Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * gendisk handling
4 *
5 * Portions Copyright (C) 2020 Christoph Hellwig
6 */
7
8#include <linux/module.h>
9#include <linux/ctype.h>
10#include <linux/fs.h>
11#include <linux/kdev_t.h>
12#include <linux/kernel.h>
13#include <linux/blkdev.h>
14#include <linux/backing-dev.h>
15#include <linux/init.h>
16#include <linux/spinlock.h>
17#include <linux/proc_fs.h>
18#include <linux/seq_file.h>
19#include <linux/slab.h>
20#include <linux/kmod.h>
21#include <linux/major.h>
22#include <linux/mutex.h>
23#include <linux/idr.h>
24#include <linux/log2.h>
25#include <linux/pm_runtime.h>
26#include <linux/badblocks.h>
27#include <linux/part_stat.h>
28#include "blk-throttle.h"
29
30#include "blk.h"
31#include "blk-mq-sched.h"
32#include "blk-rq-qos.h"
33#include "blk-cgroup.h"
34
35static struct kobject *block_depr;
36
37/*
38 * Unique, monotonically increasing sequential number associated with block
39 * devices instances (i.e. incremented each time a device is attached).
40 * Associating uevents with block devices in userspace is difficult and racy:
41 * the uevent netlink socket is lossy, and on slow and overloaded systems has
42 * a very high latency.
43 * Block devices do not have exclusive owners in userspace, any process can set
44 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
45 * can be reused again and again).
46 * A userspace process setting up a block device and watching for its events
47 * cannot thus reliably tell whether an event relates to the device it just set
48 * up or another earlier instance with the same name.
49 * This sequential number allows userspace processes to solve this problem, and
50 * uniquely associate an uevent to the lifetime to a device.
51 */
52static atomic64_t diskseq;
53
54/* for extended dynamic devt allocation, currently only one major is used */
55#define NR_EXT_DEVT (1 << MINORBITS)
56static DEFINE_IDA(ext_devt_ida);
57
58void set_capacity(struct gendisk *disk, sector_t sectors)
59{
60 struct block_device *bdev = disk->part0;
61
62 spin_lock(&bdev->bd_size_lock);
63 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
64 bdev->bd_nr_sectors = sectors;
65 spin_unlock(&bdev->bd_size_lock);
66}
67EXPORT_SYMBOL(set_capacity);
68
69/*
70 * Set disk capacity and notify if the size is not currently zero and will not
71 * be set to zero. Returns true if a uevent was sent, otherwise false.
72 */
73bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
74{
75 sector_t capacity = get_capacity(disk);
76 char *envp[] = { "RESIZE=1", NULL };
77
78 set_capacity(disk, size);
79
80 /*
81 * Only print a message and send a uevent if the gendisk is user visible
82 * and alive. This avoids spamming the log and udev when setting the
83 * initial capacity during probing.
84 */
85 if (size == capacity ||
86 !disk_live(disk) ||
87 (disk->flags & GENHD_FL_HIDDEN))
88 return false;
89
90 pr_info("%s: detected capacity change from %lld to %lld\n",
91 disk->disk_name, capacity, size);
92
93 /*
94 * Historically we did not send a uevent for changes to/from an empty
95 * device.
96 */
97 if (!capacity || !size)
98 return false;
99 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
100 return true;
101}
102EXPORT_SYMBOL_GPL(set_capacity_and_notify);
103
104static void part_stat_read_all(struct block_device *part,
105 struct disk_stats *stat)
106{
107 int cpu;
108
109 memset(stat, 0, sizeof(struct disk_stats));
110 for_each_possible_cpu(cpu) {
111 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
112 int group;
113
114 for (group = 0; group < NR_STAT_GROUPS; group++) {
115 stat->nsecs[group] += ptr->nsecs[group];
116 stat->sectors[group] += ptr->sectors[group];
117 stat->ios[group] += ptr->ios[group];
118 stat->merges[group] += ptr->merges[group];
119 }
120
121 stat->io_ticks += ptr->io_ticks;
122 }
123}
124
125static unsigned int part_in_flight(struct block_device *part)
126{
127 unsigned int inflight = 0;
128 int cpu;
129
130 for_each_possible_cpu(cpu) {
131 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
132 part_stat_local_read_cpu(part, in_flight[1], cpu);
133 }
134 if ((int)inflight < 0)
135 inflight = 0;
136
137 return inflight;
138}
139
140static void part_in_flight_rw(struct block_device *part,
141 unsigned int inflight[2])
142{
143 int cpu;
144
145 inflight[0] = 0;
146 inflight[1] = 0;
147 for_each_possible_cpu(cpu) {
148 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
149 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
150 }
151 if ((int)inflight[0] < 0)
152 inflight[0] = 0;
153 if ((int)inflight[1] < 0)
154 inflight[1] = 0;
155}
156
157/*
158 * Can be deleted altogether. Later.
159 *
160 */
161#define BLKDEV_MAJOR_HASH_SIZE 255
162static struct blk_major_name {
163 struct blk_major_name *next;
164 int major;
165 char name[16];
166#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
167 void (*probe)(dev_t devt);
168#endif
169} *major_names[BLKDEV_MAJOR_HASH_SIZE];
170static DEFINE_MUTEX(major_names_lock);
171static DEFINE_SPINLOCK(major_names_spinlock);
172
173/* index in the above - for now: assume no multimajor ranges */
174static inline int major_to_index(unsigned major)
175{
176 return major % BLKDEV_MAJOR_HASH_SIZE;
177}
178
179#ifdef CONFIG_PROC_FS
180void blkdev_show(struct seq_file *seqf, off_t offset)
181{
182 struct blk_major_name *dp;
183
184 spin_lock(&major_names_spinlock);
185 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
186 if (dp->major == offset)
187 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
188 spin_unlock(&major_names_spinlock);
189}
190#endif /* CONFIG_PROC_FS */
191
192/**
193 * __register_blkdev - register a new block device
194 *
195 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
196 * @major = 0, try to allocate any unused major number.
197 * @name: the name of the new block device as a zero terminated string
198 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
199 * pre-created device node is accessed. When a probe call uses
200 * add_disk() and it fails the driver must cleanup resources. This
201 * interface may soon be removed.
202 *
203 * The @name must be unique within the system.
204 *
205 * The return value depends on the @major input parameter:
206 *
207 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
208 * then the function returns zero on success, or a negative error code
209 * - if any unused major number was requested with @major = 0 parameter
210 * then the return value is the allocated major number in range
211 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
212 *
213 * See Documentation/admin-guide/devices.txt for the list of allocated
214 * major numbers.
215 *
216 * Use register_blkdev instead for any new code.
217 */
218int __register_blkdev(unsigned int major, const char *name,
219 void (*probe)(dev_t devt))
220{
221 struct blk_major_name **n, *p;
222 int index, ret = 0;
223
224 mutex_lock(&major_names_lock);
225
226 /* temporary */
227 if (major == 0) {
228 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
229 if (major_names[index] == NULL)
230 break;
231 }
232
233 if (index == 0) {
234 printk("%s: failed to get major for %s\n",
235 __func__, name);
236 ret = -EBUSY;
237 goto out;
238 }
239 major = index;
240 ret = major;
241 }
242
243 if (major >= BLKDEV_MAJOR_MAX) {
244 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
245 __func__, major, BLKDEV_MAJOR_MAX-1, name);
246
247 ret = -EINVAL;
248 goto out;
249 }
250
251 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
252 if (p == NULL) {
253 ret = -ENOMEM;
254 goto out;
255 }
256
257 p->major = major;
258#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
259 p->probe = probe;
260#endif
261 strlcpy(p->name, name, sizeof(p->name));
262 p->next = NULL;
263 index = major_to_index(major);
264
265 spin_lock(&major_names_spinlock);
266 for (n = &major_names[index]; *n; n = &(*n)->next) {
267 if ((*n)->major == major)
268 break;
269 }
270 if (!*n)
271 *n = p;
272 else
273 ret = -EBUSY;
274 spin_unlock(&major_names_spinlock);
275
276 if (ret < 0) {
277 printk("register_blkdev: cannot get major %u for %s\n",
278 major, name);
279 kfree(p);
280 }
281out:
282 mutex_unlock(&major_names_lock);
283 return ret;
284}
285EXPORT_SYMBOL(__register_blkdev);
286
287void unregister_blkdev(unsigned int major, const char *name)
288{
289 struct blk_major_name **n;
290 struct blk_major_name *p = NULL;
291 int index = major_to_index(major);
292
293 mutex_lock(&major_names_lock);
294 spin_lock(&major_names_spinlock);
295 for (n = &major_names[index]; *n; n = &(*n)->next)
296 if ((*n)->major == major)
297 break;
298 if (!*n || strcmp((*n)->name, name)) {
299 WARN_ON(1);
300 } else {
301 p = *n;
302 *n = p->next;
303 }
304 spin_unlock(&major_names_spinlock);
305 mutex_unlock(&major_names_lock);
306 kfree(p);
307}
308
309EXPORT_SYMBOL(unregister_blkdev);
310
311int blk_alloc_ext_minor(void)
312{
313 int idx;
314
315 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
316 if (idx == -ENOSPC)
317 return -EBUSY;
318 return idx;
319}
320
321void blk_free_ext_minor(unsigned int minor)
322{
323 ida_free(&ext_devt_ida, minor);
324}
325
326static char *bdevt_str(dev_t devt, char *buf)
327{
328 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
329 char tbuf[BDEVT_SIZE];
330 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
331 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
332 } else
333 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
334
335 return buf;
336}
337
338void disk_uevent(struct gendisk *disk, enum kobject_action action)
339{
340 struct block_device *part;
341 unsigned long idx;
342
343 rcu_read_lock();
344 xa_for_each(&disk->part_tbl, idx, part) {
345 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
346 continue;
347 if (!kobject_get_unless_zero(&part->bd_device.kobj))
348 continue;
349
350 rcu_read_unlock();
351 kobject_uevent(bdev_kobj(part), action);
352 put_device(&part->bd_device);
353 rcu_read_lock();
354 }
355 rcu_read_unlock();
356}
357EXPORT_SYMBOL_GPL(disk_uevent);
358
359int disk_scan_partitions(struct gendisk *disk, fmode_t mode, void *owner)
360{
361 struct block_device *bdev;
362
363 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
364 return -EINVAL;
365 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
366 return -EINVAL;
367 if (disk->open_partitions)
368 return -EBUSY;
369 /* Someone else has bdev exclusively open? */
370 if (disk->part0->bd_holder && disk->part0->bd_holder != owner)
371 return -EBUSY;
372
373 set_bit(GD_NEED_PART_SCAN, &disk->state);
374 bdev = blkdev_get_by_dev(disk_devt(disk), mode, NULL);
375 if (IS_ERR(bdev))
376 return PTR_ERR(bdev);
377 blkdev_put(bdev, mode);
378 return 0;
379}
380
381/**
382 * device_add_disk - add disk information to kernel list
383 * @parent: parent device for the disk
384 * @disk: per-device partitioning information
385 * @groups: Additional per-device sysfs groups
386 *
387 * This function registers the partitioning information in @disk
388 * with the kernel.
