path: root/block/genhd.c (plain)
blob: fcd6d4fae657cfdf118274cec31780f9e8b8c1f8

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	/**
576	* device_add_disk - add partitioning information to kernel list
577	* @parent: parent device for the disk
578	* @disk: per-device partitioning information
579	*
580	* This function registers the partitioning information in @disk
581	* with the kernel.
582	*
583	* FIXME: error handling
584	*/
585	void device_add_disk(struct device *parent, struct gendisk *disk)
586	{
587	struct backing_dev_info *bdi;
588	dev_t devt;
589	int retval;
590
591	/* minors == 0 indicates to use ext devt from part0 and should
592	* be accompanied with EXT_DEVT flag. Make sure all
593	* parameters make sense.
594	*/
595	WARN_ON(disk->minors && !(disk->major || disk->first_minor));
596	WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
597
598	disk->flags |= GENHD_FL_UP;
599
600	retval = blk_alloc_devt(&disk->part0, &devt);
601	if (retval) {
602	WARN_ON(1);
603	return;
604	}
605	disk_to_dev(disk)->devt = devt;
606
607	/* ->major and ->first_minor aren't supposed to be
608	* dereferenced from here on, but set them just in case.
609	*/
610	disk->major = MAJOR(devt);
611	disk->first_minor = MINOR(devt);
612
613	disk_alloc_events(disk);
614
615	/* Register BDI before referencing it from bdev */
616	bdi = &disk->queue->backing_dev_info;
617	bdi_register_owner(bdi, disk_to_dev(disk));
618
619	blk_register_region(disk_devt(disk), disk->minors, NULL,
620	exact_match, exact_lock, disk);
621	register_disk(parent, disk);
622	blk_register_queue(disk);
623
624	/*
625	* Take an extra ref on queue which will be put on disk_release()
626	* so that it sticks around as long as @disk is there.
627	*/
628	WARN_ON_ONCE(!blk_get_queue(disk->queue));
629
630	retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
631	"bdi");
632	WARN_ON(retval);
633
634	disk_add_events(disk);
635	blk_integrity_add(disk);
636	}
637	EXPORT_SYMBOL(device_add_disk);
638
639	void del_gendisk(struct gendisk *disk)
640	{
641	struct disk_part_iter piter;
642	struct hd_struct *part;
643
644	blk_integrity_del(disk);
645	disk_del_events(disk);
646
647	/* invalidate stuff */
648	disk_part_iter_init(&piter, disk,
649	DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
650	while ((part = disk_part_iter_next(&piter))) {
651	invalidate_partition(disk, part->partno);
652	delete_partition(disk, part->partno);
653	}
654	disk_part_iter_exit(&piter);
655
656	invalidate_partition(disk, 0);
657	set_capacity(disk, 0);
658	disk->flags &= ~GENHD_FL_UP;
659
660	sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
661	blk_unregister_queue(disk);
662	blk_unregister_region(disk_devt(disk), disk->minors);
663
664	part_stat_set_all(&disk->part0, 0);
665	disk->part0.stamp = 0;
666
667	kobject_put(disk->part0.holder_dir);
668	kobject_put(disk->slave_dir);
669	if (!sysfs_deprecated)
670	sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
671	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
672	device_del(disk_to_dev(disk));
673	}
674	EXPORT_SYMBOL(del_gendisk);
675
676	/* sysfs access to bad-blocks list. */
677	static ssize_t disk_badblocks_show(struct device *dev,
678	struct device_attribute *attr,
679	char *page)
680	{
681	struct gendisk *disk = dev_to_disk(dev);
682
683	if (!disk->bb)
684	return sprintf(page, "\n");
685
686	return badblocks_show(disk->bb, page, 0);
687	}
688
689	static ssize_t disk_badblocks_store(struct device *dev,
690	struct device_attribute *attr,
691	const char *page, size_t len)
692	{
693	struct gendisk *disk = dev_to_disk(dev);
694
695	if (!disk->bb)
696	return -ENXIO;
697
698	return badblocks_store(disk->bb, page, len, 0);
699	}
700
701	/**
702	* get_gendisk - get partitioning information for a given device
703	* @devt: device to get partitioning information for
704	* @partno: returned partition index
705	*
706	* This function gets the structure containing partitioning
707	* information for the given device @devt.
708	*/
709	struct gendisk *get_gendisk(dev_t devt, int *partno)
710	{
711	struct gendisk *disk = NULL;
712
713	if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
714	struct kobject *kobj;
715
716	kobj = kobj_lookup(bdev_map, devt, partno);
717	if (kobj)
718	disk = dev_to_disk(kobj_to_dev(kobj));
719	} else {
720	struct hd_struct *part;
721
722	spin_lock_bh(&ext_devt_lock);
723	part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
724	if (part && get_disk(part_to_disk(part))) {
725	*partno = part->partno;
726	disk = part_to_disk(part);
727	}
728	spin_unlock_bh(&ext_devt_lock);
729	}
730
731	return disk;
732	}
733	EXPORT_SYMBOL(get_gendisk);
734
735	/**
736	* bdget_disk - do bdget() by gendisk and partition number
737	* @disk: gendisk of interest
738	* @partno: partition number
739	*
740	* Find partition @partno from @disk, do bdget() on it.
