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_bh(&ext_devt_lock);
426 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
427 spin_unlock_bh(&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_bh(&ext_devt_lock);
453 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
454 spin_unlock_bh(&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 blk_unregister_queue(disk);
657 blk_unregister_region(disk_devt(disk), disk->minors);
658
659 part_stat_set_all(&disk->part0, 0);
660 disk->part0.stamp = 0;
661
662 kobject_put(disk->part0.holder_dir);
663 kobject_put(disk->slave_dir);
664 disk->driverfs_dev = NULL;
665 if (!sysfs_deprecated)
666 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
667 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
668 device_del(disk_to_dev(disk));
669 }
670 EXPORT_SYMBOL(del_gendisk);
671
672 /**
673 * get_gendisk - get partitioning information for a given device
674 * @devt: device to get partitioning information for
675 * @partno: returned partition index
676 *
677 * This function gets the structure containing partitioning
678 * information for the given device @devt.
679 */
680 struct gendisk *get_gendisk(dev_t devt, int *partno)
681 {
682 struct gendisk *disk = NULL;
683
684 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
685 struct kobject *kobj;
686
687 kobj = kobj_lookup(bdev_map, devt, partno);
688 if (kobj)
689 disk = dev_to_disk(kobj_to_dev(kobj));
690 } else {
691 struct hd_struct *part;
692
693 spin_lock_bh(&ext_devt_lock);
694 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
695 if (part && get_disk(part_to_disk(part))) {
696 *partno = part->partno;
697 disk = part_to_disk(part);
698 }
699 spin_unlock_bh(&ext_devt_lock);
700 }
701
702 return disk;
703 }
704 EXPORT_SYMBOL(get_gendisk);
705
706 /**
707 * bdget_disk - do bdget() by gendisk and partition number
708 * @disk: gendisk of interest
709 * @partno: partition number
710 *
711 * Find partition @partno from @disk, do bdget() on it.
712 *
713 * CONTEXT:
714 * Don't care.
715 *
716 * RETURNS:
717 * Resulting block_device on success, NULL on failure.
718 */
719 struct block_device *bdget_disk(struct gendisk *disk, int partno)
720 {
721 struct hd_struct *part;
722 struct block_device *bdev = NULL;
723
724 part = disk_get_part(disk, partno);
725 if (part)
726 bdev = bdget(part_devt(part));
727 disk_put_part(part);
728
729 return bdev;
730 }
731 EXPORT_SYMBOL(bdget_disk);
732
733 /*
734 * print a full list of all partitions - intended for places where the root
735 * filesystem can't be mounted and thus to give the victim some idea of what
736 * went wrong
737 */
738 void __init printk_all_partitions(void)
739 {
740 struct class_dev_iter iter;
741 struct device *dev;
742
743 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
744 while ((dev = class_dev_iter_next(&iter))) {
745 struct gendisk *disk = dev_to_disk(dev);
746 struct disk_part_iter piter;
747 struct hd_struct *part;
748 char name_buf[BDEVNAME_SIZE];
749 char devt_buf[BDEVT_SIZE];
750
751 /*
752 * Don't show empty devices or things that have been
753 * suppressed
754 */
755 if (get_capacity(disk) == 0 ||
756 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
757 continue;
758
759 /*
760 * Note, unlike /proc/partitions, I am showing the
761 * numbers in hex - the same format as the root=
762 * option takes.
