xref: /openbmc/linux/block/genhd.c (revision 7bcae826)
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 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 static struct blk_major_name {
246 	struct blk_major_name *next;
247 	int major;
248 	char name[16];
249 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
250 
251 /* index in the above - for now: assume no multimajor ranges */
252 static inline int major_to_index(unsigned major)
253 {
254 	return major % BLKDEV_MAJOR_HASH_SIZE;
255 }
256 
257 #ifdef CONFIG_PROC_FS
258 void blkdev_show(struct seq_file *seqf, off_t offset)
259 {
260 	struct blk_major_name *dp;
261 
262 	if (offset < BLKDEV_MAJOR_HASH_SIZE) {
263 		mutex_lock(&block_class_lock);
264 		for (dp = major_names[offset]; dp; dp = dp->next)
265 			seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
266 		mutex_unlock(&block_class_lock);
267 	}
268 }
269 #endif /* CONFIG_PROC_FS */
270 
271 /**
272  * register_blkdev - register a new block device
273  *
274  * @major: the requested major device number [1..255]. If @major=0, try to
275  *         allocate any unused major number.
276  * @name: the name of the new block device as a zero terminated string
277  *
278  * The @name must be unique within the system.
279  *
280  * The return value depends on the @major input parameter.
281  *  - if a major device number was requested in range [1..255] then the
282  *    function returns zero on success, or a negative error code
283  *  - if any unused major number was requested with @major=0 parameter
284  *    then the return value is the allocated major number in range
285  *    [1..255] or a negative error code otherwise
286  */
287 int register_blkdev(unsigned int major, const char *name)
288 {
289 	struct blk_major_name **n, *p;
290 	int index, ret = 0;
291 
292 	mutex_lock(&block_class_lock);
293 
294 	/* temporary */
295 	if (major == 0) {
296 		for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
297 			if (major_names[index] == NULL)
298 				break;
299 		}
300 
301 		if (index == 0) {
302 			printk("register_blkdev: failed to get major for %s\n",
303 			       name);
304 			ret = -EBUSY;
305 			goto out;
306 		}
307 		major = index;
308 		ret = major;
309 	}
310 
311 	p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
312 	if (p == NULL) {
313 		ret = -ENOMEM;
314 		goto out;
315 	}
316 
317 	p->major = major;
318 	strlcpy(p->name, name, sizeof(p->name));
319 	p->next = NULL;
320 	index = major_to_index(major);
321 
322 	for (n = &major_names[index]; *n; n = &(*n)->next) {
323 		if ((*n)->major == major)
324 			break;
325 	}
326 	if (!*n)
327 		*n = p;
328 	else
329 		ret = -EBUSY;
330 
331 	if (ret < 0) {
332 		printk("register_blkdev: cannot get major %d for %s\n",
333 		       major, name);
334 		kfree(p);
335 	}
336 out:
337 	mutex_unlock(&block_class_lock);
338 	return ret;
339 }
340 
341 EXPORT_SYMBOL(register_blkdev);
342 
343 void unregister_blkdev(unsigned int major, const char *name)
344 {
345 	struct blk_major_name **n;
346 	struct blk_major_name *p = NULL;
347 	int index = major_to_index(major);
348 
349 	mutex_lock(&block_class_lock);
350 	for (n = &major_names[index]; *n; n = &(*n)->next)
351 		if ((*n)->major == major)
352 			break;
353 	if (!*n || strcmp((*n)->name, name)) {
354 		WARN_ON(1);
355 	} else {
356 		p = *n;
357 		*n = p->next;
358 	}
359 	mutex_unlock(&block_class_lock);
360 	kfree(p);
361 }
362 
363 EXPORT_SYMBOL(unregister_blkdev);
364 
365 static struct kobj_map *bdev_map;
366 
367 /**
368  * blk_mangle_minor - scatter minor numbers apart
369  * @minor: minor number to mangle
370  *
371  * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
372  * is enabled.  Mangling twice gives the original value.
373  *
374  * RETURNS:
375  * Mangled value.
376  *
377  * CONTEXT:
378  * Don't care.
379  */
380 static int blk_mangle_minor(int minor)
381 {
382 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
383 	int i;
384 
385 	for (i = 0; i < MINORBITS / 2; i++) {
386 		int low = minor & (1 << i);
387 		int high = minor & (1 << (MINORBITS - 1 - i));
388 		int distance = MINORBITS - 1 - 2 * i;
389 
390 		minor ^= low | high;	/* clear both bits */
391 		low <<= distance;	/* swap the positions */
392 		high >>= distance;
393 		minor |= low | high;	/* and set */
394 	}
395 #endif
396 	return minor;
397 }
398 
399 /**
400  * blk_alloc_devt - allocate a dev_t for a partition
401  * @part: partition to allocate dev_t for
402  * @devt: out parameter for resulting dev_t
403  *
404  * Allocate a dev_t for block device.
405  *
406  * RETURNS:
407  * 0 on success, allocated dev_t is returned in *@devt.  -errno on
408  * failure.
409  *
410  * CONTEXT:
411  * Might sleep.
412  */
413 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
414 {
415 	struct gendisk *disk = part_to_disk(part);
416 	int idx;
417 
418 	/* in consecutive minor range? */
419 	if (part->partno < disk->minors) {
420 		*devt = MKDEV(disk->major, disk->first_minor + part->partno);
421 		return 0;
422 	}
423 
424 	/* allocate ext devt */
425 	idr_preload(GFP_KERNEL);
426 
427 	spin_lock_bh(&ext_devt_lock);
428 	idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
429 	spin_unlock_bh(&ext_devt_lock);
430 
431 	idr_preload_end();
432 	if (idx < 0)
433 		return idx == -ENOSPC ? -EBUSY : idx;
434 
435 	*devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
436 	return 0;
437 }
438 
439 /**
440  * blk_free_devt - free a dev_t
441  * @devt: dev_t to free
442  *
443  * Free @devt which was allocated using blk_alloc_devt().
444  *
445  * CONTEXT:
446  * Might sleep.
447  */
448 void blk_free_devt(dev_t devt)
449 {
450 	if (devt == MKDEV(0, 0))
451 		return;
452 
453 	if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
454 		spin_lock_bh(&ext_devt_lock);
455 		idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
456 		spin_unlock_bh(&ext_devt_lock);
457 	}
458 }
459 
460 static char *bdevt_str(dev_t devt, char *buf)
461 {
462 	if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
463 		char tbuf[BDEVT_SIZE];
464 		snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
465 		snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
466 	} else
467 		snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
468 
469 	return buf;
470 }
471 
472 /*
473  * Register device numbers dev..(dev+range-1)
474  * range must be nonzero
475  * The hash chain is sorted on range, so that subranges can override.