741	*
742	* CONTEXT:
743	* Don't care.
744	*
745	* RETURNS:
746	* Resulting block_device on success, NULL on failure.
747	*/
748	struct block_device *bdget_disk(struct gendisk *disk, int partno)
749	{
750	struct hd_struct *part;
751	struct block_device *bdev = NULL;
752
753	part = disk_get_part(disk, partno);
754	if (part)
755	bdev = bdget(part_devt(part));
756	disk_put_part(part);
757
758	return bdev;
759	}
760	EXPORT_SYMBOL(bdget_disk);
761
762	/*
763	* print a full list of all partitions - intended for places where the root
764	* filesystem can't be mounted and thus to give the victim some idea of what
765	* went wrong
766	*/
767	void __init printk_all_partitions(void)
768	{
769	struct class_dev_iter iter;
770	struct device *dev;
771
772	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
773	while ((dev = class_dev_iter_next(&iter))) {
774	struct gendisk *disk = dev_to_disk(dev);
775	struct disk_part_iter piter;
776	struct hd_struct *part;
777	char name_buf[BDEVNAME_SIZE];
778	char devt_buf[BDEVT_SIZE];
779
780	/*
781	* Don't show empty devices or things that have been
782	* suppressed
783	*/
784	if (get_capacity(disk) == 0 ||
785	(disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
786	continue;
787
788	/*
789	* Note, unlike /proc/partitions, I am showing the
790	* numbers in hex - the same format as the root=
791	* option takes.
792	*/
793	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
794	while ((part = disk_part_iter_next(&piter))) {
795	bool is_part0 = part == &disk->part0;
796
797	printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
798	bdevt_str(part_devt(part), devt_buf),
799	(unsigned long long)part_nr_sects_read(part) >> 1
800	, disk_name(disk, part->partno, name_buf),
801	part->info ? part->info->uuid : "");
802	if (is_part0) {
803	if (dev->parent && dev->parent->driver)
804	printk(" driver: %s\n",
805	dev->parent->driver->name);
806	else
807	printk(" (driver?)\n");
808	} else
809	printk("\n");
810	}
811	disk_part_iter_exit(&piter);
812	}
813	class_dev_iter_exit(&iter);
814	}
815
816	#ifdef CONFIG_PROC_FS
817	/* iterator */
818	static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
819	{
820	loff_t skip = *pos;
821	struct class_dev_iter *iter;
822	struct device *dev;
823
824	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
825	if (!iter)
826	return ERR_PTR(-ENOMEM);
827
828	seqf->private = iter;
829	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
830	do {
831	dev = class_dev_iter_next(iter);
832	if (!dev)
833	return NULL;
834	} while (skip--);
835
836	return dev_to_disk(dev);
837	}
838
839	static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
840	{
841	struct device *dev;
842
843	(*pos)++;
844	dev = class_dev_iter_next(seqf->private);
845	if (dev)
846	return dev_to_disk(dev);
847
848	return NULL;
849	}
850
851	static void disk_seqf_stop(struct seq_file *seqf, void *v)
852	{
853	struct class_dev_iter *iter = seqf->private;
854
855	/* stop is called even after start failed :-( */
856	if (iter) {
857	class_dev_iter_exit(iter);
858	kfree(iter);
859	seqf->private = NULL;
860	}
861	}
862
863	static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
864	{
865	void *p;
866
867	p = disk_seqf_start(seqf, pos);
868	if (!IS_ERR_OR_NULL(p) && !*pos)
869	seq_puts(seqf, "major minor #blocks name\n\n");
870	return p;
871	}
872
873	static int show_partition(struct seq_file *seqf, void *v)
874	{
875	struct gendisk *sgp = v;
876	struct disk_part_iter piter;
877	struct hd_struct *part;
878	char buf[BDEVNAME_SIZE];
879
880	/* Don't show non-partitionable removeable devices or empty devices */
881	if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
882	(sgp->flags & GENHD_FL_REMOVABLE)))
883	return 0;
884	if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
885	return 0;
886
887	/* show the full disk and all non-0 size partitions of it */
888	disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
889	while ((part = disk_part_iter_next(&piter)))
890	seq_printf(seqf, "%4d %7d %10llu %s\n",
891	MAJOR(part_devt(part)), MINOR(part_devt(part)),
892	(unsigned long long)part_nr_sects_read(part) >> 1,
893	disk_name(sgp, part->partno, buf));
894	disk_part_iter_exit(&piter);
895
896	return 0;
897	}
898
899	static const struct seq_operations partitions_op = {
900	.start = show_partition_start,
901	.next = disk_seqf_next,
902	.stop = disk_seqf_stop,
903	.show = show_partition
904	};
905
906	static int partitions_open(struct inode *inode, struct file *file)
907	{
908	return seq_open(file, &partitions_op);
909	}
910
911	static const struct file_operations proc_partitions_operations = {
912	.open = partitions_open,
913	.read = seq_read,
914	.llseek = seq_lseek,
915	.release = seq_release,