763 */
764 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
765 while ((part = disk_part_iter_next(&piter))) {
766 bool is_part0 = part == &disk->part0;
767
768 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
769 bdevt_str(part_devt(part), devt_buf),
770 (unsigned long long)part_nr_sects_read(part) >> 1
771 , disk_name(disk, part->partno, name_buf),
772 part->info ? part->info->uuid : "");
773 if (is_part0) {
774 if (disk->driverfs_dev != NULL &&
775 disk->driverfs_dev->driver != NULL)
776 printk(" driver: %s\n",
777 disk->driverfs_dev->driver->name);
778 else
779 printk(" (driver?)\n");
780 } else
781 printk("\n");
782 }
783 disk_part_iter_exit(&piter);
784 }
785 class_dev_iter_exit(&iter);
786 }
787
788 #ifdef CONFIG_PROC_FS
789 /* iterator */
790 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
791 {
792 loff_t skip = *pos;
793 struct class_dev_iter *iter;
794 struct device *dev;
795
796 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
797 if (!iter)
798 return ERR_PTR(-ENOMEM);
799
800 seqf->private = iter;
801 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
802 do {
803 dev = class_dev_iter_next(iter);
804 if (!dev)
805 return NULL;
806 } while (skip--);
807
808 return dev_to_disk(dev);
809 }
810
811 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
812 {
813 struct device *dev;
814
815 (*pos)++;
816 dev = class_dev_iter_next(seqf->private);
817 if (dev)
818 return dev_to_disk(dev);
819
820 return NULL;
821 }
822
823 static void disk_seqf_stop(struct seq_file *seqf, void *v)
824 {
825 struct class_dev_iter *iter = seqf->private;
826
827 /* stop is called even after start failed :-( */
828 if (iter) {
829 class_dev_iter_exit(iter);
830 kfree(iter);
831 }
832 }
833
834 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
835 {
836 void *p;
837
838 p = disk_seqf_start(seqf, pos);
839 if (!IS_ERR_OR_NULL(p) && !*pos)
840 seq_puts(seqf, "major minor #blocks name\n\n");
841 return p;
842 }
843
844 static int show_partition(struct seq_file *seqf, void *v)
845 {
846 struct gendisk *sgp = v;
847 struct disk_part_iter piter;
848 struct hd_struct *part;
849 char buf[BDEVNAME_SIZE];
850
851 /* Don't show non-partitionable removeable devices or empty devices */
852 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
853 (sgp->flags & GENHD_FL_REMOVABLE)))
854 return 0;
855 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
856 return 0;
857
858 /* show the full disk and all non-0 size partitions of it */
859 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
860 while ((part = disk_part_iter_next(&piter)))
861 seq_printf(seqf, "%4d %7d %10llu %s\n",
862 MAJOR(part_devt(part)), MINOR(part_devt(part)),
863 (unsigned long long)part_nr_sects_read(part) >> 1,
864 disk_name(sgp, part->partno, buf));
865 disk_part_iter_exit(&piter);
866
867 return 0;
868 }
869
870 static const struct seq_operations partitions_op = {
871 .start = show_partition_start,
872 .next = disk_seqf_next,
873 .stop = disk_seqf_stop,
874 .show = show_partition
875 };
876
877 static int partitions_open(struct inode *inode, struct file *file)
878 {
879 return seq_open(file, &partitions_op);
880 }
881
882 static const struct file_operations proc_partitions_operations = {
883 .open = partitions_open,
884 .read = seq_read,
885 .llseek = seq_lseek,
886 .release = seq_release,
887 };
888 #endif
889
890
891 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
892 {
893 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
894 /* Make old-style 2.4 aliases work */
895 request_module("block-major-%d", MAJOR(devt));
896 return NULL;
897 }
898
899 static int __init genhd_device_init(void)
900 {
901 int error;
902
903 block_class.dev_kobj = sysfs_dev_block_kobj;
904 error = class_register(&block_class);
905 if (unlikely(error))
906 return error;
907 bdev_map = kobj_map_init(base_probe, &block_class_lock);
908 blk_dev_init();
909
910 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
911
912 /* create top-level block dir */
913 if (!sysfs_deprecated)
914 block_depr = kobject_create_and_add("block", NULL);
915 return 0;
916 }
917
918 subsys_initcall(genhd_device_init);
919
920 static ssize_t disk_range_show(struct device *dev,
921 struct device_attribute *attr, char *buf)
922 {
923 struct gendisk *disk = dev_to_disk(dev);
924
925 return sprintf(buf, "%d\n", disk->minors);
926 }
927
928 static ssize_t disk_ext_range_show(struct device *dev,
929 struct device_attribute *attr, char *buf)
930 {
931 struct gendisk *disk = dev_to_disk(dev);
932
933 return sprintf(buf, "%d\n", disk_max_parts(disk));
934 }
935
936 static ssize_t disk_removable_show(struct device *dev,
937 struct device_attribute *attr, char *buf)
938 {
939 struct gendisk *disk = dev_to_disk(dev);
940
941 return sprintf(buf, "%d\n",
942 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
943 }
944
945 static ssize_t disk_ro_show(struct device *dev,
946 struct device_attribute *attr, char *buf)
947 {
948 struct gendisk *disk = dev_to_disk(dev);
949
950 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
951 }
952
953 static ssize_t disk_capability_show(struct device *dev,
954 struct device_attribute *attr, char *buf)
955 {
956 struct gendisk *disk = dev_to_disk(dev);
957
958 return sprintf(buf, "%x\n", disk->flags);
959 }
960
961 static ssize_t disk_alignment_offset_show(struct device *dev,
