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