389 */
390int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
391 const struct attribute_group **groups)
392
393{
394 struct device *ddev = disk_to_dev(disk);
395 int ret;
396
397 /* Only makes sense for bio-based to set ->poll_bio */
398 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
399 return -EINVAL;
400
401 /*
402 * The disk queue should now be all set with enough information about
403 * the device for the elevator code to pick an adequate default
404 * elevator if one is needed, that is, for devices requesting queue
405 * registration.
406 */
407 elevator_init_mq(disk->queue);
408
409 /*
410 * If the driver provides an explicit major number it also must provide
411 * the number of minors numbers supported, and those will be used to
412 * setup the gendisk.
413 * Otherwise just allocate the device numbers for both the whole device
414 * and all partitions from the extended dev_t space.
415 */
416 ret = -EINVAL;
417 if (disk->major) {
418 if (WARN_ON(!disk->minors))
419 goto out_exit_elevator;
420
421 if (disk->minors > DISK_MAX_PARTS) {
422 pr_err("block: can't allocate more than %d partitions\n",
423 DISK_MAX_PARTS);
424 disk->minors = DISK_MAX_PARTS;
425 }
426 if (disk->first_minor + disk->minors > MINORMASK + 1)
427 goto out_exit_elevator;
428 } else {
429 if (WARN_ON(disk->minors))
430 goto out_exit_elevator;
431
432 ret = blk_alloc_ext_minor();
433 if (ret < 0)
434 goto out_exit_elevator;
435 disk->major = BLOCK_EXT_MAJOR;
436 disk->first_minor = ret;
437 }
438
439 /* delay uevents, until we scanned partition table */
440 dev_set_uevent_suppress(ddev, 1);
441
442 ddev->parent = parent;
443 ddev->groups = groups;
444 dev_set_name(ddev, "%s", disk->disk_name);
445 if (!(disk->flags & GENHD_FL_HIDDEN))
446 ddev->devt = MKDEV(disk->major, disk->first_minor);
447 ret = device_add(ddev);
448 if (ret)
449 goto out_free_ext_minor;
450
451 ret = disk_alloc_events(disk);
452 if (ret)
453 goto out_device_del;
454
455 if (!sysfs_deprecated) {
456 ret = sysfs_create_link(block_depr, &ddev->kobj,
457 kobject_name(&ddev->kobj));
458 if (ret)
459 goto out_device_del;
460 }
461
462 /*
463 * avoid probable deadlock caused by allocating memory with
464 * GFP_KERNEL in runtime_resume callback of its all ancestor
465 * devices
466 */
467 pm_runtime_set_memalloc_noio(ddev, true);
468
469 ret = blk_integrity_add(disk);
470 if (ret)
471 goto out_del_block_link;
472
473 disk->part0->bd_holder_dir =
474 kobject_create_and_add("holders", &ddev->kobj);
475 if (!disk->part0->bd_holder_dir) {
476 ret = -ENOMEM;
477 goto out_del_integrity;
478 }
479 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
480 if (!disk->slave_dir) {
481 ret = -ENOMEM;
482 goto out_put_holder_dir;
483 }
484
485 ret = blk_register_queue(disk);
486 if (ret)
487 goto out_put_slave_dir;
488
489 if (!(disk->flags & GENHD_FL_HIDDEN)) {
490 ret = bdi_register(disk->bdi, "%u:%u",
491 disk->major, disk->first_minor);
492 if (ret)
493 goto out_unregister_queue;
494 bdi_set_owner(disk->bdi, ddev);
495 ret = sysfs_create_link(&ddev->kobj,
496 &disk->bdi->dev->kobj, "bdi");
497 if (ret)
498 goto out_unregister_bdi;
499
500 bdev_add(disk->part0, ddev->devt);
501 if (get_capacity(disk))
502 disk_scan_partitions(disk, FMODE_READ, NULL);
503
504 /*
505 * Announce the disk and partitions after all partitions are
506 * created. (for hidden disks uevents remain suppressed forever)
507 */
508 dev_set_uevent_suppress(ddev, 0);
509 disk_uevent(disk, KOBJ_ADD);
510 } else {
511 /*
512 * Even if the block_device for a hidden gendisk is not
513 * registered, it needs to have a valid bd_dev so that the
514 * freeing of the dynamic major works.
515 */
516 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
517 }
518
519 disk_update_readahead(disk);
520 disk_add_events(disk);
521 set_bit(GD_ADDED, &disk->state);
522 return 0;
523
524out_unregister_bdi:
525 if (!(disk->flags & GENHD_FL_HIDDEN))
526 bdi_unregister(disk->bdi);
527out_unregister_queue:
528 blk_unregister_queue(disk);
529 rq_qos_exit(disk->queue);
530out_put_slave_dir:
531 kobject_put(disk->slave_dir);
532 disk->slave_dir = NULL;
533out_put_holder_dir:
534 kobject_put(disk->part0->bd_holder_dir);
535out_del_integrity:
536 blk_integrity_del(disk);
537out_del_block_link:
538 if (!sysfs_deprecated)
539 sysfs_remove_link(block_depr, dev_name(ddev));
540out_device_del:
541 device_del(ddev);
542out_free_ext_minor:
543 if (disk->major == BLOCK_EXT_MAJOR)
544 blk_free_ext_minor(disk->first_minor);
545out_exit_elevator:
546 if (disk->queue->elevator)
547 elevator_exit(disk->queue);
548 return ret;
549}
550EXPORT_SYMBOL(device_add_disk);
551
552/**
553 * blk_mark_disk_dead - mark a disk as dead
554 * @disk: disk to mark as dead
555 *
556 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
557 * to this disk.
558 */
559void blk_mark_disk_dead(struct gendisk *disk)
560{
561 set_bit(GD_DEAD, &disk->state);
562 blk_queue_start_drain(disk->queue);
563
564 /*
565 * Stop buffered writers from dirtying pages that can't be written out.
566 */
567 set_capacity_and_notify(disk, 0);
568}
569EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
570
571/**
572 * del_gendisk - remove the gendisk
573 * @disk: the struct gendisk to remove
574 *
575 * Removes the gendisk and all its associated resources. This deletes the
576 * partitions associated with the gendisk, and unregisters the associated
577 * request_queue.
578 *
579 * This is the counter to the respective __device_add_disk() call.
580 *
581 * The final removal of the struct gendisk happens when its refcount reaches 0
582 * with put_disk(), which should be called after del_gendisk(), if
583 * __device_add_disk() was used.
584 *
585 * Drivers exist which depend on the release of the gendisk to be synchronous,
586 * it should not be deferred.
587 *
588 * Context: can sleep
589 */
590void del_gendisk(struct gendisk *disk)
591{
592 struct request_queue *q = disk->queue;
593
594 might_sleep();
595
596 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
597 return;
598
599 blk_integrity_del(disk);
600 disk_del_events(disk);
601
602 mutex_lock(&disk->open_mutex);
603 remove_inode_hash(disk->part0->bd_inode);
604 blk_drop_partitions(disk);
605 mutex_unlock(&disk->open_mutex);
606
607 fsync_bdev(disk->part0);
608 __invalidate_device(disk->part0, true);
609
610 /*
611 * Fail any new I/O.
612 */
613 set_bit(GD_DEAD, &disk->state);
614 if (test_bit(GD_OWNS_QUEUE, &disk->state))
615 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
616 set_capacity(disk, 0);
617
618 /*
619 * Prevent new I/O from crossing bio_queue_enter().
620 */
621 blk_queue_start_drain(q);
622
623 if (!(disk->flags & GENHD_FL_HIDDEN)) {
624 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
625
626 /*
627 * Unregister bdi before releasing device numbers (as they can
628 * get reused and we'd get clashes in sysfs).
629 */
630 bdi_unregister(disk->bdi);
631 }
632
633 blk_unregister_queue(disk);
634
635 kobject_put(disk->part0->bd_holder_dir);
636 kobject_put(disk->slave_dir);
637 disk->slave_dir = NULL;
638
639 part_stat_set_all(disk->part0, 0);
640 disk->part0->bd_stamp = 0;
641 if (!sysfs_deprecated)
642 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
643 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
644 device_del(disk_to_dev(disk));
645
646 blk_mq_freeze_queue_wait(q);
647
648 blk_throtl_cancel_bios(disk);
649
650 blk_sync_queue(q);
651 blk_flush_integrity();
652
653 if (queue_is_mq(q))
654 blk_mq_cancel_work_sync(q);
655
656 blk_mq_quiesce_queue(q);
657 if (q->elevator) {
658 mutex_lock(&q->sysfs_lock);
659 elevator_exit(q);
660 mutex_unlock(&q->sysfs_lock);
661 }
662 rq_qos_exit(q);
663 blk_mq_unquiesce_queue(q);
664
665 /*
666 * If the disk does not own the queue, allow using passthrough requests
667 * again. Else leave the queue frozen to fail all I/O.
