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
part_inc_in_flight(struct request_queue * q,struct hd_struct * part,int rw)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
part_dec_in_flight(struct request_queue * q,struct hd_struct * part,int rw)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
part_in_flight(struct request_queue * q,struct hd_struct * part)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
part_in_flight_rw(struct request_queue * q,struct hd_struct * part,unsigned int inflight[2])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
__disk_get_part(struct gendisk * disk,int partno)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 */
disk_get_part(struct gendisk * disk,int partno)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 */
disk_part_iter_init(struct disk_part_iter * piter,struct gendisk * disk,unsigned int flags)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 */
disk_part_iter_next(struct disk_part_iter * piter)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 get_device(part_to_dev(part));
226 piter->part = part;
227 if (!part_nr_sects_read(part) &&
228 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
229 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
230 piter->idx == 0)) {
231 put_device(part_to_dev(part));
232 piter->part = NULL;
233 continue;
234 }
235
236 piter->idx += inc;
237 break;
238 }
239
240 rcu_read_unlock();
241
242 return piter->part;
243 }
244 EXPORT_SYMBOL_GPL(disk_part_iter_next);
245
246 /**
247 * disk_part_iter_exit - finish up partition iteration
248 * @piter: iter of interest
249 *
250 * Called when iteration is over. Cleans up @piter.
251 *
252 * CONTEXT:
253 * Don't care.
254 */
disk_part_iter_exit(struct disk_part_iter * piter)255 void disk_part_iter_exit(struct disk_part_iter *piter)
256 {
257 disk_put_part(piter->part);
258 piter->part = NULL;
259 }
260 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
261
sector_in_part(struct hd_struct * part,sector_t sector)262 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
263 {
264 return part->start_sect <= sector &&
265 sector < part->start_sect + part_nr_sects_read(part);
266 }
267
268 /**
269 * disk_map_sector_rcu - map sector to partition
270 * @disk: gendisk of interest
271 * @sector: sector to map
272 *
273 * Find out which partition @sector maps to on @disk. This is
274 * primarily used for stats accounting.
275 *
276 * CONTEXT:
277 * RCU read locked. The returned partition pointer is valid only
278 * while preemption is disabled.
279 *
280 * RETURNS:
281 * Found partition on success, part0 is returned if no partition matches
282 */
disk_map_sector_rcu(struct gendisk * disk,sector_t sector)283 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
284 {
285 struct disk_part_tbl *ptbl;
286 struct hd_struct *part;
287 int i;
288
289 ptbl = rcu_dereference(disk->part_tbl);
290
291 part = rcu_dereference(ptbl->last_lookup);
292 if (part && sector_in_part(part, sector))
293 return part;
294
295 for (i = 1; i < ptbl->len; i++) {
296 part = rcu_dereference(ptbl->part[i]);
297
298 if (part && sector_in_part(part, sector)) {
299 rcu_assign_pointer(ptbl->last_lookup, part);
300 return part;
301 }
302 }
303 return &disk->part0;
304 }
305 EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
306
307 /*
308 * Can be deleted altogether. Later.
309 *
310 */
311 #define BLKDEV_MAJOR_HASH_SIZE 255
312 static struct blk_major_name {
313 struct blk_major_name *next;
314 int major;
315 char name[16];
316 } *major_names[BLKDEV_MAJOR_HASH_SIZE];
317
318 /* index in the above - for now: assume no multimajor ranges */
major_to_index(unsigned major)319 static inline int major_to_index(unsigned major)
320 {
321 return major % BLKDEV_MAJOR_HASH_SIZE;
322 }
323
324 #ifdef CONFIG_PROC_FS
blkdev_show(struct seq_file * seqf,off_t offset)325 void blkdev_show(struct seq_file *seqf, off_t offset)
326 {
327 struct blk_major_name *dp;
328
329 mutex_lock(&block_class_lock);
330 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
331 if (dp->major == offset)
332 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
333 mutex_unlock(&block_class_lock);
334 }
335 #endif /* CONFIG_PROC_FS */
336
337 /**
338 * register_blkdev - register a new block device
339 *
340 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
341 * @major = 0, try to allocate any unused major number.
342 * @name: the name of the new block device as a zero terminated string
343 *
344 * The @name must be unique within the system.
345 *
346 * The return value depends on the @major input parameter:
347 *
348 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
349 * then the function returns zero on success, or a negative error code
350 * - if any unused major number was requested with @major = 0 parameter
351 * then the return value is the allocated major number in range
352 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
353 *
354 * See Documentation/admin-guide/devices.txt for the list of allocated
355 * major numbers.
356 */
register_blkdev(unsigned int major,const char * name)357 int register_blkdev(unsigned int major, const char *name)
358 {
359 struct blk_major_name **n, *p;
360 int index, ret = 0;
361
362 mutex_lock(&block_class_lock);
363
364 /* temporary */
365 if (major == 0) {
366 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
367 if (major_names[index] == NULL)
368 break;
369 }
370
371 if (index == 0) {
372 printk("%s: failed to get major for %s\n",
373 __func__, name);
374 ret = -EBUSY;
375 goto out;
376 }
377 major = index;
378 ret = major;
379 }
380
381 if (major >= BLKDEV_MAJOR_MAX) {
382 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
383 __func__, major, BLKDEV_MAJOR_MAX-1, name);
384
385 ret = -EINVAL;
386 goto out;
387 }
388
389 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
390 if (p == NULL) {
391 ret = -ENOMEM;
392 goto out;
393 }
394
395 p->major = major;
396 strlcpy(p->name, name, sizeof(p->name));
397 p->next = NULL;
398 index = major_to_index(major);
399
400 for (n = &major_names[index]; *n; n = &(*n)->next) {
401 if ((*n)->major == major)
402 break;
403 }
404 if (!*n)
405 *n = p;
406 else
407 ret = -EBUSY;
408
409 if (ret < 0) {
410 printk("register_blkdev: cannot get major %u for %s\n",
411 major, name);
412 kfree(p);
413 }
414 out:
415 mutex_unlock(&block_class_lock);
416 return ret;
417 }
418
419 EXPORT_SYMBOL(register_blkdev);
420
unregister_blkdev(unsigned int major,const char * name)421 void unregister_blkdev(unsigned int major, const char *name)
422 {
423 struct blk_major_name **n;
424 struct blk_major_name *p = NULL;
425 int index = major_to_index(major);
426
427 mutex_lock(&block_class_lock);
428 for (n = &major_names[index]; *n; n = &(*n)->next)
429 if ((*n)->major == major)
430 break;
431 if (!*n || strcmp((*n)->name, name)) {
432 WARN_ON(1);
433 } else {
434 p = *n;
435 *n = p->next;
436 }
437 mutex_unlock(&block_class_lock);
438 kfree(p);
439 }
440
441 EXPORT_SYMBOL(unregister_blkdev);
442
443 static struct kobj_map *bdev_map;
444
445 /**
446 * blk_mangle_minor - scatter minor numbers apart
447 * @minor: minor number to mangle
448 *
449 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
450 * is enabled. Mangling twice gives the original value.
451 *
452 * RETURNS:
453 * Mangled value.
454 *
455 * CONTEXT:
456 * Don't care.
457 */
blk_mangle_minor(int minor)458 static int blk_mangle_minor(int minor)
459 {
460 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
461 int i;
462
463 for (i = 0; i < MINORBITS / 2; i++) {
464 int low = minor & (1 << i);
465 int high = minor & (1 << (MINORBITS - 1 - i));
466 int distance = MINORBITS - 1 - 2 * i;
467
468 minor ^= low | high; /* clear both bits */
469 low <<= distance; /* swap the positions */
470 high >>= distance;
471 minor |= low | high; /* and set */
472 }
473 #endif
474 return minor;
475 }
476
477 /**
478 * blk_alloc_devt - allocate a dev_t for a partition
479 * @part: partition to allocate dev_t for
480 * @devt: out parameter for resulting dev_t
481 *
482 * Allocate a dev_t for block device.
