xref: /kernel/linux/linux-5.10/block/genhd.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0
2/*
3 *  gendisk handling
4 */
5
6#include <linux/module.h>
7#include <linux/ctype.h>
8#include <linux/fs.h>
9#include <linux/genhd.h>
10#include <linux/kdev_t.h>
11#include <linux/kernel.h>
12#include <linux/blkdev.h>
13#include <linux/backing-dev.h>
14#include <linux/init.h>
15#include <linux/spinlock.h>
16#include <linux/proc_fs.h>
17#include <linux/seq_file.h>
18#include <linux/slab.h>
19#include <linux/kmod.h>
20#include <linux/kobj_map.h>
21#include <linux/mutex.h>
22#include <linux/idr.h>
23#include <linux/log2.h>
24#include <linux/pm_runtime.h>
25#include <linux/badblocks.h>
26
27#include "blk.h"
28
29static DEFINE_MUTEX(block_class_lock);
30static struct kobject *block_depr;
31
32/* for extended dynamic devt allocation, currently only one major is used */
33#define NR_EXT_DEVT		(1 << MINORBITS)
34
35/* For extended devt allocation.  ext_devt_lock prevents look up
36 * results from going away underneath its user.
37 */
38static DEFINE_SPINLOCK(ext_devt_lock);
39static DEFINE_IDR(ext_devt_idr);
40
41static void disk_check_events(struct disk_events *ev,
42			      unsigned int *clearing_ptr);
43static void disk_alloc_events(struct gendisk *disk);
44static void disk_add_events(struct gendisk *disk);
45static void disk_del_events(struct gendisk *disk);
46static void disk_release_events(struct gendisk *disk);
47
48/*
49 * Set disk capacity and notify if the size is not currently
50 * zero and will not be set to zero
51 */
52bool set_capacity_revalidate_and_notify(struct gendisk *disk, sector_t size,
53					bool update_bdev)
54{
55	sector_t capacity = get_capacity(disk);
56
57	set_capacity(disk, size);
58	if (update_bdev)
59		revalidate_disk_size(disk, true);
60
61	if (capacity != size && capacity != 0 && size != 0) {
62		char *envp[] = { "RESIZE=1", NULL };
63
64		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
65		return true;
66	}
67
68	return false;
69}
70
71EXPORT_SYMBOL_GPL(set_capacity_revalidate_and_notify);
72
73/*
74 * Format the device name of the indicated disk into the supplied buffer and
75 * return a pointer to that same buffer for convenience.
76 */
77char *disk_name(struct gendisk *hd, int partno, char *buf)
78{
79	if (!partno)
80		snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
81	else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
82		snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
83	else
84		snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
85
86	return buf;
87}
88
89const char *bdevname(struct block_device *bdev, char *buf)
90{
91	return disk_name(bdev->bd_disk, bdev->bd_partno, buf);
92}
93EXPORT_SYMBOL(bdevname);
94
95static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
96{
97	int cpu;
98
99	memset(stat, 0, sizeof(struct disk_stats));
100	for_each_possible_cpu(cpu) {
101		struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
102		int group;
103
104		for (group = 0; group < NR_STAT_GROUPS; group++) {
105			stat->nsecs[group] += ptr->nsecs[group];
106			stat->sectors[group] += ptr->sectors[group];
107			stat->ios[group] += ptr->ios[group];
108			stat->merges[group] += ptr->merges[group];
109		}
110
111		stat->io_ticks += ptr->io_ticks;
112	}
113}
114
115static unsigned int part_in_flight(struct hd_struct *part)
116{
117	unsigned int inflight = 0;
118	int cpu;
119
120	for_each_possible_cpu(cpu) {
121		inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
122			    part_stat_local_read_cpu(part, in_flight[1], cpu);
123	}
124	if ((int)inflight < 0)
125		inflight = 0;
126
127	return inflight;
128}
129
130static void part_in_flight_rw(struct hd_struct *part, unsigned int inflight[2])
131{
132	int cpu;
133
134	inflight[0] = 0;
135	inflight[1] = 0;
136	for_each_possible_cpu(cpu) {
137		inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
138		inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
139	}
140	if ((int)inflight[0] < 0)
141		inflight[0] = 0;
142	if ((int)inflight[1] < 0)
143		inflight[1] = 0;
144}
145
146struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
147{
148	struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
149
150	if (unlikely(partno < 0 || partno >= ptbl->len))
151		return NULL;
152	return rcu_dereference(ptbl->part[partno]);
153}
154
155/**
156 * disk_get_part - get partition
157 * @disk: disk to look partition from
158 * @partno: partition number
159 *
160 * Look for partition @partno from @disk.  If found, increment
161 * reference count and return it.
162 *
163 * CONTEXT:
164 * Don't care.
165 *
166 * RETURNS:
167 * Pointer to the found partition on success, NULL if not found.
168 */
169struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
170{
171	struct hd_struct *part;
172
173	rcu_read_lock();
174	part = __disk_get_part(disk, partno);
175	if (part)
176		get_device(part_to_dev(part));
177	rcu_read_unlock();
178
179	return part;
180}
181
182/**
183 * disk_part_iter_init - initialize partition iterator
184 * @piter: iterator to initialize
185 * @disk: disk to iterate over
186 * @flags: DISK_PITER_* flags
187 *
188 * Initialize @piter so that it iterates over partitions of @disk.
189 *
190 * CONTEXT:
191 * Don't care.
192 */
193void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
194			  unsigned int flags)
195{
196	struct disk_part_tbl *ptbl;
197
198	rcu_read_lock();
199	ptbl = rcu_dereference(disk->part_tbl);
200
201	piter->disk = disk;
202	piter->part = NULL;
203
204	if (flags & DISK_PITER_REVERSE)
205		piter->idx = ptbl->len - 1;
206	else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
207		piter->idx = 0;
208	else
209		piter->idx = 1;
210
211	piter->flags = flags;
212
213	rcu_read_unlock();
214}
215EXPORT_SYMBOL_GPL(disk_part_iter_init);
216
217/**
218 * disk_part_iter_next - proceed iterator to the next partition and return it
219 * @piter: iterator of interest
220 *
221 * Proceed @piter to the next partition and return it.
222 *
223 * CONTEXT:
224 * Don't care.
225 */
226struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
227{
228	struct disk_part_tbl *ptbl;
229	int inc, end;
230
231	/* put the last partition */
232	disk_put_part(piter->part);
233	piter->part = NULL;
234
235	/* get part_tbl */
236	rcu_read_lock();
237	ptbl = rcu_dereference(piter->disk->part_tbl);
238
239	/* determine iteration parameters */
240	if (piter->flags & DISK_PITER_REVERSE) {
241		inc = -1;
242		if (piter->flags & (DISK_PITER_INCL_PART0 |
243				    DISK_PITER_INCL_EMPTY_PART0))
244			end = -1;
245		else
246			end = 0;
247	} else {
248		inc = 1;
249		end = ptbl->len;
250	}
251
252	/* iterate to the next partition */
253	for (; piter->idx != end; piter->idx += inc) {
254		struct hd_struct *part;
255
256		part = rcu_dereference(ptbl->part[piter->idx]);
257		if (!part)
258			continue;
259		get_device(part_to_dev(part));
260		piter->part = part;
261		if (!part_nr_sects_read(part) &&
262		    !(piter->flags & DISK_PITER_INCL_EMPTY) &&
263		    !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
264		      piter->idx == 0)) {
265			put_device(part_to_dev(part));
266			piter->part = NULL;
267			continue;
268		}
269
270		piter->idx += inc;
271		break;
272	}
273
274	rcu_read_unlock();
275
276	return piter->part;
277}
278EXPORT_SYMBOL_GPL(disk_part_iter_next);
279
280/**
281 * disk_part_iter_exit - finish up partition iteration
282 * @piter: iter of interest
283 *
284 * Called when iteration is over.  Cleans up @piter.
285 *
286 * CONTEXT:
287 * Don't care.
288 */
289void disk_part_iter_exit(struct disk_part_iter *piter)
290{
291	disk_put_part(piter->part);
292	piter->part = NULL;
293}
294EXPORT_SYMBOL_GPL(disk_part_iter_exit);
295
296static inline int sector_in_part(struct hd_struct *part, sector_t sector)
297{
298	return part->start_sect <= sector &&
299		sector < part->start_sect + part_nr_sects_read(part);
300}
301
302/**
303 * disk_map_sector_rcu - map sector to partition
304 * @disk: gendisk of interest
305 * @sector: sector to map
306 *
307 * Find out which partition @sector maps to on @disk.  This is
308 * primarily used for stats accounting.
309 *
310 * CONTEXT:
311 * RCU read locked.  The returned partition pointer is always valid
312 * because its refcount is grabbed except for part0, which lifetime
313 * is same with the disk.
314 *
315 * RETURNS:
316 * Found partition on success, part0 is returned if no partition matches
317 * or the matched partition is being deleted.
