]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - block/genhd.c
s390/kvm: REPLACE barrier fixup with READ_ONCE
[mirror_ubuntu-zesty-kernel.git] / block / genhd.c
1 /*
2 * gendisk handling
3 */
4
5 #include <linux/module.h>
6 #include <linux/fs.h>
7 #include <linux/genhd.h>
8 #include <linux/kdev_t.h>
9 #include <linux/kernel.h>
10 #include <linux/blkdev.h>
11 #include <linux/init.h>
12 #include <linux/spinlock.h>
13 #include <linux/proc_fs.h>
14 #include <linux/seq_file.h>
15 #include <linux/slab.h>
16 #include <linux/kmod.h>
17 #include <linux/kobj_map.h>
18 #include <linux/mutex.h>
19 #include <linux/idr.h>
20 #include <linux/log2.h>
21 #include <linux/pm_runtime.h>
22
23 #include "blk.h"
24
25 static DEFINE_MUTEX(block_class_lock);
26 struct kobject *block_depr;
27
28 /* for extended dynamic devt allocation, currently only one major is used */
29 #define NR_EXT_DEVT (1 << MINORBITS)
30
31 /* For extended devt allocation. ext_devt_lock prevents look up
32 * results from going away underneath its user.
33 */
34 static DEFINE_SPINLOCK(ext_devt_lock);
35 static DEFINE_IDR(ext_devt_idr);
36
37 static struct device_type disk_type;
38
39 static void disk_check_events(struct disk_events *ev,
40 unsigned int *clearing_ptr);
41 static void disk_alloc_events(struct gendisk *disk);
42 static void disk_add_events(struct gendisk *disk);
43 static void disk_del_events(struct gendisk *disk);
44 static void disk_release_events(struct gendisk *disk);
45
46 /**
47 * disk_get_part - get partition
48 * @disk: disk to look partition from
49 * @partno: partition number
50 *
51 * Look for partition @partno from @disk. If found, increment
52 * reference count and return it.
53 *
54 * CONTEXT:
55 * Don't care.
56 *
57 * RETURNS:
58 * Pointer to the found partition on success, NULL if not found.
59 */
60 struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
61 {
62 struct hd_struct *part = NULL;
63 struct disk_part_tbl *ptbl;
64
65 if (unlikely(partno < 0))
66 return NULL;
67
68 rcu_read_lock();
69
70 ptbl = rcu_dereference(disk->part_tbl);
71 if (likely(partno < ptbl->len)) {
72 part = rcu_dereference(ptbl->part[partno]);
73 if (part)
74 get_device(part_to_dev(part));
75 }
76
77 rcu_read_unlock();
78
79 return part;
80 }
81 EXPORT_SYMBOL_GPL(disk_get_part);
82
83 /**
84 * disk_part_iter_init - initialize partition iterator
85 * @piter: iterator to initialize
86 * @disk: disk to iterate over
87 * @flags: DISK_PITER_* flags
88 *
89 * Initialize @piter so that it iterates over partitions of @disk.
90 *
91 * CONTEXT:
92 * Don't care.
93 */
94 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
95 unsigned int flags)
96 {
97 struct disk_part_tbl *ptbl;
98
99 rcu_read_lock();
100 ptbl = rcu_dereference(disk->part_tbl);
101
102 piter->disk = disk;
103 piter->part = NULL;
104
105 if (flags & DISK_PITER_REVERSE)
106 piter->idx = ptbl->len - 1;
107 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
108 piter->idx = 0;
109 else
110 piter->idx = 1;
111
112 piter->flags = flags;
113
114 rcu_read_unlock();
115 }
116 EXPORT_SYMBOL_GPL(disk_part_iter_init);
117
118 /**
119 * disk_part_iter_next - proceed iterator to the next partition and return it
120 * @piter: iterator of interest
121 *
122 * Proceed @piter to the next partition and return it.
123 *
124 * CONTEXT:
125 * Don't care.
126 */
127 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
128 {
129 struct disk_part_tbl *ptbl;
130 int inc, end;
131
132 /* put the last partition */
133 disk_put_part(piter->part);
134 piter->part = NULL;
135
136 /* get part_tbl */
137 rcu_read_lock();
138 ptbl = rcu_dereference(piter->disk->part_tbl);
139
140 /* determine iteration parameters */
141 if (piter->flags & DISK_PITER_REVERSE) {
142 inc = -1;
143 if (piter->flags & (DISK_PITER_INCL_PART0 |
144 DISK_PITER_INCL_EMPTY_PART0))
145 end = -1;
146 else
147 end = 0;
148 } else {
149 inc = 1;
150 end = ptbl->len;
151 }
152
153 /* iterate to the next partition */
154 for (; piter->idx != end; piter->idx += inc) {
155 struct hd_struct *part;
156
157 part = rcu_dereference(ptbl->part[piter->idx]);
158 if (!part)
159 continue;
160 if (!part_nr_sects_read(part) &&
161 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
162 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
163 piter->idx == 0))
164 continue;
165
166 get_device(part_to_dev(part));
167 piter->part = part;
168 piter->idx += inc;
169 break;
170 }
171
172 rcu_read_unlock();
173
174 return piter->part;
175 }
176 EXPORT_SYMBOL_GPL(disk_part_iter_next);
177
178 /**
179 * disk_part_iter_exit - finish up partition iteration
180 * @piter: iter of interest
181 *
182 * Called when iteration is over. Cleans up @piter.
183 *
184 * CONTEXT:
185 * Don't care.
186 */
187 void disk_part_iter_exit(struct disk_part_iter *piter)
188 {
189 disk_put_part(piter->part);
190 piter->part = NULL;
191 }
192 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
193
194 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
195 {
196 return part->start_sect <= sector &&
197 sector < part->start_sect + part_nr_sects_read(part);
198 }
199
200 /**
201 * disk_map_sector_rcu - map sector to partition
202 * @disk: gendisk of interest
203 * @sector: sector to map
204 *
205 * Find out which partition @sector maps to on @disk. This is
206 * primarily used for stats accounting.
207 *
208 * CONTEXT:
209 * RCU read locked. The returned partition pointer is valid only
210 * while preemption is disabled.
211 *
212 * RETURNS:
213 * Found partition on success, part0 is returned if no partition matches
214 */
215 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
216 {
217 struct disk_part_tbl *ptbl;
218 struct hd_struct *part;
219 int i;
220
221 ptbl = rcu_dereference(disk->part_tbl);
222
223 part = rcu_dereference(ptbl->last_lookup);
224 if (part && sector_in_part(part, sector))
225 return part;
226
227 for (i = 1; i < ptbl->len; i++) {
228 part = rcu_dereference(ptbl->part[i]);
229
230 if (part && sector_in_part(part, sector)) {
231 rcu_assign_pointer(ptbl->last_lookup, part);
232 return part;
233 }
234 }
235 return &disk->part0;
236 }
237 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
238
239 /*
240 * Can be deleted altogether. Later.
