_invalidated = 1;
567 	err = blkdev_get(bdev, FMODE_READ, NULL);
568 	if (err < 0)
569 		goto exit;
570 	blkdev_put(bdev, FMODE_READ);
571 
572 exit:
573 	/* announce disk after possible partitions are created */
574 	dev_set_uevent_suppress(ddev, 0);
575 	kobject_uevent(&ddev->kobj, KOBJ_ADD);
576 
577 	/* announce possible partitions */
578 	disk_part_iter_init(&piter, disk, 0);
579 	while ((part = disk_part_iter_next(&piter)))
580 		kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
581 	disk_part_iter_exit(&piter);
582 }
583 
584 /**
585  * device_add_disk - add partitioning information to kernel list
586  * @parent: parent device for the disk
587  * @disk: per-device partitioning information
588  *
589  * This function registers the partitioning information in @disk
590  * with the kernel.
591  *
592  * FIXME: error handling
593  */
594 void device_add_disk(struct device *parent, struct gendisk *disk)
595 {
596 	struct backing_dev_info *bdi;
597 	dev_t devt;
598 	int retval;
599 
600 	/* minors == 0 indicates to use ext devt from part0 and should
601 	 * be accompanied with EXT_DEVT flag.  Make sure all
602 	 * parameters make sense.
603 	 */
604 	WARN_ON(disk->minors && !(disk->major || disk->first_minor));
605 	WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
606 
607 	disk->flags |= GENHD_FL_UP;
608 
609 	retval = blk_alloc_devt(&disk->part0, &devt);
610 	if (retval) {
611 		WARN_ON(1);
612 		return;
613 	}
614 	disk_to_dev(disk)->devt = devt;
615 
616 	/* ->major and ->first_minor aren't supposed to be
617 	 * dereferenced from here on, but set them just in case.
618 	 */
619 	disk->major = MAJOR(devt);
620 	disk->first_minor = MINOR(devt);
621 
622 	disk_alloc_events(disk);
623 
624 	/* Register BDI before referencing it from bdev */
625 	bdi = disk->queue->backing_dev_info;
626 	bdi_register_owner(bdi, disk_to_dev(disk));
627 
628 	blk_register_region(disk_devt(disk), disk->minors, NULL,
629 			    exact_match, exact_lock, disk);
630 	register_disk(parent, disk);
631 	blk_register_queue(disk);
632 
633 	/*
634 	 * Take an extra ref on queue which will be put on disk_release()
635 	 * so that it sticks around as long as @disk is there.
636 	 */
637 	WARN_ON_ONCE(!blk_get_queue(disk->queue));
638 
639 	retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
640 				   "bdi");
641 	WARN_ON(retval);
642 
643 	disk_add_events(disk);
644 	blk_integrity_add(disk);
645 }
646 EXPORT_SYMBOL(device_add_disk);
647 
648 void del_gendisk(struct gendisk *disk)
649 {
650 	struct disk_part_iter piter;
651 	struct hd_struct *part;
652 
653 	blk_integrity_del(disk);
654 	disk_del_events(disk);
655 
656 	/* invalidate stuff */
657 	disk_part_iter_init(&piter, disk,
658 			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
659 	while ((part = disk_part_iter_next(&piter))) {
660 		invalidate_partition(disk, part->partno);
661 		bdev_unhash_inode(part_devt(part));
662 		delete_partition(disk, part->partno);
663 	}
664 	disk_part_iter_exit(&piter);
665 
666 	invalidate_partition(disk, 0);
667 	bdev_unhash_inode(disk_devt(disk));
668 	set_capacity(disk, 0);
669 	disk->flags &= ~GENHD_FL_UP;
670 
671 	sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
672 	if (disk->queue) {
673 		/*
674 		 * Unregister bdi before releasing device numbers (as they can
675 		 * get reused and we'd get clashes in sysfs).
676 		 */
677 		bdi_unregister(disk->queue->backing_dev_info);
678 		blk_unregister_queue(disk);
679 	} else {
680 		WARN_ON(1);
681 	}
682 	blk_unregister_region(disk_devt(disk), disk->minors);
683 
684 	part_stat_set_all(&disk->part0, 0);
685 	disk->part0.stamp = 0;
686 
687 	kobject_put(disk->part0.holder_dir);
688 	kobject_put(disk->slave_dir);
689 	if (!sysfs_deprecated)
690 		sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
691 	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
692 	device_del(disk_to_dev(disk));
693 }
694 EXPORT_SYMBOL(del_gendisk);
695 
696 /* sysfs access to bad-blocks list. */
697 static ssize_t disk_badblocks_show(struct device *dev,
698 					struct device_attribute *attr,
699 					char *page)
700 {
701 	struct gendisk *disk = dev_to_disk(dev);
702 
703 	if (!disk->bb)
704 		return sprintf(page, "\n");
705 
706 	return badblocks_show(disk->bb, page, 0);
707 }
708 
709 static ssize_t disk_badblocks_store(struct device *dev,
710 					struct device_attribute *attr,
711 					const char *page, size_t len)
712 {
713 	struct gendisk *disk = dev_to_disk(dev);
714 
715 	if (!disk->bb)
716 		return -ENXIO;
717 
718 	return badblocks_store(disk->bb, page, len, 0);
719 }
720 
721 /**
722  * get_gendisk - get partitioning information for a given device
723  * @devt: device to get partitioning information for
724  * @partno: returned partition index
725  *
726  * This function gets the structure containing partitioning
727  * information for the given device @devt.