668 */
669 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
670 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
671 __blk_mq_unfreeze_queue(q, true);
672 } else {
673 if (queue_is_mq(q))
674 blk_mq_exit_queue(q);
675 }
676}
677EXPORT_SYMBOL(del_gendisk);
678
679/**
680 * invalidate_disk - invalidate the disk
681 * @disk: the struct gendisk to invalidate
682 *
683 * A helper to invalidates the disk. It will clean the disk's associated
684 * buffer/page caches and reset its internal states so that the disk
685 * can be reused by the drivers.
686 *
687 * Context: can sleep
688 */
689void invalidate_disk(struct gendisk *disk)
690{
691 struct block_device *bdev = disk->part0;
692
693 invalidate_bdev(bdev);
694 bdev->bd_inode->i_mapping->wb_err = 0;
695 set_capacity(disk, 0);
696}
697EXPORT_SYMBOL(invalidate_disk);
698
699/* sysfs access to bad-blocks list. */
700static ssize_t disk_badblocks_show(struct device *dev,
701 struct device_attribute *attr,
702 char *page)
703{
704 struct gendisk *disk = dev_to_disk(dev);
705
706 if (!disk->bb)
707 return sprintf(page, "\n");
708
709 return badblocks_show(disk->bb, page, 0);
710}
711
712static ssize_t disk_badblocks_store(struct device *dev,
713 struct device_attribute *attr,
714 const char *page, size_t len)
715{
716 struct gendisk *disk = dev_to_disk(dev);
717
718 if (!disk->bb)
719 return -ENXIO;
720
721 return badblocks_store(disk->bb, page, len, 0);
722}
723
724#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
725void blk_request_module(dev_t devt)
726{
727 unsigned int major = MAJOR(devt);
728 struct blk_major_name **n;
729
730 mutex_lock(&major_names_lock);
731 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
732 if ((*n)->major == major && (*n)->probe) {
733 (*n)->probe(devt);
734 mutex_unlock(&major_names_lock);
735 return;
736 }
737 }
738 mutex_unlock(&major_names_lock);
739
740 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
741 /* Make old-style 2.4 aliases work */
742 request_module("block-major-%d", MAJOR(devt));
743}
744#endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
745
746/*
747 * print a full list of all partitions - intended for places where the root
748 * filesystem can't be mounted and thus to give the victim some idea of what
749 * went wrong
750 */
751void __init printk_all_partitions(void)
752{
753 struct class_dev_iter iter;
754 struct device *dev;
755
756 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
757 while ((dev = class_dev_iter_next(&iter))) {
758 struct gendisk *disk = dev_to_disk(dev);
759 struct block_device *part;
760 char devt_buf[BDEVT_SIZE];
761 unsigned long idx;
762
763 /*
764 * Don't show empty devices or things that have been
765 * suppressed
766 */
767 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
768 continue;
769
770 /*
771 * Note, unlike /proc/partitions, I am showing the numbers in
772 * hex - the same format as the root= option takes.
773 */
774 rcu_read_lock();
775 xa_for_each(&disk->part_tbl, idx, part) {
776 if (!bdev_nr_sectors(part))
777 continue;
778 printk("%s%s %10llu %pg %s",
779 bdev_is_partition(part) ? " " : "",
780 bdevt_str(part->bd_dev, devt_buf),
781 bdev_nr_sectors(part) >> 1, part,
782 part->bd_meta_info ?
783 part->bd_meta_info->uuid : "");
784 if (bdev_is_partition(part))
785 printk("\n");
786 else if (dev->parent && dev->parent->driver)
787 printk(" driver: %s\n",
788 dev->parent->driver->name);
789 else
790 printk(" (driver?)\n");
791 }
792 rcu_read_unlock();
793 }
794 class_dev_iter_exit(&iter);
795}
796
797#ifdef CONFIG_PROC_FS
798/* iterator */
799static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
800{
801 loff_t skip = *pos;
802 struct class_dev_iter *iter;
803 struct device *dev;
804
805 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
806 if (!iter)
807 return ERR_PTR(-ENOMEM);
808
809 seqf->private = iter;
810 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
811 do {
812 dev = class_dev_iter_next(iter);
813 if (!dev)
814 return NULL;
815 } while (skip--);
816
817 return dev_to_disk(dev);
818}
819
820static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
821{
822 struct device *dev;
823
824 (*pos)++;
825 dev = class_dev_iter_next(seqf->private);
826 if (dev)
827 return dev_to_disk(dev);
828
829 return NULL;
830}
831
832static void disk_seqf_stop(struct seq_file *seqf, void *v)
833{
834 struct class_dev_iter *iter = seqf->private;
835
836 /* stop is called even after start failed :-( */
837 if (iter) {
838 class_dev_iter_exit(iter);
839 kfree(iter);
840 seqf->private = NULL;
841 }
842}
843
844static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
845{
846 void *p;
847
848 p = disk_seqf_start(seqf, pos);
849 if (!IS_ERR_OR_NULL(p) && !*pos)
850 seq_puts(seqf, "major minor #blocks name\n\n");
851 return p;
852}
853
854static int show_partition(struct seq_file *seqf, void *v)
855{
856 struct gendisk *sgp = v;
857 struct block_device *part;
858 unsigned long idx;
859
860 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
861 return 0;
862
863 rcu_read_lock();
864 xa_for_each(&sgp->part_tbl, idx, part) {
865 if (!bdev_nr_sectors(part))
866 continue;
867 seq_printf(seqf, "%4d %7d %10llu %pg\n",
868 MAJOR(part->bd_dev), MINOR(part->bd_dev),
869 bdev_nr_sectors(part) >> 1, part);
870 }
871 rcu_read_unlock();
872 return 0;
873}
874
875static const struct seq_operations partitions_op = {
876 .start = show_partition_start,
877 .next = disk_seqf_next,
878 .stop = disk_seqf_stop,
879 .show = show_partition
880};
881#endif
882
883static int __init genhd_device_init(void)
884{
885 int error;
886
887 block_class.dev_kobj = sysfs_dev_block_kobj;
888 error = class_register(&block_class);
889 if (unlikely(error))
890 return error;
891 blk_dev_init();
892
893 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
894
895 /* create top-level block dir */
896 if (!sysfs_deprecated)
897 block_depr = kobject_create_and_add("block", NULL);
898 return 0;
899}
900
901subsys_initcall(genhd_device_init);
902
903static ssize_t disk_range_show(struct device *dev,
904 struct device_attribute *attr, char *buf)
905{
906 struct gendisk *disk = dev_to_disk(dev);
907
908 return sprintf(buf, "%d\n", disk->minors);
909}
910
911static ssize_t disk_ext_range_show(struct device *dev,
912 struct device_attribute *attr, char *buf)
913{
914 struct gendisk *disk = dev_to_disk(dev);
915
916 return sprintf(buf, "%d\n",
917 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
918}
919
920static ssize_t disk_removable_show(struct device *dev,
921 struct device_attribute *attr, char *buf)
922{
923 struct gendisk *disk = dev_to_disk(dev);
924
925 return sprintf(buf, "%d\n",
926 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
927}
928
929static ssize_t disk_hidden_show(struct device *dev,
930 struct device_attribute *attr, char *buf)
931{
932 struct gendisk *disk = dev_to_disk(dev);
933
934 return sprintf(buf, "%d\n",
935 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
936}
937
938static ssize_t disk_ro_show(struct device *dev,
939 struct device_attribute *attr, char *buf)
940{
941 struct gendisk *disk = dev_to_disk(dev);
942
943 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
944}
945
946ssize_t part_size_show(struct device *dev,
947 struct device_attribute *attr, char *buf)
948{
949 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
950}
951
952ssize_t part_stat_show(struct device *dev,
953 struct device_attribute *attr, char *buf)
954{
955 struct block_device *bdev = dev_to_bdev(dev);
956 struct request_queue *q = bdev_get_queue(bdev);
957 struct disk_stats stat;
958 unsigned int inflight;
959
960 if (queue_is_mq(q))
961 inflight = blk_mq_in_flight(q, bdev);
962 else
963 inflight = part_in_flight(bdev);
964
965 if (inflight) {
966 part_stat_lock();
967 update_io_ticks(bdev, jiffies, true);
968 part_stat_unlock();
969 }
970 part_stat_read_all(bdev, &stat);
971 return sprintf(buf,
972 "%8lu %8lu %8llu %8u "
973 "%8lu %8lu %8llu %8u "
974 "%8u %8u %8u "
975 "%8lu %8lu %8llu %8u "
976 "%8lu %8u"
977 "\n",
978 stat.ios[STAT_READ],
979 stat.merges[STAT_READ],
980 (unsigned long long)stat.sectors[STAT_READ],
981 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
982 stat.ios[STAT_WRITE],
983 stat.merges[STAT_WRITE],
984 (unsigned long long)stat.sectors[STAT_WRITE],
985 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
986 inflight,
987 jiffies_to_msecs(stat.io_ticks),
988 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
989 stat.nsecs[STAT_WRITE] +
990 stat.nsecs[STAT_DISCARD] +
991 stat.nsecs[STAT_FLUSH],
992 NSEC_PER_MSEC),
993 stat.ios[STAT_DISCARD],
994 stat.merges[STAT_DISCARD],
995 (unsigned long long)stat.sectors[STAT_DISCARD],
996 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
997 stat.ios[STAT_FLUSH],
998 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
999}
1000
1001ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1002 char *buf)
1003{
1004 struct block_device *bdev = dev_to_bdev(dev);
1005 struct request_queue *q = bdev_get_queue(bdev);
1006 unsigned int inflight[2];
1007
1008 if (queue_is_mq(q))
1009 blk_mq_in_flight_rw(q, bdev, inflight);
1010 else
1011 part_in_flight_rw(bdev, inflight);
1012
1013 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1014}
1015
1016static ssize_t disk_capability_show(struct device *dev,
1017 struct device_attribute *attr, char *buf)
1018{
1019 struct gendisk *disk = dev_to_disk(dev);
1020
1021 return sprintf(buf, "%x\n", disk->flags);
1022}
1023
1024static ssize_t disk_alignment_offset_show(struct device *dev,