241 *
242 */
243 static struct blk_major_name {
244 struct blk_major_name *next;
245 int major;
246 char name[16];
247 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
248
249 /* index in the above - for now: assume no multimajor ranges */
250 static inline int major_to_index(unsigned major)
251 {
252 return major % BLKDEV_MAJOR_HASH_SIZE;
253 }
254
255 #ifdef CONFIG_PROC_FS
256 void blkdev_show(struct seq_file *seqf, off_t offset)
257 {
258 struct blk_major_name *dp;
259
260 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
261 mutex_lock(&block_class_lock);
262 for (dp = major_names[offset]; dp; dp = dp->next)
263 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
264 mutex_unlock(&block_class_lock);
265 }
266 }
267 #endif /* CONFIG_PROC_FS */
268
269 /**
270 * register_blkdev - register a new block device
271 *
272 * @major: the requested major device number [1..255]. If @major=0, try to
273 * allocate any unused major number.
274 * @name: the name of the new block device as a zero terminated string
275 *
276 * The @name must be unique within the system.
277 *
278 * The return value depends on the @major input parameter.
279 * - if a major device number was requested in range [1..255] then the
280 * function returns zero on success, or a negative error code
281 * - if any unused major number was requested with @major=0 parameter
282 * then the return value is the allocated major number in range
283 * [1..255] or a negative error code otherwise
284 */
285 int register_blkdev(unsigned int major, const char *name)
286 {
287 struct blk_major_name **n, *p;
288 int index, ret = 0;
289
290 mutex_lock(&block_class_lock);
291
292 /* temporary */
293 if (major == 0) {
294 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
295 if (major_names[index] == NULL)
296 break;
297 }
298
299 if (index == 0) {
300 printk("register_blkdev: failed to get major for %s\n",
301 name);
302 ret = -EBUSY;
303 goto out;
304 }
305 major = index;
306 ret = major;
307 }
308
309 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
310 if (p == NULL) {
311 ret = -ENOMEM;
312 goto out;
313 }
314
315 p->major = major;
316 strlcpy(p->name, name, sizeof(p->name));
317 p->next = NULL;
318 index = major_to_index(major);
319
320 for (n = &major_names[index]; *n; n = &(*n)->next) {
321 if ((*n)->major == major)
322 break;
323 }
324 if (!*n)
325 *n = p;
326 else
327 ret = -EBUSY;
328
329 if (ret < 0) {
330 printk("register_blkdev: cannot get major %d for %s\n",
331 major, name);
332 kfree(p);
333 }
334 out:
335 mutex_unlock(&block_class_lock);
336 return ret;
337 }
338
339 EXPORT_SYMBOL(register_blkdev);
340
341 void unregister_blkdev(unsigned int major, const char *name)
342 {
343 struct blk_major_name **n;
344 struct blk_major_name *p = NULL;
345 int index = major_to_index(major);
346
347 mutex_lock(&block_class_lock);
348 for (n = &major_names[index]; *n; n = &(*n)->next)
349 if ((*n)->major == major)
350 break;
351 if (!*n || strcmp((*n)->name, name)) {
352 WARN_ON(1);
353 } else {
354 p = *n;
355 *n = p->next;
356 }
357 mutex_unlock(&block_class_lock);
358 kfree(p);
359 }
360
361 EXPORT_SYMBOL(unregister_blkdev);
362
363 static struct kobj_map *bdev_map;
364
365 /**
366 * blk_mangle_minor - scatter minor numbers apart
367 * @minor: minor number to mangle
368 *
369 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
370 * is enabled. Mangling twice gives the original value.
371 *
372 * RETURNS:
373 * Mangled value.
374 *
375 * CONTEXT:
376 * Don't care.
377 */
378 static int blk_mangle_minor(int minor)
379 {
380 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
381 int i;
382
383 for (i = 0; i < MINORBITS / 2; i++) {
384 int low = minor & (1 << i);
385 int high = minor & (1 << (MINORBITS - 1 - i));
386 int distance = MINORBITS - 1 - 2 * i;
387
388 minor ^= low | high; /* clear both bits */
389 low <<= distance; /* swap the positions */
390 high >>= distance;
391 minor |= low | high; /* and set */
392 }
393 #endif
394 return minor;
395 }
396
397 /**
398 * blk_alloc_devt - allocate a dev_t for a partition
399 * @part: partition to allocate dev_t for
400 * @devt: out parameter for resulting dev_t
401 *
402 * Allocate a dev_t for block device.
403 *
404 * RETURNS:
405 * 0 on success, allocated dev_t is returned in *@devt. -errno on
406 * failure.
407 *
408 * CONTEXT:
409 * Might sleep.
410 */
411 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
412 {
413 struct gendisk *disk = part_to_disk(part);
414 int idx;
415
416 /* in consecutive minor range? */
417 if (part->partno < disk->minors) {
418 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
419 return 0;
420 }
421
422 /* allocate ext devt */
423 idr_preload(GFP_KERNEL);
424
425 spin_lock(&ext_devt_lock);
426 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
427 spin_unlock(&ext_devt_lock);
428
429 idr_preload_end();
430 if (idx < 0)
431 return idx == -ENOSPC ? -EBUSY : idx;
432
433 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
434 return 0;
435 }
436
437 /**
438 * blk_free_devt - free a dev_t
439 * @devt: dev_t to free
440 *
441 * Free @devt which was allocated using blk_alloc_devt().
442 *
443 * CONTEXT:
444 * Might sleep.
445 */
446 void blk_free_devt(dev_t devt)
447 {
448 if (devt == MKDEV(0, 0))
449 return;
450
451 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
452 spin_lock(&ext_devt_lock);
453 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
454 spin_unlock(&ext_devt_lock);
455 }
456 }
457
458 static char *bdevt_str(dev_t devt, char *buf)
459 {
460 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
461 char tbuf[BDEVT_SIZE];
462 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
463 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
464 } else
465 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
466
467 return buf;
468 }
469
470 /*
471 * Register device numbers dev..(dev+range-1)
472 * range must be nonzero
473 * The hash chain is sorted on range, so that subranges can override.
