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_mutex prevents look up
32 * results from going away underneath its user.
33 */
34 static DEFINE_MUTEX(ext_devt_mutex);
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 */
disk_get_part(struct gendisk * disk,int partno)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 */
disk_part_iter_init(struct disk_part_iter * piter,struct gendisk * disk,unsigned int flags)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 */
disk_part_iter_next(struct disk_part_iter * piter)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 */
disk_part_iter_exit(struct disk_part_iter * piter)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
sector_in_part(struct hd_struct * part,sector_t sector)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 */
disk_map_sector_rcu(struct gendisk * disk,sector_t sector)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 */
major_to_index(unsigned major)250 static inline int major_to_index(unsigned major)
251 {
252 return major % BLKDEV_MAJOR_HASH_SIZE;
253 }
254
255 #ifdef CONFIG_PROC_FS
blkdev_show(struct seq_file * seqf,off_t offset)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 */
register_blkdev(unsigned int major,const char * name)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
unregister_blkdev(unsigned int major,const char * name)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 */
blk_mangle_minor(int minor)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 */
blk_alloc_devt(struct hd_struct * part,dev_t * devt)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 mutex_lock(&ext_devt_mutex);
424 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_KERNEL);
425 mutex_unlock(&ext_devt_mutex);
426 if (idx < 0)
427 return idx == -ENOSPC ? -EBUSY : idx;
428
429 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
430 return 0;
431 }
432
433 /**
434 * blk_free_devt - free a dev_t
435 * @devt: dev_t to free
436 *
437 * Free @devt which was allocated using blk_alloc_devt().
438 *
439 * CONTEXT:
440 * Might sleep.
441 */
blk_free_devt(dev_t devt)442 void blk_free_devt(dev_t devt)
443 {
444 might_sleep();
445
446 if (devt == MKDEV(0, 0))
447 return;
448
449 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
450 mutex_lock(&ext_devt_mutex);
451 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
452 mutex_unlock(&ext_devt_mutex);
453 }
454 }
455
bdevt_str(dev_t devt,char * buf)456 static char *bdevt_str(dev_t devt, char *buf)
457 {
458 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
459 char tbuf[BDEVT_SIZE];
460 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
461 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
462 } else
463 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
464
465 return buf;
466 }
467
468 /*
469 * Register device numbers dev..(dev+range-1)
470 * range must be nonzero
471 * The hash chain is sorted on range, so that subranges can override.
472 */
blk_register_region(dev_t devt,unsigned long range,struct module * module,struct kobject * (* probe)(dev_t,int *,void *),int (* lock)(dev_t,void *),void * data)473 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
474 struct kobject *(*probe)(dev_t, int *, void *),
475 int (*lock)(dev_t, void *), void *data)
476 {
477 kobj_map(bdev_map, devt, range, module, probe, lock, data);
478 }
479
480 EXPORT_SYMBOL(blk_register_region);
481
blk_unregister_region(dev_t devt,unsigned long range)482 void blk_unregister_region(dev_t devt, unsigned long range)
483 {
484 kobj_unmap(bdev_map, devt, range);
485 }
486
487 EXPORT_SYMBOL(blk_unregister_region);
488
exact_match(dev_t devt,int * partno,void * data)489 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
490 {
491 struct gendisk *p = data;
492
493 return &disk_to_dev(p)->kobj;
494 }
495
exact_lock(dev_t devt,void * data)496 static int exact_lock(dev_t devt, void *data)
497 {
498 struct gendisk *p = data;
499
500 if (!get_disk(p))
501 return -1;
502 return 0;
503 }
504
register_disk(struct gendisk * disk)505 static void register_disk(struct gendisk *disk)
506 {
507 struct device *ddev = disk_to_dev(disk);
508 struct block_device *bdev;
509 struct disk_part_iter piter;
510 struct hd_struct *part;
511 int err;
512
513 ddev->parent = disk->driverfs_dev;
514
515 dev_set_name(ddev, disk->disk_name);
516
517 /* delay uevents, until we scanned partition table */
518 dev_set_uevent_suppress(ddev, 1);
519
520 if (device_add(ddev))
521 return;
522 if (!sysfs_deprecated) {
523 err = sysfs_create_link(block_depr, &ddev->kobj,
524 kobject_name(&ddev->kobj));
525 if (err) {
526 device_del(ddev);
527 return;
528 }
529 }
530
531 /*
532 * avoid probable deadlock caused by allocating memory with
533 * GFP_KERNEL in runtime_resume callback of its all ancestor
534 * devices
535 */
536 pm_runtime_set_memalloc_noio(ddev, true);
537
538 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
539 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
540
541 /* No minors to use for partitions */
542 if (!disk_part_scan_enabled(disk))
543 goto exit;
544
545 /* No such device (e.g., media were just removed) */
546 if (!get_capacity(disk))
547 goto exit;
548
549 bdev = bdget_disk(disk, 0);
550 if (!bdev)
551 goto exit;
552
553 bdev->bd_invalidated = 1;
554 err = blkdev_get(bdev, FMODE_READ, NULL);
555 if (err < 0)
556 goto exit;
557 blkdev_put(bdev, FMODE_READ);
558
559 exit:
560 /* announce disk after possible partitions are created */
561 dev_set_uevent_suppress(ddev, 0);
562 kobject_uevent(&ddev->kobj, KOBJ_ADD);
563
564 /* announce possible partitions */
565 disk_part_iter_init(&piter, disk, 0);
566 while ((part = disk_part_iter_next(&piter)))
567 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
568 disk_part_iter_exit(&piter);
569 }
570
571 /**
572 * add_disk - add partitioning information to kernel list
573 * @disk: per-device partitioning information
574 *
575 * This function registers the partitioning information in @disk
576 * with the kernel.