962 struct device_attribute *attr,
963 char *buf)
964 {
965 struct gendisk *disk = dev_to_disk(dev);
966
967 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
968 }
969
970 static ssize_t disk_discard_alignment_show(struct device *dev,
971 struct device_attribute *attr,
972 char *buf)
973 {
974 struct gendisk *disk = dev_to_disk(dev);
975
976 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
977 }
978
979 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
980 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
981 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
982 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
983 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
984 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL);
985 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show,
986 NULL);
987 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
988 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
989 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL);
990 #ifdef CONFIG_FAIL_MAKE_REQUEST
991 static struct device_attribute dev_attr_fail =
992 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
993 #endif
994 #ifdef CONFIG_FAIL_IO_TIMEOUT
995 static struct device_attribute dev_attr_fail_timeout =
996 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
997 part_timeout_store);
998 #endif
999
1000 static struct attribute *disk_attrs[] = {
1001 &dev_attr_range.attr,
1002 &dev_attr_ext_range.attr,
1003 &dev_attr_removable.attr,
1004 &dev_attr_ro.attr,
1005 &dev_attr_size.attr,
1006 &dev_attr_alignment_offset.attr,
1007 &dev_attr_discard_alignment.attr,
1008 &dev_attr_capability.attr,
1009 &dev_attr_stat.attr,
1010 &dev_attr_inflight.attr,
1011 #ifdef CONFIG_FAIL_MAKE_REQUEST
1012 &dev_attr_fail.attr,
1013 #endif
1014 #ifdef CONFIG_FAIL_IO_TIMEOUT
1015 &dev_attr_fail_timeout.attr,
1016 #endif
1017 NULL
1018 };
1019
1020 static struct attribute_group disk_attr_group = {
1021 .attrs = disk_attrs,
1022 };
1023
1024 static const struct attribute_group *disk_attr_groups[] = {
1025 &disk_attr_group,
1026 NULL
1027 };
1028
1029 /**
1030 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1031 * @disk: disk to replace part_tbl for
1032 * @new_ptbl: new part_tbl to install
1033 *
1034 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1035 * original ptbl is freed using RCU callback.
1036 *
1037 * LOCKING:
1038 * Matching bd_mutx locked.
1039 */
1040 static void disk_replace_part_tbl(struct gendisk *disk,
1041 struct disk_part_tbl *new_ptbl)
1042 {
1043 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1044
1045 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1046
1047 if (old_ptbl) {
1048 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1049 kfree_rcu(old_ptbl, rcu_head);
1050 }
1051 }
1052
1053 /**
1054 * disk_expand_part_tbl - expand disk->part_tbl
1055 * @disk: disk to expand part_tbl for
1056 * @partno: expand such that this partno can fit in
1057 *
1058 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1059 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1060 *
1061 * LOCKING:
1062 * Matching bd_mutex locked, might sleep.
1063 *
1064 * RETURNS:
1065 * 0 on success, -errno on failure.
1066 */
1067 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1068 {
1069 struct disk_part_tbl *old_ptbl = disk->part_tbl;
1070 struct disk_part_tbl *new_ptbl;
1071 int len = old_ptbl ? old_ptbl->len : 0;
1072 int i, target;
1073 size_t size;
1074
1075 /*
1076 * check for int overflow, since we can get here from blkpg_ioctl()
1077 * with a user passed 'partno'.
1078 */
1079 target = partno + 1;
1080 if (target < 0)
1081 return -EINVAL;
1082
1083 /* disk_max_parts() is zero during initialization, ignore if so */
1084 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1085 return -EINVAL;
1086
1087 if (target <= len)
1088 return 0;
1089
1090 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
1091 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
1092 if (!new_ptbl)
1093 return -ENOMEM;
1094
1095 new_ptbl->len = target;
1096
1097 for (i = 0; i < len; i++)
1098 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1099
1100 disk_replace_part_tbl(disk, new_ptbl);
1101 return 0;
1102 }
1103
1104 static void disk_release(struct device *dev)
1105 {
1106 struct gendisk *disk = dev_to_disk(dev);
1107
1108 blk_free_devt(dev->devt);
1109 disk_release_events(disk);
1110 kfree(disk->random);
1111 disk_replace_part_tbl(disk, NULL);
1112 free_part_stats(&disk->part0);
1113 free_part_info(&disk->part0);
1114 if (disk->queue)
1115 blk_put_queue(disk->queue);
1116 kfree(disk);
1117 }
1118 struct class block_class = {
1119 .name = "block",
1120 };
1121
1122 static char *block_devnode(struct device *dev, umode_t *mode,
1123 kuid_t *uid, kgid_t *gid)
1124 {
1125 struct gendisk *disk = dev_to_disk(dev);
1126
1127 if (disk->devnode)
1128 return disk->devnode(disk, mode);
1129 return NULL;
1130 }
1131
1132 static struct device_type disk_type = {
1133 .name = "disk",
1134 .groups = disk_attr_groups,
1135 .release = disk_release,
1136 .devnode = block_devnode,
1137 };
1138
1139 #ifdef CONFIG_PROC_FS
1140 /*
1141 * aggregate disk stat collector. Uses the same stats that the sysfs
1142 * entries do, above, but makes them available through one seq_file.