483 *
484 * RETURNS:
485 * 0 on success, allocated dev_t is returned in *@devt. -errno on
486 * failure.
487 *
488 * CONTEXT:
489 * Might sleep.
490 */
blk_alloc_devt(struct hd_struct * part,dev_t * devt)491 int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
492 {
493 struct gendisk *disk = part_to_disk(part);
494 int idx;
495
496 /* in consecutive minor range? */
497 if (part->partno < disk->minors) {
498 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
499 return 0;
500 }
501
502 /* allocate ext devt */
503 idr_preload(GFP_KERNEL);
504
505 spin_lock_bh(&ext_devt_lock);
506 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
507 spin_unlock_bh(&ext_devt_lock);
508
509 idr_preload_end();
510 if (idx < 0)
511 return idx == -ENOSPC ? -EBUSY : idx;
512
513 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
514 return 0;
515 }
516
517 /**
518 * blk_free_devt - free a dev_t
519 * @devt: dev_t to free
520 *
521 * Free @devt which was allocated using blk_alloc_devt().
522 *
523 * CONTEXT:
524 * Might sleep.
525 */
blk_free_devt(dev_t devt)526 void blk_free_devt(dev_t devt)
527 {
528 if (devt == MKDEV(0, 0))
529 return;
530
531 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
532 spin_lock_bh(&ext_devt_lock);
533 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
534 spin_unlock_bh(&ext_devt_lock);
535 }
536 }
537
538 /*
539 * We invalidate devt by assigning NULL pointer for devt in idr.
540 */
blk_invalidate_devt(dev_t devt)541 void blk_invalidate_devt(dev_t devt)
542 {
543 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
544 spin_lock_bh(&ext_devt_lock);
545 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
546 spin_unlock_bh(&ext_devt_lock);
547 }
548 }
549
bdevt_str(dev_t devt,char * buf)550 static char *bdevt_str(dev_t devt, char *buf)
551 {
552 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
553 char tbuf[BDEVT_SIZE];
554 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
555 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
556 } else
557 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
558
559 return buf;
560 }
561
562 /*
563 * Register device numbers dev..(dev+range-1)
564 * range must be nonzero
565 * The hash chain is sorted on range, so that subranges can override.
566 */
blk_register_region(dev_t devt,unsigned long range,struct module * module,struct kobject * (* probe)(dev_t,int *,void *),int (* lock)(dev_t,void *),void * data)567 void blk_register_region(dev_t devt, unsigned long range, struct module *module,
568 struct kobject *(*probe)(dev_t, int *, void *),
569 int (*lock)(dev_t, void *), void *data)
570 {
571 kobj_map(bdev_map, devt, range, module, probe, lock, data);
572 }
573
574 EXPORT_SYMBOL(blk_register_region);
575
blk_unregister_region(dev_t devt,unsigned long range)576 void blk_unregister_region(dev_t devt, unsigned long range)
577 {
578 kobj_unmap(bdev_map, devt, range);
579 }
580
581 EXPORT_SYMBOL(blk_unregister_region);
582
exact_match(dev_t devt,int * partno,void * data)583 static struct kobject *exact_match(dev_t devt, int *partno, void *data)
584 {
585 struct gendisk *p = data;
586
587 return &disk_to_dev(p)->kobj;
588 }
589
exact_lock(dev_t devt,void * data)590 static int exact_lock(dev_t devt, void *data)
591 {
592 struct gendisk *p = data;
593
594 if (!get_disk_and_module(p))
595 return -1;
596 return 0;
597 }
598
register_disk(struct device * parent,struct gendisk * disk,const struct attribute_group ** groups)599 static void register_disk(struct device *parent, struct gendisk *disk,
600 const struct attribute_group **groups)
601 {
602 struct device *ddev = disk_to_dev(disk);
603 struct block_device *bdev;
604 struct disk_part_iter piter;
605 struct hd_struct *part;
606 int err;
607
608 ddev->parent = parent;
609
610 dev_set_name(ddev, "%s", disk->disk_name);
611
612 /* delay uevents, until we scanned partition table */
613 dev_set_uevent_suppress(ddev, 1);
614
615 if (groups) {
616 WARN_ON(ddev->groups);
617 ddev->groups = groups;
618 }
619 if (device_add(ddev))
620 return;
621 if (!sysfs_deprecated) {
622 err = sysfs_create_link(block_depr, &ddev->kobj,
623 kobject_name(&ddev->kobj));
624 if (err) {
625 device_del(ddev);
626 return;
627 }
628 }
629
630 /*
631 * avoid probable deadlock caused by allocating memory with
632 * GFP_KERNEL in runtime_resume callback of its all ancestor
633 * devices
634 */
635 pm_runtime_set_memalloc_noio(ddev, true);
636
637 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
638 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
639
640 if (disk->flags & GENHD_FL_HIDDEN)
641 return;
642
643 /* No minors to use for partitions */
644 if (!disk_part_scan_enabled(disk))
645 goto exit;
646
647 /* No such device (e.g., media were just removed) */
648 if (!get_capacity(disk))
649 goto exit;
650
651 bdev = bdget_disk(disk, 0);
652 if (!bdev)
653 goto exit;
654
655 bdev->bd_invalidated = 1;
656 err = blkdev_get(bdev, FMODE_READ, NULL);
657 if (err < 0)
658 goto exit;
659 blkdev_put(bdev, FMODE_READ);
660
661 exit:
662 /* announce disk after possible partitions are created */
663 dev_set_uevent_suppress(ddev, 0);
664 kobject_uevent(&ddev->kobj, KOBJ_ADD);
665
666 /* announce possible partitions */
667 disk_part_iter_init(&piter, disk, 0);
668 while ((part = disk_part_iter_next(&piter)))
669 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
670 disk_part_iter_exit(&piter);
671
672 if (disk->queue->backing_dev_info->dev) {
673 err = sysfs_create_link(&ddev->kobj,
674 &disk->queue->backing_dev_info->dev->kobj,
675 "bdi");
676 WARN_ON(err);
677 }
678 }
679
680 /**
681 * __device_add_disk - add disk information to kernel list
682 * @parent: parent device for the disk
683 * @disk: per-device partitioning information
684 * @groups: Additional per-device sysfs groups
685 * @register_queue: register the queue if set to true
686 *
687 * This function registers the partitioning information in @disk
688 * with the kernel.
689 *
690 * FIXME: error handling
691 */
__device_add_disk(struct device * parent,struct gendisk * disk,const struct attribute_group ** groups,bool register_queue)692 static void __device_add_disk(struct device *parent, struct gendisk *disk,
693 const struct attribute_group **groups,
694 bool register_queue)
695 {
696 dev_t devt;
697 int retval;
698
699 /*
700 * The disk queue should now be all set with enough information about
701 * the device for the elevator code to pick an adequate default
702 * elevator if one is needed, that is, for devices requesting queue
703 * registration.
704 */
705 if (register_queue)
706 elevator_init_mq(disk->queue);
707
708 /* minors == 0 indicates to use ext devt from part0 and should
709 * be accompanied with EXT_DEVT flag. Make sure all
710 * parameters make sense.