916	};
917	#endif
918
919
920	static struct kobject *base_probe(dev_t devt, int *partno, void *data)
921	{
922	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
923	/* Make old-style 2.4 aliases work */
924	request_module("block-major-%d", MAJOR(devt));
925	return NULL;
926	}
927
928	static int __init genhd_device_init(void)
929	{
930	int error;
931
932	block_class.dev_kobj = sysfs_dev_block_kobj;
933	error = class_register(&block_class);
934	if (unlikely(error))
935	return error;
936	bdev_map = kobj_map_init(base_probe, &block_class_lock);
937	blk_dev_init();
938
939	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
940
941	/* create top-level block dir */
942	if (!sysfs_deprecated)
943	block_depr = kobject_create_and_add("block", NULL);
944	return 0;
945	}
946
947	subsys_initcall(genhd_device_init);
948
949	static ssize_t disk_range_show(struct device *dev,
950	struct device_attribute *attr, char *buf)
951	{
952	struct gendisk *disk = dev_to_disk(dev);
953
954	return sprintf(buf, "%d\n", disk->minors);
955	}
956
957	static ssize_t disk_ext_range_show(struct device *dev,
958	struct device_attribute *attr, char *buf)
959	{
960	struct gendisk *disk = dev_to_disk(dev);
961
962	return sprintf(buf, "%d\n", disk_max_parts(disk));
963	}
964
965	static ssize_t disk_removable_show(struct device *dev,
966	struct device_attribute *attr, char *buf)
967	{
968	struct gendisk *disk = dev_to_disk(dev);
969
970	return sprintf(buf, "%d\n",
971	(disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
972	}
973
974	static ssize_t disk_ro_show(struct device *dev,
975	struct device_attribute *attr, char *buf)
976	{
977	struct gendisk *disk = dev_to_disk(dev);
978
979	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
980	}
981
982	static ssize_t disk_capability_show(struct device *dev,
983	struct device_attribute *attr, char *buf)
984	{
985	struct gendisk *disk = dev_to_disk(dev);
986
987	return sprintf(buf, "%x\n", disk->flags);
988	}
989
990	static ssize_t disk_alignment_offset_show(struct device *dev,
991	struct device_attribute *attr,
992	char *buf)
993	{
994	struct gendisk *disk = dev_to_disk(dev);
995
996	return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
997	}
998
999	static ssize_t disk_discard_alignment_show(struct device *dev,
1000	struct device_attribute *attr,
1001	char *buf)
1002	{
1003	struct gendisk *disk = dev_to_disk(dev);
1004
1005	return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1006	}
1007
1008	static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
1009	static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
1010	static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
1011	static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
1012	static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
1013	static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
1014	static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
1015	NULL);
1016	static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
1017	static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
1018	static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
1019	static DEVICE_ATTR(badblocks, S_IRUGO | S_IWUSR, disk_badblocks_show,
1020	disk_badblocks_store);
1021	#ifdef CONFIG_FAIL_MAKE_REQUEST
1022	static struct device_attribute dev_attr_fail =
1023	__ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
1024	#endif
1025	#ifdef CONFIG_FAIL_IO_TIMEOUT
1026	static struct device_attribute dev_attr_fail_timeout =
1027	__ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
1028	part_timeout_store);
1029	#endif
1030
1031	static struct attribute *disk_attrs[] = {
1032	&dev_attr_range.attr,
1033	&dev_attr_ext_range.attr,
1034	&dev_attr_removable.attr,
1035	&dev_attr_ro.attr,
1036	&dev_attr_size.attr,
1037	&dev_attr_alignment_offset.attr,
1038	&dev_attr_discard_alignment.attr,
1039	&dev_attr_capability.attr,
1040	&dev_attr_stat.attr,
1041	&dev_attr_inflight.attr,
1042	&dev_attr_badblocks.attr,
1043	#ifdef CONFIG_FAIL_MAKE_REQUEST
1044	&dev_attr_fail.attr,
1045	#endif
1046	#ifdef CONFIG_FAIL_IO_TIMEOUT
1047	&dev_attr_fail_timeout.attr,
1048	#endif
1049	NULL
1050	};
1051
1052	static struct attribute_group disk_attr_group = {
1053	.attrs = disk_attrs,
1054	};
1055
1056	static const struct attribute_group *disk_attr_groups[] = {
1057	&disk_attr_group,
1058	NULL
1059	};
1060
1061	/**
1062	* disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1063	* @disk: disk to replace part_tbl for
1064	* @new_ptbl: new part_tbl to install
1065	*
1066	* Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1067	* original ptbl is freed using RCU callback.
1068	*
1069	* LOCKING:
1070	* Matching bd_mutx locked.