476  */
477 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
478 			 struct kobject *(*probe)(dev_t, int *, void *),
479 			 int (*lock)(dev_t, void *), void *data)
480 {
481 	kobj_map(bdev_map, devt, range, module, probe, lock, data);
482 }
483 
484 EXPORT_SYMBOL(blk_register_region);
485 
486 void blk_unregister_region(dev_t devt, unsigned long range)
487 {
488 	kobj_unmap(bdev_map, devt, range);
489 }
490 
491 EXPORT_SYMBOL(blk_unregister_region);
492 
493 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
494 {
495 	struct gendisk *p = data;
496 
497 	return &disk_to_dev(p)->kobj;
498 }
499 
500 static int exact_lock(dev_t devt, void *data)
501 {
502 	struct gendisk *p = data;
503 
504 	if (!get_disk(p))
505 		return -1;
506 	return 0;
507 }
508 
509 static void register_disk(struct device *parent, struct gendisk *disk)
510 {
511 	struct device *ddev = disk_to_dev(disk);
512 	struct block_device *bdev;
513 	struct disk_part_iter piter;
514 	struct hd_struct *part;
515 	int err;
516 
517 	ddev->parent = parent;
518 
519 	dev_set_name(ddev, "%s", disk->disk_name);
520 
521 	/* delay uevents, until we scanned partition table */
522 	dev_set_uevent_suppress(ddev, 1);
523 
524 	if (device_add(ddev))
525 		return;
526 	if (!sysfs_deprecated) {
527 		err = sysfs_create_link(block_depr, &ddev->kobj,
528 					kobject_name(&ddev->kobj));
529 		if (err) {
530 			device_del(ddev);
531 			return;
532 		}
533 	}
534 
535 	/*
536 	 * avoid probable deadlock caused by allocating memory with
537 	 * GFP_KERNEL in runtime_resume callback of its all ancestor
538 	 * devices
539 	 */
540 	pm_runtime_set_memalloc_noio(ddev, true);
541 
542 	disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
543 	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
544 
545 	/* No minors to use for partitions */
546 	if (!disk_part_scan_enabled(disk))
547 		goto exit;
548 
549 	/* No such device (e.g., media were just removed) */
550 	if (!get_capacity(disk))
551 		goto exit;
552 
553 	bdev = bdget_disk(disk, 0);
554 	if (!bdev)
555 		goto exit;
556 
557 	bdev->bd_invalidated = 1;
558 	err = blkdev_get(bdev, FMODE_READ, NULL);
559 	if (err < 0)
560 		goto exit;
561 	blkdev_put(bdev, FMODE_READ);
562 
563 exit:
564 	/* announce disk after possible partitions are created */
565 	dev_set_uevent_suppress(ddev, 0);
566 	kobject_uevent(&ddev->kobj, KOBJ_ADD);
567 
568 	/* announce possible partitions */
569 	disk_part_iter_init(&piter, disk, 0);
570 	while ((part = disk_part_iter_next(&piter)))
571 		kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
572 	disk_part_iter_exit(&piter);
573 }
574 
575 void put_disk_devt(struct disk_devt *disk_devt)
576 {
577 	if (disk_devt && atomic_dec_and_test(&disk_devt->count))
578 		disk_devt->release(disk_devt);
579 }
580 EXPORT_SYMBOL(put_disk_devt);
581 
582 void get_disk_devt(struct disk_devt *disk_devt)
583 {
584 	if (disk_devt)
585 		atomic_inc(&disk_devt->count);
586 }
587 EXPORT_SYMBOL(get_disk_devt);
588 
589 /**
590  * device_add_disk - add partitioning information to kernel list
591  * @parent: parent device for the disk
592  * @disk: per-device partitioning information
593  *
594  * This function registers the partitioning information in @disk
595  * with the kernel.
596  *
597  * FIXME: error handling
598  */
599 void device_add_disk(struct device *parent, struct gendisk *disk)
600 {
601 	struct backing_dev_info *bdi;
602 	dev_t devt;
603 	int retval;
604 
605 	/* minors == 0 indicates to use ext devt from part0 and should
606 	 * be accompanied with EXT_DEVT flag.  Make sure all
607 	 * parameters make sense.
608 	 */
609 	WARN_ON(disk->minors && !(disk->major || disk->first_minor));
610 	WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
611 
612 	disk->flags |= GENHD_FL_UP;
613 
614 	retval = blk_alloc_devt(&disk->part0, &devt);
615 	if (retval) {
616 		WARN_ON(1);
617 		return;
618 	}
619 	disk_to_dev(disk)->devt = devt;
620 
621 	/* ->major and ->first_minor aren't supposed to be
622 	 * dereferenced from here on, but set them just in case.
623 	 */
624 	disk->major = MAJOR(devt);
625 	disk->first_minor = MINOR(devt);
626 
627 	disk_alloc_events(disk);
628 
629 	/*
630 	 * Take a reference on the devt and assign it to queue since it
631 	 * must not be reallocated while the bdi is registered
632 	 */
633 	disk->queue->disk_devt = disk->disk_devt;
634 	get_disk_devt(disk->disk_devt);
635 
636 	/* Register BDI before referencing it from bdev */
637 	bdi = disk->queue->backing_dev_info;
638 	bdi_register_owner(bdi, disk_to_dev(disk));
639 
640 	blk_register_region(disk_devt(disk), disk->minors, NULL,
641 			    exact_match, exact_lock, disk);
642 	register_disk(parent, disk);
643 	blk_register_queue(disk);
644 
645 	/*
646 	 * Take an extra ref on queue which will be put on disk_release()
647 	 * so that it sticks around as long as @disk is there.