728  */
729 struct gendisk *get_gendisk(dev_t devt, int *partno)
730 {
731 	struct gendisk *disk = NULL;
732 
733 	if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
734 		struct kobject *kobj;
735 
736 		kobj = kobj_lookup(bdev_map, devt, partno);
737 		if (kobj)
738 			disk = dev_to_disk(kobj_to_dev(kobj));
739 	} else {
740 		struct hd_struct *part;
741 
742 		spin_lock_bh(&ext_devt_lock);
743 		part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
744 		if (part && get_disk(part_to_disk(part))) {
745 			*partno = part->partno;
746 			disk = part_to_disk(part);
747 		}
748 		spin_unlock_bh(&ext_devt_lock);
749 	}
750 
751 	return disk;
752 }
753 EXPORT_SYMBOL(get_gendisk);
754 
755 /**
756  * bdget_disk - do bdget() by gendisk and partition number
757  * @disk: gendisk of interest
758  * @partno: partition number
759  *
760  * Find partition @partno from @disk, do bdget() on it.
761  *
762  * CONTEXT:
763  * Don't care.
764  *
765  * RETURNS:
766  * Resulting block_device on success, NULL on failure.
767  */
768 struct block_device *bdget_disk(struct gendisk *disk, int partno)
769 {
770 	struct hd_struct *part;
771 	struct block_device *bdev = NULL;
772 
773 	part = disk_get_part(disk, partno);
774 	if (part)
775 		bdev = bdget(part_devt(part));
776 	disk_put_part(part);
777 
778 	return bdev;
779 }
780 EXPORT_SYMBOL(bdget_disk);
781 
782 /*
783  * print a full list of all partitions - intended for places where the root
784  * filesystem can't be mounted and thus to give the victim some idea of what
785  * went wrong
786  */
787 void __init printk_all_partitions(void)
788 {
789 	struct class_dev_iter iter;
790 	struct device *dev;
791 
792 	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
793 	while ((dev = class_dev_iter_next(&iter))) {
794 		struct gendisk *disk = dev_to_disk(dev);
795 		struct disk_part_iter piter;
796 		struct hd_struct *part;
797 		char name_buf[BDEVNAME_SIZE];
798 		char devt_buf[BDEVT_SIZE];
799 
800 		/*
801 		 * Don't show empty devices or things that have been
802 		 * suppressed
803 		 */
804 		if (get_capacity(disk) == 0 ||
805 		    (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
806 			continue;
807 
808 		/*
809 		 * Note, unlike /proc/partitions, I am showing the
810 		 * numbers in hex - the same format as the root=
811 		 * option takes.
812 		 */
813 		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
814 		while ((part = disk_part_iter_next(&piter))) {
815 			bool is_part0 = part == &disk->part0;
816 
817 			printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
818 			       bdevt_str(part_devt(part), devt_buf),
819 			       (unsigned long long)part_nr_sects_read(part) >> 1
820 			       , disk_name(disk, part->partno, name_buf),
821 			       part->info ? part->info->uuid : "");
822 			if (is_part0) {
823 				if (dev->parent && dev->parent->driver)
824 					printk(" driver: %s\n",
825 					      dev->parent->driver->name);
826 				else
827 					printk(" (driver?)\n");
828 			} else
829 				printk("\n");
830 		}
831 		disk_part_iter_exit(&piter);
832 	}
833 	class_dev_iter_exit(&iter);
834 }
835 
836 #ifdef CONFIG_PROC_FS
837 /* iterator */
838 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
839 {
840 	loff_t skip = *pos;
841 	struct class_dev_iter *iter;
842 	struct device *dev;
843 
844 	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
845 	if (!iter)
846 		return ERR_PTR(-ENOMEM);
847 
848 	seqf->private = iter;
849 	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
850 	do {
851 		dev = class_dev_iter_next(iter);
852 		if (!dev)
853 			return NULL;
854 	} while (skip--);
855 
856 	return dev_to_disk(dev);
857 }
858 
859 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
860 {
861 	struct device *dev;
862 
863 	(*pos)++;
864 	dev = class_dev_iter_next(seqf->private);
865 	if (dev)
866 		return dev_to_disk(dev);
867 
868 	return NULL;
869 }
870 
871 static void disk_seqf_stop(struct seq_file *seqf, void *v)
872 {
873 	struct class_dev_iter *iter = seqf->private;
874 
875 	/* stop is called even after start failed :-( */
876 	if (iter) {
877 		class_dev_iter_exit(iter);
878 		kfree(iter);
879 		seqf->private = NULL;
880 	}
881 }
882 
883 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
884 {
885 	void *p;
886 
887 	p = disk_seqf_start(seqf, pos);
888 	if (!IS_ERR_OR_NULL(p) && !*pos)
889 		seq_puts(seqf, "major minor  #blocks  name\n\n");
890 	return p;
891 }
892 
893 static int show_partition(struct seq_file *seqf, void *v)
894 {
895 	struct gendisk *sgp = v;
896 	struct disk_part_iter piter;
897 	struct hd_struct *part;
898 	char buf[BDEVNAME_SIZE];
899 
900 	/* Don't show non-partitionable removeable devices or empty devices */
901 	if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
902 				   (sgp->flags & GENHD_FL_REMOVABLE)))
903 		return 0;
904 	if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
905 		return 0;
906 
907 	/* show the full disk and all non-0 size partitions of it */
908 	disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
909 	while ((part = disk_part_iter_next(&piter)))
910 		seq_printf(seqf, "%4d  %7d %10llu %s\n",
911 			   MAJOR(part_devt(part)), MINOR(part_devt(part)),
912 			   (unsigned long long)part_nr_sects_read(part) >> 1,
913 			   disk_name(sgp, part->partno, buf));
914 	disk_part_iter_exit(&piter);
915 
916 	return 0;
917 }
918 
919 static const struct seq_operations partitions_op = {
920 	.start	= show_partition_start,