474 */
475 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
476 struct kobject *(*probe)(dev_t, int *, void *),
477 int (*lock)(dev_t, void *), void *data)
478 {
479 kobj_map(bdev_map, devt, range, module, probe, lock, data);
480 }
481
482 EXPORT_SYMBOL(blk_register_region);
483
484 void blk_unregister_region(dev_t devt, unsigned long range)
485 {
486 kobj_unmap(bdev_map, devt, range);
487 }
488
489 EXPORT_SYMBOL(blk_unregister_region);
490
491 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
492 {
493 struct gendisk *p = data;
494
495 return &disk_to_dev(p)->kobj;
496 }
497
498 static int exact_lock(dev_t devt, void *data)
499 {
500 struct gendisk *p = data;
501
502 if (!get_disk(p))
503 return -1;
504 return 0;
505 }
506
507 static void register_disk(struct gendisk *disk)
508 {
509 struct device *ddev = disk_to_dev(disk);
510 struct block_device *bdev;
511 struct disk_part_iter piter;
512 struct hd_struct *part;
513 int err;
514
515 ddev->parent = disk->driverfs_dev;
516
517 dev_set_name(ddev, "%s", disk->disk_name);
518
519 /* delay uevents, until we scanned partition table */
520 dev_set_uevent_suppress(ddev, 1);
521
522 if (device_add(ddev))
523 return;
524 if (!sysfs_deprecated) {
525 err = sysfs_create_link(block_depr, &ddev->kobj,
526 kobject_name(&ddev->kobj));
527 if (err) {
528 device_del(ddev);
529 return;
530 }
531 }
532
533 /*
534 * avoid probable deadlock caused by allocating memory with
535 * GFP_KERNEL in runtime_resume callback of its all ancestor
536 * devices
537 */
538 pm_runtime_set_memalloc_noio(ddev, true);
539
540 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
541 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
542
543 /* No minors to use for partitions */
544 if (!disk_part_scan_enabled(disk))
545 goto exit;
546
547 /* No such device (e.g., media were just removed) */
548 if (!get_capacity(disk))
549 goto exit;
550
551 bdev = bdget_disk(disk, 0);
552 if (!bdev)
553 goto exit;
554
555 bdev->bd_invalidated = 1;
556 err = blkdev_get(bdev, FMODE_READ, NULL);
557 if (err < 0)
558 goto exit;
559 blkdev_put(bdev, FMODE_READ);
560
561 exit:
562 /* announce disk after possible partitions are created */
563 dev_set_uevent_suppress(ddev, 0);
564 kobject_uevent(&ddev->kobj, KOBJ_ADD);
565
566 /* announce possible partitions */
567 disk_part_iter_init(&piter, disk, 0);
568 while ((part = disk_part_iter_next(&piter)))
569 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
570 disk_part_iter_exit(&piter);
571 }
572
573 /**
574 * add_disk - add partitioning information to kernel list
575 * @disk: per-device partitioning information
576 *
577 * This function registers the partitioning information in @disk
578 * with the kernel.
579 *
580 * FIXME: error handling
581 */
582 void add_disk(struct gendisk *disk)
583 {
584 struct backing_dev_info *bdi;
585 dev_t devt;
586 int retval;
587
588 /* minors == 0 indicates to use ext devt from part0 and should
589 * be accompanied with EXT_DEVT flag. Make sure all
590 * parameters make sense.
591 */
592 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
593 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
594
595 disk->flags |= GENHD_FL_UP;
596
597 retval = blk_alloc_devt(&disk->part0, &devt);
598 if (retval) {
599 WARN_ON(1);
600 return;
601 }
602 disk_to_dev(disk)->devt = devt;
603
604 /* ->major and ->first_minor aren't supposed to be
605 * dereferenced from here on, but set them just in case.
606 */
607 disk->major = MAJOR(devt);
608 disk->first_minor = MINOR(devt);
609
610 disk_alloc_events(disk);
611
612 /* Register BDI before referencing it from bdev */
613 bdi = &disk->queue->backing_dev_info;
614 bdi_register_dev(bdi, disk_devt(disk));
615
616 blk_register_region(disk_devt(disk), disk->minors, NULL,
617 exact_match, exact_lock, disk);
618 register_disk(disk);
619 blk_register_queue(disk);
620
621 /*
622 * Take an extra ref on queue which will be put on disk_release()
623 * so that it sticks around as long as @disk is there.
624 */
625 WARN_ON_ONCE(!blk_get_queue(disk->queue));
626
627 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
628 "bdi");
629 WARN_ON(retval);
630
631 disk_add_events(disk);
632 }
633 EXPORT_SYMBOL(add_disk);
634
635 void del_gendisk(struct gendisk *disk)
636 {
637 struct disk_part_iter piter;
638 struct hd_struct *part;
639
640 disk_del_events(disk);
641
642 /* invalidate stuff */
643 disk_part_iter_init(&piter, disk,
644 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
645 while ((part = disk_part_iter_next(&piter))) {
646 invalidate_partition(disk, part->partno);
647 delete_partition(disk, part->partno);
648 }
649 disk_part_iter_exit(&piter);
650
651 invalidate_partition(disk, 0);
652 set_capacity(disk, 0);
653 disk->flags &= ~GENHD_FL_UP;
654
655 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
656 bdi_unregister(&disk->queue->backing_dev_info);
657 blk_unregister_queue(disk);
658 blk_unregister_region(disk_devt(disk), disk->minors);
659
660 part_stat_set_all(&disk->part0, 0);
661 disk->part0.stamp = 0;
662
663 kobject_put(disk->part0.holder_dir);
664 kobject_put(disk->slave_dir);
665 disk->driverfs_dev = NULL;
666 if (!sysfs_deprecated)
667 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
668 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
669 device_del(disk_to_dev(disk));
670 }
671 EXPORT_SYMBOL(del_gendisk);
672
673 /**
674 * get_gendisk - get partitioning information for a given device
675 * @devt: device to get partitioning information for
676 * @partno: returned partition index
677 *
678 * This function gets the structure containing partitioning
679 * information for the given device @devt.
680 */
681 struct gendisk *get_gendisk(dev_t devt, int *partno)
682 {
683 struct gendisk *disk = NULL;
684
685 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
686 struct kobject *kobj;
687
688 kobj = kobj_lookup(bdev_map, devt, partno);
689 if (kobj)
690 disk = dev_to_disk(kobj_to_dev(kobj));
691 } else {
692 struct hd_struct *part;
693
694 spin_lock(&ext_devt_lock);
695 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
696 if (part && get_disk(part_to_disk(part))) {
697 *partno = part->partno;
698 disk = part_to_disk(part);
699 }
700 spin_unlock(&ext_devt_lock);
701 }
702
703 return disk;
704 }
705 EXPORT_SYMBOL(get_gendisk);
706
707 /**
708 * bdget_disk - do bdget() by gendisk and partition number
709 * @disk: gendisk of interest
710 * @partno: partition number
711 *
712 * Find partition @partno from @disk, do bdget() on it.
713 *
714 * CONTEXT:
715 * Don't care.
716 *
717 * RETURNS:
718 * Resulting block_device on success, NULL on failure.