648 	 */
649 	WARN_ON_ONCE(!blk_get_queue(disk->queue));
650 
651 	retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
652 				   "bdi");
653 	WARN_ON(retval);
654 
655 	disk_add_events(disk);
656 	blk_integrity_add(disk);
657 }
658 EXPORT_SYMBOL(device_add_disk);
659 
660 void del_gendisk(struct gendisk *disk)
661 {
662 	struct disk_part_iter piter;
663 	struct hd_struct *part;
664 
665 	blk_integrity_del(disk);
666 	disk_del_events(disk);
667 
668 	/* invalidate stuff */
669 	disk_part_iter_init(&piter, disk,
670 			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
671 	while ((part = disk_part_iter_next(&piter))) {
672 		invalidate_partition(disk, part->partno);
673 		bdev_unhash_inode(part_devt(part));
674 		delete_partition(disk, part->partno);
675 	}
676 	disk_part_iter_exit(&piter);
677 
678 	invalidate_partition(disk, 0);
679 	bdev_unhash_inode(disk_devt(disk));
680 	set_capacity(disk, 0);
681 	disk->flags &= ~GENHD_FL_UP;
682 
683 	sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
684 	blk_unregister_queue(disk);
685 	blk_unregister_region(disk_devt(disk), disk->minors);
686 
687 	part_stat_set_all(&disk->part0, 0);
688 	disk->part0.stamp = 0;
689 
690 	kobject_put(disk->part0.holder_dir);
691 	kobject_put(disk->slave_dir);
692 	if (!sysfs_deprecated)
693 		sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
694 	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
695 	device_del(disk_to_dev(disk));
696 }
697 EXPORT_SYMBOL(del_gendisk);
698 
699 /* sysfs access to bad-blocks list. */
700 static ssize_t disk_badblocks_show(struct device *dev,
701 					struct device_attribute *attr,
702 					char *page)
703 {
704 	struct gendisk *disk = dev_to_disk(dev);
705 
706 	if (!disk->bb)
707 		return sprintf(page, "\n");
708 
709 	return badblocks_show(disk->bb, page, 0);
710 }
711 
712 static ssize_t disk_badblocks_store(struct device *dev,
713 					struct device_attribute *attr,
714 					const char *page, size_t len)
715 {
716 	struct gendisk *disk = dev_to_disk(dev);
717 
718 	if (!disk->bb)
719 		return -ENXIO;
720 
721 	return badblocks_store(disk->bb, page, len, 0);
722 }
723 
724 /**
725  * get_gendisk - get partitioning information for a given device
726  * @devt: device to get partitioning information for
727  * @partno: returned partition index
728  *
729  * This function gets the structure containing partitioning
730  * information for the given device @devt.
731  */
732 struct gendisk *get_gendisk(dev_t devt, int *partno)
733 {
734 	struct gendisk *disk = NULL;
735 
736 	if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
737 		struct kobject *kobj;
738 
739 		kobj = kobj_lookup(bdev_map, devt, partno);
740 		if (kobj)
741 			disk = dev_to_disk(kobj_to_dev(kobj));
742 	} else {
743 		struct hd_struct *part;
744 
745 		spin_lock_bh(&ext_devt_lock);
746 		part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
747 		if (part && get_disk(part_to_disk(part))) {
748 			*partno = part->partno;
749 			disk = part_to_disk(part);
750 		}
751 		spin_unlock_bh(&ext_devt_lock);
752 	}
753 
754 	return disk;
755 }
756 EXPORT_SYMBOL(get_gendisk);
757 
758 /**
759  * bdget_disk - do bdget() by gendisk and partition number
760  * @disk: gendisk of interest
761  * @partno: partition number
762  *
763  * Find partition @partno from @disk, do bdget() on it.
764  *
765  * CONTEXT:
766  * Don't care.
767  *
768  * RETURNS:
769  * Resulting block_device on success, NULL on failure.
770  */
771 struct block_device *bdget_disk(struct gendisk *disk, int partno)
772 {
773 	struct hd_struct *part;
774 	struct block_device *bdev = NULL;
775 
776 	part = disk_get_part(disk, partno);
777 	if (part)
778 		bdev = bdget(part_devt(part));
779 	disk_put_part(part);
780 
781 	return bdev;
782 }
783 EXPORT_SYMBOL(bdget_disk);
784 
785 /*
786  * print a full list of all partitions - intended for places where the root
787  * filesystem can't be mounted and thus to give the victim some idea of what
788  * went wrong
789  */
790 void __init printk_all_partitions(void)
791 {
792 	struct class_dev_iter iter;
793 	struct device *dev;
794 
795 	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
796 	while ((dev = class_dev_iter_next(&iter))) {
797 		struct gendisk *disk = dev_to_disk(dev);
798 		struct disk_part_iter piter;
799 		struct hd_struct *part;
800 		char name_buf[BDEVNAME_SIZE];
801 		char devt_buf[BDEVT_SIZE];
802 
803 		/*
804 		 * Don't show empty devices or things that have been
805 		 * suppressed
806 		 */
807 		if (get_capacity(disk) == 0 ||
808 		    (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
809 			continue;
810 
811 		/*
812 		 * Note, unlike /proc/partitions, I am showing the
813 		 * numbers in hex - the same format as the root=
814 		 * option takes.
815 		 */
816 		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
817 		while ((part = disk_part_iter_next(&piter))) {
818 			bool is_part0 = part == &disk->part0;
819 
820 			printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
821 			       bdevt_str(part_devt(part), devt_buf),
822 			       (unsigned long long)part_nr_sects_read(part) >> 1
823 			       , disk_name(disk, part->partno, name_buf),
824 			       part->info ? part->info->uuid : "");
825 			if (is_part0) {
826 				if (dev->parent && dev->parent->driver)
827 					printk(" driver: %s\n",
828 					      dev->parent->driver->name);
829 				else
830 					printk(" (driver?)\n");
831 			} else
832 				printk("\n");
833 		}
834 		disk_part_iter_exit(&piter);
835 	}
836 	class_dev_iter_exit(&iter);
837 }
838 
839 #ifdef CONFIG_PROC_FS
840 /* iterator */
841 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
842 {
843 	loff_t skip = *pos;
844 	struct class_dev_iter *iter;
845 	struct device *dev;
846 
847 	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
848 	if (!iter)
849 		return ERR_PTR(-ENOMEM);
850 
851 	seqf->private = iter;
852 	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
853 	do {
854 		dev = class_dev_iter_next(iter);
855 		if (!dev)
856 			return NULL;
857 	} while (skip--);
858 
859 	return dev_to_disk(dev);
860 }
861 
862 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
863 {
864 	struct device *dev;
865 
866 	(*pos)++;
867 	dev = class_dev_iter_next(seqf->private);
868 	if (dev)
869 		return dev_to_disk(dev);
870 
871 	return NULL;
872 }
873 
874 static void disk_seqf_stop(struct seq_file *seqf, void *v)
875 {
876 	struct class_dev_iter *iter = seqf->private;
877 
878 	/* stop is called even after start failed :-( */
879 	if (iter) {
880 		class_dev_iter_exit(iter);