711 */
712 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
713 WARN_ON(!disk->minors &&
714 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
715
716 disk->flags |= GENHD_FL_UP;
717
718 retval = blk_alloc_devt(&disk->part0, &devt);
719 if (retval) {
720 WARN_ON(1);
721 return;
722 }
723 disk->major = MAJOR(devt);
724 disk->first_minor = MINOR(devt);
725
726 disk_alloc_events(disk);
727
728 if (disk->flags & GENHD_FL_HIDDEN) {
729 /*
730 * Don't let hidden disks show up in /proc/partitions,
731 * and don't bother scanning for partitions either.
732 */
733 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
734 disk->flags |= GENHD_FL_NO_PART_SCAN;
735 } else {
736 int ret;
737
738 /* Register BDI before referencing it from bdev */
739 disk_to_dev(disk)->devt = devt;
740 ret = bdi_register_owner(disk->queue->backing_dev_info,
741 disk_to_dev(disk));
742 WARN_ON(ret);
743 blk_register_region(disk_devt(disk), disk->minors, NULL,
744 exact_match, exact_lock, disk);
745 }
746 register_disk(parent, disk, groups);
747 if (register_queue)
748 blk_register_queue(disk);
749
750 /*
751 * Take an extra ref on queue which will be put on disk_release()
752 * so that it sticks around as long as @disk is there.
753 */
754 WARN_ON_ONCE(!blk_get_queue(disk->queue));
755
756 disk_add_events(disk);
757 blk_integrity_add(disk);
758 }
759
device_add_disk(struct device * parent,struct gendisk * disk,const struct attribute_group ** groups)760 void device_add_disk(struct device *parent, struct gendisk *disk,
761 const struct attribute_group **groups)
762
763 {
764 __device_add_disk(parent, disk, groups, true);
765 }
766 EXPORT_SYMBOL(device_add_disk);
767
device_add_disk_no_queue_reg(struct device * parent,struct gendisk * disk)768 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
769 {
770 __device_add_disk(parent, disk, NULL, false);
771 }
772 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
773
del_gendisk(struct gendisk * disk)774 void del_gendisk(struct gendisk *disk)
775 {
776 struct disk_part_iter piter;
777 struct hd_struct *part;
778
779 blk_integrity_del(disk);
780 disk_del_events(disk);
781
782 /*
783 * Block lookups of the disk until all bdevs are unhashed and the
784 * disk is marked as dead (GENHD_FL_UP cleared).
785 */
786 down_write(&disk->lookup_sem);
787 /* invalidate stuff */
788 disk_part_iter_init(&piter, disk,
789 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
790 while ((part = disk_part_iter_next(&piter))) {
791 invalidate_partition(disk, part->partno);
792 bdev_unhash_inode(part_devt(part));
793 delete_partition(disk, part->partno);
794 }
795 disk_part_iter_exit(&piter);
796
797 invalidate_partition(disk, 0);
798 bdev_unhash_inode(disk_devt(disk));
799 set_capacity(disk, 0);
800 disk->flags &= ~GENHD_FL_UP;
801 up_write(&disk->lookup_sem);
802
803 if (!(disk->flags & GENHD_FL_HIDDEN))
804 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
805 if (disk->queue) {
806 /*
807 * Unregister bdi before releasing device numbers (as they can
808 * get reused and we'd get clashes in sysfs).
809 */
810 if (!(disk->flags & GENHD_FL_HIDDEN))
811 bdi_unregister(disk->queue->backing_dev_info);
812 blk_unregister_queue(disk);
813 } else {
814 WARN_ON(1);
815 }
816
817 if (!(disk->flags & GENHD_FL_HIDDEN))
818 blk_unregister_region(disk_devt(disk), disk->minors);
819 /*
820 * Remove gendisk pointer from idr so that it cannot be looked up
821 * while RCU period before freeing gendisk is running to prevent
822 * use-after-free issues. Note that the device number stays
823 * "in-use" until we really free the gendisk.
824 */
825 blk_invalidate_devt(disk_devt(disk));
826
827 kobject_put(disk->part0.holder_dir);
828 kobject_put(disk->slave_dir);
829
830 part_stat_set_all(&disk->part0, 0);
831 disk->part0.stamp = 0;
832 if (!sysfs_deprecated)
833 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
834 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
835 device_del(disk_to_dev(disk));
836 }
837 EXPORT_SYMBOL(del_gendisk);
838
839 /* sysfs access to bad-blocks list. */
disk_badblocks_show(struct device * dev,struct device_attribute * attr,char * page)840 static ssize_t disk_badblocks_show(struct device *dev,
841 struct device_attribute *attr,
842 char *page)
843 {
844 struct gendisk *disk = dev_to_disk(dev);
845
846 if (!disk->bb)
847 return sprintf(page, "\n");
848
849 return badblocks_show(disk->bb, page, 0);
850 }
851
disk_badblocks_store(struct device * dev,struct device_attribute * attr,const char * page,size_t len)852 static ssize_t disk_badblocks_store(struct device *dev,
853 struct device_attribute *attr,
854 const char *page, size_t len)
855 {
856 struct gendisk *disk = dev_to_disk(dev);
857
858 if (!disk->bb)
859 return -ENXIO;
860
861 return badblocks_store(disk->bb, page, len, 0);
862 }
863
864 /**
865 * get_gendisk - get partitioning information for a given device
866 * @devt: device to get partitioning information for
867 * @partno: returned partition index
868 *
869 * This function gets the structure containing partitioning
870 * information for the given device @devt.
871 */
get_gendisk(dev_t devt,int * partno)872 struct gendisk *get_gendisk(dev_t devt, int *partno)
873 {
874 struct gendisk *disk = NULL;
875
876 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
877 struct kobject *kobj;
878
879 kobj = kobj_lookup(bdev_map, devt, partno);
880 if (kobj)
881 disk = dev_to_disk(kobj_to_dev(kobj));
882 } else {
883 struct hd_struct *part;
884
885 spin_lock_bh(&ext_devt_lock);
886 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
887 if (part && get_disk_and_module(part_to_disk(part))) {
888 *partno = part->partno;
889 disk = part_to_disk(part);
890 }
891 spin_unlock_bh(&ext_devt_lock);
892 }
893
894 if (!disk)
895 return NULL;
896
897 /*
898 * Synchronize with del_gendisk() to not return disk that is being
899 * destroyed.
900 */
901 down_read(&disk->lookup_sem);
902 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
903 !(disk->flags & GENHD_FL_UP))) {
904 up_read(&disk->lookup_sem);
905 put_disk_and_module(disk);
906 disk = NULL;
907 } else {
908 up_read(&disk->lookup_sem);
909 }
910 return disk;
911 }
912 EXPORT_SYMBOL(get_gendisk);
913
914 /**
915 * bdget_disk - do bdget() by gendisk and partition number
916 * @disk: gendisk of interest
917 * @partno: partition number
918 *
919 * Find partition @partno from @disk, do bdget() on it.
920 *
921 * CONTEXT:
922 * Don't care.
923 *
924 * RETURNS:
925 * Resulting block_device on success, NULL on failure.
926 */
bdget_disk(struct gendisk * disk,int partno)927 struct block_device *bdget_disk(struct gendisk *disk, int partno)
928 {
929 struct hd_struct *part;
930 struct block_device *bdev = NULL;
931
932 part = disk_get_part(disk, partno);
933 if (part)
934 bdev = bdget(part_devt(part));
935 disk_put_part(part);
936
937 return bdev;
938 }
939 EXPORT_SYMBOL(bdget_disk);
940
941 /*
942 * print a full list of all partitions - intended for places where the root
943 * filesystem can't be mounted and thus to give the victim some idea of what
944 * went wrong
945 */
printk_all_partitions(void)946 void __init printk_all_partitions(void)
947 {
948 struct class_dev_iter iter;
949 struct device *dev;
950
951 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
952 while ((dev = class_dev_iter_next(&iter))) {
953 struct gendisk *disk = dev_to_disk(dev);
954 struct disk_part_iter piter;
955 struct hd_struct *part;
956 char name_buf[BDEVNAME_SIZE];
957 char devt_buf[BDEVT_SIZE];
958
959 /*
960 * Don't show empty devices or things that have been
961 * suppressed
962 */
963 if (get_capacity(disk) == 0 ||
964 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
965 continue;
966
967 /*
968 * Note, unlike /proc/partitions, I am showing the
969 * numbers in hex - the same format as the root=
970 * option takes.