318 */
319struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
320{
321	struct disk_part_tbl *ptbl;
322	struct hd_struct *part;
323	int i;
324
325	rcu_read_lock();
326	ptbl = rcu_dereference(disk->part_tbl);
327
328	part = rcu_dereference(ptbl->last_lookup);
329	if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
330		goto out_unlock;
331
332	for (i = 1; i < ptbl->len; i++) {
333		part = rcu_dereference(ptbl->part[i]);
334
335		if (part && sector_in_part(part, sector)) {
336			/*
337			 * only live partition can be cached for lookup,
338			 * so use-after-free on cached & deleting partition
339			 * can be avoided
340			 */
341			if (!hd_struct_try_get(part))
342				break;
343			rcu_assign_pointer(ptbl->last_lookup, part);
344			goto out_unlock;
345		}
346	}
347
348	part = &disk->part0;
349out_unlock:
350	rcu_read_unlock();
351	return part;
352}
353
354/**
355 * disk_has_partitions
356 * @disk: gendisk of interest
357 *
358 * Walk through the partition table and check if valid partition exists.
359 *
360 * CONTEXT:
361 * Don't care.
362 *
363 * RETURNS:
364 * True if the gendisk has at least one valid non-zero size partition.
365 * Otherwise false.
366 */
367bool disk_has_partitions(struct gendisk *disk)
368{
369	struct disk_part_tbl *ptbl;
370	int i;
371	bool ret = false;
372
373	rcu_read_lock();
374	ptbl = rcu_dereference(disk->part_tbl);
375
376	/* Iterate partitions skipping the whole device at index 0 */
377	for (i = 1; i < ptbl->len; i++) {
378		if (rcu_dereference(ptbl->part[i])) {
379			ret = true;
380			break;
381		}
382	}
383
384	rcu_read_unlock();
385
386	return ret;
387}
388EXPORT_SYMBOL_GPL(disk_has_partitions);
389
390/*
391 * Can be deleted altogether. Later.
392 *
393 */
394#define BLKDEV_MAJOR_HASH_SIZE 255
395static struct blk_major_name {
396	struct blk_major_name *next;
397	int major;
398	char name[16];
399} *major_names[BLKDEV_MAJOR_HASH_SIZE];
400
401/* index in the above - for now: assume no multimajor ranges */
402static inline int major_to_index(unsigned major)
403{
404	return major % BLKDEV_MAJOR_HASH_SIZE;
405}
406
407#ifdef CONFIG_PROC_FS
408void blkdev_show(struct seq_file *seqf, off_t offset)
409{
410	struct blk_major_name *dp;
411
412	mutex_lock(&block_class_lock);
413	for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
414		if (dp->major == offset)
415			seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
416	mutex_unlock(&block_class_lock);
417}
418#endif /* CONFIG_PROC_FS */
419
420/**
421 * register_blkdev - register a new block device
422 *
423 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
424 *         @major = 0, try to allocate any unused major number.
425 * @name: the name of the new block device as a zero terminated string
426 *
427 * The @name must be unique within the system.
428 *
429 * The return value depends on the @major input parameter:
430 *
431 *  - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
432 *    then the function returns zero on success, or a negative error code
433 *  - if any unused major number was requested with @major = 0 parameter
434 *    then the return value is the allocated major number in range
435 *    [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
436 *
437 * See Documentation/admin-guide/devices.txt for the list of allocated
438 * major numbers.
439 */
440int register_blkdev(unsigned int major, const char *name)
441{
442	struct blk_major_name **n, *p;
443	int index, ret = 0;
444
445	mutex_lock(&block_class_lock);
446
447	/* temporary */
448	if (major == 0) {
449		for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
450			if (major_names[index] == NULL)
451				break;
452		}
453
454		if (index == 0) {
455			printk("%s: failed to get major for %s\n",
456			       __func__, name);
457			ret = -EBUSY;
458			goto out;
459		}
460		major = index;
461		ret = major;
462	}
463
464	if (major >= BLKDEV_MAJOR_MAX) {
465		pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
466		       __func__, major, BLKDEV_MAJOR_MAX-1, name);
467
468		ret = -EINVAL;
469		goto out;
470	}
471
472	p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
473	if (p == NULL) {
474		ret = -ENOMEM;
475		goto out;
476	}
477
478	p->major = major;
479	strlcpy(p->name, name, sizeof(p->name));
480	p->next = NULL;
481	index = major_to_index(major);
482
483	for (n = &major_names[index]; *n; n = &(*n)->next) {
484		if ((*n)->major == major)
485			break;
486	}
487	if (!*n)
488		*n = p;
489	else
490		ret = -EBUSY;
491
492	if (ret < 0) {
493		printk("register_blkdev: cannot get major %u for %s\n",
494		       major, name);
495		kfree(p);
496	}
497out:
498	mutex_unlock(&block_class_lock);
499	return ret;
500}
501
502EXPORT_SYMBOL(register_blkdev);
503
504void unregister_blkdev(unsigned int major, const char *name)
505{
506	struct blk_major_name **n;
507	struct blk_major_name *p = NULL;
508	int index = major_to_index(major);
509
510	mutex_lock(&block_class_lock);
511	for (n = &major_names[index]; *n; n = &(*n)->next)
512		if ((*n)->major == major)
513			break;
514	if (!*n || strcmp((*n)->name, name)) {
515		WARN_ON(1);
516	} else {
517		p = *n;
518		*n = p->next;
519	}
520	mutex_unlock(&block_class_lock);
521	kfree(p);
522}
523
524EXPORT_SYMBOL(unregister_blkdev);
525
526static struct kobj_map *bdev_map;
527
528/**
529 * blk_mangle_minor - scatter minor numbers apart
530 * @minor: minor number to mangle
531 *
532 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
533 * is enabled.  Mangling twice gives the original value.
534 *
535 * RETURNS:
536 * Mangled value.
537 *
538 * CONTEXT:
539 * Don't care.
540 */
541static int blk_mangle_minor(int minor)
542{
543#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
544	int i;
545
546	for (i = 0; i < MINORBITS / 2; i++) {
547		int low = minor & (1 << i);
548		int high = minor & (1 << (MINORBITS - 1 - i));
549		int distance = MINORBITS - 1 - 2 * i;
550
551		minor ^= low | high;	/* clear both bits */
552		low <<= distance;	/* swap the positions */
553		high >>= distance;
554		minor |= low | high;	/* and set */
555	}
556#endif
557	return minor;
558}
559
560/**
561 * blk_alloc_devt - allocate a dev_t for a partition
562 * @part: partition to allocate dev_t for
563 * @devt: out parameter for resulting dev_t
564 *
565 * Allocate a dev_t for block device.
566 *
567 * RETURNS:
568 * 0 on success, allocated dev_t is returned in *@devt.  -errno on
569 * failure.
570 *
571 * CONTEXT:
572 * Might sleep.
573 */
574int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
575{
576	struct gendisk *disk = part_to_disk(part);
577	int idx;
578
579	/* in consecutive minor range? */
580	if (part->partno < disk->minors) {
581		*devt = MKDEV(disk->major, disk->first_minor + part->partno);
582		return 0;
583	}
584
585	/* allocate ext devt */
586	idr_preload(GFP_KERNEL);
587
588	spin_lock_bh(&ext_devt_lock);
589	idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
590	spin_unlock_bh(&ext_devt_lock);
591
592	idr_preload_end();
593	if (idx < 0)
594		return idx == -ENOSPC ? -EBUSY : idx;
595
596	*devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
597	return 0;
598}
599
600/**
601 * blk_free_devt - free a dev_t
602 * @devt: dev_t to free
603 *
604 * Free @devt which was allocated using blk_alloc_devt().
605 *
606 * CONTEXT:
607 * Might sleep.
608 */
609void blk_free_devt(dev_t devt)
610{
611	if (devt == MKDEV(0, 0))
612		return;
613
614	if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
615		spin_lock_bh(&ext_devt_lock);
616		idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
617		spin_unlock_bh(&ext_devt_lock);
618	}
619}
620
621/*
622 * We invalidate devt by assigning NULL pointer for devt in idr.
623 */
624void blk_invalidate_devt(dev_t devt)
625{
626	if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
627		spin_lock_bh(&ext_devt_lock);
628		idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
629		spin_unlock_bh(&ext_devt_lock);
630	}
631}
632
633static char *bdevt_str(dev_t devt, char *buf)
634{
635	if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
636		char tbuf[BDEVT_SIZE];
637		snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
638		snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
639	} else
640		snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
641
642	return buf;
643}
644
645/*
646 * Register device numbers dev..(dev+range-1)
647 * range must be nonzero
648 * The hash chain is sorted on range, so that subranges can override.