881 		kfree(iter);
882 		seqf->private = NULL;
883 	}
884 }
885 
886 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
887 {
888 	void *p;
889 
890 	p = disk_seqf_start(seqf, pos);
891 	if (!IS_ERR_OR_NULL(p) && !*pos)
892 		seq_puts(seqf, "major minor  #blocks  name\n\n");
893 	return p;
894 }
895 
896 static int show_partition(struct seq_file *seqf, void *v)
897 {
898 	struct gendisk *sgp = v;
899 	struct disk_part_iter piter;
900 	struct hd_struct *part;
901 	char buf[BDEVNAME_SIZE];
902 
903 	/* Don't show non-partitionable removeable devices or empty devices */
904 	if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
905 				   (sgp->flags & GENHD_FL_REMOVABLE)))
906 		return 0;
907 	if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
908 		return 0;
909 
910 	/* show the full disk and all non-0 size partitions of it */
911 	disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
912 	while ((part = disk_part_iter_next(&piter)))
913 		seq_printf(seqf, "%4d  %7d %10llu %s\n",
914 			   MAJOR(part_devt(part)), MINOR(part_devt(part)),
915 			   (unsigned long long)part_nr_sects_read(part) >> 1,
916 			   disk_name(sgp, part->partno, buf));
917 	disk_part_iter_exit(&piter);
918 
919 	return 0;
920 }
921 
922 static const struct seq_operations partitions_op = {
923 	.start	= show_partition_start,
924 	.next	= disk_seqf_next,
925 	.stop	= disk_seqf_stop,
926 	.show	= show_partition
927 };
928 
929 static int partitions_open(struct inode *inode, struct file *file)
930 {
931 	return seq_open(file, &partitions_op);
932 }
933 
934 static const struct file_operations proc_partitions_operations = {
935 	.open		= partitions_open,
936 	.read		= seq_read,
937 	.llseek		= seq_lseek,
938 	.release	= seq_release,
939 };
940 #endif
941 
942 
943 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
944 {
945 	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
946 		/* Make old-style 2.4 aliases work */
947 		request_module("block-major-%d", MAJOR(devt));
948 	return NULL;
949 }
950 
951 static int __init genhd_device_init(void)
952 {
953 	int error;
954 
955 	block_class.dev_kobj = sysfs_dev_block_kobj;
956 	error = class_register(&block_class);
957 	if (unlikely(error))
958 		return error;
959 	bdev_map = kobj_map_init(base_probe, &block_class_lock);
960 	blk_dev_init();
961 
962 	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
963 
964 	/* create top-level block dir */
965 	if (!sysfs_deprecated)
966 		block_depr = kobject_create_and_add("block", NULL);
967 	return 0;
968 }
969 
970 subsys_initcall(genhd_device_init);
971 
972 static ssize_t disk_range_show(struct device *dev,
973 			       struct device_attribute *attr, char *buf)
974 {
975 	struct gendisk *disk = dev_to_disk(dev);
976 
977 	return sprintf(buf, "%d\n", disk->minors);
978 }
979 
980 static ssize_t disk_ext_range_show(struct device *dev,
981 				   struct device_attribute *attr, char *buf)
982 {
983 	struct gendisk *disk = dev_to_disk(dev);
984 
985 	return sprintf(buf, "%d\n", disk_max_parts(disk));
986 }
987 
988 static ssize_t disk_removable_show(struct device *dev,
989 				   struct device_attribute *attr, char *buf)
990 {
991 	struct gendisk *disk = dev_to_disk(dev);
992 
993 	return sprintf(buf, "%d\n",
994 		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
995 }
996 
997 static ssize_t disk_ro_show(struct device *dev,
998 				   struct device_attribute *attr, char *buf)
999 {
1000 	struct gendisk *disk = dev_to_disk(dev);
1001 
1002 	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1003 }
1004 
1005 static ssize_t disk_capability_show(struct device *dev,
1006 				    struct device_attribute *attr, char *buf)
1007 {
1008 	struct gendisk *disk = dev_to_disk(dev);
1009 
1010 	return sprintf(buf, "%x\n", disk->flags);
1011 }
1012 
1013 static ssize_t disk_alignment_offset_show(struct device *dev,
1014 					  struct device_attribute *attr,
1015 					  char *buf)
1016 {
1017 	struct gendisk *disk = dev_to_disk(dev);
1018 
1019 	return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1020 }
1021 
1022 static ssize_t disk_discard_alignment_show(struct device *dev,
1023 					   struct device_attribute *attr,
1024 					   char *buf)
1025 {
1026 	struct gendisk *disk = dev_to_disk(dev);
1027 
1028 	return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1029 }
1030 
1031 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
1032 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
1033 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
1034 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
1035 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
1036 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
1037 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
1038 		   NULL);
1039 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
1040 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
1041 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
1042 static DEVICE_ATTR(badblocks, S_IRUGO | S_IWUSR, disk_badblocks_show,
1043 		disk_badblocks_store);
1044 #ifdef CONFIG_FAIL_MAKE_REQUEST
1045 static struct device_attribute dev_attr_fail =
1046 	__ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
1047 #endif
1048 #ifdef CONFIG_FAIL_IO_TIMEOUT
1049 static struct device_attribute dev_attr_fail_timeout =
1050 	__ATTR(io-timeout-fail,  S_IRUGO|S_IWUSR, part_timeout_show,
1051 		part_timeout_store);
1052 #endif
1053 
1054 static struct attribute *disk_attrs[] = {
1055 	&dev_attr_range.attr,
1056 	&dev_attr_ext_range.attr,
1057 	&dev_attr_removable.attr,
1058 	&dev_attr_ro.attr,
1059 	&dev_attr_size.attr,
1060 	&dev_attr_alignment_offset.attr,
1061 	&dev_attr_discard_alignment.attr,
1062 	&dev_attr_capability.attr,
1063 	&dev_attr_stat.attr,
1064 	&dev_attr_inflight.attr,
1065 	&dev_attr_badblocks.attr,
1066 #ifdef CONFIG_FAIL_MAKE_REQUEST
1067 	&dev_attr_fail.attr,
1068 #endif
1069 #ifdef CONFIG_FAIL_IO_TIMEOUT
1070 	&dev_attr_fail_timeout.attr,
1071 #endif
1072 	NULL
1073 };
1074 
1075 static struct attribute_group disk_attr_group = {
1076 	.attrs = disk_attrs,
1077 };
1078 
1079 static const struct attribute_group *disk_attr_groups[] = {
1080 	&disk_attr_group,
1081 	NULL
1082 };
1083 
1084 /**
1085  * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1086  * @disk: disk to replace part_tbl for
1087  * @new_ptbl: new part_tbl to install
1088  *
1089  * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
1090  * original ptbl is freed using RCU callback.
1091  *
1092  * LOCKING:
1093  * Matching bd_mutx locked.
