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