1025 struct device_attribute *attr,
1026 char *buf)
1027{
1028 struct gendisk *disk = dev_to_disk(dev);
1029
1030 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1031}
1032
1033static ssize_t disk_discard_alignment_show(struct device *dev,
1034 struct device_attribute *attr,
1035 char *buf)
1036{
1037 struct gendisk *disk = dev_to_disk(dev);
1038
1039 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1040}
1041
1042static ssize_t diskseq_show(struct device *dev,
1043 struct device_attribute *attr, char *buf)
1044{
1045 struct gendisk *disk = dev_to_disk(dev);
1046
1047 return sprintf(buf, "%llu\n", disk->diskseq);
1048}
1049
1050static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1051static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1052static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1053static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1054static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1055static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1056static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1057static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1058static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1059static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1060static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1061static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1062static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1063
1064#ifdef CONFIG_FAIL_MAKE_REQUEST
1065ssize_t part_fail_show(struct device *dev,
1066 struct device_attribute *attr, char *buf)
1067{
1068 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1069}
1070
1071ssize_t part_fail_store(struct device *dev,
1072 struct device_attribute *attr,
1073 const char *buf, size_t count)
1074{
1075 int i;
1076
1077 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1078 dev_to_bdev(dev)->bd_make_it_fail = i;
1079
1080 return count;
1081}
1082
1083static struct device_attribute dev_attr_fail =
1084 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1085#endif /* CONFIG_FAIL_MAKE_REQUEST */
1086
1087#ifdef CONFIG_FAIL_IO_TIMEOUT
1088static struct device_attribute dev_attr_fail_timeout =
1089 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1090#endif
1091
1092static struct attribute *disk_attrs[] = {
1093 &dev_attr_range.attr,
1094 &dev_attr_ext_range.attr,
1095 &dev_attr_removable.attr,
1096 &dev_attr_hidden.attr,
1097 &dev_attr_ro.attr,
1098 &dev_attr_size.attr,
1099 &dev_attr_alignment_offset.attr,
1100 &dev_attr_discard_alignment.attr,
1101 &dev_attr_capability.attr,
1102 &dev_attr_stat.attr,
1103 &dev_attr_inflight.attr,
1104 &dev_attr_badblocks.attr,
1105 &dev_attr_events.attr,
1106 &dev_attr_events_async.attr,
1107 &dev_attr_events_poll_msecs.attr,
1108 &dev_attr_diskseq.attr,
1109#ifdef CONFIG_FAIL_MAKE_REQUEST
1110 &dev_attr_fail.attr,
1111#endif
1112#ifdef CONFIG_FAIL_IO_TIMEOUT
1113 &dev_attr_fail_timeout.attr,
1114#endif
1115 NULL
1116};
1117
1118static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1119{
1120 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1121 struct gendisk *disk = dev_to_disk(dev);
1122
1123 if (a == &dev_attr_badblocks.attr && !disk->bb)
1124 return 0;
1125 return a->mode;
1126}
1127
1128static struct attribute_group disk_attr_group = {
1129 .attrs = disk_attrs,
1130 .is_visible = disk_visible,
1131};
1132
1133static const struct attribute_group *disk_attr_groups[] = {
1134 &disk_attr_group,
1135#ifdef CONFIG_BLK_DEV_IO_TRACE
1136 &blk_trace_attr_group,
1137#endif
1138 NULL
1139};
1140
1141/**
1142 * disk_release - releases all allocated resources of the gendisk
1143 * @dev: the device representing this disk
1144 *
1145 * This function releases all allocated resources of the gendisk.
1146 *
1147 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1148 * assigned. Since the request_queue sits on top of the gendisk for these
1149 * drivers we also call blk_put_queue() for them, and we expect the
1150 * request_queue refcount to reach 0 at this point, and so the request_queue
1151 * will also be freed prior to the disk.
1152 *
1153 * Context: can sleep
1154 */
1155static void disk_release(struct device *dev)
1156{
1157 struct gendisk *disk = dev_to_disk(dev);
1158
1159 might_sleep();
1160 WARN_ON_ONCE(disk_live(disk));
1161
1162 /*
1163 * To undo the all initialization from blk_mq_init_allocated_queue in
1164 * case of a probe failure where add_disk is never called we have to
1165 * call blk_mq_exit_queue here. We can't do this for the more common
1166 * teardown case (yet) as the tagset can be gone by the time the disk
1167 * is released once it was added.
1168 */
1169 if (queue_is_mq(disk->queue) &&
1170 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1171 !test_bit(GD_ADDED, &disk->state))
1172 blk_mq_exit_queue(disk->queue);
1173
1174 blkcg_exit_disk(disk);
1175
1176 bioset_exit(&disk->bio_split);
1177
1178 disk_release_events(disk);
1179 kfree(disk->random);
1180 disk_free_zone_bitmaps(disk);
1181 xa_destroy(&disk->part_tbl);
1182
1183 disk->queue->disk = NULL;
1184 blk_put_queue(disk->queue);
1185
1186 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1187 disk->fops->free_disk(disk);
1188
1189 iput(disk->part0->bd_inode); /* frees the disk */
1190}
1191
1192static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1193{
1194 const struct gendisk *disk = dev_to_disk(dev);
1195
1196 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1197}
1198
1199struct class block_class = {
1200 .name = "block",
1201 .dev_uevent = block_uevent,
1202};
1203
1204static char *block_devnode(struct device *dev, umode_t *mode,
1205 kuid_t *uid, kgid_t *gid)
1206{
1207 struct gendisk *disk = dev_to_disk(dev);
1208
1209 if (disk->fops->devnode)
1210 return disk->fops->devnode(disk, mode);
1211 return NULL;
1212}
1213
1214const struct device_type disk_type = {
1215 .name = "disk",
1216 .groups = disk_attr_groups,
1217 .release = disk_release,
1218 .devnode = block_devnode,
1219};
1220
1221#ifdef CONFIG_PROC_FS
1222/*
1223 * aggregate disk stat collector. Uses the same stats that the sysfs
1224 * entries do, above, but makes them available through one seq_file.
1225 *
1226 * The output looks suspiciously like /proc/partitions with a bunch of
1227 * extra fields.
1228 */
1229static int diskstats_show(struct seq_file *seqf, void *v)
1230{
1231 struct gendisk *gp = v;
1232 struct block_device *hd;
1233 unsigned int inflight;
1234 struct disk_stats stat;
1235 unsigned long idx;
1236
1237 /*
1238 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1239 seq_puts(seqf, "major minor name"
1240 " rio rmerge rsect ruse wio wmerge "
1241 "wsect wuse running use aveq"
1242 "\n\n");
1243 */
1244
1245 rcu_read_lock();
1246 xa_for_each(&gp->part_tbl, idx, hd) {
1247 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1248 continue;
1249 if (queue_is_mq(gp->queue))
1250 inflight = blk_mq_in_flight(gp->queue, hd);
1251 else
1252 inflight = part_in_flight(hd);
1253
1254 if (inflight) {
1255 part_stat_lock();
1256 update_io_ticks(hd, jiffies, true);
1257 part_stat_unlock();
1258 }
1259 part_stat_read_all(hd, &stat);
1260 seq_printf(seqf, "%4d %7d %pg "
1261 "%lu %lu %lu %u "
1262 "%lu %lu %lu %u "
1263 "%u %u %u "
1264 "%lu %lu %lu %u "
1265 "%lu %u"
1266 "\n",
1267 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1268 stat.ios[STAT_READ],
1269 stat.merges[STAT_READ],
1270 stat.sectors[STAT_READ],
1271 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1272 NSEC_PER_MSEC),
1273 stat.ios[STAT_WRITE],
1274 stat.merges[STAT_WRITE],
1275 stat.sectors[STAT_WRITE],
1276 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1277 NSEC_PER_MSEC),
1278 inflight,
1279 jiffies_to_msecs(stat.io_ticks),
1280 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1281 stat.nsecs[STAT_WRITE] +
1282 stat.nsecs[STAT_DISCARD] +
1283 stat.nsecs[STAT_FLUSH],
1284 NSEC_PER_MSEC),
1285 stat.ios[STAT_DISCARD],
1286 stat.merges[STAT_DISCARD],
1287 stat.sectors[STAT_DISCARD],
1288 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1289 NSEC_PER_MSEC),
1290 stat.ios[STAT_FLUSH],
1291 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1292 NSEC_PER_MSEC)
1293 );
1294 }
1295 rcu_read_unlock();
1296
1297 return 0;
1298}
1299
1300static const struct seq_operations diskstats_op = {
1301 .start = disk_seqf_start,
1302 .next = disk_seqf_next,
1303 .stop = disk_seqf_stop,
1304 .show = diskstats_show
1305};
1306
1307static int __init proc_genhd_init(void)
1308{
1309 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1310 proc_create_seq("partitions", 0, NULL, &partitions_op);
1311 return 0;
1312}
1313module_init(proc_genhd_init);
1314#endif /* CONFIG_PROC_FS */
1315
1316dev_t part_devt(struct gendisk *disk, u8 partno)
1317{
1318 struct block_device *part;
1319 dev_t devt = 0;
1320
1321 rcu_read_lock();
1322 part = xa_load(&disk->part_tbl, partno);
1323 if (part)
1324 devt = part->bd_dev;
1325 rcu_read_unlock();
1326
1327 return devt;
1328}
1329
1330dev_t blk_lookup_devt(const char *name, int partno)
1331{
1332 dev_t devt = MKDEV(0, 0);
1333 struct class_dev_iter iter;
1334 struct device *dev;
1335
1336 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1337 while ((dev = class_dev_iter_next(&iter))) {
1338 struct gendisk *disk = dev_to_disk(dev);
1339
1340 if (strcmp(dev_name(dev), name))
1341 continue;
1342
1343 if (partno < disk->minors) {
1344 /* We need to return the right devno, even
1345 * if the partition doesn't exist yet.