719 */
720 struct block_device *bdget_disk(struct gendisk *disk, int partno)
721 {
722 struct hd_struct *part;
723 struct block_device *bdev = NULL;
724
725 part = disk_get_part(disk, partno);
726 if (part)
727 bdev = bdget(part_devt(part));
728 disk_put_part(part);
729
730 return bdev;
731 }
732 EXPORT_SYMBOL(bdget_disk);
733
734 /*
735 * print a full list of all partitions - intended for places where the root
736 * filesystem can't be mounted and thus to give the victim some idea of what
737 * went wrong
738 */
739 void __init printk_all_partitions(void)
740 {
741 struct class_dev_iter iter;
742 struct device *dev;
743
744 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
745 while ((dev = class_dev_iter_next(&iter))) {
746 struct gendisk *disk = dev_to_disk(dev);
747 struct disk_part_iter piter;
748 struct hd_struct *part;
749 char name_buf[BDEVNAME_SIZE];
750 char devt_buf[BDEVT_SIZE];
751
752 /*
753 * Don't show empty devices or things that have been
754 * suppressed
755 */
756 if (get_capacity(disk) == 0 ||
757 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
758 continue;
759
760 /*
761 * Note, unlike /proc/partitions, I am showing the
762 * numbers in hex - the same format as the root=
763 * option takes.
764 */
765 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
766 while ((part = disk_part_iter_next(&piter))) {
767 bool is_part0 = part == &disk->part0;
768
769 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
770 bdevt_str(part_devt(part), devt_buf),
771 (unsigned long long)part_nr_sects_read(part) >> 1
772 , disk_name(disk, part->partno, name_buf),
773 part->info ? part->info->uuid : "");
774 if (is_part0) {
775 if (disk->driverfs_dev != NULL &&
776 disk->driverfs_dev->driver != NULL)
777 printk(" driver: %s\n",
778 disk->driverfs_dev->driver->name);
779 else
780 printk(" (driver?)\n");
781 } else
782 printk("\n");
783 }
784 disk_part_iter_exit(&piter);
785 }
786 class_dev_iter_exit(&iter);
787 }
788
789 #ifdef CONFIG_PROC_FS
790 /* iterator */
791 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
792 {
793 loff_t skip = *pos;
794 struct class_dev_iter *iter;
795 struct device *dev;
796
797 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
798 if (!iter)
799 return ERR_PTR(-ENOMEM);
800
801 seqf->private = iter;
802 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
803 do {
804 dev = class_dev_iter_next(iter);
805 if (!dev)
806 return NULL;
807 } while (skip--);
808
809 return dev_to_disk(dev);
810 }
811
812 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
813 {
814 struct device *dev;
815
816 (*pos)++;
817 dev = class_dev_iter_next(seqf->private);
818 if (dev)
819 return dev_to_disk(dev);
820
821 return NULL;
822 }
823
824 static void disk_seqf_stop(struct seq_file *seqf, void *v)
825 {
826 struct class_dev_iter *iter = seqf->private;
827
828 /* stop is called even after start failed :-( */
829 if (iter) {
830 class_dev_iter_exit(iter);
831 kfree(iter);
832 }
833 }
834
835 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
836 {
837 void *p;
838
839 p = disk_seqf_start(seqf, pos);
840 if (!IS_ERR_OR_NULL(p) && !*pos)
841 seq_puts(seqf, "major minor #blocks name\n\n");
842 return p;
843 }
844
845 static int show_partition(struct seq_file *seqf, void *v)
846 {
847 struct gendisk *sgp = v;
848 struct disk_part_iter piter;
849 struct hd_struct *part;
850 char buf[BDEVNAME_SIZE];
851
852 /* Don't show non-partitionable removeable devices or empty devices */
853 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
854 (sgp->flags & GENHD_FL_REMOVABLE)))
855 return 0;
856 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
857 return 0;
858
859 /* show the full disk and all non-0 size partitions of it */
860 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
861 while ((part = disk_part_iter_next(&piter)))
862 seq_printf(seqf, "%4d %7d %10llu %s\n",
863 MAJOR(part_devt(part)), MINOR(part_devt(part)),
864 (unsigned long long)part_nr_sects_read(part) >> 1,
865 disk_name(sgp, part->partno, buf));
866 disk_part_iter_exit(&piter);
867
868 return 0;
869 }
870
871 static const struct seq_operations partitions_op = {
872 .start = show_partition_start,
873 .next = disk_seqf_next,
874 .stop = disk_seqf_stop,
875 .show = show_partition
876 };
877
878 static int partitions_open(struct inode *inode, struct file *file)
879 {
880 return seq_open(file, &partitions_op);
881 }
882
883 static const struct file_operations proc_partitions_operations = {
884 .open = partitions_open,
885 .read = seq_read,
886 .llseek = seq_lseek,
887 .release = seq_release,
888 };
889 #endif
890
891
892 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
893 {
894 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
895 /* Make old-style 2.4 aliases work */
896 request_module("block-major-%d", MAJOR(devt));
897 return NULL;
898 }
899
900 static int __init genhd_device_init(void)
901 {
902 int error;
903
904 block_class.dev_kobj = sysfs_dev_block_kobj;
905 error = class_register(&block_class);
906 if (unlikely(error))
907 return error;
908 bdev_map = kobj_map_init(base_probe, &block_class_lock);
909 blk_dev_init();
910
911 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
912
913 /* create top-level block dir */
914 if (!sysfs_deprecated)
915 block_depr = kobject_create_and_add("block", NULL);
916 return 0;
917 }
918
919 subsys_initcall(genhd_device_init);
920
921 static ssize_t disk_range_show(struct device *dev,
922 struct device_attribute *attr, char *buf)
923 {
924 struct gendisk *disk = dev_to_disk(dev);
925
926 return sprintf(buf, "%d\n", disk->minors);
927 }
928
929 static ssize_t disk_ext_range_show(struct device *dev,
930 struct device_attribute *attr, char *buf)
931 {
932 struct gendisk *disk = dev_to_disk(dev);
933
934 return sprintf(buf, "%d\n", disk_max_parts(disk));
935 }
936
937 static ssize_t disk_removable_show(struct device *dev,
938 struct device_attribute *attr, char *buf)
939 {
940 struct gendisk *disk = dev_to_disk(dev);
941
942 return sprintf(buf, "%d\n",
943 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
944 }
945
946 static ssize_t disk_ro_show(struct device *dev,
947 struct device_attribute *attr, char *buf)
948 {
949 struct gendisk *disk = dev_to_disk(dev);
950
951 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
952 }
953
954 static ssize_t disk_capability_show(struct device *dev,
955 struct device_attribute *attr, char *buf)
956 {
957 struct gendisk *disk = dev_to_disk(dev);
958
959 return sprintf(buf, "%x\n", disk->flags);
960 }
961
962 static ssize_t disk_alignment_offset_show(struct device *dev,
963 struct device_attribute *attr,
964 char *buf)
965 {
966 struct gendisk *disk = dev_to_disk(dev);
967
968 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
969 }
970
971 static ssize_t disk_discard_alignment_show(struct device *dev,
972 struct device_attribute *attr,
973 char *buf)
974 {
975 struct gendisk *disk = dev_to_disk(dev);
976
977 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
978 }
979
980 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
981 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
982 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
983 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