1071	*/
1072	static void disk_replace_part_tbl(struct gendisk *disk,
1073	struct disk_part_tbl *new_ptbl)
1074	{
1075	struct disk_part_tbl *old_ptbl = disk->part_tbl;
1076
1077	rcu_assign_pointer(disk->part_tbl, new_ptbl);
1078
1079	if (old_ptbl) {
1080	rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1081	kfree_rcu(old_ptbl, rcu_head);
1082	}
1083	}
1084
1085	/**
1086	* disk_expand_part_tbl - expand disk->part_tbl
1087	* @disk: disk to expand part_tbl for
1088	* @partno: expand such that this partno can fit in
1089	*
1090	* Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1091	* uses RCU to allow unlocked dereferencing for stats and other stuff.
1092	*
1093	* LOCKING:
1094	* Matching bd_mutex locked, might sleep.
1095	*
1096	* RETURNS:
1097	* 0 on success, -errno on failure.
1098	*/
1099	int disk_expand_part_tbl(struct gendisk *disk, int partno)
1100	{
1101	struct disk_part_tbl *old_ptbl = disk->part_tbl;
1102	struct disk_part_tbl *new_ptbl;
1103	int len = old_ptbl ? old_ptbl->len : 0;
1104	int i, target;
1105	size_t size;
1106
1107	/*
1108	* check for int overflow, since we can get here from blkpg_ioctl()
1109	* with a user passed 'partno'.
1110	*/
1111	target = partno + 1;
1112	if (target < 0)
1113	return -EINVAL;
1114
1115	/* disk_max_parts() is zero during initialization, ignore if so */
1116	if (disk_max_parts(disk) && target > disk_max_parts(disk))
1117	return -EINVAL;
1118
1119	if (target <= len)
1120	return 0;
1121
1122	size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1123	new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1124	if (!new_ptbl)
1125	return -ENOMEM;
1126
1127	new_ptbl->len = target;
1128
1129	for (i = 0; i < len; i++)
1130	rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1131
1132	disk_replace_part_tbl(disk, new_ptbl);
1133	return 0;
1134	}
1135
1136	static void disk_release(struct device *dev)
1137	{
1138	struct gendisk *disk = dev_to_disk(dev);
1139
1140	blk_free_devt(dev->devt);
1141	disk_release_events(disk);
1142	kfree(disk->random);
1143	disk_replace_part_tbl(disk, NULL);
1144	hd_free_part(&disk->part0);
1145	if (disk->queue)
1146	blk_put_queue(disk->queue);
1147	kfree(disk);
1148	}
1149	struct class block_class = {
1150	.name = "block",
1151	};
1152
1153	static char *block_devnode(struct device *dev, umode_t *mode,
1154	kuid_t *uid, kgid_t *gid)
1155	{
1156	struct gendisk *disk = dev_to_disk(dev);
1157
1158	if (disk->devnode)
1159	return disk->devnode(disk, mode);
1160	return NULL;
1161	}
1162
1163	static struct device_type disk_type = {
1164	.name = "disk",
1165	.groups = disk_attr_groups,
1166	.release = disk_release,
1167	.devnode = block_devnode,
1168	};
1169
1170	#ifdef CONFIG_PROC_FS
1171	/*
1172	* aggregate disk stat collector. Uses the same stats that the sysfs
1173	* entries do, above, but makes them available through one seq_file.
1174	*
1175	* The output looks suspiciously like /proc/partitions with a bunch of
1176	* extra fields.
1177	*/
1178	static int diskstats_show(struct seq_file *seqf, void *v)
1179	{
1180	struct gendisk *gp = v;
1181	struct disk_part_iter piter;
1182	struct hd_struct *hd;
1183	char buf[BDEVNAME_SIZE];
1184	int cpu;
1185
1186	/*
1187	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1188	seq_puts(seqf, "major minor name"
1189	" rio rmerge rsect ruse wio wmerge "
1190	"wsect wuse running use aveq"
1191	"\n\n");
1192	*/
1193
1194	disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1195	while ((hd = disk_part_iter_next(&piter))) {
1196	cpu = part_stat_lock();
1197	part_round_stats(cpu, hd);
1198	part_stat_unlock();
1199	seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1200	"%u %lu %lu %lu %u %u %u %u\n",
1201	MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1202	disk_name(gp, hd->partno, buf),
1203	part_stat_read(hd, ios[READ]),
1204	part_stat_read(hd, merges[READ]),
1205	part_stat_read(hd, sectors[READ]),
1206	jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1207	part_stat_read(hd, ios[WRITE]),
1208	part_stat_read(hd, merges[WRITE]),
1209	part_stat_read(hd, sectors[WRITE]),
1210	jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1211	part_in_flight(hd),
1212	jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1213	jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1214	);
1215	}
1216	disk_part_iter_exit(&piter);
1217
1218	return 0;
1219	}
1220
1221	static const struct seq_operations diskstats_op = {
1222	.start = disk_seqf_start,
1223	.next = disk_seqf_next,
1224	.stop = disk_seqf_stop,
1225	.show = diskstats_show
1226	};
1227
1228	static int diskstats_open(struct inode *inode, struct file *file)
1229	{
1230	return seq_open(file, &diskstats_op);
1231	}
1232
1233	static const struct file_operations proc_diskstats_operations = {
1234	.open = diskstats_open,
1235	.read = seq_read,
1236	.llseek = seq_lseek,
1237	.release = seq_release,
1238	};
1239
1240	static int __init proc_genhd_init(void)
1241	{
1242	proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1243	proc_create("partitions", 0, NULL, &proc_partitions_operations);
1244	return 0;
1245	}
1246	module_init(proc_genhd_init);
1247	#endif /* CONFIG_PROC_FS */
1248
1249	dev_t blk_lookup_devt(const char *name, int partno)
1250	{
1251	dev_t devt = MKDEV(0, 0);
1252	struct class_dev_iter iter;
1253	struct device *dev;
1254
1255	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1256	while ((dev = class_dev_iter_next(&iter))) {
1257	struct gendisk *disk = dev_to_disk(dev);
1258	struct hd_struct *part;
1259
1260	if (strcmp(dev_name(dev), name))
1261	continue;
1262
1263	if (partno < disk->minors) {
1264	/* We need to return the right devno, even
1265	* if the partition doesn't exist yet.