1346 */
1347 devt = MKDEV(MAJOR(dev->devt),
1348 MINOR(dev->devt) + partno);
1349 } else {
1350 devt = part_devt(disk, partno);
1351 if (devt)
1352 break;
1353 }
1354 }
1355 class_dev_iter_exit(&iter);
1356 return devt;
1357}
1358
1359struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1360 struct lock_class_key *lkclass)
1361{
1362 struct gendisk *disk;
1363
1364 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1365 if (!disk)
1366 return NULL;
1367
1368 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1369 goto out_free_disk;
1370
1371 disk->bdi = bdi_alloc(node_id);
1372 if (!disk->bdi)
1373 goto out_free_bioset;
1374
1375 /* bdev_alloc() might need the queue, set before the first call */
1376 disk->queue = q;
1377
1378 disk->part0 = bdev_alloc(disk, 0);
1379 if (!disk->part0)
1380 goto out_free_bdi;
1381
1382 disk->node_id = node_id;
1383 mutex_init(&disk->open_mutex);
1384 xa_init(&disk->part_tbl);
1385 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1386 goto out_destroy_part_tbl;
1387
1388 if (blkcg_init_disk(disk))
1389 goto out_erase_part0;
1390
1391 rand_initialize_disk(disk);
1392 disk_to_dev(disk)->class = &block_class;
1393 disk_to_dev(disk)->type = &disk_type;
1394 device_initialize(disk_to_dev(disk));
1395 inc_diskseq(disk);
1396 q->disk = disk;
1397 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1398#ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1399 INIT_LIST_HEAD(&disk->slave_bdevs);
1400#endif
1401 return disk;
1402
1403out_erase_part0:
1404 xa_erase(&disk->part_tbl, 0);
1405out_destroy_part_tbl:
1406 xa_destroy(&disk->part_tbl);
1407 disk->part0->bd_disk = NULL;
1408 iput(disk->part0->bd_inode);
1409out_free_bdi:
1410 bdi_put(disk->bdi);
1411out_free_bioset:
1412 bioset_exit(&disk->bio_split);
1413out_free_disk:
1414 kfree(disk);
1415 return NULL;
1416}
1417
1418struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1419{
1420 struct request_queue *q;
1421 struct gendisk *disk;
1422
1423 q = blk_alloc_queue(node);
1424 if (!q)
1425 return NULL;
1426
1427 disk = __alloc_disk_node(q, node, lkclass);
1428 if (!disk) {
1429 blk_put_queue(q);
1430 return NULL;
1431 }
1432 set_bit(GD_OWNS_QUEUE, &disk->state);
1433 return disk;
1434}
1435EXPORT_SYMBOL(__blk_alloc_disk);
1436
1437/**
1438 * put_disk - decrements the gendisk refcount
1439 * @disk: the struct gendisk to decrement the refcount for
1440 *
1441 * This decrements the refcount for the struct gendisk. When this reaches 0
1442 * we'll have disk_release() called.
1443 *
1444 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1445 * when handling probe errors (that is before add_disk() is called).
1446 *
1447 * Context: Any context, but the last reference must not be dropped from
1448 * atomic context.
1449 */
1450void put_disk(struct gendisk *disk)
1451{
1452 if (disk)
1453 put_device(disk_to_dev(disk));
1454}
1455EXPORT_SYMBOL(put_disk);
1456
1457static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1458{
1459 char event[] = "DISK_RO=1";
1460 char *envp[] = { event, NULL };
1461
1462 if (!ro)
1463 event[8] = '0';
1464 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1465}
1466
1467/**
1468 * set_disk_ro - set a gendisk read-only
1469 * @disk: gendisk to operate on
1470 * @read_only: %true to set the disk read-only, %false set the disk read/write
1471 *
1472 * This function is used to indicate whether a given disk device should have its
1473 * read-only flag set. set_disk_ro() is typically used by device drivers to
1474 * indicate whether the underlying physical device is write-protected.
1475 */
1476void set_disk_ro(struct gendisk *disk, bool read_only)
1477{
1478 if (read_only) {
1479 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1480 return;
1481 } else {
1482 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1483 return;
1484 }
1485 set_disk_ro_uevent(disk, read_only);
1486}
1487EXPORT_SYMBOL(set_disk_ro);
1488
1489void inc_diskseq(struct gendisk *disk)
1490{
1491 disk->diskseq = atomic64_inc_return(&diskseq);
1492}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * gendisk handling
4 */
5
6#include <linux/module.h>
7#include <linux/ctype.h>
8#include <linux/fs.h>
9#include <linux/genhd.h>
10#include <linux/kdev_t.h>
11#include <linux/kernel.h>
12#include <linux/blkdev.h>
13#include <linux/backing-dev.h>
14#include <linux/init.h>
15#include <linux/spinlock.h>
16#include <linux/proc_fs.h>
17#include <linux/seq_file.h>
18#include <linux/slab.h>
19#include <linux/kmod.h>
20#include <linux/kobj_map.h>
21#include <linux/mutex.h>
22#include <linux/idr.h>
23#include <linux/log2.h>
24#include <linux/pm_runtime.h>
25#include <linux/badblocks.h>
26
27#include "blk.h"
28
29static DEFINE_MUTEX(block_class_lock);
30static struct kobject *block_depr;
31
32/* for extended dynamic devt allocation, currently only one major is used */
33#define NR_EXT_DEVT (1 << MINORBITS)
34
35/* For extended devt allocation. ext_devt_lock prevents look up
36 * results from going away underneath its user.
37 */
38static DEFINE_SPINLOCK(ext_devt_lock);
39static DEFINE_IDR(ext_devt_idr);
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
48/*
49 * Set disk capacity and notify if the size is not currently
50 * zero and will not be set to zero
51 */
52void set_capacity_revalidate_and_notify(struct gendisk *disk, sector_t size,
53 bool revalidate)
54{
55 sector_t capacity = get_capacity(disk);
56
57 set_capacity(disk, size);
58
59 if (revalidate)
60 revalidate_disk(disk);
61
62 if (capacity != size && capacity != 0 && size != 0) {
63 char *envp[] = { "RESIZE=1", NULL };
64
65 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
66 }
67}
68
69EXPORT_SYMBOL_GPL(set_capacity_revalidate_and_notify);
70
71/*
72 * Format the device name of the indicated disk into the supplied buffer and
73 * return a pointer to that same buffer for convenience.
74 */
75char *disk_name(struct gendisk *hd, int partno, char *buf)
76{
77 if (!partno)
78 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name);
79 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1]))
80 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno);
81 else
82 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno);
83
84 return buf;
85}
86
87const char *bdevname(struct block_device *bdev, char *buf)
88{
89 return disk_name(bdev->bd_disk, bdev->bd_part->partno, buf);
90}
91EXPORT_SYMBOL(bdevname);
92
93static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
94{
95 int cpu;
96
97 memset(stat, 0, sizeof(struct disk_stats));
98 for_each_possible_cpu(cpu) {
99 struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
100 int group;
101
102 for (group = 0; group < NR_STAT_GROUPS; group++) {
103 stat->nsecs[group] += ptr->nsecs[group];
104 stat->sectors[group] += ptr->sectors[group];
105 stat->ios[group] += ptr->ios[group];
106 stat->merges[group] += ptr->merges[group];
107 }
108
109 stat->io_ticks += ptr->io_ticks;
110 }
111}
112
113static unsigned int part_in_flight(struct request_queue *q,
114 struct hd_struct *part)
115{
116 unsigned int inflight = 0;
117 int cpu;
118
119 for_each_possible_cpu(cpu) {
120 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
121 part_stat_local_read_cpu(part, in_flight[1], cpu);
122 }
123 if ((int)inflight < 0)
124 inflight = 0;
125
126 return inflight;
127}
128
129static void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
130 unsigned int inflight[2])
131{
132 int cpu;
133
134 inflight[0] = 0;
135 inflight[1] = 0;
136 for_each_possible_cpu(cpu) {
137 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
138 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
139 }
140 if ((int)inflight[0] < 0)
141 inflight[0] = 0;
142 if ((int)inflight[1] < 0)
143 inflight[1] = 0;
144}
145
146struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
147{
148 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
149
150 if (unlikely(partno < 0 || partno >= ptbl->len))
151 return NULL;
152 return rcu_dereference(ptbl->part[partno]);
153}
154
155/**
156 * disk_get_part - get partition
157 * @disk: disk to look partition from
158 * @partno: partition number
159 *
160 * Look for partition @partno from @disk. If found, increment
161 * reference count and return it.
162 *
163 * CONTEXT:
164 * Don't care.
165 *
166 * RETURNS:
167 * Pointer to the found partition on success, NULL if not found.
168 */
169struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
170{
171 struct hd_struct *part;
172
173 rcu_read_lock();
174 part = __disk_get_part(disk, partno);
175 if (part)
176 get_device(part_to_dev(part));
177 rcu_read_unlock();
178
179 return part;
180}
181
182/**
183 * disk_part_iter_init - initialize partition iterator
184 * @piter: iterator to initialize
185 * @disk: disk to iterate over
186 * @flags: DISK_PITER_* flags
187 *
188 * Initialize @piter so that it iterates over partitions of @disk.
189 *
190 * CONTEXT:
191 * Don't care.
192 */
193void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
194 unsigned int flags)
195{
196 struct disk_part_tbl *ptbl;
197
198 rcu_read_lock();
199 ptbl = rcu_dereference(disk->part_tbl);
200
201 piter->disk = disk;
202 piter->part = NULL;
203
204 if (flags & DISK_PITER_REVERSE)
205 piter->idx = ptbl->len - 1;
206 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
207 piter->idx = 0;
208 else
209 piter->idx = 1;
210
211 piter->flags = flags;
212
213 rcu_read_unlock();
214}
215EXPORT_SYMBOL_GPL(disk_part_iter_init);
216
217/**
218 * disk_part_iter_next - proceed iterator to the next partition and return it
219 * @piter: iterator of interest
220 *
221 * Proceed @piter to the next partition and return it.