577 *
578 * FIXME: error handling
579 */
add_disk(struct gendisk * disk)580 void add_disk(struct gendisk *disk)
581 {
582 struct backing_dev_info *bdi;
583 dev_t devt;
584 int retval;
585
586 /* minors == 0 indicates to use ext devt from part0 and should
587 * be accompanied with EXT_DEVT flag. Make sure all
588 * parameters make sense.
589 */
590 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
591 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
592
593 disk->flags |= GENHD_FL_UP;
594
595 retval = blk_alloc_devt(&disk->part0, &devt);
596 if (retval) {
597 WARN_ON(1);
598 return;
599 }
600 disk_to_dev(disk)->devt = devt;
601
602 /* ->major and ->first_minor aren't supposed to be
603 * dereferenced from here on, but set them just in case.
604 */
605 disk->major = MAJOR(devt);
606 disk->first_minor = MINOR(devt);
607
608 disk_alloc_events(disk);
609
610 /* Register BDI before referencing it from bdev */
611 bdi = &disk->queue->backing_dev_info;
612 bdi_register_dev(bdi, disk_devt(disk));
613
614 blk_register_region(disk_devt(disk), disk->minors, NULL,
615 exact_match, exact_lock, disk);
616 register_disk(disk);
617 blk_register_queue(disk);
618
619 /*
620 * Take an extra ref on queue which will be put on disk_release()
621 * so that it sticks around as long as @disk is there.
622 */
623 WARN_ON_ONCE(!blk_get_queue(disk->queue));
624
625 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
626 "bdi");
627 WARN_ON(retval);
628
629 disk_add_events(disk);
630 }
631 EXPORT_SYMBOL(add_disk);
632
del_gendisk(struct gendisk * disk)633 void del_gendisk(struct gendisk *disk)
634 {
635 struct disk_part_iter piter;
636 struct hd_struct *part;
637
638 disk_del_events(disk);
639
640 /* invalidate stuff */
641 disk_part_iter_init(&piter, disk,
642 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
643 while ((part = disk_part_iter_next(&piter))) {
644 invalidate_partition(disk, part->partno);
645 delete_partition(disk, part->partno);
646 }
647 disk_part_iter_exit(&piter);
648
649 invalidate_partition(disk, 0);
650 set_capacity(disk, 0);
651 disk->flags &= ~GENHD_FL_UP;
652
653 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
654 bdi_unregister(&disk->queue->backing_dev_info);
655 blk_unregister_queue(disk);
656 blk_unregister_region(disk_devt(disk), disk->minors);
657
658 part_stat_set_all(&disk->part0, 0);
659 disk->part0.stamp = 0;
660
661 kobject_put(disk->part0.holder_dir);
662 kobject_put(disk->slave_dir);
663 disk->driverfs_dev = NULL;
664 if (!sysfs_deprecated)
665 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
666 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
667 device_del(disk_to_dev(disk));
668 blk_free_devt(disk_to_dev(disk)->devt);
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 */
get_gendisk(dev_t devt,int * partno)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 mutex_lock(&ext_devt_mutex);
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 mutex_unlock(&ext_devt_mutex);
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 */
bdget_disk(struct gendisk * disk,int partno)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 */
printk_all_partitions(void)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 */
disk_seqf_start(struct seq_file * seqf,loff_t * pos)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
disk_seqf_next(struct seq_file * seqf,void * v,loff_t * pos)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
disk_seqf_stop(struct seq_file * seqf,void * v)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 seqf->private = NULL;