649 */
650void blk_register_region(dev_t devt, unsigned long range, struct module *module,
651			 struct kobject *(*probe)(dev_t, int *, void *),
652			 int (*lock)(dev_t, void *), void *data)
653{
654	kobj_map(bdev_map, devt, range, module, probe, lock, data);
655}
656
657EXPORT_SYMBOL(blk_register_region);
658
659void blk_unregister_region(dev_t devt, unsigned long range)
660{
661	kobj_unmap(bdev_map, devt, range);
662}
663
664EXPORT_SYMBOL(blk_unregister_region);
665
666void blk_delete_region(dev_t devt, unsigned long range,
667			struct kobject *(*probe)(dev_t, int *, void *))
668{
669	kobj_delete(bdev_map, devt, range, probe);
670}
671
672EXPORT_SYMBOL(blk_delete_region);
673
674static struct kobject *exact_match(dev_t devt, int *partno, void *data)
675{
676	struct gendisk *p = data;
677
678	return &disk_to_dev(p)->kobj;
679}
680
681static int exact_lock(dev_t devt, void *data)
682{
683	struct gendisk *p = data;
684
685	if (!get_disk_and_module(p))
686		return -1;
687	return 0;
688}
689
690static void disk_scan_partitions(struct gendisk *disk)
691{
692	struct block_device *bdev;
693
694	if (!get_capacity(disk) || !disk_part_scan_enabled(disk))
695		return;
696
697	set_bit(GD_NEED_PART_SCAN, &disk->state);
698	bdev = blkdev_get_by_dev(disk_devt(disk), FMODE_READ, NULL);
699	if (!IS_ERR(bdev))
700		blkdev_put(bdev, FMODE_READ);
701}
702
703static void register_disk(struct device *parent, struct gendisk *disk,
704			  const struct attribute_group **groups)
705{
706	struct device *ddev = disk_to_dev(disk);
707	struct disk_part_iter piter;
708	struct hd_struct *part;
709	int err;
710
711	ddev->parent = parent;
712
713	dev_set_name(ddev, "%s", disk->disk_name);
714
715	/* delay uevents, until we scanned partition table */
716	dev_set_uevent_suppress(ddev, 1);
717
718	if (groups) {
719		WARN_ON(ddev->groups);
720		ddev->groups = groups;
721	}
722	if (device_add(ddev))
723		return;
724	if (!sysfs_deprecated) {
725		err = sysfs_create_link(block_depr, &ddev->kobj,
726					kobject_name(&ddev->kobj));
727		if (err) {
728			device_del(ddev);
729			return;
730		}
731	}
732
733	/*
734	 * avoid probable deadlock caused by allocating memory with
735	 * GFP_KERNEL in runtime_resume callback of its all ancestor
736	 * devices
737	 */
738	pm_runtime_set_memalloc_noio(ddev, true);
739
740	disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
741	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
742
743	if (disk->flags & GENHD_FL_HIDDEN)
744		return;
745
746	disk_scan_partitions(disk);
747
748	/* announce disk after possible partitions are created */
749	dev_set_uevent_suppress(ddev, 0);
750	kobject_uevent(&ddev->kobj, KOBJ_ADD);
751
752	/* announce possible partitions */
753	disk_part_iter_init(&piter, disk, 0);
754	while ((part = disk_part_iter_next(&piter)))
755		kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
756	disk_part_iter_exit(&piter);
757
758	if (disk->queue->backing_dev_info->dev) {
759		err = sysfs_create_link(&ddev->kobj,
760			  &disk->queue->backing_dev_info->dev->kobj,
761			  "bdi");
762		WARN_ON(err);
763	}
764}
765
766/**
767 * __device_add_disk - add disk information to kernel list
768 * @parent: parent device for the disk
769 * @disk: per-device partitioning information
770 * @groups: Additional per-device sysfs groups
771 * @register_queue: register the queue if set to true
772 *
773 * This function registers the partitioning information in @disk
774 * with the kernel.
775 *
776 * FIXME: error handling
777 */
778static void __device_add_disk(struct device *parent, struct gendisk *disk,
779			      const struct attribute_group **groups,
780			      bool register_queue)
781{
782	dev_t devt;
783	int retval;
784
785	/*
786	 * The disk queue should now be all set with enough information about
787	 * the device for the elevator code to pick an adequate default
788	 * elevator if one is needed, that is, for devices requesting queue
789	 * registration.
790	 */
791	if (register_queue)
792		elevator_init_mq(disk->queue);
793
794	/* minors == 0 indicates to use ext devt from part0 and should
795	 * be accompanied with EXT_DEVT flag.  Make sure all
796	 * parameters make sense.
797	 */
798	WARN_ON(disk->minors && !(disk->major || disk->first_minor));
799	WARN_ON(!disk->minors &&
800		!(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
801
802	disk->flags |= GENHD_FL_UP;
803
804	retval = blk_alloc_devt(&disk->part0, &devt);
805	if (retval) {
806		WARN_ON(1);
807		return;
808	}
809	disk->major = MAJOR(devt);
810	disk->first_minor = MINOR(devt);
811
812	disk_alloc_events(disk);
813
814	if (disk->flags & GENHD_FL_HIDDEN) {
815		/*
816		 * Don't let hidden disks show up in /proc/partitions,
817		 * and don't bother scanning for partitions either.
818		 */
819		disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
820		disk->flags |= GENHD_FL_NO_PART_SCAN;
821	} else {
822		struct backing_dev_info *bdi = disk->queue->backing_dev_info;
823		struct device *dev = disk_to_dev(disk);
824		int ret;
825
826		/* Register BDI before referencing it from bdev */
827		dev->devt = devt;
828		ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
829		WARN_ON(ret);
830		bdi_set_owner(bdi, dev);
831		blk_register_region(disk_devt(disk), disk->minors, NULL,
832				    exact_match, exact_lock, disk);
833	}
834	register_disk(parent, disk, groups);
835	if (register_queue)
836		blk_register_queue(disk);
837
838	/*
839	 * Take an extra ref on queue which will be put on disk_release()
840	 * so that it sticks around as long as @disk is there.
841	 */
842	WARN_ON_ONCE(!blk_get_queue(disk->queue));
843
844	disk_add_events(disk);
845	blk_integrity_add(disk);
846}
847
848void device_add_disk(struct device *parent, struct gendisk *disk,
849		     const struct attribute_group **groups)
850
851{
852	__device_add_disk(parent, disk, groups, true);
853}
854EXPORT_SYMBOL(device_add_disk);
855
856void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
857{
858	__device_add_disk(parent, disk, NULL, false);
859}
860EXPORT_SYMBOL(device_add_disk_no_queue_reg);
861
862static void invalidate_partition(struct gendisk *disk, int partno)
863{
864	struct block_device *bdev;
865
866	bdev = bdget_disk(disk, partno);
867	if (!bdev)
868		return;
869
870	fsync_bdev(bdev);
871	__invalidate_device(bdev, true);
872
873	/*
874	 * Unhash the bdev inode for this device so that it gets evicted as soon
875	 * as last inode reference is dropped.
876	 */
877	remove_inode_hash(bdev->bd_inode);
878	bdput(bdev);
879}
880
881/**
882 * del_gendisk - remove the gendisk
883 * @disk: the struct gendisk to remove
884 *
885 * Removes the gendisk and all its associated resources. This deletes the
886 * partitions associated with the gendisk, and unregisters the associated
887 * request_queue.
888 *
889 * This is the counter to the respective __device_add_disk() call.
890 *
891 * The final removal of the struct gendisk happens when its refcount reaches 0
892 * with put_disk(), which should be called after del_gendisk(), if
893 * __device_add_disk() was used.
894 *
895 * Drivers exist which depend on the release of the gendisk to be synchronous,
896 * it should not be deferred.
897 *
898 * Context: can sleep
899 */
900void del_gendisk(struct gendisk *disk)
901{
902	struct disk_part_iter piter;
903	struct hd_struct *part;
904
905	might_sleep();
906
907	blk_integrity_del(disk);
908	disk_del_events(disk);
909
910	/*
911	 * Block lookups of the disk until all bdevs are unhashed and the
912	 * disk is marked as dead (GENHD_FL_UP cleared).
913	 */
914	down_write(&disk->lookup_sem);
915	/* invalidate stuff */
916	disk_part_iter_init(&piter, disk,
917			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
918	while ((part = disk_part_iter_next(&piter))) {
919		invalidate_partition(disk, part->partno);
920		delete_partition(part);
921	}
922	disk_part_iter_exit(&piter);
923
924	invalidate_partition(disk, 0);
925	set_capacity(disk, 0);
926	disk->flags &= ~GENHD_FL_UP;
927	up_write(&disk->lookup_sem);
928
929	if (!(disk->flags & GENHD_FL_HIDDEN))
930		sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
931	if (disk->queue) {
932		/*
933		 * Unregister bdi before releasing device numbers (as they can
934		 * get reused and we'd get clashes in sysfs).
935		 */
936		if (!(disk->flags & GENHD_FL_HIDDEN))
937			bdi_unregister(disk->queue->backing_dev_info);
938		blk_unregister_queue(disk);
939	} else {
940		WARN_ON(1);
941	}
942
943	if (!(disk->flags & GENHD_FL_HIDDEN))
944		blk_unregister_region(disk_devt(disk), disk->minors);
945	/*
946	 * Remove gendisk pointer from idr so that it cannot be looked up
947	 * while RCU period before freeing gendisk is running to prevent
948	 * use-after-free issues. Note that the device number stays
949	 * "in-use" until we really free the gendisk.