222 *
223 * CONTEXT:
224 * Don't care.
225 */
226struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
227{
228 struct disk_part_tbl *ptbl;
229 int inc, end;
230
231 /* put the last partition */
232 disk_put_part(piter->part);
233 piter->part = NULL;
234
235 /* get part_tbl */
236 rcu_read_lock();
237 ptbl = rcu_dereference(piter->disk->part_tbl);
238
239 /* determine iteration parameters */
240 if (piter->flags & DISK_PITER_REVERSE) {
241 inc = -1;
242 if (piter->flags & (DISK_PITER_INCL_PART0 |
243 DISK_PITER_INCL_EMPTY_PART0))
244 end = -1;
245 else
246 end = 0;
247 } else {
248 inc = 1;
249 end = ptbl->len;
250 }
251
252 /* iterate to the next partition */
253 for (; piter->idx != end; piter->idx += inc) {
254 struct hd_struct *part;
255
256 part = rcu_dereference(ptbl->part[piter->idx]);
257 if (!part)
258 continue;
259 if (!part_nr_sects_read(part) &&
260 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
261 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
262 piter->idx == 0))
263 continue;
264
265 get_device(part_to_dev(part));
266 piter->part = part;
267 piter->idx += inc;
268 break;
269 }
270
271 rcu_read_unlock();
272
273 return piter->part;
274}
275EXPORT_SYMBOL_GPL(disk_part_iter_next);
276
277/**
278 * disk_part_iter_exit - finish up partition iteration
279 * @piter: iter of interest
280 *
281 * Called when iteration is over. Cleans up @piter.
282 *
283 * CONTEXT:
284 * Don't care.
285 */
286void disk_part_iter_exit(struct disk_part_iter *piter)
287{
288 disk_put_part(piter->part);
289 piter->part = NULL;
290}
291EXPORT_SYMBOL_GPL(disk_part_iter_exit);
292
293static inline int sector_in_part(struct hd_struct *part, sector_t sector)
294{
295 return part->start_sect <= sector &&
296 sector < part->start_sect + part_nr_sects_read(part);
297}
298
299/**
300 * disk_map_sector_rcu - map sector to partition
301 * @disk: gendisk of interest
302 * @sector: sector to map
303 *
304 * Find out which partition @sector maps to on @disk. This is
305 * primarily used for stats accounting.
306 *
307 * CONTEXT:
308 * RCU read locked. The returned partition pointer is always valid
309 * because its refcount is grabbed except for part0, which lifetime
310 * is same with the disk.
311 *
312 * RETURNS:
313 * Found partition on success, part0 is returned if no partition matches
314 * or the matched partition is being deleted.
315 */
316struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
317{
318 struct disk_part_tbl *ptbl;
319 struct hd_struct *part;
320 int i;
321
322 rcu_read_lock();
323 ptbl = rcu_dereference(disk->part_tbl);
324
325 part = rcu_dereference(ptbl->last_lookup);
326 if (part && sector_in_part(part, sector) && hd_struct_try_get(part))
327 goto out_unlock;
328
329 for (i = 1; i < ptbl->len; i++) {
330 part = rcu_dereference(ptbl->part[i]);
331
332 if (part && sector_in_part(part, sector)) {
333 /*
334 * only live partition can be cached for lookup,
335 * so use-after-free on cached & deleting partition
336 * can be avoided
337 */
338 if (!hd_struct_try_get(part))
339 break;
340 rcu_assign_pointer(ptbl->last_lookup, part);
341 goto out_unlock;
342 }
343 }
344
345 part = &disk->part0;
346out_unlock:
347 rcu_read_unlock();
348 return part;
349}
350
351/**
352 * disk_has_partitions
353 * @disk: gendisk of interest
354 *
355 * Walk through the partition table and check if valid partition exists.
356 *
357 * CONTEXT:
358 * Don't care.
359 *
360 * RETURNS:
361 * True if the gendisk has at least one valid non-zero size partition.
362 * Otherwise false.
363 */
364bool disk_has_partitions(struct gendisk *disk)
365{
366 struct disk_part_tbl *ptbl;
367 int i;
368 bool ret = false;
369
370 rcu_read_lock();
371 ptbl = rcu_dereference(disk->part_tbl);
372
373 /* Iterate partitions skipping the whole device at index 0 */
374 for (i = 1; i < ptbl->len; i++) {
375 if (rcu_dereference(ptbl->part[i])) {
376 ret = true;
377 break;
378 }
379 }
380
381 rcu_read_unlock();
382
383 return ret;
384}
385EXPORT_SYMBOL_GPL(disk_has_partitions);
386
387/*
388 * Can be deleted altogether. Later.
389 *
390 */
391#define BLKDEV_MAJOR_HASH_SIZE 255
392static struct blk_major_name {
393 struct blk_major_name *next;
394 int major;
395 char name[16];
396} *major_names[BLKDEV_MAJOR_HASH_SIZE];
397
398/* index in the above - for now: assume no multimajor ranges */
399static inline int major_to_index(unsigned major)
400{
401 return major % BLKDEV_MAJOR_HASH_SIZE;
402}
403
404#ifdef CONFIG_PROC_FS
405void blkdev_show(struct seq_file *seqf, off_t offset)
406{
407 struct blk_major_name *dp;
408
409 mutex_lock(&block_class_lock);
410 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
411 if (dp->major == offset)
412 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
413 mutex_unlock(&block_class_lock);
414}
415#endif /* CONFIG_PROC_FS */
416
417/**
418 * register_blkdev - register a new block device
419 *
420 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
421 * @major = 0, try to allocate any unused major number.
422 * @name: the name of the new block device as a zero terminated string
423 *
424 * The @name must be unique within the system.
425 *
426 * The return value depends on the @major input parameter:
427 *
428 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
429 * then the function returns zero on success, or a negative error code
430 * - if any unused major number was requested with @major = 0 parameter
431 * then the return value is the allocated major number in range
432 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
433 *
434 * See Documentation/admin-guide/devices.txt for the list of allocated
435 * major numbers.
436 */
437int register_blkdev(unsigned int major, const char *name)
438{
439 struct blk_major_name **n, *p;
440 int index, ret = 0;
441
442 mutex_lock(&block_class_lock);
443
444 /* temporary */
445 if (major == 0) {
446 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
447 if (major_names[index] == NULL)
448 break;
449 }
450
451 if (index == 0) {
452 printk("%s: failed to get major for %s\n",
453 __func__, name);
454 ret = -EBUSY;
455 goto out;
456 }
457 major = index;
458 ret = major;
459 }
460
461 if (major >= BLKDEV_MAJOR_MAX) {
462 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
463 __func__, major, BLKDEV_MAJOR_MAX-1, name);
464
465 ret = -EINVAL;
466 goto out;
467 }
468
469 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
470 if (p == NULL) {
471 ret = -ENOMEM;
472 goto out;
473 }
474
475 p->major = major;
476 strlcpy(p->name, name, sizeof(p->name));
477 p->next = NULL;
478 index = major_to_index(major);
479
480 for (n = &major_names[index]; *n; n = &(*n)->next) {
481 if ((*n)->major == major)
482 break;
483 }
484 if (!*n)
485 *n = p;
486 else
487 ret = -EBUSY;
488
489 if (ret < 0) {
490 printk("register_blkdev: cannot get major %u for %s\n",
491 major, name);
492 kfree(p);
493 }
494out:
495 mutex_unlock(&block_class_lock);
496 return ret;
497}
498
499EXPORT_SYMBOL(register_blkdev);
500
501void unregister_blkdev(unsigned int major, const char *name)
502{
503 struct blk_major_name **n;
504 struct blk_major_name *p = NULL;
505 int index = major_to_index(major);
506
507 mutex_lock(&block_class_lock);
508 for (n = &major_names[index]; *n; n = &(*n)->next)
509 if ((*n)->major == major)
510 break;
511 if (!*n || strcmp((*n)->name, name)) {
512 WARN_ON(1);
513 } else {
514 p = *n;
515 *n = p->next;
516 }
517 mutex_unlock(&block_class_lock);
518 kfree(p);
519}
520
521EXPORT_SYMBOL(unregister_blkdev);
522
523static struct kobj_map *bdev_map;
524
525/**
526 * blk_mangle_minor - scatter minor numbers apart
527 * @minor: minor number to mangle
528 *
529 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
530 * is enabled. Mangling twice gives the original value.
531 *
532 * RETURNS:
533 * Mangled value.
534 *
535 * CONTEXT:
536 * Don't care.
537 */
538static int blk_mangle_minor(int minor)
539{
540#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
541 int i;
542
543 for (i = 0; i < MINORBITS / 2; i++) {
544 int low = minor & (1 << i);
545 int high = minor & (1 << (MINORBITS - 1 - i));
546 int distance = MINORBITS - 1 - 2 * i;
547
548 minor ^= low | high; /* clear both bits */
549 low <<= distance; /* swap the positions */
550 high >>= distance;
551 minor |= low | high; /* and set */
552 }
553#endif
554 return minor;
555}
556
557/**
558 * blk_alloc_devt - allocate a dev_t for a partition
559 * @part: partition to allocate dev_t for
560 * @devt: out parameter for resulting dev_t
561 *
562 * Allocate a dev_t for block device.
563 *
564 * RETURNS:
565 * 0 on success, allocated dev_t is returned in *@devt. -errno on
566 * failure.
567 *
568 * CONTEXT:
569 * Might sleep.