1266	*/
1267	devt = MKDEV(MAJOR(dev->devt),
1268	MINOR(dev->devt) + partno);
1269	break;
1270	}
1271	part = disk_get_part(disk, partno);
1272	if (part) {
1273	devt = part_devt(part);
1274	disk_put_part(part);
1275	break;
1276	}
1277	disk_put_part(part);
1278	}
1279	class_dev_iter_exit(&iter);
1280	return devt;
1281	}
1282	EXPORT_SYMBOL(blk_lookup_devt);
1283
1284	struct gendisk *alloc_disk(int minors)
1285	{
1286	return alloc_disk_node(minors, NUMA_NO_NODE);
1287	}
1288	EXPORT_SYMBOL(alloc_disk);
1289
1290	struct gendisk *alloc_disk_node(int minors, int node_id)
1291	{
1292	struct gendisk *disk;
1293
1294	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1295	if (disk) {
1296	if (!init_part_stats(&disk->part0)) {
1297	kfree(disk);
1298	return NULL;
1299	}
1300	disk->node_id = node_id;
1301	if (disk_expand_part_tbl(disk, 0)) {
1302	free_part_stats(&disk->part0);
1303	kfree(disk);
1304	return NULL;
1305	}
1306	disk->part_tbl->part[0] = &disk->part0;
1307
1308	/*
1309	* set_capacity() and get_capacity() currently don't use
1310	* seqcounter to read/update the part0->nr_sects. Still init
1311	* the counter as we can read the sectors in IO submission
1312	* patch using seqence counters.
1313	*
1314	* TODO: Ideally set_capacity() and get_capacity() should be
1315	* converted to make use of bd_mutex and sequence counters.
1316	*/
1317	seqcount_init(&disk->part0.nr_sects_seq);
1318	if (hd_ref_init(&disk->part0)) {
1319	hd_free_part(&disk->part0);
1320	kfree(disk);
1321	return NULL;
1322	}
1323
1324	disk->minors = minors;
1325	rand_initialize_disk(disk);
1326	disk_to_dev(disk)->class = &block_class;
1327	disk_to_dev(disk)->type = &disk_type;
1328	device_initialize(disk_to_dev(disk));
1329	}
1330	return disk;
1331	}
1332	EXPORT_SYMBOL(alloc_disk_node);
1333
1334	struct kobject *get_disk(struct gendisk *disk)
1335	{
1336	struct module *owner;
1337	struct kobject *kobj;
1338
1339	if (!disk->fops)
1340	return NULL;
1341	owner = disk->fops->owner;
1342	if (owner && !try_module_get(owner))
1343	return NULL;
1344	kobj = kobject_get(&disk_to_dev(disk)->kobj);
1345	if (kobj == NULL) {
1346	module_put(owner);
1347	return NULL;
1348	}
1349	return kobj;
1350
1351	}
1352
1353	EXPORT_SYMBOL(get_disk);
1354
1355	void put_disk(struct gendisk *disk)
1356	{
1357	if (disk)
1358	kobject_put(&disk_to_dev(disk)->kobj);
1359	}
1360
1361	EXPORT_SYMBOL(put_disk);
1362
1363	static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1364	{
1365	char event[] = "DISK_RO=1";
1366	char *envp[] = { event, NULL };
1367
1368	if (!ro)
1369	event[8] = '0';
1370	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1371	}
1372
1373	void set_device_ro(struct block_device *bdev, int flag)
1374	{
1375	bdev->bd_part->policy = flag;
1376	}
1377
1378	EXPORT_SYMBOL(set_device_ro);
1379
1380	void set_disk_ro(struct gendisk *disk, int flag)
1381	{
1382	struct disk_part_iter piter;
1383	struct hd_struct *part;
1384
1385	if (disk->part0.policy != flag) {
1386	set_disk_ro_uevent(disk, flag);
1387	disk->part0.policy = flag;
1388	}
1389
1390	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1391	while ((part = disk_part_iter_next(&piter)))
1392	part->policy = flag;
1393	disk_part_iter_exit(&piter);
1394	}
1395
1396	EXPORT_SYMBOL(set_disk_ro);
1397
1398	int bdev_read_only(struct block_device *bdev)
1399	{
1400	if (!bdev)
1401	return 0;
1402	return bdev->bd_part->policy;
1403	}
1404
1405	EXPORT_SYMBOL(bdev_read_only);
1406
1407	int invalidate_partition(struct gendisk *disk, int partno)
1408	{
1409	int res = 0;
1410	struct block_device *bdev = bdget_disk(disk, partno);
1411	if (bdev) {
1412	fsync_bdev(bdev);
1413	res = __invalidate_device(bdev, true);
1414	bdput(bdev);
1415	}
1416	return res;
1417	}
1418
1419	EXPORT_SYMBOL(invalidate_partition);
1420
1421	/*
1422	* Disk events - monitor disk events like media change and eject request.