832 }
833 }
834
show_partition_start(struct seq_file * seqf,loff_t * pos)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
show_partition(struct seq_file * seqf,void * v)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
partitions_open(struct inode * inode,struct file * file)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
base_probe(dev_t devt,int * partno,void * data)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
genhd_device_init(void)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
disk_range_show(struct device * dev,struct device_attribute * attr,char * buf)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
disk_ext_range_show(struct device * dev,struct device_attribute * attr,char * buf)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
disk_removable_show(struct device * dev,struct device_attribute * attr,char * buf)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
disk_ro_show(struct device * dev,struct device_attribute * attr,char * buf)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
disk_capability_show(struct device * dev,struct device_attribute * attr,char * buf)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
disk_alignment_offset_show(struct device * dev,struct device_attribute * attr,char * buf)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
disk_discard_alignment_show(struct device * dev,struct device_attribute * attr,char * buf)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 */
disk_replace_part_tbl(struct gendisk * disk,struct disk_part_tbl * new_ptbl)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 */
disk_expand_part_tbl(struct gendisk * disk,int partno)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
disk_release(struct device * dev)1098 static void disk_release(struct device *dev)
1099 {
1100 struct gendisk *disk = dev_to_disk(dev);
1101
1102 disk_release_events(disk);
1103 kfree(disk->random);
1104 disk_replace_part_tbl(disk, NULL);
1105 free_part_stats(&disk->part0);
1106 free_part_info(&disk->part0);
1107 if (disk->queue)
1108 blk_put_queue(disk->queue);
1109 kfree(disk);
1110 }
1111
disk_uevent(struct device * dev,struct kobj_uevent_env * env)1112 static int disk_uevent(struct device *dev, struct kobj_uevent_env *env)
1113 {
1114 struct gendisk *disk = dev_to_disk(dev);
1115 struct disk_part_iter piter;
1116 struct hd_struct *part;
1117 int cnt = 0;
1118
1119 disk_part_iter_init(&piter, disk, 0);
1120 while((part = disk_part_iter_next(&piter)))
1121 cnt++;
1122 disk_part_iter_exit(&piter);
1123 add_uevent_var(env, "NPARTS=%u", cnt);
1124 return 0;
1125 }
1126
1127 struct class block_class = {
1128 .name = "block",
1129 };
1130
block_devnode(struct device * dev,umode_t * mode,kuid_t * uid,kgid_t * gid)1131 static char *block_devnode(struct device *dev, umode_t *mode,
1132 kuid_t *uid, kgid_t *gid)
1133 {
1134 struct gendisk *disk = dev_to_disk(dev);
1135
1136 if (disk->devnode)
1137 return disk->devnode(disk, mode);
1138 return NULL;
1139 }
1140
1141 static struct device_type disk_type = {
1142 .name = "disk",
1143 .groups = disk_attr_groups,
1144 .release = disk_release,
1145 .devnode = block_devnode,
1146 .uevent = disk_uevent,
1147 };
1148
1149 #ifdef CONFIG_PROC_FS
1150 /*
1151 * aggregate disk stat collector. Uses the same stats that the sysfs
1152 * entries do, above, but makes them available through one seq_file.
1153 *
1154 * The output looks suspiciously like /proc/partitions with a bunch of
1155 * extra fields.