921 	.next	= disk_seqf_next,
922 	.stop	= disk_seqf_stop,
923 	.show	= show_partition
924 };
925 
926 static int partitions_open(struct inode *inode, struct file *file)
927 {
928 	return seq_open(file, &partitions_op);
929 }
930 
931 static const struct file_operations proc_partitions_operations = {
932 	.open		= partitions_open,
933 	.read		= seq_read,
934 	.llseek		= seq_lseek,
935 	.release	= seq_release,
936 };
937 #endif
938 
939 
940 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
941 {
942 	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
943 		/* Make old-style 2.4 aliases work */
944 		request_module("block-major-%d", MAJOR(devt));
945 	return NULL;
946 }
947 
948 static int __init genhd_device_init(void)
949 {
950 	int error;
951 
952 	block_class.dev_kobj = sysfs_dev_block_kobj;
953 	error = class_register(&block_class);
954 	if (unlikely(error))
955 		return error;
956 	bdev_map = kobj_map_init(base_probe, &block_class_lock);
957 	blk_dev_init();
958 
959 	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
960 
961 	/* create top-level block dir */
962 	if (!sysfs_deprecated)
963 		block_depr = kobject_create_and_add("block", NULL);
964 	return 0;
965 }
966 
967 subsys_initcall(genhd_device_init);
968 
969 static ssize_t disk_range_show(struct device *dev,
970 			       struct device_attribute *attr, char *buf)
971 {
972 	struct gendisk *disk = dev_to_disk(dev);
973 
974 	return sprintf(buf, "%d\n", disk->minors);
975 }
976 
977 static ssize_t disk_ext_range_show(struct device *dev,
978 				   struct device_attribute *attr, char *buf)
979 {
980 	struct gendisk *disk = dev_to_disk(dev);
981 
982 	return sprintf(buf, "%d\n", disk_max_parts(disk));
983 }
984 
985 static ssize_t disk_removable_show(struct device *dev,
986 				   struct device_attribute *attr, char *buf)
987 {
988 	struct gendisk *disk = dev_to_disk(dev);
989 
990 	return sprintf(buf, "%d\n",
991 		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
992 }
993 
994 static ssize_t disk_ro_show(struct device *dev,
995 				   struct device_attribute *attr, char *buf)
996 {
997 	struct gendisk *disk = dev_to_disk(dev);
998 
999 	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1000 }
1001 
1002 static ssize_t disk_capability_show(struct device *dev,
1003 				    struct device_attribute *attr, char *buf)
1004 {
1005 	struct gendisk *disk = dev_to_disk(dev);
1006 
1007 	return sprintf(buf, "%x\n", disk->flags);
1008 }
1009 
1010 static ssize_t disk_alignment_offset_show(struct device *dev,
1011 					  struct device_attribute *attr,
1012 					  char *buf)
1013 {
1014 	struct gendisk *disk = dev_to_disk(dev);
1015 
1016 	return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1017 }
1018 
1019 static ssize_t disk_discard_alignment_show(struct device *dev,
1020 					   struct device_attribute *attr,
1021 					   char *buf)
1022 {
1023 	struct gendisk *disk = dev_to_disk(dev);
1024 
1025 	return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1026 }
1027 
1028 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
1029 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
1030 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
1031 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
1032 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
1033 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
1034 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
1035 		   NULL);
1036 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
1037 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
1038 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
1039 static DEVICE_ATTR(badblocks, S_IRUGO | S_IWUSR, disk_badblocks_show,
1040 		disk_badblocks_store);
1041 #ifdef CONFIG_FAIL_MAKE_REQUEST
1042 static struct device_attribute dev_attr_fail =
1043 	__ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
1044 #endif
1045 #ifdef CONFIG_FAIL_IO_TIMEOUT
1046 static struct device_attribute dev_attr_fail_timeout =
1047 	__ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
1048 		part_timeout_store);
1049 #endif
1050 
1051 static struct attribute *disk_attrs[] = {
1052 	&dev_attr_range.attr,
1053 	&dev_attr_ext_range.attr,
1054 	&dev_attr_removable.attr,
1055 	&dev_attr_ro.attr,
1056 	&dev_attr_size.attr,
1057 	&dev_attr_alignment_offset.attr,
1058 	&dev_attr_discard_alignment.attr,
1059 	&dev_attr_capability.attr,
1060 	&dev_attr_stat.attr,
1061 	&dev_attr_inflight.attr,
1062 	&dev_attr_badblocks.attr,
1063 #ifdef CONFIG_FAIL_MAKE_REQUEST
1064 	&dev_attr_fail.attr,
1065 #endif
1066 #ifdef CONFIG_FAIL_IO_TIMEOUT
1067 	&dev_attr_fail_timeout.attr,
1068 #endif
1069 	NULL
1070 };
1071 
1072 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1073 {
1074 	struct device *dev = container_of(kobj, typeof(*dev), kobj);
1075 	struct gendisk *disk = dev_to_disk(dev);
1076 
1077 	if (a == &dev_attr_badblocks.attr && !disk->bb)
1078 		return 0;
1079 	return a->mode;
1080 }
1081 
1082 static struct attribute_group disk_attr_group = {
1083 	.attrs = disk_attrs,
1084 	.is_visible = disk_visible,
1085 };
1086 
1087 static const struct attribute_group *disk_attr_groups[] = {
1088 	&disk_attr_group,
1089 	NULL
1090 };
1091 
1092 /**
1093  * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1094  * @disk: disk to replace part_tbl for
1095  * @new_ptbl: new part_tbl to install
1096  *
1097  * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
1098  * original ptbl is freed using RCU callback.