1094  */
1095 static void disk_replace_part_tbl(struct gendisk *disk,
1096 				  struct disk_part_tbl *new_ptbl)
1097 {
1098 	struct disk_part_tbl *old_ptbl = disk->part_tbl;
1099 
1100 	rcu_assign_pointer(disk->part_tbl, new_ptbl);
1101 
1102 	if (old_ptbl) {
1103 		rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1104 		kfree_rcu(old_ptbl, rcu_head);
1105 	}
1106 }
1107 
1108 /**
1109  * disk_expand_part_tbl - expand disk->part_tbl
1110  * @disk: disk to expand part_tbl for
1111  * @partno: expand such that this partno can fit in
1112  *
1113  * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
1114  * uses RCU to allow unlocked dereferencing for stats and other stuff.
1115  *
1116  * LOCKING:
1117  * Matching bd_mutex locked, might sleep.
1118  *
1119  * RETURNS:
1120  * 0 on success, -errno on failure.
1121  */
1122 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1123 {
1124 	struct disk_part_tbl *old_ptbl = disk->part_tbl;
1125 	struct disk_part_tbl *new_ptbl;
1126 	int len = old_ptbl ? old_ptbl->len : 0;
1127 	int i, target;
1128 	size_t size;
1129 
1130 	/*
1131 	 * check for int overflow, since we can get here from blkpg_ioctl()
1132 	 * with a user passed 'partno'.
1133 	 */
1134 	target = partno + 1;
1135 	if (target < 0)
1136 		return -EINVAL;
1137 
1138 	/* disk_max_parts() is zero during initialization, ignore if so */
1139 	if (disk_max_parts(disk) && target > disk_max_parts(disk))
1140 		return -EINVAL;
1141 
1142 	if (target <= len)
1143 		return 0;
1144 
1145 	size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1146 	new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1147 	if (!new_ptbl)
1148 		return -ENOMEM;
1149 
1150 	new_ptbl->len = target;
1151 
1152 	for (i = 0; i < len; i++)
1153 		rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1154 
1155 	disk_replace_part_tbl(disk, new_ptbl);
1156 	return 0;
1157 }
1158 
1159 static void disk_release(struct device *dev)
1160 {
1161 	struct gendisk *disk = dev_to_disk(dev);
1162 
1163 	blk_free_devt(dev->devt);
1164 	disk_release_events(disk);
1165 	kfree(disk->random);
1166 	disk_replace_part_tbl(disk, NULL);
1167 	hd_free_part(&disk->part0);
1168 	if (disk->queue)
1169 		blk_put_queue(disk->queue);
1170 	kfree(disk);
1171 }
1172 struct class block_class = {
1173 	.name		= "block",
1174 };
1175 
1176 static char *block_devnode(struct device *dev, umode_t *mode,
1177 			   kuid_t *uid, kgid_t *gid)
1178 {
1179 	struct gendisk *disk = dev_to_disk(dev);
1180 
1181 	if (disk->devnode)
1182 		return disk->devnode(disk, mode);
1183 	return NULL;
1184 }
1185 
1186 static struct device_type disk_type = {
1187 	.name		= "disk",
1188 	.groups		= disk_attr_groups,
1189 	.release	= disk_release,
1190 	.devnode	= block_devnode,
1191 };
1192 
1193 #ifdef CONFIG_PROC_FS
1194 /*
1195  * aggregate disk stat collector.  Uses the same stats that the sysfs
1196  * entries do, above, but makes them available through one seq_file.
1197  *
1198  * The output looks suspiciously like /proc/partitions with a bunch of
1199  * extra fields.
1200  */
1201 static int diskstats_show(struct seq_file *seqf, void *v)
1202 {
1203 	struct gendisk *gp = v;
1204 	struct disk_part_iter piter;
1205 	struct hd_struct *hd;
1206 	char buf[BDEVNAME_SIZE];
1207 	int cpu;
1208 
1209 	/*
1210 	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1211 		seq_puts(seqf,	"major minor name"
1212 				"     rio rmerge rsect ruse wio wmerge "
1213 				"wsect wuse running use aveq"
1214 				"\n\n");
1215 	*/
1216 
1217 	disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1218 	while ((hd = disk_part_iter_next(&piter))) {
1219 		cpu = part_stat_lock();
1220 		part_round_stats(cpu, hd);
1221 		part_stat_unlock();
1222 		seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1223 			   "%u %lu %lu %lu %u %u %u %u\n",
1224 			   MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1225 			   disk_name(gp, hd->partno, buf),
1226 			   part_stat_read(hd, ios[READ]),
1227 			   part_stat_read(hd, merges[READ]),
1228 			   part_stat_read(hd, sectors[READ]),
1229 			   jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1230 			   part_stat_read(hd, ios[WRITE]),
1231 			   part_stat_read(hd, merges[WRITE]),
1232 			   part_stat_read(hd, sectors[WRITE]),
1233 			   jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1234 			   part_in_flight(hd),
1235 			   jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1236 			   jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1237 			);
1238 	}
1239 	disk_part_iter_exit(&piter);
1240 
1241 	return 0;
1242 }
1243 
1244 static const struct seq_operations diskstats_op = {
1245 	.start	= disk_seqf_start,
1246 	.next	= disk_seqf_next,
1247 	.stop	= disk_seqf_stop,
1248 	.show	= diskstats_show
1249 };
1250 
1251 static int diskstats_open(struct inode *inode, struct file *file)
1252 {
1253 	return seq_open(file, &diskstats_op);
1254 }
1255 
1256 static const struct file_operations proc_diskstats_operations = {
1257 	.open		= diskstats_open,
1258 	.read		= seq_read,
1259 	.llseek		= seq_lseek,
1260 	.release	= seq_release,
1261 };
1262 
1263 static int __init proc_genhd_init(void)
1264 {
1265 	proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1266 	proc_create("partitions", 0, NULL, &proc_partitions_operations);
1267 	return 0;
1268 }
1269 module_init(proc_genhd_init);
1270 #endif /* CONFIG_PROC_FS */
1271 
1272 dev_t blk_lookup_devt(const char *name, int partno)
1273 {
1274 	dev_t devt = MKDEV(0, 0);
1275 	struct class_dev_iter iter;
1276 	struct device *dev;
1277 
1278 	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1279 	while ((dev = class_dev_iter_next(&iter))) {
1280 		struct gendisk *disk = dev_to_disk(dev);
1281 		struct hd_struct *part;
1282 
1283 		if (strcmp(dev_name(dev), name))
1284 			continue;
1285 
1286 		if (partno < disk->minors) {
1287 			/* We need to return the right devno, even
1288 			 * if the partition doesn't exist yet.