950	 */
951	blk_invalidate_devt(disk_devt(disk));
952
953	kobject_put(disk->part0.holder_dir);
954	kobject_put(disk->slave_dir);
955
956	part_stat_set_all(&disk->part0, 0);
957	disk->part0.stamp = 0;
958	if (!sysfs_deprecated)
959		sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
960	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
961	device_del(disk_to_dev(disk));
962}
963EXPORT_SYMBOL(del_gendisk);
964
965/* sysfs access to bad-blocks list. */
966static ssize_t disk_badblocks_show(struct device *dev,
967					struct device_attribute *attr,
968					char *page)
969{
970	struct gendisk *disk = dev_to_disk(dev);
971
972	if (!disk->bb)
973		return sprintf(page, "\n");
974
975	return badblocks_show(disk->bb, page, 0);
976}
977
978static ssize_t disk_badblocks_store(struct device *dev,
979					struct device_attribute *attr,
980					const char *page, size_t len)
981{
982	struct gendisk *disk = dev_to_disk(dev);
983
984	if (!disk->bb)
985		return -ENXIO;
986
987	return badblocks_store(disk->bb, page, len, 0);
988}
989
990/**
991 * get_gendisk - get partitioning information for a given device
992 * @devt: device to get partitioning information for
993 * @partno: returned partition index
994 *
995 * This function gets the structure containing partitioning
996 * information for the given device @devt.
997 *
998 * Context: can sleep
999 */
1000struct gendisk *get_gendisk(dev_t devt, int *partno)
1001{
1002	struct gendisk *disk = NULL;
1003
1004	might_sleep();
1005
1006	if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
1007		struct kobject *kobj;
1008
1009		kobj = kobj_lookup(bdev_map, devt, partno);
1010		if (kobj)
1011			disk = dev_to_disk(kobj_to_dev(kobj));
1012	} else {
1013		struct hd_struct *part;
1014
1015		spin_lock_bh(&ext_devt_lock);
1016		part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
1017		if (part && get_disk_and_module(part_to_disk(part))) {
1018			*partno = part->partno;
1019			disk = part_to_disk(part);
1020		}
1021		spin_unlock_bh(&ext_devt_lock);
1022	}
1023
1024	if (!disk)
1025		return NULL;
1026
1027	/*
1028	 * Synchronize with del_gendisk() to not return disk that is being
1029	 * destroyed.
1030	 */
1031	down_read(&disk->lookup_sem);
1032	if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1033		     !(disk->flags & GENHD_FL_UP))) {
1034		up_read(&disk->lookup_sem);
1035		put_disk_and_module(disk);
1036		disk = NULL;
1037	} else {
1038		up_read(&disk->lookup_sem);
1039	}
1040	return disk;
1041}
1042
1043/**
1044 * bdget_disk - do bdget() by gendisk and partition number
1045 * @disk: gendisk of interest
1046 * @partno: partition number
1047 *
1048 * Find partition @partno from @disk, do bdget() on it.
1049 *
1050 * CONTEXT:
1051 * Don't care.
1052 *
1053 * RETURNS:
1054 * Resulting block_device on success, NULL on failure.
1055 */
1056struct block_device *bdget_disk(struct gendisk *disk, int partno)
1057{
1058	struct hd_struct *part;
1059	struct block_device *bdev = NULL;
1060
1061	part = disk_get_part(disk, partno);
1062	if (part)
1063		bdev = bdget_part(part);
1064	disk_put_part(part);
1065
1066	return bdev;
1067}
1068EXPORT_SYMBOL(bdget_disk);
1069
1070/*
1071 * print a full list of all partitions - intended for places where the root
1072 * filesystem can't be mounted and thus to give the victim some idea of what
1073 * went wrong
1074 */
1075void __init printk_all_partitions(void)
1076{
1077	struct class_dev_iter iter;
1078	struct device *dev;
1079
1080	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1081	while ((dev = class_dev_iter_next(&iter))) {
1082		struct gendisk *disk = dev_to_disk(dev);
1083		struct disk_part_iter piter;
1084		struct hd_struct *part;
1085		char name_buf[BDEVNAME_SIZE];
1086		char devt_buf[BDEVT_SIZE];
1087
1088		/*
1089		 * Don't show empty devices or things that have been
1090		 * suppressed
1091		 */
1092		if (get_capacity(disk) == 0 ||
1093		    (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1094			continue;
1095
1096		/*
1097		 * Note, unlike /proc/partitions, I am showing the
1098		 * numbers in hex - the same format as the root=
1099		 * option takes.
1100		 */
1101		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1102		while ((part = disk_part_iter_next(&piter))) {
1103			bool is_part0 = part == &disk->part0;
1104
1105			printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
1106			       bdevt_str(part_devt(part), devt_buf),
1107			       (unsigned long long)part_nr_sects_read(part) >> 1
1108			       , disk_name(disk, part->partno, name_buf),
1109			       part->info ? part->info->uuid : "");
1110			if (is_part0) {
1111				if (dev->parent && dev->parent->driver)
1112					printk(" driver: %s\n",
1113					      dev->parent->driver->name);
1114				else
1115					printk(" (driver?)\n");
1116			} else
1117				printk("\n");
1118		}
1119		disk_part_iter_exit(&piter);
1120	}
1121	class_dev_iter_exit(&iter);
1122}
1123
1124#ifdef CONFIG_PROC_FS
1125/* iterator */
1126static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1127{
1128	loff_t skip = *pos;
1129	struct class_dev_iter *iter;
1130	struct device *dev;
1131
1132	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1133	if (!iter)
1134		return ERR_PTR(-ENOMEM);
1135
1136	seqf->private = iter;
1137	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1138	do {
1139		dev = class_dev_iter_next(iter);
1140		if (!dev)
1141			return NULL;
1142	} while (skip--);
1143
1144	return dev_to_disk(dev);
1145}
1146
1147static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1148{
1149	struct device *dev;
1150
1151	(*pos)++;
1152	dev = class_dev_iter_next(seqf->private);
1153	if (dev)
1154		return dev_to_disk(dev);
1155
1156	return NULL;
1157}
1158
1159static void disk_seqf_stop(struct seq_file *seqf, void *v)
1160{
1161	struct class_dev_iter *iter = seqf->private;
1162
1163	/* stop is called even after start failed :-( */
1164	if (iter) {
1165		class_dev_iter_exit(iter);
1166		kfree(iter);
1167		seqf->private = NULL;
1168	}
1169}
1170
1171static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1172{
1173	void *p;
1174
1175	p = disk_seqf_start(seqf, pos);
1176	if (!IS_ERR_OR_NULL(p) && !*pos)
1177		seq_puts(seqf, "major minor  #blocks  name\n\n");
1178	return p;
1179}
1180
1181static int show_partition(struct seq_file *seqf, void *v)
1182{
1183	struct gendisk *sgp = v;
1184	struct disk_part_iter piter;
1185	struct hd_struct *part;
1186	char buf[BDEVNAME_SIZE];
1187
1188	/* Don't show non-partitionable removeable devices or empty devices */
1189	if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1190				   (sgp->flags & GENHD_FL_REMOVABLE)))
1191		return 0;
1192	if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1193		return 0;
1194
1195	/* show the full disk and all non-0 size partitions of it */
1196	disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1197	while ((part = disk_part_iter_next(&piter)))
1198		seq_printf(seqf, "%4d  %7d %10llu %s\n",
1199			   MAJOR(part_devt(part)), MINOR(part_devt(part)),
1200			   (unsigned long long)part_nr_sects_read(part) >> 1,
1201			   disk_name(sgp, part->partno, buf));
1202	disk_part_iter_exit(&piter);
1203
1204	return 0;
1205}
1206
1207static const struct seq_operations partitions_op = {
1208	.start	= show_partition_start,
1209	.next	= disk_seqf_next,
1210	.stop	= disk_seqf_stop,
1211	.show	= show_partition
1212};
1213#endif
1214
1215
1216static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1217{
1218	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1219		/* Make old-style 2.4 aliases work */
1220		request_module("block-major-%d", MAJOR(devt));
1221	return NULL;
1222}
1223
1224static int __init genhd_device_init(void)
1225{
1226	int error;
1227
1228	block_class.dev_kobj = sysfs_dev_block_kobj;
1229	error = class_register(&block_class);
1230	if (unlikely(error))
1231		return error;
1232	bdev_map = kobj_map_init(base_probe, &block_class_lock);
1233	blk_dev_init();
1234
1235	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1236
1237	/* create top-level block dir */
1238	if (!sysfs_deprecated)
1239		block_depr = kobject_create_and_add("block", NULL);
1240	return 0;
1241}
1242
1243subsys_initcall(genhd_device_init);
1244
1245static ssize_t disk_range_show(struct device *dev,
1246			       struct device_attribute *attr, char *buf)
1247{
1248	struct gendisk *disk = dev_to_disk(dev);
1249