971 */
972 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
973 while ((part = disk_part_iter_next(&piter))) {
974 bool is_part0 = part == &disk->part0;
975
976 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
977 bdevt_str(part_devt(part), devt_buf),
978 (unsigned long long)part_nr_sects_read(part) >> 1
979 , disk_name(disk, part->partno, name_buf),
980 part->info ? part->info->uuid : "");
981 if (is_part0) {
982 if (dev->parent && dev->parent->driver)
983 printk(" driver: %s\n",
984 dev->parent->driver->name);
985 else
986 printk(" (driver?)\n");
987 } else
988 printk("\n");
989 }
990 disk_part_iter_exit(&piter);
991 }
992 class_dev_iter_exit(&iter);
993 }
994
995 #ifdef CONFIG_PROC_FS
996 /* iterator */
disk_seqf_start(struct seq_file * seqf,loff_t * pos)997 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
998 {
999 loff_t skip = *pos;
1000 struct class_dev_iter *iter;
1001 struct device *dev;
1002
1003 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1004 if (!iter)
1005 return ERR_PTR(-ENOMEM);
1006
1007 seqf->private = iter;
1008 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1009 do {
1010 dev = class_dev_iter_next(iter);
1011 if (!dev)
1012 return NULL;
1013 } while (skip--);
1014
1015 return dev_to_disk(dev);
1016 }
1017
disk_seqf_next(struct seq_file * seqf,void * v,loff_t * pos)1018 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1019 {
1020 struct device *dev;
1021
1022 (*pos)++;
1023 dev = class_dev_iter_next(seqf->private);
1024 if (dev)
1025 return dev_to_disk(dev);
1026
1027 return NULL;
1028 }
1029
disk_seqf_stop(struct seq_file * seqf,void * v)1030 static void disk_seqf_stop(struct seq_file *seqf, void *v)
1031 {
1032 struct class_dev_iter *iter = seqf->private;
1033
1034 /* stop is called even after start failed :-( */
1035 if (iter) {
1036 class_dev_iter_exit(iter);
1037 kfree(iter);
1038 seqf->private = NULL;
1039 }
1040 }
1041
show_partition_start(struct seq_file * seqf,loff_t * pos)1042 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1043 {
1044 void *p;
1045
1046 p = disk_seqf_start(seqf, pos);
1047 if (!IS_ERR_OR_NULL(p) && !*pos)
1048 seq_puts(seqf, "major minor #blocks name\n\n");
1049 return p;
1050 }
1051
show_partition(struct seq_file * seqf,void * v)1052 static int show_partition(struct seq_file *seqf, void *v)
1053 {
1054 struct gendisk *sgp = v;
1055 struct disk_part_iter piter;
1056 struct hd_struct *part;
1057 char buf[BDEVNAME_SIZE];
1058
1059 /* Don't show non-partitionable removeable devices or empty devices */
1060 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1061 (sgp->flags & GENHD_FL_REMOVABLE)))
1062 return 0;
1063 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1064 return 0;
1065
1066 /* show the full disk and all non-0 size partitions of it */
1067 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1068 while ((part = disk_part_iter_next(&piter)))
1069 seq_printf(seqf, "%4d %7d %10llu %s\n",
1070 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1071 (unsigned long long)part_nr_sects_read(part) >> 1,
1072 disk_name(sgp, part->partno, buf));
1073 disk_part_iter_exit(&piter);
1074
1075 return 0;
1076 }
1077
1078 static const struct seq_operations partitions_op = {
1079 .start = show_partition_start,
1080 .next = disk_seqf_next,
1081 .stop = disk_seqf_stop,
1082 .show = show_partition
1083 };
1084 #endif
1085
1086
base_probe(dev_t devt,int * partno,void * data)1087 static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1088 {
1089 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1090 /* Make old-style 2.4 aliases work */
1091 request_module("block-major-%d", MAJOR(devt));
1092 return NULL;
1093 }
1094
genhd_device_init(void)1095 static int __init genhd_device_init(void)
1096 {
1097 int error;
1098
1099 block_class.dev_kobj = sysfs_dev_block_kobj;
1100 error = class_register(&block_class);
1101 if (unlikely(error))
1102 return error;
1103 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1104 blk_dev_init();
1105
1106 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1107
1108 /* create top-level block dir */
1109 if (!sysfs_deprecated)
1110 block_depr = kobject_create_and_add("block", NULL);
1111 return 0;
1112 }
1113
1114 subsys_initcall(genhd_device_init);
1115
disk_range_show(struct device * dev,struct device_attribute * attr,char * buf)1116 static ssize_t disk_range_show(struct device *dev,
1117 struct device_attribute *attr, char *buf)
1118 {
1119 struct gendisk *disk = dev_to_disk(dev);
1120
1121 return sprintf(buf, "%d\n", disk->minors);
1122 }
1123
disk_ext_range_show(struct device * dev,struct device_attribute * attr,char * buf)1124 static ssize_t disk_ext_range_show(struct device *dev,
1125 struct device_attribute *attr, char *buf)
1126 {
1127 struct gendisk *disk = dev_to_disk(dev);
1128
1129 return sprintf(buf, "%d\n", disk_max_parts(disk));
1130 }
1131
disk_removable_show(struct device * dev,struct device_attribute * attr,char * buf)1132 static ssize_t disk_removable_show(struct device *dev,
1133 struct device_attribute *attr, char *buf)
1134 {
1135 struct gendisk *disk = dev_to_disk(dev);
1136
1137 return sprintf(buf, "%d\n",
1138 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1139 }
1140
disk_hidden_show(struct device * dev,struct device_attribute * attr,char * buf)1141 static ssize_t disk_hidden_show(struct device *dev,
1142 struct device_attribute *attr, char *buf)
1143 {
1144 struct gendisk *disk = dev_to_disk(dev);
1145
1146 return sprintf(buf, "%d\n",
1147 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1148 }
1149
disk_ro_show(struct device * dev,struct device_attribute * attr,char * buf)1150 static ssize_t disk_ro_show(struct device *dev,
1151 struct device_attribute *attr, char *buf)
1152 {
1153 struct gendisk *disk = dev_to_disk(dev);
1154
1155 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1156 }
1157
disk_capability_show(struct device * dev,struct device_attribute * attr,char * buf)1158 static ssize_t disk_capability_show(struct device *dev,
1159 struct device_attribute *attr, char *buf)
1160 {
1161 struct gendisk *disk = dev_to_disk(dev);
1162
1163 return sprintf(buf, "%x\n", disk->flags);
1164 }
1165
disk_alignment_offset_show(struct device * dev,struct device_attribute * attr,char * buf)1166 static ssize_t disk_alignment_offset_show(struct device *dev,