1423	*/
1424	struct disk_events {
1425	struct list_head node; /* all disk_event's */
1426	struct gendisk *disk; /* the associated disk */
1427	spinlock_t lock;
1428
1429	struct mutex block_mutex; /* protects blocking */
1430	int block; /* event blocking depth */
1431	unsigned int pending; /* events already sent out */
1432	unsigned int clearing; /* events being cleared */
1433
1434	long poll_msecs; /* interval, -1 for default */
1435	struct delayed_work dwork;
1436	};
1437
1438	static const char *disk_events_strs[] = {
1439	[ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1440	[ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1441	};
1442
1443	static char *disk_uevents[] = {
1444	[ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1445	[ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1446	};
1447
1448	/* list of all disk_events */
1449	static DEFINE_MUTEX(disk_events_mutex);
1450	static LIST_HEAD(disk_events);
1451
1452	/* disable in-kernel polling by default */
1453	static unsigned long disk_events_dfl_poll_msecs;
1454
1455	static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1456	{
1457	struct disk_events *ev = disk->ev;
1458	long intv_msecs = 0;
1459
1460	/*
1461	* If device-specific poll interval is set, always use it. If
1462	* the default is being used, poll iff there are events which
1463	* can't be monitored asynchronously.
1464	*/
1465	if (ev->poll_msecs >= 0)
1466	intv_msecs = ev->poll_msecs;
1467	else if (disk->events & ~disk->async_events)
1468	intv_msecs = disk_events_dfl_poll_msecs;
1469
1470	return msecs_to_jiffies(intv_msecs);
1471	}
1472
1473	/**
1474	* disk_block_events - block and flush disk event checking
1475	* @disk: disk to block events for
1476	*
1477	* On return from this function, it is guaranteed that event checking
1478	* isn't in progress and won't happen until unblocked by
1479	* disk_unblock_events(). Events blocking is counted and the actual
1480	* unblocking happens after the matching number of unblocks are done.
1481	*
1482	* Note that this intentionally does not block event checking from
1483	* disk_clear_events().
1484	*
1485	* CONTEXT:
1486	* Might sleep.
1487	*/
1488	void disk_block_events(struct gendisk *disk)
1489	{
1490	struct disk_events *ev = disk->ev;
1491	unsigned long flags;
1492	bool cancel;
1493
1494	if (!ev)
1495	return;
1496
1497	/*
1498	* Outer mutex ensures that the first blocker completes canceling
1499	* the event work before further blockers are allowed to finish.
1500	*/
1501	mutex_lock(&ev->block_mutex);
1502
1503	spin_lock_irqsave(&ev->lock, flags);
1504	cancel = !ev->block++;
1505	spin_unlock_irqrestore(&ev->lock, flags);
1506
1507	if (cancel)
1508	cancel_delayed_work_sync(&disk->ev->dwork);
1509
1510	mutex_unlock(&ev->block_mutex);
1511	}
1512
1513	static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1514	{
1515	struct disk_events *ev = disk->ev;
1516	unsigned long intv;
1517	unsigned long flags;
1518
1519	spin_lock_irqsave(&ev->lock, flags);
1520
1521	if (WARN_ON_ONCE(ev->block <= 0))
1522	goto out_unlock;
1523
1524	if (--ev->block)
1525	goto out_unlock;
1526
1527	intv = disk_events_poll_jiffies(disk);
1528	if (check_now)
1529	queue_delayed_work(system_freezable_power_efficient_wq,
1530	&ev->dwork, 0);
1531	else if (intv)
1532	queue_delayed_work(system_freezable_power_efficient_wq,
1533	&ev->dwork, intv);
1534	out_unlock:
1535	spin_unlock_irqrestore(&ev->lock, flags);
1536	}
1537
1538	/**
1539	* disk_unblock_events - unblock disk event checking
1540	* @disk: disk to unblock events for
1541	*
1542	* Undo disk_block_events(). When the block count reaches zero, it
1543	* starts events polling if configured.
1544	*
1545	* CONTEXT:
1546	* Don't care. Safe to call from irq context.
1547	*/
1548	void disk_unblock_events(struct gendisk *disk)
1549	{
1550	if (disk->ev)
1551	__disk_unblock_events(disk, false);
1552	}
1553
1554	/**
1555	* disk_flush_events - schedule immediate event checking and flushing
1556	* @disk: disk to check and flush events for
1557	* @mask: events to flush
1558	*
1559	* Schedule immediate event checking on @disk if not blocked. Events in
1560	* @mask are scheduled to be cleared from the driver. Note that this
1561	* doesn't clear the events from @disk->ev.