1250	return sprintf(buf, "%d\n", disk->minors);
1251}
1252
1253static ssize_t disk_ext_range_show(struct device *dev,
1254				   struct device_attribute *attr, char *buf)
1255{
1256	struct gendisk *disk = dev_to_disk(dev);
1257
1258	return sprintf(buf, "%d\n", disk_max_parts(disk));
1259}
1260
1261static ssize_t disk_removable_show(struct device *dev,
1262				   struct device_attribute *attr, char *buf)
1263{
1264	struct gendisk *disk = dev_to_disk(dev);
1265
1266	return sprintf(buf, "%d\n",
1267		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1268}
1269
1270static ssize_t disk_hidden_show(struct device *dev,
1271				   struct device_attribute *attr, char *buf)
1272{
1273	struct gendisk *disk = dev_to_disk(dev);
1274
1275	return sprintf(buf, "%d\n",
1276		       (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1277}
1278
1279static ssize_t disk_ro_show(struct device *dev,
1280				   struct device_attribute *attr, char *buf)
1281{
1282	struct gendisk *disk = dev_to_disk(dev);
1283
1284	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1285}
1286
1287ssize_t part_size_show(struct device *dev,
1288		       struct device_attribute *attr, char *buf)
1289{
1290	struct hd_struct *p = dev_to_part(dev);
1291
1292	return sprintf(buf, "%llu\n",
1293		(unsigned long long)part_nr_sects_read(p));
1294}
1295
1296ssize_t part_stat_show(struct device *dev,
1297		       struct device_attribute *attr, char *buf)
1298{
1299	struct hd_struct *p = dev_to_part(dev);
1300	struct request_queue *q = part_to_disk(p)->queue;
1301	struct disk_stats stat;
1302	unsigned int inflight;
1303
1304	part_stat_read_all(p, &stat);
1305	if (queue_is_mq(q))
1306		inflight = blk_mq_in_flight(q, p);
1307	else
1308		inflight = part_in_flight(p);
1309
1310	return sprintf(buf,
1311		"%8lu %8lu %8llu %8u "
1312		"%8lu %8lu %8llu %8u "
1313		"%8u %8u %8u "
1314		"%8lu %8lu %8llu %8u "
1315		"%8lu %8u"
1316		"\n",
1317		stat.ios[STAT_READ],
1318		stat.merges[STAT_READ],
1319		(unsigned long long)stat.sectors[STAT_READ],
1320		(unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1321		stat.ios[STAT_WRITE],
1322		stat.merges[STAT_WRITE],
1323		(unsigned long long)stat.sectors[STAT_WRITE],
1324		(unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1325		inflight,
1326		jiffies_to_msecs(stat.io_ticks),
1327		(unsigned int)div_u64(stat.nsecs[STAT_READ] +
1328				      stat.nsecs[STAT_WRITE] +
1329				      stat.nsecs[STAT_DISCARD] +
1330				      stat.nsecs[STAT_FLUSH],
1331						NSEC_PER_MSEC),
1332		stat.ios[STAT_DISCARD],
1333		stat.merges[STAT_DISCARD],
1334		(unsigned long long)stat.sectors[STAT_DISCARD],
1335		(unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1336		stat.ios[STAT_FLUSH],
1337		(unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1338}
1339
1340ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1341			   char *buf)
1342{
1343	struct hd_struct *p = dev_to_part(dev);
1344	struct request_queue *q = part_to_disk(p)->queue;
1345	unsigned int inflight[2];
1346
1347	if (queue_is_mq(q))
1348		blk_mq_in_flight_rw(q, p, inflight);
1349	else
1350		part_in_flight_rw(p, inflight);
1351
1352	return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1353}
1354
1355static ssize_t disk_capability_show(struct device *dev,
1356				    struct device_attribute *attr, char *buf)
1357{
1358	struct gendisk *disk = dev_to_disk(dev);
1359
1360	return sprintf(buf, "%x\n", disk->flags);
1361}
1362
1363static ssize_t disk_alignment_offset_show(struct device *dev,
1364					  struct device_attribute *attr,
1365					  char *buf)
1366{
1367	struct gendisk *disk = dev_to_disk(dev);
1368
1369	return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1370}
1371
1372static ssize_t disk_discard_alignment_show(struct device *dev,
1373					   struct device_attribute *attr,
1374					   char *buf)
1375{
1376	struct gendisk *disk = dev_to_disk(dev);
1377
1378	return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1379}
1380
1381static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1382static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1383static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1384static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1385static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1386static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1387static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1388static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1389static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1390static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1391static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1392static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1393
1394#ifdef CONFIG_FAIL_MAKE_REQUEST
1395ssize_t part_fail_show(struct device *dev,
1396		       struct device_attribute *attr, char *buf)
1397{
1398	struct hd_struct *p = dev_to_part(dev);
1399
1400	return sprintf(buf, "%d\n", p->make_it_fail);
1401}
1402
1403ssize_t part_fail_store(struct device *dev,
1404			struct device_attribute *attr,
1405			const char *buf, size_t count)
1406{
1407	struct hd_struct *p = dev_to_part(dev);
1408	int i;
1409
1410	if (count > 0 && sscanf(buf, "%d", &i) > 0)
1411		p->make_it_fail = (i == 0) ? 0 : 1;
1412
1413	return count;
1414}
1415
1416static struct device_attribute dev_attr_fail =
1417	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1418#endif /* CONFIG_FAIL_MAKE_REQUEST */
1419
1420#ifdef CONFIG_FAIL_IO_TIMEOUT
1421static struct device_attribute dev_attr_fail_timeout =
1422	__ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1423#endif
1424
1425static struct attribute *disk_attrs[] = {
1426	&dev_attr_range.attr,
1427	&dev_attr_ext_range.attr,
1428	&dev_attr_removable.attr,
1429	&dev_attr_hidden.attr,
1430	&dev_attr_ro.attr,
1431	&dev_attr_size.attr,
1432	&dev_attr_alignment_offset.attr,
1433	&dev_attr_discard_alignment.attr,
1434	&dev_attr_capability.attr,
1435	&dev_attr_stat.attr,
1436	&dev_attr_inflight.attr,
1437	&dev_attr_badblocks.attr,
1438#ifdef CONFIG_FAIL_MAKE_REQUEST
1439	&dev_attr_fail.attr,
1440#endif
1441#ifdef CONFIG_FAIL_IO_TIMEOUT
1442	&dev_attr_fail_timeout.attr,
1443#endif
1444	NULL
1445};
1446
1447static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1448{
1449	struct device *dev = container_of(kobj, typeof(*dev), kobj);
1450	struct gendisk *disk = dev_to_disk(dev);
1451
1452	if (a == &dev_attr_badblocks.attr && !disk->bb)
1453		return 0;
1454	return a->mode;
1455}
1456
1457static struct attribute_group disk_attr_group = {
1458	.attrs = disk_attrs,
1459	.is_visible = disk_visible,
1460};
1461
1462static const struct attribute_group *disk_attr_groups[] = {
1463	&disk_attr_group,
1464	NULL
1465};
1466
1467/**
1468 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1469 * @disk: disk to replace part_tbl for
1470 * @new_ptbl: new part_tbl to install
1471 *
1472 * Replace disk->part_tbl with @new_ptbl in RCU-safe way.  The
1473 * original ptbl is freed using RCU callback.
1474 *
1475 * LOCKING:
1476 * Matching bd_mutex locked or the caller is the only user of @disk.
1477 */
1478static void disk_replace_part_tbl(struct gendisk *disk,
1479				  struct disk_part_tbl *new_ptbl)
1480{
1481	struct disk_part_tbl *old_ptbl =
1482		rcu_dereference_protected(disk->part_tbl, 1);
1483
1484	rcu_assign_pointer(disk->part_tbl, new_ptbl);
1485
1486	if (old_ptbl) {
1487		rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1488		kfree_rcu(old_ptbl, rcu_head);
1489	}
1490}
1491
1492/**
1493 * disk_expand_part_tbl - expand disk->part_tbl
1494 * @disk: disk to expand part_tbl for
1495 * @partno: expand such that this partno can fit in
1496 *
1497 * Expand disk->part_tbl such that @partno can fit in.  disk->part_tbl
1498 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1499 *
1500 * LOCKING:
1501 * Matching bd_mutex locked or the caller is the only user of @disk.
1502 * Might sleep.
1503 *
1504 * RETURNS:
1505 * 0 on success, -errno on failure.
1506 */
1507int disk_expand_part_tbl(struct gendisk *disk, int partno)
1508{
1509	struct disk_part_tbl *old_ptbl =
1510		rcu_dereference_protected(disk->part_tbl, 1);
1511	struct disk_part_tbl *new_ptbl;
1512	int len = old_ptbl ? old_ptbl->len : 0;
1513	int i, target;
1514
1515	/*
1516	 * check for int overflow, since we can get here from blkpg_ioctl()
1517	 * with a user passed 'partno'.