1167 struct device_attribute *attr,
1168 char *buf)
1169 {
1170 struct gendisk *disk = dev_to_disk(dev);
1171
1172 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1173 }
1174
disk_discard_alignment_show(struct device * dev,struct device_attribute * attr,char * buf)1175 static ssize_t disk_discard_alignment_show(struct device *dev,
1176 struct device_attribute *attr,
1177 char *buf)
1178 {
1179 struct gendisk *disk = dev_to_disk(dev);
1180
1181 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1182 }
1183
1184 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1185 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1186 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1187 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1188 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1189 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1190 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1191 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1192 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1193 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1194 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1195 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1196 #ifdef CONFIG_FAIL_MAKE_REQUEST
1197 static struct device_attribute dev_attr_fail =
1198 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1199 #endif
1200 #ifdef CONFIG_FAIL_IO_TIMEOUT
1201 static struct device_attribute dev_attr_fail_timeout =
1202 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1203 #endif
1204
1205 static struct attribute *disk_attrs[] = {
1206 &dev_attr_range.attr,
1207 &dev_attr_ext_range.attr,
1208 &dev_attr_removable.attr,
1209 &dev_attr_hidden.attr,
1210 &dev_attr_ro.attr,
1211 &dev_attr_size.attr,
1212 &dev_attr_alignment_offset.attr,
1213 &dev_attr_discard_alignment.attr,
1214 &dev_attr_capability.attr,
1215 &dev_attr_stat.attr,
1216 &dev_attr_inflight.attr,
1217 &dev_attr_badblocks.attr,
1218 #ifdef CONFIG_FAIL_MAKE_REQUEST
1219 &dev_attr_fail.attr,
1220 #endif
1221 #ifdef CONFIG_FAIL_IO_TIMEOUT
1222 &dev_attr_fail_timeout.attr,
1223 #endif
1224 NULL
1225 };
1226
disk_visible(struct kobject * kobj,struct attribute * a,int n)1227 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1228 {
1229 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1230 struct gendisk *disk = dev_to_disk(dev);
1231
1232 if (a == &dev_attr_badblocks.attr && !disk->bb)
1233 return 0;
1234 return a->mode;
1235 }
1236
1237 static struct attribute_group disk_attr_group = {
1238 .attrs = disk_attrs,
1239 .is_visible = disk_visible,
1240 };
1241
1242 static const struct attribute_group *disk_attr_groups[] = {
1243 &disk_attr_group,
1244 NULL
1245 };
1246
1247 /**
1248 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1249 * @disk: disk to replace part_tbl for
1250 * @new_ptbl: new part_tbl to install
1251 *
1252 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1253 * original ptbl is freed using RCU callback.
1254 *
1255 * LOCKING:
1256 * Matching bd_mutex locked or the caller is the only user of @disk.
1257 */
disk_replace_part_tbl(struct gendisk * disk,struct disk_part_tbl * new_ptbl)1258 static void disk_replace_part_tbl(struct gendisk *disk,
1259 struct disk_part_tbl *new_ptbl)
1260 {
1261 struct disk_part_tbl *old_ptbl =
1262 rcu_dereference_protected(disk->part_tbl, 1);
1263
1264 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1265
1266 if (old_ptbl) {
1267 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1268 kfree_rcu(old_ptbl, rcu_head);
1269 }
1270 }
1271
1272 /**
1273 * disk_expand_part_tbl - expand disk->part_tbl
1274 * @disk: disk to expand part_tbl for
1275 * @partno: expand such that this partno can fit in
1276 *
1277 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1278 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1279 *
1280 * LOCKING:
1281 * Matching bd_mutex locked or the caller is the only user of @disk.
1282 * Might sleep.
1283 *
1284 * RETURNS:
1285 * 0 on success, -errno on failure.
1286 */
disk_expand_part_tbl(struct gendisk * disk,int partno)1287 int disk_expand_part_tbl(struct gendisk *disk, int partno)
1288 {
1289 struct disk_part_tbl *old_ptbl =
1290 rcu_dereference_protected(disk->part_tbl, 1);
1291 struct disk_part_tbl *new_ptbl;
1292 int len = old_ptbl ? old_ptbl->len : 0;
1293 int i, target;
1294
1295 /*
1296 * check for int overflow, since we can get here from blkpg_ioctl()
1297 * with a user passed 'partno'.
1298 */
1299 target = partno + 1;
1300 if (target < 0)
1301 return -EINVAL;
1302
1303 /* disk_max_parts() is zero during initialization, ignore if so */
1304 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1305 return -EINVAL;
1306
1307 if (target <= len)
1308 return 0;
1309
1310 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1311 disk->node_id);
1312 if (!new_ptbl)
1313 return -ENOMEM;
1314
1315 new_ptbl->len = target;
1316
1317 for (i = 0; i < len; i++)
1318 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1319
1320 disk_replace_part_tbl(disk, new_ptbl);
1321 return 0;
1322 }
1323
disk_release(struct device * dev)1324 static void disk_release(struct device *dev)
1325 {
1326 struct gendisk *disk = dev_to_disk(dev);
1327
1328 blk_free_devt(dev->devt);
1329 disk_release_events(disk);
1330 kfree(disk->random);
1331 disk_replace_part_tbl(disk, NULL);
1332 hd_free_part(&disk->part0);
1333 if (disk->queue)
1334 blk_put_queue(disk->queue);
1335 kfree(disk);
1336 }
1337 struct class block_class = {
1338 .name = "block",
1339 };
1340
block_devnode(struct device * dev,umode_t * mode,kuid_t * uid,kgid_t * gid)1341 static char *block_devnode(struct device *dev, umode_t *mode,
1342 kuid_t *uid, kgid_t *gid)
1343 {
1344 struct gendisk *disk = dev_to_disk(dev);
1345
1346 if (disk->devnode)
1347 return disk->devnode(disk, mode);
1348 return NULL;
1349 }
1350
1351 static const struct device_type disk_type = {
1352 .name = "disk",
1353 .groups = disk_attr_groups,
1354 .release = disk_release,
1355 .devnode = block_devnode,
1356 };
1357
1358 #ifdef CONFIG_PROC_FS
1359 /*
1360 * aggregate disk stat collector. Uses the same stats that the sysfs
1361 * entries do, above, but makes them available through one seq_file.
1362 *
1363 * The output looks suspiciously like /proc/partitions with a bunch of
1364 * extra fields.