1156 */
diskstats_show(struct seq_file * seqf,void * v)1157 static int diskstats_show(struct seq_file *seqf, void *v)
1158 {
1159 struct gendisk *gp = v;
1160 struct disk_part_iter piter;
1161 struct hd_struct *hd;
1162 char buf[BDEVNAME_SIZE];
1163 int cpu;
1164
1165 /*
1166 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1167 seq_puts(seqf, "major minor name"
1168 " rio rmerge rsect ruse wio wmerge "
1169 "wsect wuse running use aveq"
1170 "\n\n");
1171 */
1172
1173 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1174 while ((hd = disk_part_iter_next(&piter))) {
1175 cpu = part_stat_lock();
1176 part_round_stats(cpu, hd);
1177 part_stat_unlock();
1178 seq_printf(seqf, "%4d %7d %s %lu %lu %lu "
1179 "%u %lu %lu %lu %u %u %u %u\n",
1180 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1181 disk_name(gp, hd->partno, buf),
1182 part_stat_read(hd, ios[READ]),
1183 part_stat_read(hd, merges[READ]),
1184 part_stat_read(hd, sectors[READ]),
1185 jiffies_to_msecs(part_stat_read(hd, ticks[READ])),
1186 part_stat_read(hd, ios[WRITE]),
1187 part_stat_read(hd, merges[WRITE]),
1188 part_stat_read(hd, sectors[WRITE]),
1189 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])),
1190 part_in_flight(hd),
1191 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1192 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1193 );
1194 }
1195 disk_part_iter_exit(&piter);
1196
1197 return 0;
1198 }
1199
1200 static const struct seq_operations diskstats_op = {
1201 .start = disk_seqf_start,
1202 .next = disk_seqf_next,
1203 .stop = disk_seqf_stop,
1204 .show = diskstats_show
1205 };
1206
diskstats_open(struct inode * inode,struct file * file)1207 static int diskstats_open(struct inode *inode, struct file *file)
1208 {
1209 return seq_open(file, &diskstats_op);
1210 }
1211
1212 static const struct file_operations proc_diskstats_operations = {
1213 .open = diskstats_open,
1214 .read = seq_read,
1215 .llseek = seq_lseek,
1216 .release = seq_release,
1217 };
1218
proc_genhd_init(void)1219 static int __init proc_genhd_init(void)
1220 {
1221 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
1222 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1223 return 0;
1224 }
1225 module_init(proc_genhd_init);
1226 #endif /* CONFIG_PROC_FS */
1227
blk_lookup_devt(const char * name,int partno)1228 dev_t blk_lookup_devt(const char *name, int partno)
1229 {
1230 dev_t devt = MKDEV(0, 0);
1231 struct class_dev_iter iter;
1232 struct device *dev;
1233
1234 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1235 while ((dev = class_dev_iter_next(&iter))) {
1236 struct gendisk *disk = dev_to_disk(dev);
1237 struct hd_struct *part;
1238
1239 if (strcmp(dev_name(dev), name))
1240 continue;
1241
1242 if (partno < disk->minors) {
1243 /* We need to return the right devno, even
1244 * if the partition doesn't exist yet.
1245 */
1246 devt = MKDEV(MAJOR(dev->devt),
1247 MINOR(dev->devt) + partno);
1248 break;
1249 }
1250 part = disk_get_part(disk, partno);
1251 if (part) {
1252 devt = part_devt(part);
1253 disk_put_part(part);
1254 break;
1255 }
1256 disk_put_part(part);
1257 }
1258 class_dev_iter_exit(&iter);
1259 return devt;
1260 }
1261 EXPORT_SYMBOL(blk_lookup_devt);
1262
alloc_disk(int minors)1263 struct gendisk *alloc_disk(int minors)
1264 {
1265 return alloc_disk_node(minors, NUMA_NO_NODE);
1266 }
1267 EXPORT_SYMBOL(alloc_disk);
1268
alloc_disk_node(int minors,int node_id)1269 struct gendisk *alloc_disk_node(int minors, int node_id)
1270 {
1271 struct gendisk *disk;
1272
1273 disk = kmalloc_node(sizeof(struct gendisk),
1274 GFP_KERNEL | __GFP_ZERO, node_id);
1275 if (disk) {
1276 if (!init_part_stats(&disk->part0)) {
1277 kfree(disk);
1278 return NULL;
1279 }
1280 disk->node_id = node_id;
1281 if (disk_expand_part_tbl(disk, 0)) {
1282 free_part_stats(&disk->part0);
1283 kfree(disk);
1284 return NULL;
1285 }
1286 disk->part_tbl->part[0] = &disk->part0;
1287
1288 /*
1289 * set_capacity() and get_capacity() currently don't use
1290 * seqcounter to read/update the part0->nr_sects. Still init
1291 * the counter as we can read the sectors in IO submission
1292 * patch using seqence counters.
1293 *
1294 * TODO: Ideally set_capacity() and get_capacity() should be
1295 * converted to make use of bd_mutex and sequence counters.