1518	 */
1519	target = partno + 1;
1520	if (target < 0)
1521		return -EINVAL;
1522
1523	/* disk_max_parts() is zero during initialization, ignore if so */
1524	if (disk_max_parts(disk) && target > disk_max_parts(disk))
1525		return -EINVAL;
1526
1527	if (target <= len)
1528		return 0;
1529
1530	new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1531				disk->node_id);
1532	if (!new_ptbl)
1533		return -ENOMEM;
1534
1535	new_ptbl->len = target;
1536
1537	for (i = 0; i < len; i++)
1538		rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1539
1540	disk_replace_part_tbl(disk, new_ptbl);
1541	return 0;
1542}
1543
1544/**
1545 * disk_release - releases all allocated resources of the gendisk
1546 * @dev: the device representing this disk
1547 *
1548 * This function releases all allocated resources of the gendisk.
1549 *
1550 * The struct gendisk refcount is incremented with get_gendisk() or
1551 * get_disk_and_module(), and its refcount is decremented with
1552 * put_disk_and_module() or put_disk(). Once the refcount reaches 0 this
1553 * function is called.
1554 *
1555 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1556 * assigned. Since the request_queue sits on top of the gendisk for these
1557 * drivers we also call blk_put_queue() for them, and we expect the
1558 * request_queue refcount to reach 0 at this point, and so the request_queue
1559 * will also be freed prior to the disk.
1560 *
1561 * Context: can sleep
1562 */
1563static void disk_release(struct device *dev)
1564{
1565	struct gendisk *disk = dev_to_disk(dev);
1566
1567	might_sleep();
1568	blk_mq_cancel_work_sync(disk->queue);
1569
1570	blk_free_devt(dev->devt);
1571	disk_release_events(disk);
1572	kfree(disk->random);
1573	disk_replace_part_tbl(disk, NULL);
1574	hd_free_part(&disk->part0);
1575	if (disk->queue)
1576		blk_put_queue(disk->queue);
1577	kfree(disk);
1578}
1579struct class block_class = {
1580	.name		= "block",
1581};
1582
1583static char *block_devnode(struct device *dev, umode_t *mode,
1584			   kuid_t *uid, kgid_t *gid)
1585{
1586	struct gendisk *disk = dev_to_disk(dev);
1587
1588	if (disk->fops->devnode)
1589		return disk->fops->devnode(disk, mode);
1590	return NULL;
1591}
1592
1593const struct device_type disk_type = {
1594	.name		= "disk",
1595	.groups		= disk_attr_groups,
1596	.release	= disk_release,
1597	.devnode	= block_devnode,
1598};
1599
1600#ifdef CONFIG_PROC_FS
1601/*
1602 * aggregate disk stat collector.  Uses the same stats that the sysfs
1603 * entries do, above, but makes them available through one seq_file.
1604 *
1605 * The output looks suspiciously like /proc/partitions with a bunch of
1606 * extra fields.
1607 */
1608static int diskstats_show(struct seq_file *seqf, void *v)
1609{
1610	struct gendisk *gp = v;
1611	struct disk_part_iter piter;
1612	struct hd_struct *hd;
1613	char buf[BDEVNAME_SIZE];
1614	unsigned int inflight;
1615	struct disk_stats stat;
1616
1617	/*
1618	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1619		seq_puts(seqf,	"major minor name"
1620				"     rio rmerge rsect ruse wio wmerge "
1621				"wsect wuse running use aveq"
1622				"\n\n");
1623	*/
1624
1625	disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1626	while ((hd = disk_part_iter_next(&piter))) {
1627		part_stat_read_all(hd, &stat);
1628		if (queue_is_mq(gp->queue))
1629			inflight = blk_mq_in_flight(gp->queue, hd);
1630		else
1631			inflight = part_in_flight(hd);
1632
1633		seq_printf(seqf, "%4d %7d %s "
1634			   "%lu %lu %lu %u "
1635			   "%lu %lu %lu %u "
1636			   "%u %u %u "
1637			   "%lu %lu %lu %u "
1638			   "%lu %u"
1639			   "\n",
1640			   MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1641			   disk_name(gp, hd->partno, buf),
1642			   stat.ios[STAT_READ],
1643			   stat.merges[STAT_READ],
1644			   stat.sectors[STAT_READ],
1645			   (unsigned int)div_u64(stat.nsecs[STAT_READ],
1646							NSEC_PER_MSEC),
1647			   stat.ios[STAT_WRITE],
1648			   stat.merges[STAT_WRITE],
1649			   stat.sectors[STAT_WRITE],
1650			   (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1651							NSEC_PER_MSEC),
1652			   inflight,
1653			   jiffies_to_msecs(stat.io_ticks),
1654			   (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1655						 stat.nsecs[STAT_WRITE] +
1656						 stat.nsecs[STAT_DISCARD] +
1657						 stat.nsecs[STAT_FLUSH],
1658							NSEC_PER_MSEC),
1659			   stat.ios[STAT_DISCARD],
1660			   stat.merges[STAT_DISCARD],
1661			   stat.sectors[STAT_DISCARD],
1662			   (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1663						 NSEC_PER_MSEC),
1664			   stat.ios[STAT_FLUSH],
1665			   (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1666						 NSEC_PER_MSEC)
1667			);
1668	}
1669	disk_part_iter_exit(&piter);
1670
1671	return 0;
1672}
1673
1674static const struct seq_operations diskstats_op = {
1675	.start	= disk_seqf_start,
1676	.next	= disk_seqf_next,
1677	.stop	= disk_seqf_stop,
1678	.show	= diskstats_show
1679};
1680
1681static int __init proc_genhd_init(void)
1682{
1683	proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1684	proc_create_seq("partitions", 0, NULL, &partitions_op);
1685	return 0;
1686}
1687module_init(proc_genhd_init);
1688#endif /* CONFIG_PROC_FS */
1689
1690dev_t blk_lookup_devt(const char *name, int partno)
1691{
1692	dev_t devt = MKDEV(0, 0);
1693	struct class_dev_iter iter;
1694	struct device *dev;
1695
1696	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1697	while ((dev = class_dev_iter_next(&iter))) {
1698		struct gendisk *disk = dev_to_disk(dev);
1699		struct hd_struct *part;
1700
1701		if (strcmp(dev_name(dev), name))
1702			continue;
1703
1704		if (partno < disk->minors) {
1705			/* We need to return the right devno, even
1706			 * if the partition doesn't exist yet.
1707			 */
1708			devt = MKDEV(MAJOR(dev->devt),
1709				     MINOR(dev->devt) + partno);
1710			break;
1711		}
1712		part = disk_get_part(disk, partno);
1713		if (part) {
1714			devt = part_devt(part);
1715			disk_put_part(part);
1716			break;
1717		}
1718		disk_put_part(part);
1719	}
1720	class_dev_iter_exit(&iter);
1721	return devt;
1722}
1723
1724struct gendisk *__alloc_disk_node(int minors, int node_id)
1725{
1726	struct gendisk *disk;
1727	struct disk_part_tbl *ptbl;
1728
1729	if (minors > DISK_MAX_PARTS) {
1730		printk(KERN_ERR
1731			"block: can't allocate more than %d partitions\n",
1732			DISK_MAX_PARTS);
1733		minors = DISK_MAX_PARTS;
1734	}
1735
1736	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1737	if (!disk)
1738		return NULL;
1739
1740	disk->part0.dkstats = alloc_percpu(struct disk_stats);
1741	if (!disk->part0.dkstats)
1742		goto out_free_disk;
1743
1744	init_rwsem(&disk->lookup_sem);
1745	disk->node_id = node_id;
1746	if (disk_expand_part_tbl(disk, 0)) {
1747		free_percpu(disk->part0.dkstats);
1748		goto out_free_disk;
1749	}
1750
1751	ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1752	rcu_assign_pointer(ptbl->part[0], &disk->part0);
1753
1754	/*
1755	 * set_capacity() and get_capacity() currently don't use
1756	 * seqcounter to read/update the part0->nr_sects. Still init
1757	 * the counter as we can read the sectors in IO submission
1758	 * patch using seqence counters.
1759	 *
1760	 * TODO: Ideally set_capacity() and get_capacity() should be
1761	 * converted to make use of bd_mutex and sequence counters.
1762	 */
1763	hd_sects_seq_init(&disk->part0);
1764	if (hd_ref_init(&disk->part0))
1765		goto out_free_part0;
1766
1767	disk->minors = minors;
1768	rand_initialize_disk(disk);
1769	disk_to_dev(disk)->class = &block_class;
1770	disk_to_dev(disk)->type = &disk_type;
1771	device_initialize(disk_to_dev(disk));
1772	return disk;
1773
1774out_free_part0:
1775	hd_free_part(&disk->part0);
1776out_free_disk:
1777	kfree(disk);
1778	return NULL;
1779}
1780EXPORT_SYMBOL(__alloc_disk_node);
1781
1782/**
1783 * get_disk_and_module - increments the gendisk and gendisk fops module refcount
1784 * @disk: the struct gendisk to increment the refcount for
1785 *
1786 * This increments the refcount for the struct gendisk, and the gendisk's
1787 * fops module owner.
1788 *
1789 * Context: Any context.