984 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
985 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
986 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
987 NULL);
988 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
989 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
990 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
991 #ifdef CONFIG_FAIL_MAKE_REQUEST
992 static struct device_attribute dev_attr_fail =
993 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
994 #endif
995 #ifdef CONFIG_FAIL_IO_TIMEOUT
996 static struct device_attribute dev_attr_fail_timeout =
997 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
998 part_timeout_store);
999 #endif
1000
1001 static struct attribute *disk_attrs[] = {
1002 &dev_attr_range.attr,
1003 &dev_attr_ext_range.attr,
1004 &dev_attr_removable.attr,
1005 &dev_attr_ro.attr,
1006 &dev_attr_size.attr,
1007 &dev_attr_alignment_offset.attr,
1008 &dev_attr_discard_alignment.attr,
1009 &dev_attr_capability.attr,
1010 &dev_attr_stat.attr,
1011 &dev_attr_inflight.attr,
1012 #ifdef CONFIG_FAIL_MAKE_REQUEST
1013 &dev_attr_fail.attr,
1014 #endif
1015 #ifdef CONFIG_FAIL_IO_TIMEOUT
1016 &dev_attr_fail_timeout.attr,
1017 #endif
1018 NULL
1019 };
1020
1021 static struct attribute_group disk_attr_group = {
1022 .attrs = disk_attrs,
1023 };
1024
1025 static const struct attribute_group *disk_attr_groups[] = {
1026 &disk_attr_group,
1027 NULL
1028 };
1029
1030 /**
1031 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1032 * @disk: disk to replace part_tbl for
1033 * @new_ptbl: new part_tbl to install
1034 *
1035 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1036 * original ptbl is freed using RCU callback.
1037 *
1038 * LOCKING:
1039 * Matching bd_mutx locked.
1040 */
1041 static void disk_replace_part_tbl(struct gendisk *disk,
1042 struct disk_part_tbl *new_ptbl)
1043 {
1044 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1045
1046 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1047
1048 if (old_ptbl) {
1049 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1050 kfree_rcu(old_ptbl, rcu_head);
1051 }
1052 }
1053
1054 /**
1055 * disk_expand_part_tbl - expand disk->part_tbl
1056 * @disk: disk to expand part_tbl for
1057 * @partno: expand such that this partno can fit in
1058 *
1059 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1060 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1061 *
1062 * LOCKING:
1063 * Matching bd_mutex locked, might sleep.
1064 *
1065 * RETURNS:
1066 * 0 on success, -errno on failure.
1067 */
1068 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1069 {
1070 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1071 struct disk_part_tbl *new_ptbl;
1072 int len = old_ptbl ? old_ptbl->len : 0;
1073 int target = partno + 1;
1074 size_t size;
1075 int i;
1076
1077 /* disk_max_parts() is zero during initialization, ignore if so */
1078 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1079 return -EINVAL;
1080
1081 if (target <= len)
1082 return 0;
1083
1084 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1085 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1086 if (!new_ptbl)
1087 return -ENOMEM;
1088
1089 new_ptbl->len = target;
1090
1091 for (i = 0; i < len; i++)
1092 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1093
1094 disk_replace_part_tbl(disk, new_ptbl);
1095 return 0;
1096 }
1097
1098 static void disk_release(struct device *dev)
1099 {
1100 struct gendisk *disk = dev_to_disk(dev);
1101
1102 blk_free_devt(dev->devt);
1103 disk_release_events(disk);
1104 kfree(disk->random);
1105 disk_replace_part_tbl(disk, NULL);
1106 free_part_stats(&disk->part0);
1107 free_part_info(&disk->part0);
1108 if (disk->queue)
1109 blk_put_queue(disk->queue);
1110 kfree(disk);
1111 }
1112 struct class block_class = {
1113 .name = "block",
1114 };
1115
1116 static char *block_devnode(struct device *dev, umode_t *mode,
1117 kuid_t *uid, kgid_t *gid)
1118 {
1119 struct gendisk *disk = dev_to_disk(dev);
1120
1121 if (disk->devnode)
1122 return disk->devnode(disk, mode);
1123 return NULL;
1124 }
1125
1126 static struct device_type disk_type = {
1127 .name = "disk",
1128 .groups = disk_attr_groups,
1129 .release = disk_release,
1130 .devnode = block_devnode,
1131 };
1132
1133 #ifdef CONFIG_PROC_FS
1134 /*
1135 * aggregate disk stat collector. Uses the same stats that the sysfs
1136 * entries do, above, but makes them available through one seq_file.
1137 *
1138 * The output looks suspiciously like /proc/partitions with a bunch of
1139 * extra fields.
1140 */
1141 static int diskstats_show(struct seq_file *seqf, void *v)
1142 {
1143 struct gendisk *gp = v;
1144 struct disk_part_iter piter;
1145 struct hd_struct *hd;
1146 char buf[BDEVNAME_SIZE];
1147 int cpu;
1148
1149 /*
1150 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1151 seq_puts(seqf, "major minor name"
1152 " rio rmerge rsect ruse wio wmerge "
1153 "wsect wuse running use aveq"
1154 "\n\n");
1155 */
1156
1157 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1158 while ((hd = disk_part_iter_next(&piter))) {
1159 cpu = part_stat_lock();
1160 part_round_stats(cpu, hd);
1161 part_stat_unlock();
1162 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1163 "%u %lu %lu %lu %u %u %u %u\n",
1164 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1165 disk_name(gp, hd->partno, buf),
1166 part_stat_read(hd, ios[READ]),
1167 part_stat_read(hd, merges[READ]),
1168 part_stat_read(hd, sectors[READ]),
1169 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1170 part_stat_read(hd, ios[WRITE]),
1171 part_stat_read(hd, merges[WRITE]),
1172 part_stat_read(hd, sectors[WRITE]),
1173 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1174 part_in_flight(hd),
1175 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1176 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1177 );
1178 }
1179 disk_part_iter_exit(&piter);
1180
1181 return 0;
1182 }
1183
1184 static const struct seq_operations diskstats_op = {
1185 .start = disk_seqf_start,
1186 .next = disk_seqf_next,
1187 .stop = disk_seqf_stop,
1188 .show = diskstats_show
1189 };
1190
1191 static int diskstats_open(struct inode *inode, struct file *file)
1192 {
1193 return seq_open(file, &diskstats_op);
1194 }
1195
1196 static const struct file_operations proc_diskstats_operations = {
1197 .open = diskstats_open,
1198 .read = seq_read,
1199 .llseek = seq_lseek,
1200 .release = seq_release,
1201 };
1202
1203 static int __init proc_genhd_init(void)
1204 {
1205 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1206 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1207 return 0;
1208 }
1209 module_init(proc_genhd_init);
1210 #endif /* CONFIG_PROC_FS */
1211
1212 dev_t blk_lookup_devt(const char *name, int partno)
1213 {
1214 dev_t devt = MKDEV(0, 0);
1215 struct class_dev_iter iter;
1216 struct device *dev;
1217
1218 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1219 while ((dev = class_dev_iter_next(&iter))) {
1220 struct gendisk *disk = dev_to_disk(dev);
1221 struct hd_struct *part;
1222
1223 if (strcmp(dev_name(dev), name))
1224 continue;
1225
1226 if (partno < disk->minors) {
1227 /* We need to return the right devno, even
1228 * if the partition doesn't exist yet.