1143 *
1144 * The output looks suspiciously like /proc/partitions with a bunch of
1145 * extra fields.
1146 */
1147 static int diskstats_show(struct seq_file *seqf, void *v)
1148 {
1149 struct gendisk *gp = v;
1150 struct disk_part_iter piter;
1151 struct hd_struct *hd;
1152 char buf[BDEVNAME_SIZE];
1153 int cpu;
1154
1155 /*
1156 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1157 seq_puts(seqf, "major minor name"
1158 " rio rmerge rsect ruse wio wmerge "
1159 "wsect wuse running use aveq"
1160 "\n\n");
1161 */
1162
1163 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1164 while ((hd = disk_part_iter_next(&piter))) {
1165 cpu = part_stat_lock();
1166 part_round_stats(cpu, hd);
1167 part_stat_unlock();
1168 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1169 "%u %lu %lu %lu %u %u %u %u\n",
1170 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1171 disk_name(gp, hd->partno, buf),
1172 part_stat_read(hd, ios[READ]),
1173 part_stat_read(hd, merges[READ]),
1174 part_stat_read(hd, sectors[READ]),
1175 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1176 part_stat_read(hd, ios[WRITE]),
1177 part_stat_read(hd, merges[WRITE]),
1178 part_stat_read(hd, sectors[WRITE]),
1179 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1180 part_in_flight(hd),
1181 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1182 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1183 );
1184 }
1185 disk_part_iter_exit(&piter);
1186
1187 return 0;
1188 }
1189
1190 static const struct seq_operations diskstats_op = {
1191 .start = disk_seqf_start,
1192 .next = disk_seqf_next,
1193 .stop = disk_seqf_stop,
1194 .show = diskstats_show
1195 };
1196
1197 static int diskstats_open(struct inode *inode, struct file *file)
1198 {
1199 return seq_open(file, &diskstats_op);
1200 }
1201
1202 static const struct file_operations proc_diskstats_operations = {
1203 .open = diskstats_open,
1204 .read = seq_read,
1205 .llseek = seq_lseek,
1206 .release = seq_release,
1207 };
1208
1209 static int __init proc_genhd_init(void)
1210 {
1211 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1212 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1213 return 0;
1214 }
1215 module_init(proc_genhd_init);
1216 #endif /* CONFIG_PROC_FS */
1217
1218 dev_t blk_lookup_devt(const char *name, int partno)
1219 {
1220 dev_t devt = MKDEV(0, 0);
1221 struct class_dev_iter iter;
1222 struct device *dev;
1223
1224 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1225 while ((dev = class_dev_iter_next(&iter))) {
1226 struct gendisk *disk = dev_to_disk(dev);
1227 struct hd_struct *part;
1228
1229 if (strcmp(dev_name(dev), name))
1230 continue;
1231
1232 if (partno < disk->minors) {
1233 /* We need to return the right devno, even
1234 * if the partition doesn't exist yet.
1235 */
1236 devt = MKDEV(MAJOR(dev->devt),
1237 MINOR(dev->devt) + partno);
1238 break;
1239 }
1240 part = disk_get_part(disk, partno);
1241 if (part) {
1242 devt = part_devt(part);
1243 disk_put_part(part);
1244 break;
1245 }
1246 disk_put_part(part);
1247 }
1248 class_dev_iter_exit(&iter);
1249 return devt;
1250 }
1251 EXPORT_SYMBOL(blk_lookup_devt);
1252
1253 struct gendisk *alloc_disk(int minors)
1254 {
1255 return alloc_disk_node(minors, NUMA_NO_NODE);
1256 }
1257 EXPORT_SYMBOL(alloc_disk);
1258
1259 struct gendisk *alloc_disk_node(int minors, int node_id)
1260 {
1261 struct gendisk *disk;
1262
1263 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1264 if (disk) {
1265 if (!init_part_stats(&disk->part0)) {
1266 kfree(disk);
1267 return NULL;
1268 }
1269 disk->node_id = node_id;
1270 if (disk_expand_part_tbl(disk, 0)) {
1271 free_part_stats(&disk->part0);
1272 kfree(disk);
1273 return NULL;
1274 }
1275 disk->part_tbl->part[0] = &disk->part0;
1276
1277 /*
1278 * set_capacity() and get_capacity() currently don't use
1279 * seqcounter to read/update the part0->nr_sects. Still init
1280 * the counter as we can read the sectors in IO submission
1281 * patch using seqence counters.