1790 */
1791struct kobject *get_disk_and_module(struct gendisk *disk)
1792{
1793	struct module *owner;
1794	struct kobject *kobj;
1795
1796	if (!disk->fops)
1797		return NULL;
1798	owner = disk->fops->owner;
1799	if (owner && !try_module_get(owner))
1800		return NULL;
1801	kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1802	if (kobj == NULL) {
1803		module_put(owner);
1804		return NULL;
1805	}
1806	return kobj;
1807
1808}
1809EXPORT_SYMBOL(get_disk_and_module);
1810
1811/**
1812 * put_disk - decrements the gendisk refcount
1813 * @disk: the struct gendisk to decrement the refcount for
1814 *
1815 * This decrements the refcount for the struct gendisk. When this reaches 0
1816 * we'll have disk_release() called.
1817 *
1818 * Context: Any context, but the last reference must not be dropped from
1819 *          atomic context.
1820 */
1821void put_disk(struct gendisk *disk)
1822{
1823	if (disk)
1824		kobject_put(&disk_to_dev(disk)->kobj);
1825}
1826EXPORT_SYMBOL(put_disk);
1827
1828/**
1829 * put_disk_and_module - decrements the module and gendisk refcount
1830 * @disk: the struct gendisk to decrement the refcount for
1831 *
1832 * This is a counterpart of get_disk_and_module() and thus also of
1833 * get_gendisk().
1834 *
1835 * Context: Any context, but the last reference must not be dropped from
1836 *          atomic context.
1837 */
1838void put_disk_and_module(struct gendisk *disk)
1839{
1840	if (disk) {
1841		struct module *owner = disk->fops->owner;
1842
1843		put_disk(disk);
1844		module_put(owner);
1845	}
1846}
1847EXPORT_SYMBOL(put_disk_and_module);
1848
1849static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1850{
1851	char event[] = "DISK_RO=1";
1852	char *envp[] = { event, NULL };
1853
1854	if (!ro)
1855		event[8] = '0';
1856	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1857}
1858
1859void set_device_ro(struct block_device *bdev, int flag)
1860{
1861	bdev->bd_part->policy = flag;
1862}
1863
1864EXPORT_SYMBOL(set_device_ro);
1865
1866void set_disk_ro(struct gendisk *disk, int flag)
1867{
1868	struct disk_part_iter piter;
1869	struct hd_struct *part;
1870
1871	if (disk->part0.policy != flag) {
1872		set_disk_ro_uevent(disk, flag);
1873		disk->part0.policy = flag;
1874	}
1875
1876	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1877	while ((part = disk_part_iter_next(&piter)))
1878		part->policy = flag;
1879	disk_part_iter_exit(&piter);
1880}
1881
1882EXPORT_SYMBOL(set_disk_ro);
1883
1884int bdev_read_only(struct block_device *bdev)
1885{
1886	if (!bdev)
1887		return 0;
1888	return bdev->bd_part->policy;
1889}
1890
1891EXPORT_SYMBOL(bdev_read_only);
1892
1893/*
1894 * Disk events - monitor disk events like media change and eject request.
1895 */
1896struct disk_events {
1897	struct list_head	node;		/* all disk_event's */
1898	struct gendisk		*disk;		/* the associated disk */
1899	spinlock_t		lock;
1900
1901	struct mutex		block_mutex;	/* protects blocking */
1902	int			block;		/* event blocking depth */
1903	unsigned int		pending;	/* events already sent out */
1904	unsigned int		clearing;	/* events being cleared */
1905
1906	long			poll_msecs;	/* interval, -1 for default */
1907	struct delayed_work	dwork;
1908};
1909
1910static const char *disk_events_strs[] = {
1911	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "media_change",
1912	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "eject_request",
1913};
1914
1915static char *disk_uevents[] = {
1916	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "DISK_MEDIA_CHANGE=1",
1917	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "DISK_EJECT_REQUEST=1",
1918};
1919
1920/* list of all disk_events */
1921static DEFINE_MUTEX(disk_events_mutex);
1922static LIST_HEAD(disk_events);
1923
1924/* disable in-kernel polling by default */
1925static unsigned long disk_events_dfl_poll_msecs;
1926
1927static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1928{
1929	struct disk_events *ev = disk->ev;
1930	long intv_msecs = 0;
1931
1932	/*
1933	 * If device-specific poll interval is set, always use it.  If
1934	 * the default is being used, poll if the POLL flag is set.
1935	 */
1936	if (ev->poll_msecs >= 0)
1937		intv_msecs = ev->poll_msecs;
1938	else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1939		intv_msecs = disk_events_dfl_poll_msecs;
1940
1941	return msecs_to_jiffies(intv_msecs);
1942}
1943
1944/**
1945 * disk_block_events - block and flush disk event checking
1946 * @disk: disk to block events for
1947 *
1948 * On return from this function, it is guaranteed that event checking
1949 * isn't in progress and won't happen until unblocked by
1950 * disk_unblock_events().  Events blocking is counted and the actual
1951 * unblocking happens after the matching number of unblocks are done.
1952 *
1953 * Note that this intentionally does not block event checking from
1954 * disk_clear_events().
1955 *
1956 * CONTEXT:
1957 * Might sleep.
1958 */
1959void disk_block_events(struct gendisk *disk)
1960{
1961	struct disk_events *ev = disk->ev;
1962	unsigned long flags;
1963	bool cancel;
1964
1965	if (!ev)
1966		return;
1967
1968	/*
1969	 * Outer mutex ensures that the first blocker completes canceling
1970	 * the event work before further blockers are allowed to finish.
1971	 */
1972	mutex_lock(&ev->block_mutex);
1973
1974	spin_lock_irqsave(&ev->lock, flags);
1975	cancel = !ev->block++;
1976	spin_unlock_irqrestore(&ev->lock, flags);
1977
1978	if (cancel)
1979		cancel_delayed_work_sync(&disk->ev->dwork);
1980
1981	mutex_unlock(&ev->block_mutex);
1982}
1983
1984static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1985{
1986	struct disk_events *ev = disk->ev;
1987	unsigned long intv;
1988	unsigned long flags;
1989
1990	spin_lock_irqsave(&ev->lock, flags);
1991
1992	if (WARN_ON_ONCE(ev->block <= 0))
1993		goto out_unlock;
1994
1995	if (--ev->block)
1996		goto out_unlock;
1997
1998	intv = disk_events_poll_jiffies(disk);
1999	if (check_now)
2000		queue_delayed_work(system_freezable_power_efficient_wq,
2001				&ev->dwork, 0);
2002	else if (intv)
2003		queue_delayed_work(system_freezable_power_efficient_wq,
2004				&ev->dwork, intv);
2005out_unlock:
2006	spin_unlock_irqrestore(&ev->lock, flags);
2007}
2008
2009/**
2010 * disk_unblock_events - unblock disk event checking
2011 * @disk: disk to unblock events for
2012 *
2013 * Undo disk_block_events().  When the block count reaches zero, it
2014 * starts events polling if configured.
2015 *
2016 * CONTEXT:
2017 * Don't care.  Safe to call from irq context.
2018 */
2019void disk_unblock_events(struct gendisk *disk)
2020{
2021	if (disk->ev)
2022		__disk_unblock_events(disk, false);
2023}
2024
2025/**
2026 * disk_flush_events - schedule immediate event checking and flushing
2027 * @disk: disk to check and flush events for
2028 * @mask: events to flush
2029 *
2030 * Schedule immediate event checking on @disk if not blocked.  Events in
2031 * @mask are scheduled to be cleared from the driver.  Note that this
2032 * doesn't clear the events from @disk->ev.
2033 *
2034 * CONTEXT:
2035 * If @mask is non-zero must be called with bdev->bd_mutex held.
2036 */
2037void disk_flush_events(struct gendisk *disk, unsigned int mask)
2038{
2039	struct disk_events *ev = disk->ev;
2040
2041	if (!ev)
2042		return;
2043
2044	spin_lock_irq(&ev->lock);
2045	ev->clearing |= mask;
2046	if (!ev->block)
2047		mod_delayed_work(system_freezable_power_efficient_wq,
2048				&ev->dwork, 0);
2049	spin_unlock_irq(&ev->lock);
2050}
2051
2052/**
2053 * disk_clear_events - synchronously check, clear and return pending events
2054 * @disk: disk to fetch and clear events from
2055 * @mask: mask of events to be fetched and cleared
2056 *
2057 * Disk events are synchronously checked and pending events in @mask
2058 * are cleared and returned.  This ignores the block count.
2059 *
2060 * CONTEXT:
2061 * Might sleep.
2062 */
2063static unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
2064{
2065	struct disk_events *ev = disk->ev;
2066	unsigned int pending;
2067	unsigned int clearing = mask;
2068
2069	if (!ev)
2070		return 0;
2071
2072	disk_block_events(disk);
2073
2074	/*
2075	 * store the union of mask and ev->clearing on the stack so that the
2076	 * race with disk_flush_events does not cause ambiguity (ev->clearing
2077	 * can still be modified even if events are blocked).