1229 */
1230 devt = MKDEV(MAJOR(dev->devt),
1231 MINOR(dev->devt) + partno);
1232 break;
1233 }
1234 part = disk_get_part(disk, partno);
1235 if (part) {
1236 devt = part_devt(part);
1237 disk_put_part(part);
1238 break;
1239 }
1240 disk_put_part(part);
1241 }
1242 class_dev_iter_exit(&iter);
1243 return devt;
1244 }
1245 EXPORT_SYMBOL(blk_lookup_devt);
1246
1247 struct gendisk *alloc_disk(int minors)
1248 {
1249 return alloc_disk_node(minors, NUMA_NO_NODE);
1250 }
1251 EXPORT_SYMBOL(alloc_disk);
1252
1253 struct gendisk *alloc_disk_node(int minors, int node_id)
1254 {
1255 struct gendisk *disk;
1256
1257 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1258 if (disk) {
1259 if (!init_part_stats(&disk->part0)) {
1260 kfree(disk);
1261 return NULL;
1262 }
1263 disk->node_id = node_id;
1264 if (disk_expand_part_tbl(disk, 0)) {
1265 free_part_stats(&disk->part0);
1266 kfree(disk);
1267 return NULL;
1268 }
1269 disk->part_tbl->part[0] = &disk->part0;
1270
1271 /*
1272 * set_capacity() and get_capacity() currently don't use
1273 * seqcounter to read/update the part0->nr_sects. Still init
1274 * the counter as we can read the sectors in IO submission
1275 * patch using seqence counters.
1276 *
1277 * TODO: Ideally set_capacity() and get_capacity() should be
1278 * converted to make use of bd_mutex and sequence counters.
1279 */
1280 seqcount_init(&disk->part0.nr_sects_seq);
1281 hd_ref_init(&disk->part0);
1282
1283 disk->minors = minors;
1284 rand_initialize_disk(disk);
1285 disk_to_dev(disk)->class = &block_class;
1286 disk_to_dev(disk)->type = &disk_type;
1287 device_initialize(disk_to_dev(disk));
1288 }
1289 return disk;
1290 }
1291 EXPORT_SYMBOL(alloc_disk_node);
1292
1293 struct kobject *get_disk(struct gendisk *disk)
1294 {
1295 struct module *owner;
1296 struct kobject *kobj;
1297
1298 if (!disk->fops)
1299 return NULL;
1300 owner = disk->fops->owner;
1301 if (owner && !try_module_get(owner))
1302 return NULL;
1303 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1304 if (kobj == NULL) {
1305 module_put(owner);
1306 return NULL;
1307 }
1308 return kobj;
1309
1310 }
1311
1312 EXPORT_SYMBOL(get_disk);
1313
1314 void put_disk(struct gendisk *disk)
1315 {
1316 if (disk)
1317 kobject_put(&disk_to_dev(disk)->kobj);
1318 }
1319
1320 EXPORT_SYMBOL(put_disk);
1321
1322 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1323 {
1324 char event[] = "DISK_RO=1";
1325 char *envp[] = { event, NULL };
1326
1327 if (!ro)
1328 event[8] = '0';
1329 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1330 }
1331
1332 void set_device_ro(struct block_device *bdev, int flag)
1333 {
1334 bdev->bd_part->policy = flag;
1335 }
1336
1337 EXPORT_SYMBOL(set_device_ro);
1338
1339 void set_disk_ro(struct gendisk *disk, int flag)
1340 {
1341 struct disk_part_iter piter;
1342 struct hd_struct *part;
1343
1344 if (disk->part0.policy != flag) {
1345 set_disk_ro_uevent(disk, flag);
1346 disk->part0.policy = flag;
1347 }
1348
1349 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1350 while ((part = disk_part_iter_next(&piter)))
1351 part->policy = flag;
1352 disk_part_iter_exit(&piter);
1353 }
1354
1355 EXPORT_SYMBOL(set_disk_ro);
1356
1357 int bdev_read_only(struct block_device *bdev)
1358 {
1359 if (!bdev)
1360 return 0;
1361 return bdev->bd_part->policy;
1362 }
1363
1364 EXPORT_SYMBOL(bdev_read_only);
1365
1366 int invalidate_partition(struct gendisk *disk, int partno)
1367 {
1368 int res = 0;
1369 struct block_device *bdev = bdget_disk(disk, partno);
1370 if (bdev) {
1371 fsync_bdev(bdev);
1372 res = __invalidate_device(bdev, true);
1373 bdput(bdev);
1374 }
1375 return res;
1376 }
1377
1378 EXPORT_SYMBOL(invalidate_partition);
1379
1380 /*
1381 * Disk events - monitor disk events like media change and eject request.
1382 */
1383 struct disk_events {
1384 struct list_head node; /* all disk_event's */
1385 struct gendisk *disk; /* the associated disk */
1386 spinlock_t lock;
1387
1388 struct mutex block_mutex; /* protects blocking */
1389 int block; /* event blocking depth */
1390 unsigned int pending; /* events already sent out */
1391 unsigned int clearing; /* events being cleared */
1392
1393 long poll_msecs; /* interval, -1 for default */
1394 struct delayed_work dwork;
1395 };
1396
1397 static const char *disk_events_strs[] = {
1398 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1399 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1400 };
1401
1402 static char *disk_uevents[] = {
1403 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1404 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1405 };
1406
1407 /* list of all disk_events */
1408 static DEFINE_MUTEX(disk_events_mutex);
1409 static LIST_HEAD(disk_events);
1410
1411 /* disable in-kernel polling by default */
1412 static unsigned long disk_events_dfl_poll_msecs = 0;
1413
1414 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1415 {
1416 struct disk_events *ev = disk->ev;
1417 long intv_msecs = 0;
1418
1419 /*
1420 * If device-specific poll interval is set, always use it. If
1421 * the default is being used, poll iff there are events which
1422 * can't be monitored asynchronously.