570 */
571int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
572{
573 struct gendisk *disk = part_to_disk(part);
574 int idx;
575
576 /* in consecutive minor range? */
577 if (part->partno < disk->minors) {
578 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
579 return 0;
580 }
581
582 /* allocate ext devt */
583 idr_preload(GFP_KERNEL);
584
585 spin_lock_bh(&ext_devt_lock);
586 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
587 spin_unlock_bh(&ext_devt_lock);
588
589 idr_preload_end();
590 if (idx < 0)
591 return idx == -ENOSPC ? -EBUSY : idx;
592
593 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
594 return 0;
595}
596
597/**
598 * blk_free_devt - free a dev_t
599 * @devt: dev_t to free
600 *
601 * Free @devt which was allocated using blk_alloc_devt().
602 *
603 * CONTEXT:
604 * Might sleep.
605 */
606void blk_free_devt(dev_t devt)
607{
608 if (devt == MKDEV(0, 0))
609 return;
610
611 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
612 spin_lock_bh(&ext_devt_lock);
613 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
614 spin_unlock_bh(&ext_devt_lock);
615 }
616}
617
618/*
619 * We invalidate devt by assigning NULL pointer for devt in idr.
620 */
621void blk_invalidate_devt(dev_t devt)
622{
623 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
624 spin_lock_bh(&ext_devt_lock);
625 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
626 spin_unlock_bh(&ext_devt_lock);
627 }
628}
629
630static char *bdevt_str(dev_t devt, char *buf)
631{
632 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
633 char tbuf[BDEVT_SIZE];
634 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
635 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
636 } else
637 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
638
639 return buf;
640}
641
642/*
643 * Register device numbers dev..(dev+range-1)
644 * range must be nonzero
645 * The hash chain is sorted on range, so that subranges can override.
646 */
647void blk_register_region(dev_t devt, unsigned long range, struct module *module,
648 struct kobject *(*probe)(dev_t, int *, void *),
649 int (*lock)(dev_t, void *), void *data)
650{
651 kobj_map(bdev_map, devt, range, module, probe, lock, data);
652}
653
654EXPORT_SYMBOL(blk_register_region);
655
656void blk_unregister_region(dev_t devt, unsigned long range)
657{
658 kobj_unmap(bdev_map, devt, range);
659}
660
661EXPORT_SYMBOL(blk_unregister_region);
662
663static struct kobject *exact_match(dev_t devt, int *partno, void *data)
664{
665 struct gendisk *p = data;
666
667 return &disk_to_dev(p)->kobj;
668}
669
670static int exact_lock(dev_t devt, void *data)
671{
672 struct gendisk *p = data;
673
674 if (!get_disk_and_module(p))
675 return -1;
676 return 0;
677}
678
679static void register_disk(struct device *parent, struct gendisk *disk,
680 const struct attribute_group **groups)
681{
682 struct device *ddev = disk_to_dev(disk);
683 struct block_device *bdev;
684 struct disk_part_iter piter;
685 struct hd_struct *part;
686 int err;
687
688 ddev->parent = parent;
689
690 dev_set_name(ddev, "%s", disk->disk_name);
691
692 /* delay uevents, until we scanned partition table */
693 dev_set_uevent_suppress(ddev, 1);
694
695 if (groups) {
696 WARN_ON(ddev->groups);
697 ddev->groups = groups;
698 }
699 if (device_add(ddev))
700 return;
701 if (!sysfs_deprecated) {
702 err = sysfs_create_link(block_depr, &ddev->kobj,
703 kobject_name(&ddev->kobj));
704 if (err) {
705 device_del(ddev);
706 return;
707 }
708 }
709
710 /*
711 * avoid probable deadlock caused by allocating memory with
712 * GFP_KERNEL in runtime_resume callback of its all ancestor
713 * devices
714 */
715 pm_runtime_set_memalloc_noio(ddev, true);
716
717 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
718 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
719
720 if (disk->flags & GENHD_FL_HIDDEN) {
721 dev_set_uevent_suppress(ddev, 0);
722 return;
723 }
724
725 /* No minors to use for partitions */
726 if (!disk_part_scan_enabled(disk))
727 goto exit;
728
729 /* No such device (e.g., media were just removed) */
730 if (!get_capacity(disk))
731 goto exit;
732
733 bdev = bdget_disk(disk, 0);
734 if (!bdev)
735 goto exit;
736
737 bdev->bd_invalidated = 1;
738 err = blkdev_get(bdev, FMODE_READ, NULL);
739 if (err < 0)
740 goto exit;
741 blkdev_put(bdev, FMODE_READ);
742
743exit:
744 /* announce disk after possible partitions are created */
745 dev_set_uevent_suppress(ddev, 0);
746 kobject_uevent(&ddev->kobj, KOBJ_ADD);
747
748 /* announce possible partitions */
749 disk_part_iter_init(&piter, disk, 0);
750 while ((part = disk_part_iter_next(&piter)))
751 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
752 disk_part_iter_exit(&piter);
753
754 if (disk->queue->backing_dev_info->dev) {
755 err = sysfs_create_link(&ddev->kobj,
756 &disk->queue->backing_dev_info->dev->kobj,
757 "bdi");
758 WARN_ON(err);
759 }
760}
761
762/**
763 * __device_add_disk - add disk information to kernel list
764 * @parent: parent device for the disk
765 * @disk: per-device partitioning information
766 * @groups: Additional per-device sysfs groups
767 * @register_queue: register the queue if set to true
768 *
769 * This function registers the partitioning information in @disk
770 * with the kernel.
771 *
772 * FIXME: error handling
773 */
774static void __device_add_disk(struct device *parent, struct gendisk *disk,
775 const struct attribute_group **groups,
776 bool register_queue)
777{
778 dev_t devt;
779 int retval;
780
781 /*
782 * The disk queue should now be all set with enough information about
783 * the device for the elevator code to pick an adequate default
784 * elevator if one is needed, that is, for devices requesting queue
785 * registration.
786 */
787 if (register_queue)
788 elevator_init_mq(disk->queue);
789
790 /* minors == 0 indicates to use ext devt from part0 and should
791 * be accompanied with EXT_DEVT flag. Make sure all
792 * parameters make sense.
793 */
794 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
795 WARN_ON(!disk->minors &&
796 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
797
798 disk->flags |= GENHD_FL_UP;
799
800 retval = blk_alloc_devt(&disk->part0, &devt);
801 if (retval) {
802 WARN_ON(1);
803 return;
804 }
805 disk->major = MAJOR(devt);
806 disk->first_minor = MINOR(devt);
807
808 disk_alloc_events(disk);
809
810 if (disk->flags & GENHD_FL_HIDDEN) {
811 /*
812 * Don't let hidden disks show up in /proc/partitions,
813 * and don't bother scanning for partitions either.
814 */
815 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
816 disk->flags |= GENHD_FL_NO_PART_SCAN;
817 } else {
818 struct backing_dev_info *bdi = disk->queue->backing_dev_info;
819 struct device *dev = disk_to_dev(disk);
820 int ret;
821
822 /* Register BDI before referencing it from bdev */
823 dev->devt = devt;
824 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt));
825 WARN_ON(ret);
826 bdi_set_owner(bdi, dev);
827 blk_register_region(disk_devt(disk), disk->minors, NULL,
828 exact_match, exact_lock, disk);
829 }
830 register_disk(parent, disk, groups);
831 if (register_queue)
832 blk_register_queue(disk);
833
834 /*
835 * Take an extra ref on queue which will be put on disk_release()
836 * so that it sticks around as long as @disk is there.
837 */
838 WARN_ON_ONCE(!blk_get_queue(disk->queue));
839
840 disk_add_events(disk);
841 blk_integrity_add(disk);
842}
843
844void device_add_disk(struct device *parent, struct gendisk *disk,
845 const struct attribute_group **groups)
846
847{
848 __device_add_disk(parent, disk, groups, true);
849}
850EXPORT_SYMBOL(device_add_disk);
851
852void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
853{
854 __device_add_disk(parent, disk, NULL, false);
855}
856EXPORT_SYMBOL(device_add_disk_no_queue_reg);
857
858static void invalidate_partition(struct gendisk *disk, int partno)
859{
860 struct block_device *bdev;
861
862 bdev = bdget_disk(disk, partno);
863 if (!bdev)
864 return;
865
866 fsync_bdev(bdev);
867 __invalidate_device(bdev, true);
868
869 /*
870 * Unhash the bdev inode for this device so that it gets evicted as soon
871 * as last inode reference is dropped.
872 */
873 remove_inode_hash(bdev->bd_inode);
874 bdput(bdev);
875}
876
877/**
878 * del_gendisk - remove the gendisk
879 * @disk: the struct gendisk to remove
880 *
881 * Removes the gendisk and all its associated resources. This deletes the
882 * partitions associated with the gendisk, and unregisters the associated
883 * request_queue.
884 *
885 * This is the counter to the respective __device_add_disk() call.
886 *
887 * The final removal of the struct gendisk happens when its refcount reaches 0
888 * with put_disk(), which should be called after del_gendisk(), if
889 * __device_add_disk() was used.
890 *
891 * Drivers exist which depend on the release of the gendisk to be synchronous,
892 * it should not be deferred.
893 *
894 * Context: can sleep
895 */
896void del_gendisk(struct gendisk *disk)
897{
898 struct disk_part_iter piter;
899 struct hd_struct *part;
900
901 might_sleep();
902
903 blk_integrity_del(disk);
904 disk_del_events(disk);
905
906 /*
907 * Block lookups of the disk until all bdevs are unhashed and the
908 * disk is marked as dead (GENHD_FL_UP cleared).
909 */
910 down_write(&disk->lookup_sem);
911 /* invalidate stuff */
912 disk_part_iter_init(&piter, disk,
913 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
914 while ((part = disk_part_iter_next(&piter))) {
915 invalidate_partition(disk, part->partno);
916 delete_partition(disk, part);
917 }
918 disk_part_iter_exit(&piter);
919
920 invalidate_partition(disk, 0);
921 set_capacity(disk, 0);
922 disk->flags &= ~GENHD_FL_UP;
923 up_write(&disk->lookup_sem);
924
925 if (!(disk->flags & GENHD_FL_HIDDEN))
926 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
927 if (disk->queue) {
928 /*
929 * Unregister bdi before releasing device numbers (as they can
930 * get reused and we'd get clashes in sysfs).