2078	 */
2079	spin_lock_irq(&ev->lock);
2080	clearing |= ev->clearing;
2081	ev->clearing = 0;
2082	spin_unlock_irq(&ev->lock);
2083
2084	disk_check_events(ev, &clearing);
2085	/*
2086	 * if ev->clearing is not 0, the disk_flush_events got called in the
2087	 * middle of this function, so we want to run the workfn without delay.
2088	 */
2089	__disk_unblock_events(disk, ev->clearing ? true : false);
2090
2091	/* then, fetch and clear pending events */
2092	spin_lock_irq(&ev->lock);
2093	pending = ev->pending & mask;
2094	ev->pending &= ~mask;
2095	spin_unlock_irq(&ev->lock);
2096	WARN_ON_ONCE(clearing & mask);
2097
2098	return pending;
2099}
2100
2101/**
2102 * bdev_check_media_change - check if a removable media has been changed
2103 * @bdev: block device to check
2104 *
2105 * Check whether a removable media has been changed, and attempt to free all
2106 * dentries and inodes and invalidates all block device page cache entries in
2107 * that case.
2108 *
2109 * Returns %true if the block device changed, or %false if not.
2110 */
2111bool bdev_check_media_change(struct block_device *bdev)
2112{
2113	unsigned int events;
2114
2115	events = disk_clear_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE |
2116				   DISK_EVENT_EJECT_REQUEST);
2117	if (!(events & DISK_EVENT_MEDIA_CHANGE))
2118		return false;
2119
2120	if (__invalidate_device(bdev, true))
2121		pr_warn("VFS: busy inodes on changed media %s\n",
2122			bdev->bd_disk->disk_name);
2123	set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
2124	return true;
2125}
2126EXPORT_SYMBOL(bdev_check_media_change);
2127
2128/*
2129 * Separate this part out so that a different pointer for clearing_ptr can be
2130 * passed in for disk_clear_events.
2131 */
2132static void disk_events_workfn(struct work_struct *work)
2133{
2134	struct delayed_work *dwork = to_delayed_work(work);
2135	struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2136
2137	disk_check_events(ev, &ev->clearing);
2138}
2139
2140static void disk_check_events(struct disk_events *ev,
2141			      unsigned int *clearing_ptr)
2142{
2143	struct gendisk *disk = ev->disk;
2144	char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2145	unsigned int clearing = *clearing_ptr;
2146	unsigned int events;
2147	unsigned long intv;
2148	int nr_events = 0, i;
2149
2150	/* check events */
2151	events = disk->fops->check_events(disk, clearing);
2152
2153	/* accumulate pending events and schedule next poll if necessary */
2154	spin_lock_irq(&ev->lock);
2155
2156	events &= ~ev->pending;
2157	ev->pending |= events;
2158	*clearing_ptr &= ~clearing;
2159
2160	intv = disk_events_poll_jiffies(disk);
2161	if (!ev->block && intv)
2162		queue_delayed_work(system_freezable_power_efficient_wq,
2163				&ev->dwork, intv);
2164
2165	spin_unlock_irq(&ev->lock);
2166
2167	/*
2168	 * Tell userland about new events.  Only the events listed in
2169	 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2170	 * is set. Otherwise, events are processed internally but never
2171	 * get reported to userland.
2172	 */
2173	for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2174		if ((events & disk->events & (1 << i)) &&
2175		    (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2176			envp[nr_events++] = disk_uevents[i];
2177
2178	if (nr_events)
2179		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2180}
2181
2182/*
2183 * A disk events enabled device has the following sysfs nodes under
2184 * its /sys/block/X/ directory.
2185 *
2186 * events		: list of all supported events
2187 * events_async		: list of events which can be detected w/o polling
2188 *			  (always empty, only for backwards compatibility)
2189 * events_poll_msecs	: polling interval, 0: disable, -1: system default
2190 */
2191static ssize_t __disk_events_show(unsigned int events, char *buf)
2192{
2193	const char *delim = "";
2194	ssize_t pos = 0;
2195	int i;
2196
2197	for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2198		if (events & (1 << i)) {
2199			pos += sprintf(buf + pos, "%s%s",
2200				       delim, disk_events_strs[i]);
2201			delim = " ";
2202		}
2203	if (pos)
2204		pos += sprintf(buf + pos, "\n");
2205	return pos;
2206}
2207
2208static ssize_t disk_events_show(struct device *dev,
2209				struct device_attribute *attr, char *buf)
2210{
2211	struct gendisk *disk = dev_to_disk(dev);
2212
2213	if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2214		return 0;
2215
2216	return __disk_events_show(disk->events, buf);
2217}
2218
2219static ssize_t disk_events_async_show(struct device *dev,
2220				      struct device_attribute *attr, char *buf)
2221{
2222	return 0;
2223}
2224
2225static ssize_t disk_events_poll_msecs_show(struct device *dev,
2226					   struct device_attribute *attr,
2227					   char *buf)
2228{
2229	struct gendisk *disk = dev_to_disk(dev);
2230
2231	if (!disk->ev)
2232		return sprintf(buf, "-1\n");
2233
2234	return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2235}
2236
2237static ssize_t disk_events_poll_msecs_store(struct device *dev,
2238					    struct device_attribute *attr,
2239					    const char *buf, size_t count)
2240{
2241	struct gendisk *disk = dev_to_disk(dev);
2242	long intv;
2243
2244	if (!count || !sscanf(buf, "%ld", &intv))
2245		return -EINVAL;
2246
2247	if (intv < 0 && intv != -1)
2248		return -EINVAL;
2249
2250	if (!disk->ev)
2251		return -ENODEV;
2252
2253	disk_block_events(disk);
2254	disk->ev->poll_msecs = intv;
2255	__disk_unblock_events(disk, true);
2256
2257	return count;
2258}
2259
2260static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2261static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2262static const DEVICE_ATTR(events_poll_msecs, 0644,
2263			 disk_events_poll_msecs_show,
2264			 disk_events_poll_msecs_store);
2265
2266static const struct attribute *disk_events_attrs[] = {
2267	&dev_attr_events.attr,
2268	&dev_attr_events_async.attr,
2269	&dev_attr_events_poll_msecs.attr,
2270	NULL,
2271};
2272
2273/*
2274 * The default polling interval can be specified by the kernel
2275 * parameter block.events_dfl_poll_msecs which defaults to 0
2276 * (disable).  This can also be modified runtime by writing to
2277 * /sys/module/block/parameters/events_dfl_poll_msecs.
2278 */
2279static int disk_events_set_dfl_poll_msecs(const char *val,
2280					  const struct kernel_param *kp)
2281{
2282	struct disk_events *ev;
2283	int ret;
2284
2285	ret = param_set_ulong(val, kp);
2286	if (ret < 0)
2287		return ret;
2288
2289	mutex_lock(&disk_events_mutex);
2290
2291	list_for_each_entry(ev, &disk_events, node)
2292		disk_flush_events(ev->disk, 0);
2293
2294	mutex_unlock(&disk_events_mutex);
2295
2296	return 0;
2297}
2298
2299static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2300	.set	= disk_events_set_dfl_poll_msecs,
2301	.get	= param_get_ulong,
2302};
2303
2304#undef MODULE_PARAM_PREFIX
2305#define MODULE_PARAM_PREFIX	"block."
2306
2307module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2308		&disk_events_dfl_poll_msecs, 0644);
2309
2310/*
2311 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2312 */
2313static void disk_alloc_events(struct gendisk *disk)
2314{
2315	struct disk_events *ev;
2316
2317	if (!disk->fops->check_events || !disk->events)
2318		return;
2319
2320	ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2321	if (!ev) {
2322		pr_warn("%s: failed to initialize events\n", disk->disk_name);
2323		return;
2324	}
2325
2326	INIT_LIST_HEAD(&ev->node);
2327	ev->disk = disk;
2328	spin_lock_init(&ev->lock);
2329	mutex_init(&ev->block_mutex);
2330	ev->block = 1;
2331	ev->poll_msecs = -1;
2332	INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2333
2334	disk->ev = ev;
2335}
2336
2337static void disk_add_events(struct gendisk *disk)
2338{
2339	/* FIXME: error handling */
2340	if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2341		pr_warn("%s: failed to create sysfs files for events\n",
2342			disk->disk_name);
2343
2344	if (!disk->ev)
2345		return;
2346
2347	mutex_lock(&disk_events_mutex);
2348	list_add_tail(&disk->ev->node, &disk_events);
2349	mutex_unlock(&disk_events_mutex);
2350
2351	/*
2352	 * Block count is initialized to 1 and the following initial
2353	 * unblock kicks it into action.
2354	 */
2355	__disk_unblock_events(disk, true);
2356}
2357
2358static void disk_del_events(struct gendisk *disk)
2359{
2360	if (disk->ev) {
2361		disk_block_events(disk);
2362
2363		mutex_lock(&disk_events_mutex);
2364		list_del_init(&disk->ev->node);
2365		mutex_unlock(&disk_events_mutex);
2366	}
2367
2368	sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2369}
2370
2371static void disk_release_events(struct gendisk *disk)
2372{
2373	/* the block count should be 1 from disk_del_events() */
2374	WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
2375	kfree(disk->ev);
2376}
2377