1099  *
1100  * LOCKING:
1101  * Matching bd_mutx locked.
1102  */
1103 static void disk_replace_part_tbl(struct gendisk *disk,
1104 				  struct disk_part_tbl *new_ptbl)
1105 {
1106 	struct disk_part_tbl *old_ptbl = disk->part_tbl;
1107 
1108 	rcu_assign_pointer(disk->part_tbl, new_ptbl);
1109 
1110 	if (old_ptbl) {
1111 		rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1112 		kfree_rcu(old_ptbl, rcu_head);
1113 	}
1114 }
1115 
1116 /**
1117  * disk_expand_part_tbl - expand disk->part_tbl
1118  * @disk: disk to expand part_tbl for
1119  * @partno: expand such that this partno can fit in
1120  *
1121  * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
1122  * uses RCU to allow unlocked dereferencing for stats and other stuff.
1123  *
1124  * LOCKING:
1125  * Matching bd_mutex locked, might sleep.
1126  *
1127  * RETURNS:
1128  * 0 on success, -errno on failure.
1129  */
1130 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1131 {
1132 	struct disk_part_tbl *old_ptbl = disk->part_tbl;
1133 	struct disk_part_tbl *new_ptbl;
1134 	int len = old_ptbl ? old_ptbl->len : 0;
1135 	int i, target;
1136 	size_t size;
1137 
1138 	/*
1139 	 * check for int overflow, since we can get here from blkpg_ioctl()
1140 	 * with a user passed 'partno'.
1141 	 */
1142 	target = partno + 1;
1143 	if (target < 0)
1144 		return -EINVAL;
1145 
1146 	/* disk_max_parts() is zero during initialization, ignore if so */
1147 	if (disk_max_parts(disk) && target > disk_max_parts(disk))
1148 		return -EINVAL;
1149 
1150 	if (target <= len)
1151 		return 0;
1152 
1153 	size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1154 	new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1155 	if (!new_ptbl)
1156 		return -ENOMEM;
1157 
1158 	new_ptbl->len = target;
1159 
1160 	for (i = 0; i < len; i++)
1161 		rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1162 
1163 	disk_replace_part_tbl(disk, new_ptbl);
1164 	return 0;
1165 }
1166 
1167 static void disk_release(struct device *dev)
1168 {
1169 	struct gendisk *disk = dev_to_disk(dev);
1170 
1171 	blk_free_devt(dev->devt);
1172 	disk_release_events(disk);
1173 	kfree(disk->random);
1174 	disk_replace_part_tbl(disk, NULL);
1175 	hd_free_part(&disk->part0);
1176 	if (disk->queue)
1177 		blk_put_queue(disk->queue);
1178 	kfree(disk);
1179 }
1180 struct class block_class = {
1181 	.name		= "block",
1182 };
1183 
1184 static char *block_devnode(struct device *dev, umode_t *mode,
1185 			   kuid_t *uid, kgid_t *gid)
1186 {
1187 	struct gendisk *disk = dev_to_disk(dev);
1188 
1189 	if (disk->devnode)
1190 		return disk->devnode(disk, mode);
1191 	return NULL;
1192 }
1193 
1194 static const struct device_type disk_type = {
1195 	.name		= "disk",
1196 	.groups		= disk_attr_groups,
1197 	.release	= disk_release,
1198 	.devnode	= block_devnode,
1199 };
1200 
1201 #ifdef CONFIG_PROC_FS
1202 /*
1203  * aggregate disk stat collector.  Uses the same stats that the sysfs
1204  * entries do, above, but makes them available through one seq_file.
1205  *
1206  * The output looks suspiciously like /proc/partitions with a bunch of
1207  * extra fields.