1365 */
diskstats_show(struct seq_file * seqf,void * v)1366 static int diskstats_show(struct seq_file *seqf, void *v)
1367 {
1368 struct gendisk *gp = v;
1369 struct disk_part_iter piter;
1370 struct hd_struct *hd;
1371 char buf[BDEVNAME_SIZE];
1372 unsigned int inflight;
1373
1374 /*
1375 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1376 seq_puts(seqf, "major minor name"
1377 " rio rmerge rsect ruse wio wmerge "
1378 "wsect wuse running use aveq"
1379 "\n\n");
1380 */
1381
1382 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1383 while ((hd = disk_part_iter_next(&piter))) {
1384 inflight = part_in_flight(gp->queue, hd);
1385 seq_printf(seqf, "%4d %7d %s "
1386 "%lu %lu %lu %u "
1387 "%lu %lu %lu %u "
1388 "%u %u %u "
1389 "%lu %lu %lu %u\n",
1390 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1391 disk_name(gp, hd->partno, buf),
1392 part_stat_read(hd, ios[STAT_READ]),
1393 part_stat_read(hd, merges[STAT_READ]),
1394 part_stat_read(hd, sectors[STAT_READ]),
1395 (unsigned int)part_stat_read_msecs(hd, STAT_READ),
1396 part_stat_read(hd, ios[STAT_WRITE]),
1397 part_stat_read(hd, merges[STAT_WRITE]),
1398 part_stat_read(hd, sectors[STAT_WRITE]),
1399 (unsigned int)part_stat_read_msecs(hd, STAT_WRITE),
1400 inflight,
1401 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1402 jiffies_to_msecs(part_stat_read(hd, time_in_queue)),
1403 part_stat_read(hd, ios[STAT_DISCARD]),
1404 part_stat_read(hd, merges[STAT_DISCARD]),
1405 part_stat_read(hd, sectors[STAT_DISCARD]),
1406 (unsigned int)part_stat_read_msecs(hd, STAT_DISCARD)
1407 );
1408 }
1409 disk_part_iter_exit(&piter);
1410
1411 return 0;
1412 }
1413
1414 static const struct seq_operations diskstats_op = {
1415 .start = disk_seqf_start,
1416 .next = disk_seqf_next,
1417 .stop = disk_seqf_stop,
1418 .show = diskstats_show
1419 };
1420
proc_genhd_init(void)1421 static int __init proc_genhd_init(void)
1422 {
1423 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1424 proc_create_seq("partitions", 0, NULL, &partitions_op);
1425 return 0;
1426 }
1427 module_init(proc_genhd_init);
1428 #endif /* CONFIG_PROC_FS */
1429
blk_lookup_devt(const char * name,int partno)1430 dev_t blk_lookup_devt(const char *name, int partno)
1431 {
1432 dev_t devt = MKDEV(0, 0);
1433 struct class_dev_iter iter;
1434 struct device *dev;
1435
1436 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1437 while ((dev = class_dev_iter_next(&iter))) {
1438 struct gendisk *disk = dev_to_disk(dev);
1439 struct hd_struct *part;
1440
1441 if (strcmp(dev_name(dev), name))
1442 continue;
1443
1444 if (partno < disk->minors) {
1445 /* We need to return the right devno, even
1446 * if the partition doesn't exist yet.
1447 */
1448 devt = MKDEV(MAJOR(dev->devt),
1449 MINOR(dev->devt) + partno);
1450 break;
1451 }
1452 part = disk_get_part(disk, partno);
1453 if (part) {
1454 devt = part_devt(part);
1455 disk_put_part(part);
1456 break;
1457 }
1458 disk_put_part(part);
1459 }
1460 class_dev_iter_exit(&iter);
1461 return devt;
1462 }
1463 EXPORT_SYMBOL(blk_lookup_devt);
1464
__alloc_disk_node(int minors,int node_id)1465 struct gendisk *__alloc_disk_node(int minors, int node_id)
1466 {
1467 struct gendisk *disk;
1468 struct disk_part_tbl *ptbl;
1469
1470 if (minors > DISK_MAX_PARTS) {
1471 printk(KERN_ERR
1472 "block: can't allocate more than %d partitions\n",
1473 DISK_MAX_PARTS);
1474 minors = DISK_MAX_PARTS;
1475 }
1476
1477 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1478 if (disk) {
1479 if (!init_part_stats(&disk->part0)) {
1480 kfree(disk);
1481 return NULL;
1482 }
1483 init_rwsem(&disk->lookup_sem);
1484 disk->node_id = node_id;
1485 if (disk_expand_part_tbl(disk, 0)) {
1486 free_part_stats(&disk->part0);
1487 kfree(disk);
1488 return NULL;
1489 }
1490 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1491 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1492
1493 /*
1494 * set_capacity() and get_capacity() currently don't use
1495 * seqcounter to read/update the part0->nr_sects. Still init
1496 * the counter as we can read the sectors in IO submission
1497 * patch using seqence counters.
1498 *
1499 * TODO: Ideally set_capacity() and get_capacity() should be
1500 * converted to make use of bd_mutex and sequence counters.
1501 */
1502 seqcount_init(&disk->part0.nr_sects_seq);
1503 if (hd_ref_init(&disk->part0)) {
1504 hd_free_part(&disk->part0);
1505 kfree(disk);
1506 return NULL;
1507 }
1508
1509 disk->minors = minors;
1510 rand_initialize_disk(disk);
1511 disk_to_dev(disk)->class = &block_class;
1512 disk_to_dev(disk)->type = &disk_type;
1513 device_initialize(disk_to_dev(disk));
1514 }
1515 return disk;
1516 }
1517 EXPORT_SYMBOL(__alloc_disk_node);
1518
get_disk_and_module(struct gendisk * disk)1519 struct kobject *get_disk_and_module(struct gendisk *disk)
1520 {
1521 struct module *owner;
1522 struct kobject *kobj;
1523
1524 if (!disk->fops)
1525 return NULL;
1526 owner = disk->fops->owner;
1527 if (owner && !try_module_get(owner))
1528 return NULL;
1529 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1530 if (kobj == NULL) {
1531 module_put(owner);
1532 return NULL;
1533 }
1534 return kobj;
1535
1536 }
1537 EXPORT_SYMBOL(get_disk_and_module);
1538
put_disk(struct gendisk * disk)1539 void put_disk(struct gendisk *disk)
1540 {
1541 if (disk)
1542 kobject_put(&disk_to_dev(disk)->kobj);
1543 }
1544 EXPORT_SYMBOL(put_disk);
1545
1546 /*
1547 * This is a counterpart of get_disk_and_module() and thus also of
1548 * get_gendisk().
1549 */
put_disk_and_module(struct gendisk * disk)1550 void put_disk_and_module(struct gendisk *disk)
1551 {
1552 if (disk) {
1553 struct module *owner = disk->fops->owner;
1554
1555 put_disk(disk);
1556 module_put(owner);
1557 }
1558 }
1559 EXPORT_SYMBOL(put_disk_and_module);
1560
set_disk_ro_uevent(struct gendisk * gd,int ro)1561 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1562 {
1563 char event[] = "DISK_RO=1";
1564 char *envp[] = { event, NULL };
1565
1566 if (!ro)
1567 event[8] = '0';
1568 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1569 }
1570
set_device_ro(struct block_device * bdev,int flag)1571 void set_device_ro(struct block_device *bdev, int flag)
1572 {
1573 bdev->bd_part->policy = flag;
1574 }
1575
1576 EXPORT_SYMBOL(set_device_ro);
1577
set_disk_ro(struct gendisk * disk,int flag)1578 void set_disk_ro(struct gendisk *disk, int flag)
1579 {
1580 struct disk_part_iter piter;
1581 struct hd_struct *part;
1582
1583 if (disk->part0.policy != flag) {
1584 set_disk_ro_uevent(disk, flag);
1585 disk->part0.policy = flag;
1586 }
1587
1588 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1589 while ((part = disk_part_iter_next(&piter)))
1590 part->policy = flag;
1591 disk_part_iter_exit(&piter);
1592 }
1593
1594 EXPORT_SYMBOL(set_disk_ro);
1595
bdev_read_only(struct block_device * bdev)1596 int bdev_read_only(struct block_device *bdev)
1597 {
1598 if (!bdev)
1599 return 0;
1600 return bdev->bd_part->policy;
1601 }
1602
1603 EXPORT_SYMBOL(bdev_read_only);
1604
invalidate_partition(struct gendisk * disk,int partno)1605 int invalidate_partition(struct gendisk *disk, int partno)
1606 {
1607 int res = 0;
1608 struct block_device *bdev = bdget_disk(disk, partno);
1609 if (bdev) {
1610 fsync_bdev(bdev);
1611 res = __invalidate_device(bdev, true);
1612 bdput(bdev);
1613 }
1614 return res;
1615 }
1616
1617 EXPORT_SYMBOL(invalidate_partition);
1618
1619 /*
1620 * Disk events - monitor disk events like media change and eject request.