1296 */
1297 seqcount_init(&disk->part0.nr_sects_seq);
1298 hd_ref_init(&disk->part0);
1299
1300 disk->minors = minors;
1301 rand_initialize_disk(disk);
1302 disk_to_dev(disk)->class = &block_class;
1303 disk_to_dev(disk)->type = &disk_type;
1304 device_initialize(disk_to_dev(disk));
1305 }
1306 return disk;
1307 }
1308 EXPORT_SYMBOL(alloc_disk_node);
1309
get_disk(struct gendisk * disk)1310 struct kobject *get_disk(struct gendisk *disk)
1311 {
1312 struct module *owner;
1313 struct kobject *kobj;
1314
1315 if (!disk->fops)
1316 return NULL;
1317 owner = disk->fops->owner;
1318 if (owner && !try_module_get(owner))
1319 return NULL;
1320 kobj = kobject_get(&disk_to_dev(disk)->kobj);
1321 if (kobj == NULL) {
1322 module_put(owner);
1323 return NULL;
1324 }
1325 return kobj;
1326
1327 }
1328
1329 EXPORT_SYMBOL(get_disk);
1330
put_disk(struct gendisk * disk)1331 void put_disk(struct gendisk *disk)
1332 {
1333 if (disk)
1334 kobject_put(&disk_to_dev(disk)->kobj);
1335 }
1336
1337 EXPORT_SYMBOL(put_disk);
1338
set_disk_ro_uevent(struct gendisk * gd,int ro)1339 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1340 {
1341 char event[] = "DISK_RO=1";
1342 char *envp[] = { event, NULL };
1343
1344 if (!ro)
1345 event[8] = '0';
1346 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1347 }
1348
set_device_ro(struct block_device * bdev,int flag)1349 void set_device_ro(struct block_device *bdev, int flag)
1350 {
1351 bdev->bd_part->policy = flag;
1352 }
1353
1354 EXPORT_SYMBOL(set_device_ro);
1355
set_disk_ro(struct gendisk * disk,int flag)1356 void set_disk_ro(struct gendisk *disk, int flag)
1357 {
1358 struct disk_part_iter piter;
1359 struct hd_struct *part;
1360
1361 if (disk->part0.policy != flag) {
1362 set_disk_ro_uevent(disk, flag);
1363 disk->part0.policy = flag;
1364 }
1365
1366 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1367 while ((part = disk_part_iter_next(&piter)))
1368 part->policy = flag;
1369 disk_part_iter_exit(&piter);
1370 }
1371
1372 EXPORT_SYMBOL(set_disk_ro);
1373
bdev_read_only(struct block_device * bdev)1374 int bdev_read_only(struct block_device *bdev)
1375 {
1376 if (!bdev)
1377 return 0;
1378 return bdev->bd_part->policy;
1379 }
1380
1381 EXPORT_SYMBOL(bdev_read_only);
1382
invalidate_partition(struct gendisk * disk,int partno)1383 int invalidate_partition(struct gendisk *disk, int partno)
1384 {
1385 int res = 0;
1386 struct block_device *bdev = bdget_disk(disk, partno);
1387 if (bdev) {
1388 fsync_bdev(bdev);
1389 res = __invalidate_device(bdev, true);
1390 bdput(bdev);
1391 }
1392 return res;
1393 }
1394
1395 EXPORT_SYMBOL(invalidate_partition);
1396
1397 /*
1398 * Disk events - monitor disk events like media change and eject request.
1399 */
1400 struct disk_events {
1401 struct list_head node; /* all disk_event's */
1402 struct gendisk *disk; /* the associated disk */
1403 spinlock_t lock;
1404
1405 struct mutex block_mutex; /* protects blocking */
1406 int block; /* event blocking depth */
1407 unsigned int pending; /* events already sent out */
1408 unsigned int clearing; /* events being cleared */
1409
1410 long poll_msecs; /* interval, -1 for default */
1411 struct delayed_work dwork;
1412 };
1413
1414 static const char *disk_events_strs[] = {
1415 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1416 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1417 };
1418
1419 static char *disk_uevents[] = {
1420 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1421 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1422 };
1423
1424 /* list of all disk_events */
1425 static DEFINE_MUTEX(disk_events_mutex);
1426 static LIST_HEAD(disk_events);
1427
1428 /* disable in-kernel polling by default */
1429 static unsigned long disk_events_dfl_poll_msecs = 0;
1430
disk_events_poll_jiffies(struct gendisk * disk)1431 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1432 {
1433 struct disk_events *ev = disk->ev;
1434 long intv_msecs = 0;
1435
1436 /*
1437 * If device-specific poll interval is set, always use it. If
1438 * the default is being used, poll iff there are events which
1439 * can't be monitored asynchronously.
1440 */
1441 if (ev->poll_msecs >= 0)
1442 intv_msecs = ev->poll_msecs;
1443 else if (disk->events & ~disk->async_events)
1444 intv_msecs = disk_events_dfl_poll_msecs;
1445
1446 return msecs_to_jiffies(intv_msecs);
1447 }
1448
1449 /**
1450 * disk_block_events - block and flush disk event checking
1451 * @disk: disk to block events for
1452 *
1453 * On return from this function, it is guaranteed that event checking
1454 * isn't in progress and won't happen until unblocked by
1455 * disk_unblock_events(). Events blocking is counted and the actual
1456 * unblocking happens after the matching number of unblocks are done.