1208  */
1209 static int diskstats_show(struct seq_file *seqf, void *v)
1210 {
1211 	struct gendisk *gp = v;
1212 	struct disk_part_iter piter;
1213 	struct hd_struct *hd;
1214 	char buf[BDEVNAME_SIZE];
1215 	int cpu;
1216 
1217 	/*
1218 	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1219 		seq_puts(seqf,	"major minor name"
1220 				"     rio rmerge rsect ruse wio wmerge "
1221 				"wsect wuse running use aveq"
1222 				"\n\n");
1223 	*/
1224 
1225 	disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1226 	while ((hd = disk_part_iter_next(&piter))) {
1227 		cpu = part_stat_lock();
1228 		part_round_stats(cpu, hd);
1229 		part_stat_unlock();
1230 		seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1231 			   "%u %lu %lu %lu %u %u %u %u\n",
1232 			   MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1233 			   disk_name(gp, hd->partno, buf),
1234 			   part_stat_read(hd, ios[READ]),
1235 			   part_stat_read(hd, merges[READ]),
1236 			   part_stat_read(hd, sectors[READ]),
1237 			   jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1238 			   part_stat_read(hd, ios[WRITE]),
1239 			   part_stat_read(hd, merges[WRITE]),
1240 			   part_stat_read(hd, sectors[WRITE]),
1241 			   jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1242 			   part_in_flight(hd),
1243 			   jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1244 			   jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1245 			);
1246 	}
1247 	disk_part_iter_exit(&piter);
1248 
1249 	return 0;
1250 }
1251 
1252 static const struct seq_operations diskstats_op = {
1253 	.start	= disk_seqf_start,
1254 	.next	= disk_seqf_next,
1255 	.stop	= disk_seqf_stop,
1256 	.show	= diskstats_show
1257 };
1258 
1259 static int diskstats_open(struct inode *inode, struct file *file)
1260 {
1261 	return seq_open(file, &diskstats_op);
1262 }
1263 
1264 static const struct file_operations proc_diskstats_operations = {
1265 	.open		= diskstats_open,
1266 	.read		= seq_read,
1267 	.llseek		= seq_lseek,
1268 	.release	= seq_release,
1269 };
1270 
1271 static int __init proc_genhd_init(void)
1272 {
1273 	proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1274 	proc_create("partitions", 0, NULL, &proc_partitions_operations);
1275 	return 0;
1276 }
1277 module_init(proc_genhd_init);
1278 #endif /* CONFIG_PROC_FS */
1279 
1280 dev_t blk_lookup_devt(const char *name, int partno)
1281 {
1282 	dev_t devt = MKDEV(0, 0);
1283 	struct class_dev_iter iter;
1284 	struct device *dev;
1285 
1286 	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1287 	while ((dev = class_dev_iter_next(&iter))) {
1288 		struct gendisk *disk = dev_to_disk(dev);
1289 		struct hd_struct *part;
1290 
1291 		if (strcmp(dev_name(dev), name))
1292 			continue;
1293 
1294 		if (partno < disk->minors) {
1295 			/* We need to return the right devno, even
1296 			 * if the partition doesn't exist yet.
1297 			 */
1298 			devt = MKDEV(MAJOR(dev->devt),
1299 				     MINOR(dev->devt) + partno);
1300 			break;
1301 		}
1302 		part = disk_get_part(disk, partno);
1303 		if (part) {
1304 			devt = part_devt(part);
1305 			disk_put_part(part);
1306 			break;
1307 		}
1308 		disk_put_part(part);
1309 	}
1310 	class_dev_iter_exit(&iter);
1311 	return devt;
1312 }
1313 EXPORT_SYMBOL(blk_lookup_devt);
1314 
1315 struct gendisk *alloc_disk(int minors)
1316 {
1317 	return alloc_disk_node(minors, NUMA_NO_NODE);
1318 }
1319 EXPORT_SYMBOL(alloc_disk);
1320 
1321 struct gendisk *alloc_disk_node(int minors, int node_id)
1322 {
1323 	struct gendisk *disk;
1324 
1325 	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1326 	if (disk) {
1327 		if (!init_part_stats(&disk->part0)) {
1328 			kfree(disk);
1329 			return NULL;
1330 		}
1331 		disk->node_id = node_id;
1332 		if (disk_expand_part_tbl(disk, 0)) {
1333 			free_part_stats(&disk->part0);
1334 			kfree(disk);
1335 			return NULL;
1336 		}
1337 		disk->part_tbl->part[0] = &disk->part0;
1338 
1339 		/*
1340 		 * set_capacity() and get_capacity() currently don't use
1341 		 * seqcounter to read/update the part0->nr_sects. Still init
1342 		 * the counter as we can read the sectors in IO submission
1343 		 * patch using seqence counters.
1344 		 *
1345 		 * TODO: Ideally set_capacity() and get_capacity() should be
1346 		 * converted to make use of bd_mutex and sequence counters.
1347 		 */
1348 		seqcount_init(&disk->part0.nr_sects_seq);
1349 		if (hd_ref_init(&disk->part0)) {
1350 			hd_free_part(&disk->part0);
1351 			kfree(disk);
1352 			return NULL;
1353 		}
1354 
1355 		disk->minors = minors;
1356 		rand_initialize_disk(disk);
1357 		disk_to_dev(disk)->class = &block_class;
1358 		disk_to_dev(disk)->type = &disk_type;
1359 		device_initialize(disk_to_dev(disk));
1360 	}
1361 	return disk;
1362 }
1363 EXPORT_SYMBOL(alloc_disk_node);
1364 
1365 struct kobject *get_disk(struct gendisk *disk)
1366 {
1367 	struct module *owner;
1368 	struct kobject *kobj;
1369 
1370 	if (!disk->fops)
1371 		return NULL;
1372 	owner = disk->fops->owner;
1373 	if (owner && !try_module_get(owner))
1374 		return NULL;
1375 	kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1376 	if (kobj == NULL) {
1377 		module_put(owner);
1378 		return NULL;
1379 	}
1380 	return kobj;
1381 
1382 }
1383 
1384 EXPORT_SYMBOL(get_disk);
1385 
1386 void put_disk(struct gendisk *disk)
1387 {
1388 	if (disk)
1389 		kobject_put(&disk_to_dev(disk)->kobj);
1390 }
1391 
1392 EXPORT_SYMBOL(put_disk);
1393 
1394 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1395 {
1396 	char event[] = "DISK_RO=1";
1397 	char *envp[] = { event, NULL };
1398 
1399 	if (!ro)
1400 		event[8] = '0';
1401 	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1402 }
1403 
1404 void set_device_ro(struct block_device *bdev, int flag)
1405 {
1406 	bdev->bd_part->policy = flag;
1407 }
1408 
1409 EXPORT_SYMBOL(set_device_ro);
1410 
1411 void set_disk_ro(struct gendisk *disk, int flag)
1412 {
1413 	struct disk_part_iter piter;
1414 	struct hd_struct *part;
1415 
1416 	if (disk->part0.policy != flag) {
1417 		set_disk_ro_uevent(disk, flag);
1418 		disk->part0.policy = flag;
1419 	}
1420 
1421 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1422 	while ((part = disk_part_iter_next(&piter)))
1423 		part->policy = flag;
1424 	disk_part_iter_exit(&piter);
1425 }
1426 
1427 EXPORT_SYMBOL(set_disk_ro);
1428 
1429 int bdev_read_only(struct block_device *bdev)
1430 {
1431 	if (!bdev)
1432 		return 0;
1433 	return bdev->bd_part->policy;
1434 }
1435 
1436 EXPORT_SYMBOL(bdev_read_only);
1437 
1438 int invalidate_partition(struct gendisk *disk, int partno)
1439 {
1440 	int res = 0;
1441 	struct block_device *bdev = bdget_disk(disk, partno);
1442 	if (bdev) {
1443 		fsync_bdev(bdev);
1444 		res = __invalidate_device(bdev, true);
1445 		bdput(bdev);
1446 	}
1447 	return res;
1448 }
1449 
1450 EXPORT_SYMBOL(invalidate_partition);
1451 
1452 /*
1453  * Disk events - monitor disk events like media change and eject request.