1289 			 */
1290 			devt = MKDEV(MAJOR(dev->devt),
1291 				     MINOR(dev->devt) + partno);
1292 			break;
1293 		}
1294 		part = disk_get_part(disk, partno);
1295 		if (part) {
1296 			devt = part_devt(part);
1297 			disk_put_part(part);
1298 			break;
1299 		}
1300 		disk_put_part(part);
1301 	}
1302 	class_dev_iter_exit(&iter);
1303 	return devt;
1304 }
1305 EXPORT_SYMBOL(blk_lookup_devt);
1306 
1307 struct gendisk *alloc_disk(int minors)
1308 {
1309 	return alloc_disk_node(minors, NUMA_NO_NODE);
1310 }
1311 EXPORT_SYMBOL(alloc_disk);
1312 
1313 struct gendisk *alloc_disk_node(int minors, int node_id)
1314 {
1315 	struct gendisk *disk;
1316 
1317 	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1318 	if (disk) {
1319 		if (!init_part_stats(&disk->part0)) {
1320 			kfree(disk);
1321 			return NULL;
1322 		}
1323 		disk->node_id = node_id;
1324 		if (disk_expand_part_tbl(disk, 0)) {
1325 			free_part_stats(&disk->part0);
1326 			kfree(disk);
1327 			return NULL;
1328 		}
1329 		disk->part_tbl->part[0] = &disk->part0;
1330 
1331 		/*
1332 		 * set_capacity() and get_capacity() currently don't use
1333 		 * seqcounter to read/update the part0->nr_sects. Still init
1334 		 * the counter as we can read the sectors in IO submission
1335 		 * patch using seqence counters.
1336 		 *
1337 		 * TODO: Ideally set_capacity() and get_capacity() should be
1338 		 * converted to make use of bd_mutex and sequence counters.
1339 		 */
1340 		seqcount_init(&disk->part0.nr_sects_seq);
1341 		if (hd_ref_init(&disk->part0)) {
1342 			hd_free_part(&disk->part0);
1343 			kfree(disk);
1344 			return NULL;
1345 		}
1346 
1347 		disk->minors = minors;
1348 		rand_initialize_disk(disk);
1349 		disk_to_dev(disk)->class = &block_class;
1350 		disk_to_dev(disk)->type = &disk_type;
1351 		device_initialize(disk_to_dev(disk));
1352 	}
1353 	return disk;
1354 }
1355 EXPORT_SYMBOL(alloc_disk_node);
1356 
1357 struct kobject *get_disk(struct gendisk *disk)
1358 {
1359 	struct module *owner;
1360 	struct kobject *kobj;
1361 
1362 	if (!disk->fops)
1363 		return NULL;
1364 	owner = disk->fops->owner;
1365 	if (owner && !try_module_get(owner))
1366 		return NULL;
1367 	kobj = kobject_get(&disk_to_dev(disk)->kobj);
1368 	if (kobj == NULL) {
1369 		module_put(owner);
1370 		return NULL;
1371 	}
1372 	return kobj;
1373 
1374 }
1375 
1376 EXPORT_SYMBOL(get_disk);
1377 
1378 void put_disk(struct gendisk *disk)
1379 {
1380 	if (disk)
1381 		kobject_put(&disk_to_dev(disk)->kobj);
1382 }
1383 
1384 EXPORT_SYMBOL(put_disk);
1385 
1386 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1387 {
1388 	char event[] = "DISK_RO=1";
1389 	char *envp[] = { event, NULL };
1390 
1391 	if (!ro)
1392 		event[8] = '0';
1393 	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1394 }
1395 
1396 void set_device_ro(struct block_device *bdev, int flag)
1397 {
1398 	bdev->bd_part->policy = flag;
1399 }
1400 
1401 EXPORT_SYMBOL(set_device_ro);
1402 
1403 void set_disk_ro(struct gendisk *disk, int flag)
1404 {
1405 	struct disk_part_iter piter;
1406 	struct hd_struct *part;
1407 
1408 	if (disk->part0.policy != flag) {
1409 		set_disk_ro_uevent(disk, flag);
1410 		disk->part0.policy = flag;
1411 	}
1412 
1413 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1414 	while ((part = disk_part_iter_next(&piter)))
1415 		part->policy = flag;
1416 	disk_part_iter_exit(&piter);
1417 }
1418 
1419 EXPORT_SYMBOL(set_disk_ro);
1420 
1421 int bdev_read_only(struct block_device *bdev)
1422 {
1423 	if (!bdev)
1424 		return 0;
1425 	return bdev->bd_part->policy;
1426 }
1427 
1428 EXPORT_SYMBOL(bdev_read_only);
1429 
1430 int invalidate_partition(struct gendisk *disk, int partno)
1431 {
1432 	int res = 0;
1433 	struct block_device *bdev = bdget_disk(disk, partno);
1434 	if (bdev) {
1435 		fsync_bdev(bdev);
1436 		res = __invalidate_device(bdev, true);
1437 		bdput(bdev);
1438 	}
1439 	return res;
1440 }
1441 
1442 EXPORT_SYMBOL(invalidate_partition);
1443 
1444 /*
1445  * Disk events - monitor disk events like media change and eject request.
1446  */
1447 struct disk_events {
1448 	struct list_head	node;		/* all disk_event's */
1449 	struct gendisk		*disk;		/* the associated disk */
1450 	spinlock_t		lock;
1451 
1452 	struct mutex		block_mutex;	/* protects blocking */
1453 	int			block;		/* event blocking depth */
1454 	unsigned int		pending;	/* events already sent out */
1455 	unsigned int		clearing;	/* events being cleared */
1456 
1457 	long			poll_msecs;	/* interval, -1 for default */
1458 	struct delayed_work	dwork;
1459 };
1460 
1461 static const char *disk_events_strs[] = {
1462 	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "media_change",
1463 	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "eject_request",
1464 };
1465 
1466 static char *disk_uevents[] = {
1467 	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "DISK_MEDIA_CHANGE=1",
1468 	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "DISK_EJECT_REQUEST=1",
1469 };
1470 
1471 /* list of all disk_events */
1472 static DEFINE_MUTEX(disk_events_mutex);
1473 static LIST_HEAD(disk_events);
1474 
1475 /* disable in-kernel polling by default */
1476 static unsigned long disk_events_dfl_poll_msecs;
1477 
1478 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1479 {
1480 	struct disk_events *ev = disk->ev;
1481 	long intv_msecs = 0;
1482 
1483 	/*
1484 	 * If device-specific poll interval is set, always use it.  If
1485 	 * the default is being used, poll iff there are events which
1486 	 * can't be monitored asynchronously.