1457 *
1458 * Note that this intentionally does not block event checking from
1459 * disk_clear_events().
1460 *
1461 * CONTEXT:
1462 * Might sleep.
1463 */
disk_block_events(struct gendisk * disk)1464 void disk_block_events(struct gendisk *disk)
1465 {
1466 struct disk_events *ev = disk->ev;
1467 unsigned long flags;
1468 bool cancel;
1469
1470 if (!ev)
1471 return;
1472
1473 /*
1474 * Outer mutex ensures that the first blocker completes canceling
1475 * the event work before further blockers are allowed to finish.
1476 */
1477 mutex_lock(&ev->block_mutex);
1478
1479 spin_lock_irqsave(&ev->lock, flags);
1480 cancel = !ev->block++;
1481 spin_unlock_irqrestore(&ev->lock, flags);
1482
1483 if (cancel)
1484 cancel_delayed_work_sync(&disk->ev->dwork);
1485
1486 mutex_unlock(&ev->block_mutex);
1487 }
1488
__disk_unblock_events(struct gendisk * disk,bool check_now)1489 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1490 {
1491 struct disk_events *ev = disk->ev;
1492 unsigned long intv;
1493 unsigned long flags;
1494
1495 spin_lock_irqsave(&ev->lock, flags);
1496
1497 if (WARN_ON_ONCE(ev->block <= 0))
1498 goto out_unlock;
1499
1500 if (--ev->block)
1501 goto out_unlock;
1502
1503 /*
1504 * Not exactly a latency critical operation, set poll timer
1505 * slack to 25% and kick event check.
1506 */
1507 intv = disk_events_poll_jiffies(disk);
1508 set_timer_slack(&ev->dwork.timer, intv / 4);
1509 if (check_now)
1510 queue_delayed_work(system_freezable_wq, &ev->dwork, 0);
1511 else if (intv)
1512 queue_delayed_work(system_freezable_wq, &ev->dwork, intv);
1513 out_unlock:
1514 spin_unlock_irqrestore(&ev->lock, flags);
1515 }
1516
1517 /**
1518 * disk_unblock_events - unblock disk event checking
1519 * @disk: disk to unblock events for
1520 *
1521 * Undo disk_block_events(). When the block count reaches zero, it
1522 * starts events polling if configured.
1523 *
1524 * CONTEXT:
1525 * Don't care. Safe to call from irq context.
1526 */
disk_unblock_events(struct gendisk * disk)1527 void disk_unblock_events(struct gendisk *disk)
1528 {
1529 if (disk->ev)
1530 __disk_unblock_events(disk, false);
1531 }
1532
1533 /**
1534 * disk_flush_events - schedule immediate event checking and flushing
1535 * @disk: disk to check and flush events for
1536 * @mask: events to flush
1537 *
1538 * Schedule immediate event checking on @disk if not blocked. Events in
1539 * @mask are scheduled to be cleared from the driver. Note that this
1540 * doesn't clear the events from @disk->ev.
1541 *
1542 * CONTEXT:
1543 * If @mask is non-zero must be called with bdev->bd_mutex held.
1544 */
disk_flush_events(struct gendisk * disk,unsigned int mask)1545 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1546 {
1547 struct disk_events *ev = disk->ev;
1548
1549 if (!ev)
1550 return;
1551
1552 spin_lock_irq(&ev->lock);
1553 ev->clearing |= mask;
1554 if (!ev->block)
1555 mod_delayed_work(system_freezable_wq, &ev->dwork, 0);
1556 spin_unlock_irq(&ev->lock);
1557 }
1558
1559 /**
1560 * disk_clear_events - synchronously check, clear and return pending events
1561 * @disk: disk to fetch and clear events from
1562 * @mask: mask of events to be fetched and clearted
1563 *
1564 * Disk events are synchronously checked and pending events in @mask
1565 * are cleared and returned. This ignores the block count.
1566 *
1567 * CONTEXT:
1568 * Might sleep.
1569 */
disk_clear_events(struct gendisk * disk,unsigned int mask)1570 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1571 {
1572 const struct block_device_operations *bdops = disk->fops;
1573 struct disk_events *ev = disk->ev;
1574 unsigned int pending;
1575 unsigned int clearing = mask;
1576
1577 if (!ev) {
1578 /* for drivers still using the old ->media_changed method */
1579 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1580 bdops->media_changed && bdops->media_changed(disk))
1581 return DISK_EVENT_MEDIA_CHANGE;
1582 return 0;
1583 }
1584
1585 disk_block_events(disk);
1586
1587 /*
1588 * store the union of mask and ev->clearing on the stack so that the
1589 * race with disk_flush_events does not cause ambiguity (ev->clearing
1590 * can still be modified even if events are blocked).