1423 */
1424 if (ev->poll_msecs >= 0)
1425 intv_msecs = ev->poll_msecs;
1426 else if (disk->events & ~disk->async_events)
1427 intv_msecs = disk_events_dfl_poll_msecs;
1428
1429 return msecs_to_jiffies(intv_msecs);
1430 }
1431
1432 /**
1433 * disk_block_events - block and flush disk event checking
1434 * @disk: disk to block events for
1435 *
1436 * On return from this function, it is guaranteed that event checking
1437 * isn't in progress and won't happen until unblocked by
1438 * disk_unblock_events(). Events blocking is counted and the actual
1439 * unblocking happens after the matching number of unblocks are done.
1440 *
1441 * Note that this intentionally does not block event checking from
1442 * disk_clear_events().
1443 *
1444 * CONTEXT:
1445 * Might sleep.
1446 */
1447 void disk_block_events(struct gendisk *disk)
1448 {
1449 struct disk_events *ev = disk->ev;
1450 unsigned long flags;
1451 bool cancel;
1452
1453 if (!ev)
1454 return;
1455
1456 /*
1457 * Outer mutex ensures that the first blocker completes canceling
1458 * the event work before further blockers are allowed to finish.
1459 */
1460 mutex_lock(&ev->block_mutex);
1461
1462 spin_lock_irqsave(&ev->lock, flags);
1463 cancel = !ev->block++;
1464 spin_unlock_irqrestore(&ev->lock, flags);
1465
1466 if (cancel)
1467 cancel_delayed_work_sync(&disk->ev->dwork);
1468
1469 mutex_unlock(&ev->block_mutex);
1470 }
1471
1472 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1473 {
1474 struct disk_events *ev = disk->ev;
1475 unsigned long intv;
1476 unsigned long flags;
1477
1478 spin_lock_irqsave(&ev->lock, flags);
1479
1480 if (WARN_ON_ONCE(ev->block <= 0))
1481 goto out_unlock;
1482
1483 if (--ev->block)
1484 goto out_unlock;
1485
1486 /*
1487 * Not exactly a latency critical operation, set poll timer
1488 * slack to 25% and kick event check.
1489 */
1490 intv = disk_events_poll_jiffies(disk);
1491 set_timer_slack(&ev->dwork.timer, intv / 4);
1492 if (check_now)
1493 queue_delayed_work(system_freezable_power_efficient_wq,
1494 &ev->dwork, 0);
1495 else if (intv)
1496 queue_delayed_work(system_freezable_power_efficient_wq,
1497 &ev->dwork, intv);
1498 out_unlock:
1499 spin_unlock_irqrestore(&ev->lock, flags);
1500 }
1501
1502 /**
1503 * disk_unblock_events - unblock disk event checking
1504 * @disk: disk to unblock events for
1505 *
1506 * Undo disk_block_events(). When the block count reaches zero, it
1507 * starts events polling if configured.
1508 *
1509 * CONTEXT:
1510 * Don't care. Safe to call from irq context.
1511 */
1512 void disk_unblock_events(struct gendisk *disk)
1513 {
1514 if (disk->ev)
1515 __disk_unblock_events(disk, false);
1516 }
1517
1518 /**
1519 * disk_flush_events - schedule immediate event checking and flushing
1520 * @disk: disk to check and flush events for
1521 * @mask: events to flush
1522 *
1523 * Schedule immediate event checking on @disk if not blocked. Events in
1524 * @mask are scheduled to be cleared from the driver. Note that this
1525 * doesn't clear the events from @disk->ev.
1526 *
1527 * CONTEXT:
1528 * If @mask is non-zero must be called with bdev->bd_mutex held.
1529 */
1530 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1531 {
1532 struct disk_events *ev = disk->ev;
1533
1534 if (!ev)
1535 return;
1536
1537 spin_lock_irq(&ev->lock);
1538 ev->clearing |= mask;
1539 if (!ev->block)
1540 mod_delayed_work(system_freezable_power_efficient_wq,
1541 &ev->dwork, 0);
1542 spin_unlock_irq(&ev->lock);
1543 }
1544
1545 /**
1546 * disk_clear_events - synchronously check, clear and return pending events
1547 * @disk: disk to fetch and clear events from
1548 * @mask: mask of events to be fetched and cleared
1549 *
1550 * Disk events are synchronously checked and pending events in @mask
1551 * are cleared and returned. This ignores the block count.
1552 *
1553 * CONTEXT:
1554 * Might sleep.
1555 */
1556 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1557 {
1558 const struct block_device_operations *bdops = disk->fops;
1559 struct disk_events *ev = disk->ev;
1560 unsigned int pending;
1561 unsigned int clearing = mask;
1562
1563 if (!ev) {
1564 /* for drivers still using the old ->media_changed method */
1565 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1566 bdops->media_changed && bdops->media_changed(disk))
1567 return DISK_EVENT_MEDIA_CHANGE;
1568 return 0;
1569 }
1570
1571 disk_block_events(disk);
1572
1573 /*
1574 * store the union of mask and ev->clearing on the stack so that the
1575 * race with disk_flush_events does not cause ambiguity (ev->clearing
1576 * can still be modified even if events are blocked).
1577 */
1578 spin_lock_irq(&ev->lock);
1579 clearing |= ev->clearing;
1580 ev->clearing = 0;
1581 spin_unlock_irq(&ev->lock);
1582
1583 disk_check_events(ev, &clearing);
1584 /*
1585 * if ev->clearing is not 0, the disk_flush_events got called in the
1586 * middle of this function, so we want to run the workfn without delay.
1587 */
1588 __disk_unblock_events(disk, ev->clearing ? true : false);
1589
1590 /* then, fetch and clear pending events */
1591 spin_lock_irq(&ev->lock);
1592 pending = ev->pending & mask;
1593 ev->pending &= ~mask;
1594 spin_unlock_irq(&ev->lock);
1595 WARN_ON_ONCE(clearing & mask);
1596
1597 return pending;
1598 }
1599
1600 /*
1601 * Separate this part out so that a different pointer for clearing_ptr can be
1602 * passed in for disk_clear_events.