1487 	 */
1488 	if (ev->poll_msecs >= 0)
1489 		intv_msecs = ev->poll_msecs;
1490 	else if (disk->events & ~disk->async_events)
1491 		intv_msecs = disk_events_dfl_poll_msecs;
1492 
1493 	return msecs_to_jiffies(intv_msecs);
1494 }
1495 
1496 /**
1497  * disk_block_events - block and flush disk event checking
1498  * @disk: disk to block events for
1499  *
1500  * On return from this function, it is guaranteed that event checking
1501  * isn't in progress and won't happen until unblocked by
1502  * disk_unblock_events().  Events blocking is counted and the actual
1503  * unblocking happens after the matching number of unblocks are done.
1504  *
1505  * Note that this intentionally does not block event checking from
1506  * disk_clear_events().
1507  *
1508  * CONTEXT:
1509  * Might sleep.
1510  */
1511 void disk_block_events(struct gendisk *disk)
1512 {
1513 	struct disk_events *ev = disk->ev;
1514 	unsigned long flags;
1515 	bool cancel;
1516 
1517 	if (!ev)
1518 		return;
1519 
1520 	/*
1521 	 * Outer mutex ensures that the first blocker completes canceling
1522 	 * the event work before further blockers are allowed to finish.
1523 	 */
1524 	mutex_lock(&ev->block_mutex);
1525 
1526 	spin_lock_irqsave(&ev->lock, flags);
1527 	cancel = !ev->block++;
1528 	spin_unlock_irqrestore(&ev->lock, flags);
1529 
1530 	if (cancel)
1531 		cancel_delayed_work_sync(&disk->ev->dwork);
1532 
1533 	mutex_unlock(&ev->block_mutex);
1534 }
1535 
1536 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1537 {
1538 	struct disk_events *ev = disk->ev;
1539 	unsigned long intv;
1540 	unsigned long flags;
1541 
1542 	spin_lock_irqsave(&ev->lock, flags);
1543 
1544 	if (WARN_ON_ONCE(ev->block <= 0))
1545 		goto out_unlock;
1546 
1547 	if (--ev->block)
1548 		goto out_unlock;
1549 
1550 	intv = disk_events_poll_jiffies(disk);
1551 	if (check_now)
1552 		queue_delayed_work(system_freezable_power_efficient_wq,
1553 				&ev->dwork, 0);
1554 	else if (intv)
1555 		queue_delayed_work(system_freezable_power_efficient_wq,
1556 				&ev->dwork, intv);
1557 out_unlock:
1558 	spin_unlock_irqrestore(&ev->lock, flags);
1559 }
1560 
1561 /**
1562  * disk_unblock_events - unblock disk event checking
1563  * @disk: disk to unblock events for
1564  *
1565  * Undo disk_block_events().  When the block count reaches zero, it
1566  * starts events polling if configured.
1567  *
1568  * CONTEXT:
1569  * Don't care.  Safe to call from irq context.
1570  */
1571 void disk_unblock_events(struct gendisk *disk)
1572 {
1573 	if (disk->ev)
1574 		__disk_unblock_events(disk, false);
1575 }
1576 
1577 /**
1578  * disk_flush_events - schedule immediate event checking and flushing
1579  * @disk: disk to check and flush events for
1580  * @mask: events to flush
1581  *
1582  * Schedule immediate event checking on @disk if not blocked.  Events in
1583  * @mask are scheduled to be cleared from the driver.  Note that this
1584  * doesn't clear the events from @disk->ev.
1585  *
1586  * CONTEXT:
1587  * If @mask is non-zero must be called with bdev->bd_mutex held.
1588  */
1589 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1590 {
1591 	struct disk_events *ev = disk->ev;
1592 
1593 	if (!ev)
1594 		return;
1595 
1596 	spin_lock_irq(&ev->lock);
1597 	ev->clearing |= mask;
1598 	if (!ev->block)
1599 		mod_delayed_work(system_freezable_power_efficient_wq,
1600 				&ev->dwork, 0);
1601 	spin_unlock_irq(&ev->lock);
1602 }
1603 
1604 /**
1605  * disk_clear_events - synchronously check, clear and return pending events
1606  * @disk: disk to fetch and clear events from
1607  * @mask: mask of events to be fetched and cleared
1608  *
1609  * Disk events are synchronously checked and pending events in @mask
1610  * are cleared and returned.  This ignores the block count.
1611  *
1612  * CONTEXT:
1613  * Might sleep.
1614  */
1615 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1616 {
1617 	const struct block_device_operations *bdops = disk->fops;
1618 	struct disk_events *ev = disk->ev;
1619 	unsigned int pending;
1620 	unsigned int clearing = mask;
1621 
1622 	if (!ev) {
1623 		/* for drivers still using the old ->media_changed method */
1624 		if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1625 		    bdops->media_changed && bdops->media_changed(disk))
1626 			return DISK_EVENT_MEDIA_CHANGE;
1627 		return 0;
1628 	}
1629 
1630 	disk_block_events(disk);
1631 
1632 	/*
1633 	 * store the union of mask and ev->clearing on the stack so that the
1634 	 * race with disk_flush_events does not cause ambiguity (ev->clearing
1635 	 * can still be modified even if events are blocked).
1636 	 */
1637 	spin_lock_irq(&ev->lock);
1638 	clearing |= ev->clearing;
1639 	ev->clearing = 0;
1640 	spin_unlock_irq(&ev->lock);
1641 
1642 	disk_check_events(ev, &clearing);
1643 	/*
1644 	 * if ev->clearing is not 0, the disk_flush_events got called in the
1645 	 * middle of this function, so we want to run the workfn without delay.
1646 	 */
1647 	__disk_unblock_events(disk, ev->clearing ? true : false);
1648 
1649 	/* then, fetch and clear pending events */
1650 	spin_lock_irq(&ev->lock);
1651 	pending = ev->pending & mask;
1652 	ev->pending &= ~mask;
1653 	spin_unlock_irq(&ev->lock);
1654 	WARN_ON_ONCE(clearing & mask);
1655 
1656 	return pending;
1657 }
1658 
1659 /*
1660  * Separate this part out so that a different pointer for clearing_ptr can be
1661  * passed in for disk_clear_events.