1562	*
1563	* CONTEXT:
1564	* If @mask is non-zero must be called with bdev->bd_mutex held.
1565	*/
1566	void disk_flush_events(struct gendisk *disk, unsigned int mask)
1567	{
1568	struct disk_events *ev = disk->ev;
1569
1570	if (!ev)
1571	return;
1572
1573	spin_lock_irq(&ev->lock);
1574	ev->clearing |= mask;
1575	if (!ev->block)
1576	mod_delayed_work(system_freezable_power_efficient_wq,
1577	&ev->dwork, 0);
1578	spin_unlock_irq(&ev->lock);
1579	}
1580
1581	/**
1582	* disk_clear_events - synchronously check, clear and return pending events
1583	* @disk: disk to fetch and clear events from
1584	* @mask: mask of events to be fetched and cleared
1585	*
1586	* Disk events are synchronously checked and pending events in @mask
1587	* are cleared and returned. This ignores the block count.
1588	*
1589	* CONTEXT:
1590	* Might sleep.
1591	*/
1592	unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1593	{
1594	const struct block_device_operations *bdops = disk->fops;
1595	struct disk_events *ev = disk->ev;
1596	unsigned int pending;
1597	unsigned int clearing = mask;
1598
1599	if (!ev) {
1600	/* for drivers still using the old ->media_changed method */
1601	if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1602	bdops->media_changed && bdops->media_changed(disk))
1603	return DISK_EVENT_MEDIA_CHANGE;
1604	return 0;
1605	}
1606
1607	disk_block_events(disk);
1608
1609	/*
1610	* store the union of mask and ev->clearing on the stack so that the
1611	* race with disk_flush_events does not cause ambiguity (ev->clearing
1612	* can still be modified even if events are blocked).
1613	*/
1614	spin_lock_irq(&ev->lock);
1615	clearing |= ev->clearing;
1616	ev->clearing = 0;
1617	spin_unlock_irq(&ev->lock);
1618
1619	disk_check_events(ev, &clearing);
1620	/*
1621	* if ev->clearing is not 0, the disk_flush_events got called in the
1622	* middle of this function, so we want to run the workfn without delay.
1623	*/
1624	__disk_unblock_events(disk, ev->clearing ? true : false);
1625
1626	/* then, fetch and clear pending events */
1627	spin_lock_irq(&ev->lock);
1628	pending = ev->pending & mask;
1629	ev->pending &= ~mask;
1630	spin_unlock_irq(&ev->lock);
1631	WARN_ON_ONCE(clearing & mask);
1632
1633	return pending;
1634	}
1635
1636	/*
1637	* Separate this part out so that a different pointer for clearing_ptr can be
1638	* passed in for disk_clear_events.
1639	*/
1640	static void disk_events_workfn(struct work_struct *work)
1641	{
1642	struct delayed_work *dwork = to_delayed_work(work);
1643	struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1644
1645	disk_check_events(ev, &ev->clearing);
1646	}
1647
1648	static void disk_check_events(struct disk_events *ev,
1649	unsigned int *clearing_ptr)
1650	{
1651	struct gendisk *disk = ev->disk;
1652	char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1653	unsigned int clearing = *clearing_ptr;
1654	unsigned int events;
1655	unsigned long intv;
1656	int nr_events = 0, i;
1657
1658	/* check events */
1659	events = disk->fops->check_events(disk, clearing);
1660
1661	/* accumulate pending events and schedule next poll if necessary */
1662	spin_lock_irq(&ev->lock);
1663
1664	events &= ~ev->pending;
1665	ev->pending |= events;
1666	*clearing_ptr &= ~clearing;
1667
1668	intv = disk_events_poll_jiffies(disk);
1669	if (!ev->block && intv)
1670	queue_delayed_work(system_freezable_power_efficient_wq,
1671	&ev->dwork, intv);
1672
1673	spin_unlock_irq(&ev->lock);
1674
1675	/*
1676	* Tell userland about new events. Only the events listed in
1677	* @disk->events are reported. Unlisted events are processed the
1678	* same internally but never get reported to userland.
1679	*/
1680	for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1681	if (events & disk->events & (1 << i))
1682	envp[nr_events++] = disk_uevents[i];
1683
1684	if (nr_events)
1685	kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1686	}
1687
1688	/*
1689	* A disk events enabled device has the following sysfs nodes under
1690	* its /sys/block/X/ directory.