1621 */
1622 struct disk_events {
1623 struct list_head node; /* all disk_event's */
1624 struct gendisk *disk; /* the associated disk */
1625 spinlock_t lock;
1626
1627 struct mutex block_mutex; /* protects blocking */
1628 int block; /* event blocking depth */
1629 unsigned int pending; /* events already sent out */
1630 unsigned int clearing; /* events being cleared */
1631
1632 long poll_msecs; /* interval, -1 for default */
1633 struct delayed_work dwork;
1634 };
1635
1636 static const char *disk_events_strs[] = {
1637 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1638 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1639 };
1640
1641 static char *disk_uevents[] = {
1642 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1643 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1644 };
1645
1646 /* list of all disk_events */
1647 static DEFINE_MUTEX(disk_events_mutex);
1648 static LIST_HEAD(disk_events);
1649
1650 /* disable in-kernel polling by default */
1651 static unsigned long disk_events_dfl_poll_msecs;
1652
disk_events_poll_jiffies(struct gendisk * disk)1653 static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1654 {
1655 struct disk_events *ev = disk->ev;
1656 long intv_msecs = 0;
1657
1658 /*
1659 * If device-specific poll interval is set, always use it. If
1660 * the default is being used, poll if the POLL flag is set.
1661 */
1662 if (ev->poll_msecs >= 0)
1663 intv_msecs = ev->poll_msecs;
1664 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1665 intv_msecs = disk_events_dfl_poll_msecs;
1666
1667 return msecs_to_jiffies(intv_msecs);
1668 }
1669
1670 /**
1671 * disk_block_events - block and flush disk event checking
1672 * @disk: disk to block events for
1673 *
1674 * On return from this function, it is guaranteed that event checking
1675 * isn't in progress and won't happen until unblocked by
1676 * disk_unblock_events(). Events blocking is counted and the actual
1677 * unblocking happens after the matching number of unblocks are done.
1678 *
1679 * Note that this intentionally does not block event checking from
1680 * disk_clear_events().
1681 *
1682 * CONTEXT:
1683 * Might sleep.
1684 */
disk_block_events(struct gendisk * disk)1685 void disk_block_events(struct gendisk *disk)
1686 {
1687 struct disk_events *ev = disk->ev;
1688 unsigned long flags;
1689 bool cancel;
1690
1691 if (!ev)
1692 return;
1693
1694 /*
1695 * Outer mutex ensures that the first blocker completes canceling
1696 * the event work before further blockers are allowed to finish.
1697 */
1698 mutex_lock(&ev->block_mutex);
1699
1700 spin_lock_irqsave(&ev->lock, flags);
1701 cancel = !ev->block++;
1702 spin_unlock_irqrestore(&ev->lock, flags);
1703
1704 if (cancel)
1705 cancel_delayed_work_sync(&disk->ev->dwork);
1706
1707 mutex_unlock(&ev->block_mutex);
1708 }
1709
__disk_unblock_events(struct gendisk * disk,bool check_now)1710 static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1711 {
1712 struct disk_events *ev = disk->ev;
1713 unsigned long intv;
1714 unsigned long flags;
1715
1716 spin_lock_irqsave(&ev->lock, flags);
1717
1718 if (WARN_ON_ONCE(ev->block <= 0))
1719 goto out_unlock;
1720
1721 if (--ev->block)
1722 goto out_unlock;
1723
1724 intv = disk_events_poll_jiffies(disk);
1725 if (check_now)
1726 queue_delayed_work(system_freezable_power_efficient_wq,
1727 &ev->dwork, 0);
1728 else if (intv)
1729 queue_delayed_work(system_freezable_power_efficient_wq,
1730 &ev->dwork, intv);
1731 out_unlock:
1732 spin_unlock_irqrestore(&ev->lock, flags);
1733 }
1734
1735 /**
1736 * disk_unblock_events - unblock disk event checking
1737 * @disk: disk to unblock events for
1738 *
1739 * Undo disk_block_events(). When the block count reaches zero, it
1740 * starts events polling if configured.
1741 *
1742 * CONTEXT:
1743 * Don't care. Safe to call from irq context.
1744 */
disk_unblock_events(struct gendisk * disk)1745 void disk_unblock_events(struct gendisk *disk)
1746 {
1747 if (disk->ev)
1748 __disk_unblock_events(disk, false);
1749 }
1750
1751 /**
1752 * disk_flush_events - schedule immediate event checking and flushing
1753 * @disk: disk to check and flush events for
1754 * @mask: events to flush
1755 *
1756 * Schedule immediate event checking on @disk if not blocked. Events in
1757 * @mask are scheduled to be cleared from the driver. Note that this
1758 * doesn't clear the events from @disk->ev.
1759 *
1760 * CONTEXT:
1761 * If @mask is non-zero must be called with bdev->bd_mutex held.
1762 */
disk_flush_events(struct gendisk * disk,unsigned int mask)1763 void disk_flush_events(struct gendisk *disk, unsigned int mask)
1764 {
1765 struct disk_events *ev = disk->ev;
1766
1767 if (!ev)
1768 return;
1769
1770 spin_lock_irq(&ev->lock);
1771 ev->clearing |= mask;
1772 if (!ev->block)
1773 mod_delayed_work(system_freezable_power_efficient_wq,
1774 &ev->dwork, 0);
1775 spin_unlock_irq(&ev->lock);
1776 }
1777
1778 /**
1779 * disk_clear_events - synchronously check, clear and return pending events
1780 * @disk: disk to fetch and clear events from
1781 * @mask: mask of events to be fetched and cleared
1782 *
1783 * Disk events are synchronously checked and pending events in @mask
1784 * are cleared and returned. This ignores the block count.
1785 *
1786 * CONTEXT:
1787 * Might sleep.
1788 */
disk_clear_events(struct gendisk * disk,unsigned int mask)1789 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1790 {
1791 const struct block_device_operations *bdops = disk->fops;
1792 struct disk_events *ev = disk->ev;
1793 unsigned int pending;
1794 unsigned int clearing = mask;
1795
1796 if (!ev) {
1797 /* for drivers still using the old ->media_changed method */
1798 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1799 bdops->media_changed && bdops->media_changed(disk))
1800 return DISK_EVENT_MEDIA_CHANGE;
1801 return 0;
1802 }
1803
1804 disk_block_events(disk);
1805
1806 /*
1807 * store the union of mask and ev->clearing on the stack so that the
1808 * race with disk_flush_events does not cause ambiguity (ev->clearing
1809 * can still be modified even if events are blocked).
1810 */
1811 spin_lock_irq(&ev->lock);
1812 clearing |= ev->clearing;
1813 ev->clearing = 0;
1814 spin_unlock_irq(&ev->lock);
1815
1816 disk_check_events(ev, &clearing);
1817 /*
1818 * if ev->clearing is not 0, the disk_flush_events got called in the
1819 * middle of this function, so we want to run the workfn without delay.
1820 */
1821 __disk_unblock_events(disk, ev->clearing ? true : false);
1822
1823 /* then, fetch and clear pending events */
1824 spin_lock_irq(&ev->lock);
1825 pending = ev->pending & mask;
1826 ev->pending &= ~mask;
1827 spin_unlock_irq(&ev->lock);
1828 WARN_ON_ONCE(clearing & mask);
1829
1830 return pending;
1831 }
1832
1833 /*
1834 * Separate this part out so that a different pointer for clearing_ptr can be
1835 * passed in for disk_clear_events.