1454  */
1455 struct disk_events {
1456 	struct list_head	node;		/* all disk_event's */
1457 	struct gendisk		*disk;		/* the associated disk */
1458 	spinlock_t		lock;
1459 
1460 	struct mutex		block_mutex;	/* protects blocking */
1461 	int			block;		/* event blocking depth */
1462 	unsigned int		pending;	/* events already sent out */
1463 	unsigned int		clearing;	/* events being cleared */
1464 
1465 	long			poll_msecs;	/* interval, -1 for default */
1466 	struct delayed_work	dwork;
1467 };
1468 
1469 static const char *disk_events_strs[] = {
1470 	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "media_change",
1471 	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "eject_request",
1472 };
1473 
1474 static char *disk_uevents[] = {
1475 	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "DISK_MEDIA_CHANGE=1",
1476 	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "DISK_EJECT_REQUEST=1",
1477 };
1478 
1479 /* list of all disk_events */
1480 static DEFINE_MUTEX(disk_events_mutex);
1481 static LIST_HEAD(disk_events);
1482 
1483 /* disable in-kernel polling by default */
1484 static unsigned long disk_events_dfl_poll_msecs;
1485 
1486 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1487 {
1488 	struct disk_events *ev = disk->ev;
1489 	long intv_msecs = 0;
1490 
1491 	/*
1492 	 * If device-specific poll interval is set, always use it.  If
1493 	 * the default is being used, poll iff there are events which
1494 	 * can't be monitored asynchronously.
1495 	 */
1496 	if (ev->poll_msecs >= 0)
1497 		intv_msecs = ev->poll_msecs;
1498 	else if (disk->events & ~disk->async_events)
1499 		intv_msecs = disk_events_dfl_poll_msecs;
1500 
1501 	return msecs_to_jiffies(intv_msecs);
1502 }
1503 
1504 /**
1505  * disk_block_events - block and flush disk event checking
1506  * @disk: disk to block events for
1507  *
1508  * On return from this function, it is guaranteed that event checking
1509  * isn't in progress and won't happen until unblocked by
1510  * disk_unblock_events().  Events blocking is counted and the actual
1511  * unblocking happens after the matching number of unblocks are done.
1512  *
1513  * Note that this intentionally does not block event checking from
1514  * disk_clear_events().
1515  *
1516  * CONTEXT:
1517  * Might sleep.
1518  */
1519 void disk_block_events(struct gendisk *disk)
1520 {
1521 	struct disk_events *ev = disk->ev;
1522 	unsigned long flags;
1523 	bool cancel;
1524 
1525 	if (!ev)
1526 		return;
1527 
1528 	/*
1529 	 * Outer mutex ensures that the first blocker completes canceling
1530 	 * the event work before further blockers are allowed to finish.
1531 	 */
1532 	mutex_lock(&ev->block_mutex);
1533 
1534 	spin_lock_irqsave(&ev->lock, flags);
1535 	cancel = !ev->block++;
1536 	spin_unlock_irqrestore(&ev->lock, flags);
1537 
1538 	if (cancel)
1539 		cancel_delayed_work_sync(&disk->ev->dwork);
1540 
1541 	mutex_unlock(&ev->block_mutex);
1542 }
1543 
1544 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1545 {
1546 	struct disk_events *ev = disk->ev;
1547 	unsigned long intv;
1548 	unsigned long flags;
1549 
1550 	spin_lock_irqsave(&ev->lock, flags);
1551 
1552 	if (WARN_ON_ONCE(ev->block <= 0))
1553 		goto out_unlock;
1554 
1555 	if (--ev->block)
1556 		goto out_unlock;
1557 
1558 	intv = disk_events_poll_jiffies(disk);
1559 	if (check_now)
1560 		queue_delayed_work(system_freezable_power_efficient_wq,
1561 				&ev->dwork, 0);
1562 	else if (intv)
1563 		queue_delayed_work(system_freezable_power_efficient_wq,
1564 				&ev->dwork, intv);
1565 out_unlock:
1566 	spin_unlock_irqrestore(&ev->lock, flags);
1567 }
1568 
1569 /**
1570  * disk_unblock_events - unblock disk event checking
1571  * @disk: disk to unblock events for
1572  *
1573  * Undo disk_block_events().  When the block count reaches zero, it
1574  * starts events polling if configured.