1591 */
1592 spin_lock_irq(&ev->lock);
1593 clearing |= ev->clearing;
1594 ev->clearing = 0;
1595 spin_unlock_irq(&ev->lock);
1596
1597 disk_check_events(ev, &clearing);
1598 /*
1599 * if ev->clearing is not 0, the disk_flush_events got called in the
1600 * middle of this function, so we want to run the workfn without delay.
1601 */
1602 __disk_unblock_events(disk, ev->clearing ? true : false);
1603
1604 /* then, fetch and clear pending events */
1605 spin_lock_irq(&ev->lock);
1606 pending = ev->pending & mask;
1607 ev->pending &= ~mask;
1608 spin_unlock_irq(&ev->lock);
1609 WARN_ON_ONCE(clearing & mask);
1610
1611 return pending;
1612 }
1613
1614 /*
1615 * Separate this part out so that a different pointer for clearing_ptr can be
1616 * passed in for disk_clear_events.
1617 */
disk_events_workfn(struct work_struct * work)1618 static void disk_events_workfn(struct work_struct *work)
1619 {
1620 struct delayed_work *dwork = to_delayed_work(work);
1621 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1622
1623 disk_check_events(ev, &ev->clearing);
1624 }
1625
disk_check_events(struct disk_events * ev,unsigned int * clearing_ptr)1626 static void disk_check_events(struct disk_events *ev,
1627 unsigned int *clearing_ptr)
1628 {
1629 struct gendisk *disk = ev->disk;
1630 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1631 unsigned int clearing = *clearing_ptr;
1632 unsigned int events;
1633 unsigned long intv;
1634 int nr_events = 0, i;
1635
1636 /* check events */
1637 events = disk->fops->check_events(disk, clearing);
1638
1639 /* accumulate pending events and schedule next poll if necessary */
1640 spin_lock_irq(&ev->lock);
1641
1642 events &= ~ev->pending;
1643 ev->pending |= events;
1644 *clearing_ptr &= ~clearing;
1645
1646 intv = disk_events_poll_jiffies(disk);
1647 if (!ev->block && intv)
1648 queue_delayed_work(system_freezable_wq, &ev->dwork, intv);
1649
1650 spin_unlock_irq(&ev->lock);
1651
1652 /*
1653 * Tell userland about new events. Only the events listed in
1654 * @disk->events are reported. Unlisted events are processed the
1655 * same internally but never get reported to userland.
1656 */
1657 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1658 if (events & disk->events & (1 << i))
1659 envp[nr_events++] = disk_uevents[i];
1660
1661 if (nr_events)
1662 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1663 }
1664
1665 /*
1666 * A disk events enabled device has the following sysfs nodes under
1667 * its /sys/block/X/ directory.
1668 *
1669 * events : list of all supported events
1670 * events_async : list of events which can be detected w/o polling
1671 * events_poll_msecs : polling interval, 0: disable, -1: system default
1672 */
__disk_events_show(unsigned int events,char * buf)1673 static ssize_t __disk_events_show(unsigned int events, char *buf)
1674 {
1675 const char *delim = "";
1676 ssize_t pos = 0;
1677 int i;
1678
1679 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1680 if (events & (1 << i)) {
1681 pos += sprintf(buf + pos, "%s%s",
1682 delim, disk_events_strs[i]);
1683 delim = " ";
1684 }
1685 if (pos)
1686 pos += sprintf(buf + pos, "\n");
1687 return pos;
1688 }
1689
disk_events_show(struct device * dev,struct device_attribute * attr,char * buf)1690 static ssize_t disk_events_show(struct device *dev,
1691 struct device_attribute *attr, char *buf)
1692 {
1693 struct gendisk *disk = dev_to_disk(dev);
1694
1695 return __disk_events_show(disk->events, buf);
1696 }
1697
disk_events_async_show(struct device * dev,struct device_attribute * attr,char * buf)1698 static ssize_t disk_events_async_show(struct device *dev,
1699 struct device_attribute *attr, char *buf)
1700 {
1701 struct gendisk *disk = dev_to_disk(dev);
1702
1703 return __disk_events_show(disk->async_events, buf);
1704 }
1705
disk_events_poll_msecs_show(struct device * dev,struct device_attribute * attr,char * buf)1706 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1707 struct device_attribute *attr,
1708 char *buf)
1709 {
1710 struct gendisk *disk = dev_to_disk(dev);
1711