1836 */
disk_events_workfn(struct work_struct * work)1837 static void disk_events_workfn(struct work_struct *work)
1838 {
1839 struct delayed_work *dwork = to_delayed_work(work);
1840 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1841
1842 disk_check_events(ev, &ev->clearing);
1843 }
1844
disk_check_events(struct disk_events * ev,unsigned int * clearing_ptr)1845 static void disk_check_events(struct disk_events *ev,
1846 unsigned int *clearing_ptr)
1847 {
1848 struct gendisk *disk = ev->disk;
1849 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1850 unsigned int clearing = *clearing_ptr;
1851 unsigned int events;
1852 unsigned long intv;
1853 int nr_events = 0, i;
1854
1855 /* check events */
1856 events = disk->fops->check_events(disk, clearing);
1857
1858 /* accumulate pending events and schedule next poll if necessary */
1859 spin_lock_irq(&ev->lock);
1860
1861 events &= ~ev->pending;
1862 ev->pending |= events;
1863 *clearing_ptr &= ~clearing;
1864
1865 intv = disk_events_poll_jiffies(disk);
1866 if (!ev->block && intv)
1867 queue_delayed_work(system_freezable_power_efficient_wq,
1868 &ev->dwork, intv);
1869
1870 spin_unlock_irq(&ev->lock);
1871
1872 /*
1873 * Tell userland about new events. Only the events listed in
1874 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
1875 * is set. Otherwise, events are processed internally but never
1876 * get reported to userland.
1877 */
1878 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1879 if ((events & disk->events & (1 << i)) &&
1880 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
1881 envp[nr_events++] = disk_uevents[i];
1882
1883 if (nr_events)
1884 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1885 }
1886
1887 /*
1888 * A disk events enabled device has the following sysfs nodes under
1889 * its /sys/block/X/ directory.
1890 *
1891 * events : list of all supported events
1892 * events_async : list of events which can be detected w/o polling
1893 * (always empty, only for backwards compatibility)
1894 * events_poll_msecs : polling interval, 0: disable, -1: system default
1895 */
__disk_events_show(unsigned int events,char * buf)1896 static ssize_t __disk_events_show(unsigned int events, char *buf)
1897 {
1898 const char *delim = "";
1899 ssize_t pos = 0;
1900 int i;
1901
1902 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1903 if (events & (1 << i)) {
1904 pos += sprintf(buf + pos, "%s%s",
1905 delim, disk_events_strs[i]);
1906 delim = " ";
1907 }
1908 if (pos)
1909 pos += sprintf(buf + pos, "\n");
1910 return pos;
1911 }
1912
disk_events_show(struct device * dev,struct device_attribute * attr,char * buf)1913 static ssize_t disk_events_show(struct device *dev,
1914 struct device_attribute *attr, char *buf)
1915 {
1916 struct gendisk *disk = dev_to_disk(dev);
1917
1918 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
1919 return 0;
1920
1921 return __disk_events_show(disk->events, buf);
1922 }
1923
disk_events_async_show(struct device * dev,struct device_attribute * attr,char * buf)1924 static ssize_t disk_events_async_show(struct device *dev,
1925 struct device_attribute *attr, char *buf)
1926 {
1927 return 0;
1928 }
1929
disk_events_poll_msecs_show(struct device * dev,struct device_attribute * attr,char * buf)1930 static ssize_t disk_events_poll_msecs_show(struct device *dev,
1931 struct device_attribute *attr,
1932 char *buf)
1933 {
1934 struct gendisk *disk = dev_to_disk(dev);
1935
1936 if (!disk->ev)
1937 return sprintf(buf, "-1\n");
1938
1939 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1940 }
1941
disk_events_poll_msecs_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)1942 static ssize_t disk_events_poll_msecs_store(struct device *dev,
1943 struct device_attribute *attr,
1944 const char *buf, size_t count)
1945 {
1946 struct gendisk *disk = dev_to_disk(dev);
1947 long intv;
1948
1949 if (!count || !sscanf(buf, "%ld", &intv))
1950 return -EINVAL;
1951
1952 if (intv < 0 && intv != -1)
1953 return -EINVAL;
1954
1955 if (!disk->ev)
1956 return -ENODEV;
1957
1958 disk_block_events(disk);
1959 disk->ev->poll_msecs = intv;
1960 __disk_unblock_events(disk, true);
1961
1962 return count;
1963 }
1964
1965 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
1966 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
1967 static const DEVICE_ATTR(events_poll_msecs, 0644,
1968 disk_events_poll_msecs_show,
1969 disk_events_poll_msecs_store);
1970
1971 static const struct attribute *disk_events_attrs[] = {
1972 &dev_attr_events.attr,
1973 &dev_attr_events_async.attr,
1974 &dev_attr_events_poll_msecs.attr,
1975 NULL,
1976 };
1977
1978 /*
1979 * The default polling interval can be specified by the kernel
1980 * parameter block.events_dfl_poll_msecs which defaults to 0
1981 * (disable). This can also be modified runtime by writing to
1982 * /sys/module/block/parameters/events_dfl_poll_msecs.
1983 */
disk_events_set_dfl_poll_msecs(const char * val,const struct kernel_param * kp)1984 static int disk_events_set_dfl_poll_msecs(const char *val,
1985 const struct kernel_param *kp)
1986 {
1987 struct disk_events *ev;
1988 int ret;
1989
1990 ret = param_set_ulong(val, kp);
1991 if (ret < 0)
1992 return ret;
1993
1994 mutex_lock(&disk_events_mutex);
1995
1996 list_for_each_entry(ev, &disk_events, node)
1997 disk_flush_events(ev->disk, 0);
1998
1999 mutex_unlock(&disk_events_mutex);
2000
2001 return 0;
2002 }
2003
2004 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2005 .set = disk_events_set_dfl_poll_msecs,
2006 .get = param_get_ulong,
2007 };
2008
2009 #undef MODULE_PARAM_PREFIX
2010 #define MODULE_PARAM_PREFIX "block."
2011
2012 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2013 &disk_events_dfl_poll_msecs, 0644);
2014
2015 /*
2016 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2017 */
disk_alloc_events(struct gendisk * disk)2018 static void disk_alloc_events(struct gendisk *disk)
2019 {
2020 struct disk_events *ev;
2021
2022 if (!disk->fops->check_events || !disk->events)
2023 return;
2024
2025 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2026 if (!ev) {
2027 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2028 return;
2029 }
2030
2031 INIT_LIST_HEAD(&ev->node);
2032 ev->disk = disk;
2033 spin_lock_init(&ev->lock);
2034 mutex_init(&ev->block_mutex);
2035 ev->block = 1;
2036 ev->poll_msecs = -1;
2037 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2038
2039 disk->ev = ev;
2040 }
2041
disk_add_events(struct gendisk * disk)2042 static void disk_add_events(struct gendisk *disk)
2043 {
2044 /* FIXME: error handling */
2045 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2046 pr_warn("%s: failed to create sysfs files for events\n",
2047 disk->disk_name);
2048
2049 if (!disk->ev)
2050 return;
2051
2052 mutex_lock(&disk_events_mutex);
2053 list_add_tail(&disk->ev->node, &disk_events);
2054 mutex_unlock(&disk_events_mutex);
2055
2056 /*
2057 * Block count is initialized to 1 and the following initial
2058 * unblock kicks it into action.
2059 */
2060 __disk_unblock_events(disk, true);
2061 }
2062
disk_del_events(struct gendisk * disk)2063 static void disk_del_events(struct gendisk *disk)
2064 {
2065 if (disk->ev) {
2066 disk_block_events(disk);
2067
2068 mutex_lock(&disk_events_mutex);
2069 list_del_init(&disk->ev->node);
2070 mutex_unlock(&disk_events_mutex);
2071 }
2072
2073 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2074 }
2075
disk_release_events(struct gendisk * disk)2076 static void disk_release_events(struct gendisk *disk)
2077 {
2078 /* the block count should be 1 from disk_del_events() */
2079 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
2080 kfree(disk->ev);
2081 }
2082