1603 */
1604 static void disk_events_workfn(struct work_struct *work)
1605 {
1606 struct delayed_work *dwork = to_delayed_work(work);
1607 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1608
1609 disk_check_events(ev, &ev->clearing);
1610 }
1611
1612 static void disk_check_events(struct disk_events *ev,
1613 unsigned int *clearing_ptr)
1614 {
1615 struct gendisk *disk = ev->disk;
1616 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1617 unsigned int clearing = *clearing_ptr;
1618 unsigned int events;
1619 unsigned long intv;
1620 int nr_events = 0, i;
1621
1622 /* check events */
1623 events = disk->fops->check_events(disk, clearing);
1624
1625 /* accumulate pending events and schedule next poll if necessary */
1626 spin_lock_irq(&ev->lock);
1627
1628 events &= ~ev->pending;
1629 ev->pending |= events;
1630 *clearing_ptr &= ~clearing;
1631
1632 intv = disk_events_poll_jiffies(disk);
1633 if (!ev->block && intv)
1634 queue_delayed_work(system_freezable_power_efficient_wq,
1635 &ev->dwork, intv);
1636
1637 spin_unlock_irq(&ev->lock);
1638
1639 /*
1640 * Tell userland about new events. Only the events listed in
1641 * @disk->events are reported. Unlisted events are processed the
1642 * same internally but never get reported to userland.
1643 */
1644 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1645 if (events & disk->events & (1 << i))
1646 envp[nr_events++] = disk_uevents[i];
1647
1648 if (nr_events)
1649 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1650 }
1651
1652 /*
1653 * A disk events enabled device has the following sysfs nodes under
1654 * its /sys/block/X/ directory.
1655 *
1656 * events : list of all supported events
1657 * events_async : list of events which can be detected w/o polling
1658 * events_poll_msecs : polling interval, 0: disable, -1: system default
1659 */
1660 static ssize_t __disk_events_show(unsigned int events, char *buf)
1661 {
1662 const char *delim = "";
1663 ssize_t pos = 0;
1664 int i;
1665
1666 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1667 if (events & (1 << i)) {
1668 pos += sprintf(buf + pos, "%s%s",
1669 delim, disk_events_strs[i]);
1670 delim = " ";
1671 }
1672 if (pos)
1673 pos += sprintf(buf + pos, "\n");
1674 return pos;
1675 }
1676
1677 static ssize_t disk_events_show(struct device *dev,
1678 struct device_attribute *attr, char *buf)
1679 {
1680 struct gendisk *disk = dev_to_disk(dev);
1681
1682 return __disk_events_show(disk->events, buf);
1683 }
1684
1685 static ssize_t disk_events_async_show(struct device *dev,
1686 struct device_attribute *attr, char *buf)
1687 {
1688 struct gendisk *disk = dev_to_disk(dev);
1689
1690 return __disk_events_show(disk->async_events, buf);
1691 }
1692
1693 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1694 struct device_attribute *attr,
1695 char *buf)
1696 {
1697 struct gendisk *disk = dev_to_disk(dev);
1698
1699 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1700 }
1701
1702 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1703 struct device_attribute *attr,
1704 const char *buf, size_t count)
1705 {
1706 struct gendisk *disk = dev_to_disk(dev);
1707 long intv;
1708
1709 if (!count || !sscanf(buf, "%ld", &intv))
1710 return -EINVAL;
1711
1712 if (intv < 0 && intv != -1)
1713 return -EINVAL;
1714
1715 disk_block_events(disk);
1716 disk->ev->poll_msecs = intv;
1717 __disk_unblock_events(disk, true);
1718
1719 return count;
1720 }
1721
1722 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1723 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1724 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1725 disk_events_poll_msecs_show,
1726 disk_events_poll_msecs_store);
1727
1728 static const struct attribute *disk_events_attrs[] = {
1729 &dev_attr_events.attr,
1730 &dev_attr_events_async.attr,
1731 &dev_attr_events_poll_msecs.attr,
1732 NULL,
1733 };
1734
1735 /*
1736 * The default polling interval can be specified by the kernel
1737 * parameter block.events_dfl_poll_msecs which defaults to 0
1738 * (disable). This can also be modified runtime by writing to
1739 * /sys/module/block/events_dfl_poll_msecs.
1740 */
1741 static int disk_events_set_dfl_poll_msecs(const char *val,
1742 const struct kernel_param *kp)
1743 {
1744 struct disk_events *ev;
1745 int ret;
1746
1747 ret = param_set_ulong(val, kp);
1748 if (ret < 0)
1749 return ret;
1750
1751 mutex_lock(&disk_events_mutex);
1752
1753 list_for_each_entry(ev, &disk_events, node)
1754 disk_flush_events(ev->disk, 0);
1755
1756 mutex_unlock(&disk_events_mutex);
1757
1758 return 0;
1759 }
1760
1761 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1762 .set = disk_events_set_dfl_poll_msecs,
1763 .get = param_get_ulong,
1764 };
1765
1766 #undef MODULE_PARAM_PREFIX
1767 #define MODULE_PARAM_PREFIX "block."
1768
1769 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1770 &disk_events_dfl_poll_msecs, 0644);
1771
1772 /*
1773 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1774 */
1775 static void disk_alloc_events(struct gendisk *disk)
1776 {
1777 struct disk_events *ev;
1778
1779 if (!disk->fops->check_events)
1780 return;
1781
1782 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1783 if (!ev) {
1784 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1785 return;
1786 }
1787
1788 INIT_LIST_HEAD(&ev->node);
1789 ev->disk = disk;
1790 spin_lock_init(&ev->lock);
1791 mutex_init(&ev->block_mutex);
1792 ev->block = 1;
1793 ev->poll_msecs = -1;
1794 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1795
1796 disk->ev = ev;
1797 }
1798
1799 static void disk_add_events(struct gendisk *disk)
1800 {
1801 if (!disk->ev)
1802 return;
1803
1804 /* FIXME: error handling */
1805 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1806 pr_warn("%s: failed to create sysfs files for events\n",
1807 disk->disk_name);
1808
1809 mutex_lock(&disk_events_mutex);
1810 list_add_tail(&disk->ev->node, &disk_events);
1811 mutex_unlock(&disk_events_mutex);
1812
1813 /*
1814 * Block count is initialized to 1 and the following initial
1815 * unblock kicks it into action.
1816 */
1817 __disk_unblock_events(disk, true);
1818 }
1819
1820 static void disk_del_events(struct gendisk *disk)
1821 {
1822 if (!disk->ev)
1823 return;
1824
1825 disk_block_events(disk);
1826
1827 mutex_lock(&disk_events_mutex);
1828 list_del_init(&disk->ev->node);
1829 mutex_unlock(&disk_events_mutex);
1830
1831 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1832 }
1833
1834 static void disk_release_events(struct gendisk *disk)
1835 {
1836 /* the block count should be 1 from disk_del_events() */
1837 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1838 kfree(disk->ev);
1839 }