1691	*
1692	* events : list of all supported events
1693	* events_async : list of events which can be detected w/o polling
1694	* events_poll_msecs : polling interval, 0: disable, -1: system default
1695	*/
1696	static ssize_t __disk_events_show(unsigned int events, char *buf)
1697	{
1698	const char *delim = "";
1699	ssize_t pos = 0;
1700	int i;
1701
1702	for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1703	if (events & (1 << i)) {
1704	pos += sprintf(buf + pos, "%s%s",
1705	delim, disk_events_strs[i]);
1706	delim = " ";
1707	}
1708	if (pos)
1709	pos += sprintf(buf + pos, "\n");
1710	return pos;
1711	}
1712
1713	static ssize_t disk_events_show(struct device *dev,
1714	struct device_attribute *attr, char *buf)
1715	{
1716	struct gendisk *disk = dev_to_disk(dev);
1717
1718	return __disk_events_show(disk->events, buf);
1719	}
1720
1721	static ssize_t disk_events_async_show(struct device *dev,
1722	struct device_attribute *attr, char *buf)
1723	{
1724	struct gendisk *disk = dev_to_disk(dev);
1725
1726	return __disk_events_show(disk->async_events, buf);
1727	}
1728
1729	static ssize_t disk_events_poll_msecs_show(struct device *dev,
1730	struct device_attribute *attr,
1731	char *buf)
1732	{
1733	struct gendisk *disk = dev_to_disk(dev);
1734
1735	return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1736	}
1737
1738	static ssize_t disk_events_poll_msecs_store(struct device *dev,
1739	struct device_attribute *attr,
1740	const char *buf, size_t count)
1741	{
1742	struct gendisk *disk = dev_to_disk(dev);
1743	long intv;
1744
1745	if (!count || !sscanf(buf, "%ld", &intv))
1746	return -EINVAL;
1747
1748	if (intv < 0 && intv != -1)
1749	return -EINVAL;
1750
1751	disk_block_events(disk);
1752	disk->ev->poll_msecs = intv;
1753	__disk_unblock_events(disk, true);
1754
1755	return count;
1756	}
1757
1758	static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1759	static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1760	static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1761	disk_events_poll_msecs_show,
1762	disk_events_poll_msecs_store);
1763
1764	static const struct attribute *disk_events_attrs[] = {
1765	&dev_attr_events.attr,
1766	&dev_attr_events_async.attr,
1767	&dev_attr_events_poll_msecs.attr,
1768	NULL,
1769	};
1770
1771	/*
1772	* The default polling interval can be specified by the kernel
1773	* parameter block.events_dfl_poll_msecs which defaults to 0
1774	* (disable). This can also be modified runtime by writing to
1775	* /sys/module/block/events_dfl_poll_msecs.
1776	*/
1777	static int disk_events_set_dfl_poll_msecs(const char *val,
1778	const struct kernel_param *kp)
1779	{
1780	struct disk_events *ev;
1781	int ret;
1782
1783	ret = param_set_ulong(val, kp);
1784	if (ret < 0)
1785	return ret;
1786
1787	mutex_lock(&disk_events_mutex);
1788
1789	list_for_each_entry(ev, &disk_events, node)
1790	disk_flush_events(ev->disk, 0);
1791
1792	mutex_unlock(&disk_events_mutex);
1793
1794	return 0;
1795	}
1796
1797	static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1798	.set = disk_events_set_dfl_poll_msecs,
1799	.get = param_get_ulong,
1800	};
1801
1802	#undef MODULE_PARAM_PREFIX
1803	#define MODULE_PARAM_PREFIX "block."
1804
1805	module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1806	&disk_events_dfl_poll_msecs, 0644);
1807
1808	/*
1809	* disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1810	*/
1811	static void disk_alloc_events(struct gendisk *disk)
1812	{
1813	struct disk_events *ev;
1814
1815	if (!disk->fops->check_events)
1816	return;
1817
1818	ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1819	if (!ev) {
1820	pr_warn("%s: failed to initialize events\n", disk->disk_name);
1821	return;
1822	}
1823
1824	INIT_LIST_HEAD(&ev->node);
1825	ev->disk = disk;
1826	spin_lock_init(&ev->lock);
1827	mutex_init(&ev->block_mutex);
1828	ev->block = 1;
1829	ev->poll_msecs = -1;
1830	INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1831
1832	disk->ev = ev;
1833	}
1834
1835	static void disk_add_events(struct gendisk *disk)
1836	{
1837	if (!disk->ev)
1838	return;
1839
1840	/* FIXME: error handling */
1841	if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1842	pr_warn("%s: failed to create sysfs files for events\n",
1843	disk->disk_name);
1844
1845	mutex_lock(&disk_events_mutex);
1846	list_add_tail(&disk->ev->node, &disk_events);
1847	mutex_unlock(&disk_events_mutex);
1848
1849	/*
1850	* Block count is initialized to 1 and the following initial
1851	* unblock kicks it into action.
1852	*/
1853	__disk_unblock_events(disk, true);
1854	}
1855
1856	static void disk_del_events(struct gendisk *disk)
1857	{
1858	if (!disk->ev)
1859	return;
1860
1861	disk_block_events(disk);
1862
1863	mutex_lock(&disk_events_mutex);
1864	list_del_init(&disk->ev->node);
1865	mutex_unlock(&disk_events_mutex);
1866
1867	sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1868	}
1869
1870	static void disk_release_events(struct gendisk *disk)
1871	{
1872	/* the block count should be 1 from disk_del_events() */
1873	WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1874	kfree(disk->ev);
1875	}
1876