1712 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1713 }
1714
disk_events_poll_msecs_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)1715 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1716 struct device_attribute *attr,
1717 const char *buf, size_t count)
1718 {
1719 struct gendisk *disk = dev_to_disk(dev);
1720 long intv;
1721
1722 if (!count || !sscanf(buf, "%ld", &intv))
1723 return -EINVAL;
1724
1725 if (intv < 0 && intv != -1)
1726 return -EINVAL;
1727
1728 disk_block_events(disk);
1729 disk->ev->poll_msecs = intv;
1730 __disk_unblock_events(disk, true);
1731
1732 return count;
1733 }
1734
1735 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
1736 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
1737 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
1738 disk_events_poll_msecs_show,
1739 disk_events_poll_msecs_store);
1740
1741 static const struct attribute *disk_events_attrs[] = {
1742 &dev_attr_events.attr,
1743 &dev_attr_events_async.attr,
1744 &dev_attr_events_poll_msecs.attr,
1745 NULL,
1746 };
1747
1748 /*
1749 * The default polling interval can be specified by the kernel
1750 * parameter block.events_dfl_poll_msecs which defaults to 0
1751 * (disable). This can also be modified runtime by writing to
1752 * /sys/module/block/events_dfl_poll_msecs.
1753 */
disk_events_set_dfl_poll_msecs(const char * val,const struct kernel_param * kp)1754 static int disk_events_set_dfl_poll_msecs(const char *val,
1755 const struct kernel_param *kp)
1756 {
1757 struct disk_events *ev;
1758 int ret;
1759
1760 ret = param_set_ulong(val, kp);
1761 if (ret < 0)
1762 return ret;
1763
1764 mutex_lock(&disk_events_mutex);
1765
1766 list_for_each_entry(ev, &disk_events, node)
1767 disk_flush_events(ev->disk, 0);
1768
1769 mutex_unlock(&disk_events_mutex);
1770
1771 return 0;
1772 }
1773
1774 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
1775 .set = disk_events_set_dfl_poll_msecs,
1776 .get = param_get_ulong,
1777 };
1778
1779 #undef MODULE_PARAM_PREFIX
1780 #define MODULE_PARAM_PREFIX "block."
1781
1782 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
1783 &disk_events_dfl_poll_msecs, 0644);
1784
1785 /*
1786 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
1787 */
disk_alloc_events(struct gendisk * disk)1788 static void disk_alloc_events(struct gendisk *disk)
1789 {
1790 struct disk_events *ev;
1791
1792 if (!disk->fops->check_events)
1793 return;
1794
1795 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
1796 if (!ev) {
1797 pr_warn("%s: failed to initialize events\n", disk->disk_name);
1798 return;
1799 }
1800
1801 INIT_LIST_HEAD(&ev->node);
1802 ev->disk = disk;
1803 spin_lock_init(&ev->lock);
1804 mutex_init(&ev->block_mutex);
1805 ev->block = 1;
1806 ev->poll_msecs = -1;
1807 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
1808
1809 disk->ev = ev;
1810 }
1811
disk_add_events(struct gendisk * disk)1812 static void disk_add_events(struct gendisk *disk)
1813 {
1814 if (!disk->ev)
1815 return;
1816
1817 /* FIXME: error handling */
1818 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
1819 pr_warn("%s: failed to create sysfs files for events\n",
1820 disk->disk_name);
1821
1822 mutex_lock(&disk_events_mutex);
1823 list_add_tail(&disk->ev->node, &disk_events);
1824 mutex_unlock(&disk_events_mutex);
1825
1826 /*
1827 * Block count is initialized to 1 and the following initial
1828 * unblock kicks it into action.
1829 */
1830 __disk_unblock_events(disk, true);
1831 }
1832
disk_del_events(struct gendisk * disk)1833 static void disk_del_events(struct gendisk *disk)
1834 {
1835 if (!disk->ev)
1836 return;
1837
1838 disk_block_events(disk);
1839
1840 mutex_lock(&disk_events_mutex);
1841 list_del_init(&disk->ev->node);
1842 mutex_unlock(&disk_events_mutex);
1843
1844 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
1845 }
1846
disk_release_events(struct gendisk * disk)1847 static void disk_release_events(struct gendisk *disk)
1848 {
1849 /* the block count should be 1 from disk_del_events() */
1850 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
1851 kfree(disk->ev);
1852 }
1853