1282 *
1283 * TODO: Ideally set_capacity() and get_capacity() should be
1284 * converted to make use of bd_mutex and sequence counters.
1285 */
1286 seqcount_init(&disk->part0.nr_sects_seq);
1287 hd_ref_init(&disk->part0);
1288
1289 disk->minors = minors;
1290 rand_initialize_disk(disk);
1291 disk_to_dev(disk)->class = &block_class;
1292 disk_to_dev(disk)->type = &disk_type;
1293 device_initialize(disk_to_dev(disk));
1294 }
1295 return disk;
1296 }
1297 EXPORT_SYMBOL(alloc_disk_node);
1298
1299 struct kobject *get_disk(struct gendisk *disk)
1300 {
1301 struct module *owner;
1302 struct kobject *kobj;
1303
1304 if (!disk->fops)
1305 return NULL;
1306 owner = disk->fops->owner;
1307 if (owner && !try_module_get(owner))
1308 return NULL;
1309 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1310 if (kobj == NULL) {
1311 module_put(owner);
1312 return NULL;
1313 }
1314 return kobj;
1315
1316 }
1317
1318 EXPORT_SYMBOL(get_disk);
1319
1320 void put_disk(struct gendisk *disk)
1321 {
1322 if (disk)
1323 kobject_put(&disk_to_dev(disk)->kobj);
1324 }
1325
1326 EXPORT_SYMBOL(put_disk);
1327
1328 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1329 {
1330 char event[] = "DISK_RO=1";
1331 char *envp[] = { event, NULL };
1332
1333 if (!ro)
1334 event[8] = '';
1335 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1336 }
1337
1338 void set_device_ro(struct block_device *bdev, int flag)
1339 {
1340 bdev->bd_part->policy = flag;
1341 }
1342
1343 EXPORT_SYMBOL(set_device_ro);
1344
1345 void set_disk_ro(struct gendisk *disk, int flag)
1346 {
1347 struct disk_part_iter piter;
1348 struct hd_struct *part;
1349
1350 if (disk->part0.policy != flag) {
1351 set_disk_ro_uevent(disk, flag);
1352 disk->part0.policy = flag;
1353 }
1354
1355 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1356 while ((part = disk_part_iter_next(&piter)))
1357 part->policy = flag;
1358 disk_part_iter_exit(&piter);
1359 }
1360
1361 EXPORT_SYMBOL(set_disk_ro);
1362
1363 int bdev_read_only(struct block_device *bdev)
1364 {
1365 if (!bdev)
1366 return 0;
1367 return bdev->bd_part->policy;
1368 }
1369
1370 EXPORT_SYMBOL(bdev_read_only);
1371
1372 int invalidate_partition(struct gendisk *disk, int partno)
1373 {
1374 int res = 0;
1375 struct block_device *bdev = bdget_disk(disk, partno);
1376 if (bdev) {
1377 fsync_bdev(bdev);
1378 res = __invalidate_device(bdev, true);
1379 bdput(bdev);
1380 }
1381 return res;
1382 }
1383
1384 EXPORT_SYMBOL(invalidate_partition);
1385
1386 /*
1387 * Disk events - monitor disk events like media change and eject request.
1388 */
1389 struct disk_events {
1390 struct list_head node; /* all disk_event's */
1391 struct gendisk *disk; /* the associated disk */
1392 spinlock_t lock;
1393
1394 struct mutex block_mutex; /* protects blocking */
1395 int block; /* event blocking depth */
1396 unsigned int pending; /* events already sent out */
1397 unsigned int clearing; /* events being cleared */
1398
1399 long poll_msecs; /* interval, -1 for default */
1400 struct delayed_work dwork;
1401 };
1402
1403 static const char *disk_events_strs[] = {
1404 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1405 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1406 };
1407
1408 static char *disk_uevents[] = {
1409 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1410 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1411 };
1412
1413 /* list of all disk_events */
1414 static DEFINE_MUTEX(disk_events_mutex);
1415 static LIST_HEAD(disk_events);
1416
1417 /* disable in-kernel polling by default */
1418 static unsigned long disk_events_dfl_poll_msecs = 0;
1419
1420 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1421 {
1422 struct disk_events *ev = disk->ev;
1423 long intv_msecs = 0;
1424
1425 /*
1426 * If device-specific poll interval is set, always use it. If
1427 * the default is being used, poll iff there are events which
1428 * can't be monitored asynchronously.