1575  *
1576  * CONTEXT:
1577  * Don't care.  Safe to call from irq context.
1578  */
1579 void disk_unblock_events(struct gendisk *disk)
1580 {
1581 	if (disk->ev)
1582 		__disk_unblock_events(disk, false);
1583 }
1584 
1585 /**
1586  * disk_flush_events - schedule immediate event checking and flushing
1587  * @disk: disk to check and flush events for
1588  * @mask: events to flush
1589  *
1590  * Schedule immediate event checking on @disk if not blocked.  Events in
1591  * @mask are scheduled to be cleared from the driver.  Note that this
1592  * doesn't clear the events from @disk->ev.
1593  *
1594  * CONTEXT:
1595  * If @mask is non-zero must be called with bdev->bd_mutex held.
1596  */
1597 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1598 {
1599 	struct disk_events *ev = disk->ev;
1600 
1601 	if (!ev)
1602 		return;
1603 
1604 	spin_lock_irq(&ev->lock);
1605 	ev->clearing |= mask;
1606 	if (!ev->block)
1607 		mod_delayed_work(system_freezable_power_efficient_wq,
1608 				&ev->dwork, 0);
1609 	spin_unlock_irq(&ev->lock);
1610 }
1611 
1612 /**
1613  * disk_clear_events - synchronously check, clear and return pending events
1614  * @disk: disk to fetch and clear events from
1615  * @mask: mask of events to be fetched and cleared
1616  *
1617  * Disk events are synchronously checked and pending events in @mask
1618  * are cleared and returned.  This ignores the block count.
1619  *
1620  * CONTEXT:
1621  * Might sleep.
1622  */
1623 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1624 {
1625 	const struct block_device_operations *bdops = disk->fops;
1626 	struct disk_events *ev = disk->ev;
1627 	unsigned int pending;
1628 	unsigned int clearing = mask;
1629 
1630 	if (!ev) {
1631 		/* for drivers still using the old ->media_changed method */
1632 		if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1633 		    bdops->media_changed && bdops->media_changed(disk))
1634 			return DISK_EVENT_MEDIA_CHANGE;
1635 		return 0;
1636 	}
1637 
1638 	disk_block_events(disk);
1639 
1640 	/*
1641 	 * store the union of mask and ev->clearing on the stack so that the
1642 	 * race with disk_flush_events does not cause ambiguity (ev->clearing
1643 	 * can still be modified even if events are blocked).
1644 	 */
1645 	spin_lock_irq(&ev->lock);
1646 	clearing |= ev->clearing;
1647 	ev->clearing = 0;
1648 	spin_unlock_irq(&ev->lock);
1649 
1650 	disk_check_events(ev, &clearing);
1651 	/*
1652 	 * if ev->clearing is not 0, the disk_flush_events got called in the
1653 	 * middle of this function, so we want to run the workfn without delay.
1654 	 */
1655 	__disk_unblock_events(disk, ev->clearing ? true : false);
1656 
1657 	/* then, fetch and clear pending events */
1658 	spin_lock_irq(&ev->lock);
1659 	pending = ev->pending & mask;
1660 	ev->pending &= ~mask;
1661 	spin_unlock_irq(&ev->lock);
1662 	WARN_ON_ONCE(clearing & mask);
1663 
1664 	return pending;
1665 }
1666 
1667 /*
1668  * Separate this part out so that a different pointer for clearing_ptr can be
1669  * passed in for disk_clear_events.
1670  */
1671 static void disk_events_workfn(struct work_struct *work)
1672 {
1673 	struct delayed_work *dwork = to_delayed_work(work);
1674 	struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1675 
1676 	disk_check_events(ev, &ev->clearing);
1677 }
1678 
1679 static void disk_check_events(struct disk_events *ev,
1680 			      unsigned int *clearing_ptr)
1681 {
1682 	struct gendisk *disk = ev->disk;
1683 	char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1684 	unsigned int clearing = *clearing_ptr;
1685 	unsigned int events;
1686 	unsigned long intv;
1687 	int nr_events = 0, i;
1688 
1689 	/* check events */
1690 	events = disk->fops->check_events(disk, clearing);
1691 
1692 	/* accumulate pending events and schedule next poll if necessary */
1693 	spin_lock_irq(&ev->lock);
1694 
1695 	events &= ~ev->pending;
1696 	ev->pending |= events;
1697 	*clearing_ptr &= ~clearing;
1698 
1699 	intv = disk_events_poll_jiffies(disk);
1700 	if (!ev->block && intv)
1701 		queue_delayed_work(system_freezable_power_efficient_wq,
1702 				&ev->dwork, intv);
1703 
1704 	spin_unlock_irq(&ev->lock);
1705 
1706 	/*
1707 	 * Tell userland about new events.  Only the events listed in
1708 	 * @disk->events are reported.  Unlisted events are processed the
1709 	 * same internally but never get reported to userland.
1710 	 */
1711 	for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1712 		if (events & disk->events & (1 << i))
1713 			envp[nr_events++] = disk_uevents[i];
1714 
1715 	if (nr_events)
1716 		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1717 }
1718 
1719 /*
1720  * A disk events enabled device has the following sysfs nodes under
1721  * its /sys/block/X/ directory.