1662  */
1663 static void disk_events_workfn(struct work_struct *work)
1664 {
1665 	struct delayed_work *dwork = to_delayed_work(work);
1666 	struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1667 
1668 	disk_check_events(ev, &ev->clearing);
1669 }
1670 
1671 static void disk_check_events(struct disk_events *ev,
1672 			      unsigned int *clearing_ptr)
1673 {
1674 	struct gendisk *disk = ev->disk;
1675 	char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1676 	unsigned int clearing = *clearing_ptr;
1677 	unsigned int events;
1678 	unsigned long intv;
1679 	int nr_events = 0, i;
1680 
1681 	/* check events */
1682 	events = disk->fops->check_events(disk, clearing);
1683 
1684 	/* accumulate pending events and schedule next poll if necessary */
1685 	spin_lock_irq(&ev->lock);
1686 
1687 	events &= ~ev->pending;
1688 	ev->pending |= events;
1689 	*clearing_ptr &= ~clearing;
1690 
1691 	intv = disk_events_poll_jiffies(disk);
1692 	if (!ev->block && intv)
1693 		queue_delayed_work(system_freezable_power_efficient_wq,
1694 				&ev->dwork, intv);
1695 
1696 	spin_unlock_irq(&ev->lock);
1697 
1698 	/*
1699 	 * Tell userland about new events.  Only the events listed in
1700 	 * @disk->events are reported.  Unlisted events are processed the
1701 	 * same internally but never get reported to userland.
1702 	 */
1703 	for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1704 		if (events & disk->events & (1 << i))
1705 			envp[nr_events++] = disk_uevents[i];
1706 
1707 	if (nr_events)
1708 		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1709 }
1710 
1711 /*
1712  * A disk events enabled device has the following sysfs nodes under
1713  * its /sys/block/X/ directory.
1714  *
1715  * events		: list of all supported events
1716  * events_async		: list of events which can be detected w/o polling
1717  * events_poll_msecs	: polling interval, 0: disable, -1: system default
1718  */
1719 static ssize_t __disk_events_show(unsigned int events, char *buf)
1720 {
1721 	const char *delim = "";
1722 	ssize_t pos = 0;
1723 	int i;
1724 
1725 	for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1726 		if (events & (1 << i)) {
1727 			pos += sprintf(buf + pos, "%s%s",
1728 				       delim, disk_events_strs[i]);
1729 			delim = " ";
1730 		}
1731 	if (pos)
1732 		pos += sprintf(buf + pos, "\n");
1733 	return pos;
1734 }
1735 
1736 static ssize_t disk_events_show(struct device *dev,
1737 				struct device_attribute *attr, char *buf)
1738 {
1739 	struct gendisk *disk = dev_to_disk(dev);
1740 
1741 	return __disk_events_show(disk->events, buf);
1742 }
1743 
1744 static ssize_t disk_events_async_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->async_events, buf);
1750 }
1751 
1752 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1753 					   struct device_attribute *attr,
1754 					   char *buf)
1755 {
1756 	struct gendisk *disk = dev_to_disk(dev);
1757 
1758 	return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1759 }
1760 
1761 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1762 					    struct device_attribute *attr,
1763 					    const char *buf, size_t count)
1764 {
1765 	struct gendisk *disk = dev_to_disk(dev);
1766 	long intv;
1767 
1768 	if (!count || !sscanf(buf, "%ld", &intv))
1769 		return -EINVAL;
1770 
1771 	if (intv < 0 && intv != -1)
1772 		return -EINVAL;
1773 
1774 	disk_block_events(disk);
1775 	disk->ev->poll_msecs = intv;
1776 	__disk_unblock_events(disk, true);
1777 
1778 	return count;
1779 }
1780 
1781 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1782 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1783 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1784 			 disk_events_poll_msecs_show,
1785 			 disk_events_poll_msecs_store);
1786 
1787 static const struct attribute *disk_events_attrs[] = {
1788 	&dev_attr_events.attr,
1789 	&dev_attr_events_async.attr,
1790 	&dev_attr_events_poll_msecs.attr,
1791 	NULL,
1792 };
1793 
1794 /*
1795  * The default polling interval can be specified by the kernel
1796  * parameter block.events_dfl_poll_msecs which defaults to 0
1797  * (disable).  This can also be modified runtime by writing to
1798  * /sys/module/block/events_dfl_poll_msecs.
1799  */
1800 static int disk_events_set_dfl_poll_msecs(const char *val,
1801 					  const struct kernel_param *kp)
1802 {
1803 	struct disk_events *ev;
1804 	int ret;
1805 
1806 	ret = param_set_ulong(val, kp);
1807 	if (ret < 0)
1808 		return ret;
1809 
1810 	mutex_lock(&disk_events_mutex);
1811 
1812 	list_for_each_entry(ev, &disk_events, node)
1813 		disk_flush_events(ev->disk, 0);
1814 
1815 	mutex_unlock(&disk_events_mutex);
1816 
1817 	return 0;
1818 }
1819 
1820 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1821 	.set	= disk_events_set_dfl_poll_msecs,
1822 	.get	= param_get_ulong,
1823 };
1824 
1825 #undef MODULE_PARAM_PREFIX
1826 #define MODULE_PARAM_PREFIX	"block."
1827 
1828 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1829 		&disk_events_dfl_poll_msecs, 0644);
1830 
1831 /*
1832  * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1833  */
1834 static void disk_alloc_events(struct gendisk *disk)
1835 {
1836 	struct disk_events *ev;
1837 
1838 	if (!disk->fops->check_events)
1839 		return;
1840 
1841 	ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1842 	if (!ev) {
1843 		pr_warn("%s: failed to initialize events\n", disk->disk_name);
1844 		return;
1845 	}
1846 
1847 	INIT_LIST_HEAD(&ev->node);
1848 	ev->disk = disk;
1849 	spin_lock_init(&ev->lock);
1850 	mutex_init(&ev->block_mutex);
1851 	ev->block = 1;
1852 	ev->poll_msecs = -1;
1853 	INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1854 
1855 	disk->ev = ev;
1856 }
1857 
1858 static void disk_add_events(struct gendisk *disk)
1859 {
1860 	if (!disk->ev)
1861 		return;
1862 
1863 	/* FIXME: error handling */
1864 	if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1865 		pr_warn("%s: failed to create sysfs files for events\n",
1866 			disk->disk_name);
1867 
1868 	mutex_lock(&disk_events_mutex);
1869 	list_add_tail(&disk->ev->node, &disk_events);
1870 	mutex_unlock(&disk_events_mutex);
1871 
1872 	/*
1873 	 * Block count is initialized to 1 and the following initial
1874 	 * unblock kicks it into action.
1875 	 */
1876 	__disk_unblock_events(disk, true);
1877 }
1878 
1879 static void disk_del_events(struct gendisk *disk)
1880 {
1881 	if (!disk->ev)
1882 		return;
1883 
1884 	disk_block_events(disk);
1885 
1886 	mutex_lock(&disk_events_mutex);
1887 	list_del_init(&disk->ev->node);
1888 	mutex_unlock(&disk_events_mutex);
1889 
1890 	sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1891 }
1892 
1893 static void disk_release_events(struct gendisk *disk)
1894 {
1895 	/* the block count should be 1 from disk_del_events() */
1896 	WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1897 	kfree(disk->ev);
1898 }
1899