1429 */
1430 if (ev->poll_msecs >= 0)
1431 intv_msecs = ev->poll_msecs;
1432 else if (disk->events & ~disk->async_events)
1433 intv_msecs = disk_events_dfl_poll_msecs;
1434
1435 return msecs_to_jiffies(intv_msecs);
1436 }
1437
1438 /**
1439 * disk_block_events - block and flush disk event checking
1440 * @disk: disk to block events for
1441 *
1442 * On return from this function, it is guaranteed that event checking
1443 * isn't in progress and won't happen until unblocked by
1444 * disk_unblock_events(). Events blocking is counted and the actual
1445 * unblocking happens after the matching number of unblocks are done.
1446 *
1447 * Note that this intentionally does not block event checking from
1448 * disk_clear_events().
1449 *
1450 * CONTEXT:
1451 * Might sleep.
1452 */
1453 void disk_block_events(struct gendisk *disk)
1454 {
1455 struct disk_events *ev = disk->ev;
1456 unsigned long flags;
1457 bool cancel;
1458
1459 if (!ev)
1460 return;
1461
1462 /*
1463 * Outer mutex ensures that the first blocker completes canceling
1464 * the event work before further blockers are allowed to finish.
1465 */
1466 mutex_lock(&ev->block_mutex);
1467
1468 spin_lock_irqsave(&ev->lock, flags);
1469 cancel = !ev->block++;
1470 spin_unlock_irqrestore(&ev->lock, flags);
1471
1472 if (cancel)
1473 cancel_delayed_work_sync(&disk->ev->dwork);
1474
1475 mutex_unlock(&ev->block_mutex);
1476 }
1477
1478 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1479 {
1480 struct disk_events *ev = disk->ev;
1481 unsigned long intv;
1482 unsigned long flags;
1483
1484 spin_lock_irqsave(&ev->lock, flags);
1485
1486 if (WARN_ON_ONCE(ev->block <= 0))
1487 goto out_unlock;
1488
1489 if (--ev->block)
1490 goto out_unlock;
1491
1492 /*
1493 * Not exactly a latency critical operation, set poll timer
1494 * slack to 25% and kick event check.
1495 */
1496 intv = disk_events_poll_jiffies(disk);
1497 set_timer_slack(&ev->dwork.timer, intv / 4);
1498 if (check_now)
1499 queue_delayed_work(system_freezable_power_efficient_wq,
1500 &ev->dwork, 0);
1501 else if (intv)
1502 queue_delayed_work(system_freezable_power_efficient_wq,
1503 &ev->dwork, intv);
1504 out_unlock:
1505 spin_unlock_irqrestore(&ev->lock, flags);
1506 }
1507
1508 /**
1509 * disk_unblock_events - unblock disk event checking
1510 * @disk: disk to unblock events for
1511 *
1512 * Undo disk_block_events(). When the block count reaches zero, it
1513 * starts events polling if configured.
1514 *
1515 * CONTEXT:
1516 * Don't care. Safe to call from irq context.
1517 */
1518 void disk_unblock_events(struct gendisk *disk)
1519 {
1520 if (disk->ev)
1521 __disk_unblock_events(disk, false);
1522 }
1523
1524 /**
1525 * disk_flush_events - schedule immediate event checking and flushing
1526 * @disk: disk to check and flush events for
1527 * @mask: events to flush
1528 *
1529 * Schedule immediate event checking on @disk if not blocked. Events in
1530 * @mask are scheduled to be cleared from the driver. Note that this
1531 * doesn't clear the events from @disk->ev.
1532 *
1533 * CONTEXT:
1534 * If @mask is non-zero must be called with bdev->bd_mutex held.
1535 */
1536 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1537 {
1538 struct disk_events *ev = disk->ev;
1539
1540 if (!ev)
1541 return;
1542
1543 spin_lock_irq(&ev->lock);
1544 ev->clearing |= mask;
1545 if (!ev->block)
1546 mod_delayed_work(system_freezable_power_efficient_wq,
1547 &ev->dwork, 0);
1548 spin_unlock_irq(&ev->lock);
1549 }
1550
1551 /**
1552 * disk_clear_events - synchronously check, clear and return pending events
1553 * @disk: disk to fetch and clear events from
1554 * @mask: mask of events to be fetched and cleared
1555 *
1556 * Disk events are synchronously checked and pending events in @mask
1557 * are cleared and returned. This ignores the block count.