1722  *
1723  * events		: list of all supported events
1724  * events_async		: list of events which can be detected w/o polling
1725  * events_poll_msecs	: polling interval, 0: disable, -1: system default
1726  */
1727 static ssize_t __disk_events_show(unsigned int events, char *buf)
1728 {
1729 	const char *delim = "";
1730 	ssize_t pos = 0;
1731 	int i;
1732 
1733 	for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1734 		if (events & (1 << i)) {
1735 			pos += sprintf(buf + pos, "%s%s",
1736 				       delim, disk_events_strs[i]);
1737 			delim = " ";
1738 		}
1739 	if (pos)
1740 		pos += sprintf(buf + pos, "\n");
1741 	return pos;
1742 }
1743 
1744 static ssize_t disk_events_show(struct device *dev,
1745 				struct device_attribute *attr, char *buf)
1746 {
1747 	struct gendisk *disk = dev_to_disk(dev);
1748 
1749 	return __disk_events_show(disk->events, buf);
1750 }
1751 
1752 static ssize_t disk_events_async_show(struct device *dev,
1753 				      struct device_attribute *attr, char *buf)
1754 {
1755 	struct gendisk *disk = dev_to_disk(dev);
1756 
1757 	return __disk_events_show(disk->async_events, buf);
1758 }
1759 
1760 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1761 					   struct device_attribute *attr,
1762 					   char *buf)
1763 {
1764 	struct gendisk *disk = dev_to_disk(dev);
1765 
1766 	return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1767 }
1768 
1769 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1770 					    struct device_attribute *attr,
1771 					    const char *buf, size_t count)
1772 {
1773 	struct gendisk *disk = dev_to_disk(dev);
1774 	long intv;
1775 
1776 	if (!count || !sscanf(buf, "%ld", &intv))
1777 		return -EINVAL;
1778 
1779 	if (intv < 0 && intv != -1)
1780 		return -EINVAL;
1781 
1782 	disk_block_events(disk);
1783 	disk->ev->poll_msecs = intv;
1784 	__disk_unblock_events(disk, true);
1785 
1786 	return count;
1787 }
1788 
1789 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1790 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1791 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1792 			 disk_events_poll_msecs_show,
1793 			 disk_events_poll_msecs_store);
1794 
1795 static const struct attribute *disk_events_attrs[] = {
1796 	&dev_attr_events.attr,
1797 	&dev_attr_events_async.attr,
1798 	&dev_attr_events_poll_msecs.attr,
1799 	NULL,
1800 };
1801 
1802 /*
1803  * The default polling interval can be specified by the kernel
1804  * parameter block.events_dfl_poll_msecs which defaults to 0
1805  * (disable).  This can also be modified runtime by writing to
1806  * /sys/module/block/events_dfl_poll_msecs.
1807  */
1808 static int disk_events_set_dfl_poll_msecs(const char *val,
1809 					  const struct kernel_param *kp)
1810 {
1811 	struct disk_events *ev;
1812 	int ret;
1813 
1814 	ret = param_set_ulong(val, kp);
1815 	if (ret < 0)
1816 		return ret;
1817 
1818 	mutex_lock(&disk_events_mutex);
1819 
1820 	list_for_each_entry(ev, &disk_events, node)
1821 		disk_flush_events(ev->disk, 0);
1822 
1823 	mutex_unlock(&disk_events_mutex);
1824 
1825 	return 0;
1826 }
1827 
1828 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1829 	.set	= disk_events_set_dfl_poll_msecs,
1830 	.get	= param_get_ulong,
1831 };
1832 
1833 #undef MODULE_PARAM_PREFIX
1834 #define MODULE_PARAM_PREFIX	"block."
1835 
1836 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1837 		&disk_events_dfl_poll_msecs, 0644);
1838 
1839 /*
1840  * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1841  */
1842 static void disk_alloc_events(struct gendisk *disk)
1843 {
1844 	struct disk_events *ev;
1845 
1846 	if (!disk->fops->check_events)
1847 		return;
1848 
1849 	ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1850 	if (!ev) {
1851 		pr_warn("%s: failed to initialize events\n", disk->disk_name);
1852 		return;
1853 	}
1854 
1855 	INIT_LIST_HEAD(&ev->node);
1856 	ev->disk = disk;
1857 	spin_lock_init(&ev->lock);
1858 	mutex_init(&ev->block_mutex);
1859 	ev->block = 1;
1860 	ev->poll_msecs = -1;
1861 	INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1862 
1863 	disk->ev = ev;
1864 }
1865 
1866 static void disk_add_events(struct gendisk *disk)
1867 {
1868 	if (!disk->ev)
1869 		return;
1870 
1871 	/* FIXME: error handling */
1872 	if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1873 		pr_warn("%s: failed to create sysfs files for events\n",
1874 			disk->disk_name);
1875 
1876 	mutex_lock(&disk_events_mutex);
1877 	list_add_tail(&disk->ev->node, &disk_events);
1878 	mutex_unlock(&disk_events_mutex);
1879 
1880 	/*
1881 	 * Block count is initialized to 1 and the following initial
1882 	 * unblock kicks it into action.
1883 	 */
1884 	__disk_unblock_events(disk, true);
1885 }
1886 
1887 static void disk_del_events(struct gendisk *disk)
1888 {
1889 	if (!disk->ev)
1890 		return;
1891 
1892 	disk_block_events(disk);
1893 
1894 	mutex_lock(&disk_events_mutex);
1895 	list_del_init(&disk->ev->node);
1896 	mutex_unlock(&disk_events_mutex);
1897 
1898 	sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1899 }
1900 
1901 static void disk_release_events(struct gendisk *disk)
1902 {
1903 	/* the block count should be 1 from disk_del_events() */
1904 	WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1905 	kfree(disk->ev);
1906 }
1907