1558 *
1559 * CONTEXT:
1560 * Might sleep.
1561 */
1562 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1563 {
1564 const struct block_device_operations *bdops = disk->fops;
1565 struct disk_events *ev = disk->ev;
1566 unsigned int pending;
1567 unsigned int clearing = mask;
1568
1569 if (!ev) {
1570 /* for drivers still using the old ->media_changed method */
1571 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1572 bdops->media_changed && bdops->media_changed(disk))
1573 return DISK_EVENT_MEDIA_CHANGE;
1574 return 0;
1575 }
1576
1577 disk_block_events(disk);
1578
1579 /*
1580 * store the union of mask and ev->clearing on the stack so that the
1581 * race with disk_flush_events does not cause ambiguity (ev->clearing
1582 * can still be modified even if events are blocked).
1583 */
1584 spin_lock_irq(&ev->lock);
1585 clearing |= ev->clearing;
1586 ev->clearing = 0;
1587 spin_unlock_irq(&ev->lock);
1588
1589 disk_check_events(ev, &clearing);
1590 /*
1591 * if ev->clearing is not 0, the disk_flush_events got called in the
1592 * middle of this function, so we want to run the workfn without delay.
1593 */
1594 __disk_unblock_events(disk, ev->clearing ? true : false);
1595
1596 /* then, fetch and clear pending events */
1597 spin_lock_irq(&ev->lock);
1598 pending = ev->pending & mask;
1599 ev->pending &= ~mask;
1600 spin_unlock_irq(&ev->lock);
1601 WARN_ON_ONCE(clearing & mask);
1602
1603 return pending;
1604 }
1605
1606 /*
1607 * Separate this part out so that a different pointer for clearing_ptr can be
1608 * passed in for disk_clear_events.
1609 */
1610 static void disk_events_workfn(struct work_struct *work)
1611 {
1612 struct delayed_work *dwork = to_delayed_work(work);
1613 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1614
1615 disk_check_events(ev, &ev->clearing);
1616 }
1617
1618 static void disk_check_events(struct disk_events *ev,
1619 unsigned int *clearing_ptr)
1620 {
1621 struct gendisk *disk = ev->disk;
1622 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1623 unsigned int clearing = *clearing_ptr;
1624 unsigned int events;
1625 unsigned long intv;
1626 int nr_events = 0, i;
1627
1628 /* check events */
1629 events = disk->fops->check_events(disk, clearing);
1630
1631 /* accumulate pending events and schedule next poll if necessary */
1632 spin_lock_irq(&ev->lock);
1633
1634 events &= ~ev->pending;
1635 ev->pending |= events;
1636 *clearing_ptr &= ~clearing;
1637
1638 intv = disk_events_poll_jiffies(disk);
1639 if (!ev->block && intv)
1640 queue_delayed_work(system_freezable_power_efficient_wq,
1641 &ev->dwork, intv);
1642
1643 spin_unlock_irq(&ev->lock);
1644
1645 /*
1646 * Tell userland about new events. Only the events listed in
1647 * @disk->events are reported. Unlisted events are processed the
1648 * same internally but never get reported to userland.
1649 */
1650 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1651 if (events & disk->events & (1 << i))
1652 envp[nr_events++] = disk_uevents[i];
1653
1654 if (nr_events)
1655 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1656 }
1657
1658 /*
1659 * A disk events enabled device has the following sysfs nodes under
1660 * its /sys/block/X/ directory.
1661 *
1662 * events : list of all supported events
1663 * events_async : list of events which can be detected w/o polling
1664 * events_poll_msecs : polling interval, 0: disable, -1: system default
1665 */
1666 static ssize_t __disk_events_show(unsigned int events, char *buf)
1667 {
1668 const char *delim = "";
1669 ssize_t pos = 0;
1670 int i;
1671
1672 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1673 if (events & (1 << i)) {
1674 pos += sprintf(buf + pos, "%s%s",
1675 delim, disk_events_strs[i]);
1676 delim = " ";
1677 }
1678 if (pos)
1679 pos += sprintf(buf + pos, "\n");
1680 return pos;
1681 }
1682
1683 static ssize_t disk_events_show(struct device *dev,
1684 struct device_attribute *attr, char *buf)
1685 {
1686 struct gendisk *disk = dev_to_disk(dev);
1687
1688 return __disk_events_show(disk->events, buf);
1689 }
1690
1691 static ssize_t disk_events_async_show(struct device *dev,
1692 struct device_attribute *attr, char *buf)
1693 {
1694 struct gendisk *disk = dev_to_disk(dev);
1695
1696 return __disk_events_show(disk->async_events, buf);
1697 }
1698
1699 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1700 struct device_attribute *attr,
1701 char *buf)
1702 {
1703 struct gendisk *disk = dev_to_disk(dev);
1704
1705 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1706 }
1707
1708 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1709 struct device_attribute *attr,
1710 const char *buf, size_t count)
1711 {
1712 struct gendisk *disk = dev_to_disk(dev);
1713 long intv;
1714
1715 if (!count || !sscanf(buf, "%ld", &intv))
1716 return -EINVAL;
1717
1718 if (intv < 0 && intv != -1)
1719 return -EINVAL;
1720
1721 disk_block_events(disk);
1722 disk->ev->poll_msecs = intv;
1723 __disk_unblock_events(disk, true);
1724
1725 return count;
1726 }
1727
1728 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1729 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1730 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1731 disk_events_poll_msecs_show,
1732 disk_events_poll_msecs_store);
1733
1734 static const struct attribute *disk_events_attrs[] = {
1735 &dev_attr_events.attr,
1736 &dev_attr_events_async.attr,
1737 &dev_attr_events_poll_msecs.attr,
1738 NULL,
1739 };
1740
1741 /*
1742 * The default polling interval can be specified by the kernel
1743 * parameter block.events_dfl_poll_msecs which defaults to 0
1744 * (disable). This can also be modified runtime by writing to
1745 * /sys/module/block/events_dfl_poll_msecs.
1746 */
1747 static int disk_events_set_dfl_poll_msecs(const char *val,
1748 const struct kernel_param *kp)
1749 {
1750 struct disk_events *ev;
1751 int ret;
1752
1753 ret = param_set_ulong(val, kp);
1754 if (ret < 0)
1755 return ret;
1756
1757 mutex_lock(&disk_events_mutex);
1758
1759 list_for_each_entry(ev, &disk_events, node)
1760 disk_flush_events(ev->disk, 0);
1761
1762 mutex_unlock(&disk_events_mutex);
1763
1764 return 0;
1765 }
1766
1767 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1768 .set = disk_events_set_dfl_poll_msecs,
1769 .get = param_get_ulong,
1770 };
1771
1772 #undef MODULE_PARAM_PREFIX
1773 #define MODULE_PARAM_PREFIX "block."
1774
1775 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1776 &disk_events_dfl_poll_msecs, 0644);
1777
1778 /*
1779 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1780 */
1781 static void disk_alloc_events(struct gendisk *disk)
1782 {
1783 struct disk_events *ev;
1784
1785 if (!disk->fops->check_events)
1786 return;
1787
1788 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1789 if (!ev) {
1790 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1791 return;
1792 }
1793
1794 INIT_LIST_HEAD(&ev->node);
1795 ev->disk = disk;
1796 spin_lock_init(&ev->lock);
1797 mutex_init(&ev->block_mutex);
1798 ev->block = 1;
1799 ev->poll_msecs = -1;
1800 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1801
1802 disk->ev = ev;
1803 }
1804
1805 static void disk_add_events(struct gendisk *disk)
1806 {
1807 if (!disk->ev)
1808 return;
1809
1810 /* FIXME: error handling */
1811 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1812 pr_warn("%s: failed to create sysfs files for events\n",
1813 disk->disk_name);
1814
1815 mutex_lock(&disk_events_mutex);
1816 list_add_tail(&disk->ev->node, &disk_events);
1817 mutex_unlock(&disk_events_mutex);
1818
1819 /*
1820 * Block count is initialized to 1 and the following initial
1821 * unblock kicks it into action.
1822 */
1823 __disk_unblock_events(disk, true);
1824 }
1825
1826 static void disk_del_events(struct gendisk *disk)
1827 {
1828 if (!disk->ev)
1829 return;
1830
1831 disk_block_events(disk);
1832
1833 mutex_lock(&disk_events_mutex);
1834 list_del_init(&disk->ev->node);
1835 mutex_unlock(&disk_events_mutex);
1836
1837 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1838 }
1839
1840 static void disk_release_events(struct gendisk *disk)
1841 {
1842 /* the block count should be 1 from disk_del_events() */
1843 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1844 kfree(disk->ev);
1845 }
1846
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.