931 */
932 if (!(disk->flags & GENHD_FL_HIDDEN))
933 bdi_unregister(disk->queue->backing_dev_info);
934 blk_unregister_queue(disk);
935 } else {
936 WARN_ON(1);
937 }
938
939 if (!(disk->flags & GENHD_FL_HIDDEN))
940 blk_unregister_region(disk_devt(disk), disk->minors);
941 /*
942 * Remove gendisk pointer from idr so that it cannot be looked up
943 * while RCU period before freeing gendisk is running to prevent
944 * use-after-free issues. Note that the device number stays
945 * "in-use" until we really free the gendisk.
946 */
947 blk_invalidate_devt(disk_devt(disk));
948
949 kobject_put(disk->part0.holder_dir);
950 kobject_put(disk->slave_dir);
951
952 part_stat_set_all(&disk->part0, 0);
953 disk->part0.stamp = 0;
954 if (!sysfs_deprecated)
955 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
956 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
957 device_del(disk_to_dev(disk));
958}
959EXPORT_SYMBOL(del_gendisk);
960
961/* sysfs access to bad-blocks list. */
962static ssize_t disk_badblocks_show(struct device *dev,
963 struct device_attribute *attr,
964 char *page)
965{
966 struct gendisk *disk = dev_to_disk(dev);
967
968 if (!disk->bb)
969 return sprintf(page, "\n");
970
971 return badblocks_show(disk->bb, page, 0);
972}
973
974static ssize_t disk_badblocks_store(struct device *dev,
975 struct device_attribute *attr,
976 const char *page, size_t len)
977{
978 struct gendisk *disk = dev_to_disk(dev);
979
980 if (!disk->bb)
981 return -ENXIO;
982
983 return badblocks_store(disk->bb, page, len, 0);
984}
985
986/**
987 * get_gendisk - get partitioning information for a given device
988 * @devt: device to get partitioning information for
989 * @partno: returned partition index
990 *
991 * This function gets the structure containing partitioning
992 * information for the given device @devt.
993 *
994 * Context: can sleep
995 */
996struct gendisk *get_gendisk(dev_t devt, int *partno)
997{
998 struct gendisk *disk = NULL;
999
1000 might_sleep();
1001
1002 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
1003 struct kobject *kobj;
1004
1005 kobj = kobj_lookup(bdev_map, devt, partno);
1006 if (kobj)
1007 disk = dev_to_disk(kobj_to_dev(kobj));
1008 } else {
1009 struct hd_struct *part;
1010
1011 spin_lock_bh(&ext_devt_lock);
1012 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
1013 if (part && get_disk_and_module(part_to_disk(part))) {
1014 *partno = part->partno;
1015 disk = part_to_disk(part);
1016 }
1017 spin_unlock_bh(&ext_devt_lock);
1018 }
1019
1020 if (!disk)
1021 return NULL;
1022
1023 /*
1024 * Synchronize with del_gendisk() to not return disk that is being
1025 * destroyed.
1026 */
1027 down_read(&disk->lookup_sem);
1028 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
1029 !(disk->flags & GENHD_FL_UP))) {
1030 up_read(&disk->lookup_sem);
1031 put_disk_and_module(disk);
1032 disk = NULL;
1033 } else {
1034 up_read(&disk->lookup_sem);
1035 }
1036 return disk;
1037}
1038
1039/**
1040 * bdget_disk - do bdget() by gendisk and partition number
1041 * @disk: gendisk of interest
1042 * @partno: partition number
1043 *
1044 * Find partition @partno from @disk, do bdget() on it.
1045 *
1046 * CONTEXT:
1047 * Don't care.
1048 *
1049 * RETURNS:
1050 * Resulting block_device on success, NULL on failure.
1051 */
1052struct block_device *bdget_disk(struct gendisk *disk, int partno)
1053{
1054 struct hd_struct *part;
1055 struct block_device *bdev = NULL;
1056
1057 part = disk_get_part(disk, partno);
1058 if (part)
1059 bdev = bdget(part_devt(part));
1060 disk_put_part(part);
1061
1062 return bdev;
1063}
1064EXPORT_SYMBOL(bdget_disk);
1065
1066/*
1067 * print a full list of all partitions - intended for places where the root
1068 * filesystem can't be mounted and thus to give the victim some idea of what
1069 * went wrong
1070 */
1071void __init printk_all_partitions(void)
1072{
1073 struct class_dev_iter iter;
1074 struct device *dev;
1075
1076 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1077 while ((dev = class_dev_iter_next(&iter))) {
1078 struct gendisk *disk = dev_to_disk(dev);
1079 struct disk_part_iter piter;
1080 struct hd_struct *part;
1081 char name_buf[BDEVNAME_SIZE];
1082 char devt_buf[BDEVT_SIZE];
1083
1084 /*
1085 * Don't show empty devices or things that have been
1086 * suppressed
1087 */
1088 if (get_capacity(disk) == 0 ||
1089 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
1090 continue;
1091
1092 /*
1093 * Note, unlike /proc/partitions, I am showing the
1094 * numbers in hex - the same format as the root=
1095 * option takes.
1096 */
1097 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
1098 while ((part = disk_part_iter_next(&piter))) {
1099 bool is_part0 = part == &disk->part0;
1100
1101 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
1102 bdevt_str(part_devt(part), devt_buf),
1103 (unsigned long long)part_nr_sects_read(part) >> 1
1104 , disk_name(disk, part->partno, name_buf),
1105 part->info ? part->info->uuid : "");
1106 if (is_part0) {
1107 if (dev->parent && dev->parent->driver)
1108 printk(" driver: %s\n",
1109 dev->parent->driver->name);
1110 else
1111 printk(" (driver?)\n");
1112 } else
1113 printk("\n");
1114 }
1115 disk_part_iter_exit(&piter);
1116 }
1117 class_dev_iter_exit(&iter);
1118}
1119
1120#ifdef CONFIG_PROC_FS
1121/* iterator */
1122static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
1123{
1124 loff_t skip = *pos;
1125 struct class_dev_iter *iter;
1126 struct device *dev;
1127
1128 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1129 if (!iter)
1130 return ERR_PTR(-ENOMEM);
1131
1132 seqf->private = iter;
1133 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1134 do {
1135 dev = class_dev_iter_next(iter);
1136 if (!dev)
1137 return NULL;
1138 } while (skip--);
1139
1140 return dev_to_disk(dev);
1141}
1142
1143static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1144{
1145 struct device *dev;
1146
1147 (*pos)++;
1148 dev = class_dev_iter_next(seqf->private);
1149 if (dev)
1150 return dev_to_disk(dev);
1151
1152 return NULL;
1153}
1154
1155static void disk_seqf_stop(struct seq_file *seqf, void *v)
1156{
1157 struct class_dev_iter *iter = seqf->private;
1158
1159 /* stop is called even after start failed :-( */
1160 if (iter) {
1161 class_dev_iter_exit(iter);
1162 kfree(iter);
1163 seqf->private = NULL;
1164 }
1165}
1166
1167static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1168{
1169 void *p;
1170
1171 p = disk_seqf_start(seqf, pos);
1172 if (!IS_ERR_OR_NULL(p) && !*pos)
1173 seq_puts(seqf, "major minor #blocks name\n\n");
1174 return p;
1175}
1176
1177static int show_partition(struct seq_file *seqf, void *v)
1178{
1179 struct gendisk *sgp = v;
1180 struct disk_part_iter piter;
1181 struct hd_struct *part;
1182 char buf[BDEVNAME_SIZE];
1183
1184 /* Don't show non-partitionable removeable devices or empty devices */
1185 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1186 (sgp->flags & GENHD_FL_REMOVABLE)))
1187 return 0;
1188 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1189 return 0;
1190
1191 /* show the full disk and all non-0 size partitions of it */
1192 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1193 while ((part = disk_part_iter_next(&piter)))
1194 seq_printf(seqf, "%4d %7d %10llu %s\n",
1195 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1196 (unsigned long long)part_nr_sects_read(part) >> 1,
1197 disk_name(sgp, part->partno, buf));
1198 disk_part_iter_exit(&piter);
1199
1200 return 0;
1201}
1202
1203static const struct seq_operations partitions_op = {
1204 .start = show_partition_start,
1205 .next = disk_seqf_next,
1206 .stop = disk_seqf_stop,
1207 .show = show_partition
1208};
1209#endif
1210
1211
1212static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1213{
1214 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1215 /* Make old-style 2.4 aliases work */
1216 request_module("block-major-%d", MAJOR(devt));
1217 return NULL;
1218}
1219
1220static int __init genhd_device_init(void)
1221{
1222 int error;
1223
1224 block_class.dev_kobj = sysfs_dev_block_kobj;
1225 error = class_register(&block_class);
1226 if (unlikely(error))
1227 return error;
1228 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1229 blk_dev_init();
1230
1231 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1232
1233 /* create top-level block dir */
1234 if (!sysfs_deprecated)
1235 block_depr = kobject_create_and_add("block", NULL);
1236 return 0;
1237}
1238
1239subsys_initcall(genhd_device_init);
1240
1241static ssize_t disk_range_show(struct device *dev,
1242 struct device_attribute *attr, char *buf)
1243{
1244 struct gendisk *disk = dev_to_disk(dev);
1245
1246 return sprintf(buf, "%d\n", disk->minors);
1247}
1248
1249static ssize_t disk_ext_range_show(struct device *dev,
1250 struct device_attribute *attr, char *buf)
1251{
1252 struct gendisk *disk = dev_to_disk(dev);
1253
1254 return sprintf(buf, "%d\n", disk_max_parts(disk));
1255}
1256
1257static ssize_t disk_removable_show(struct device *dev,
1258 struct device_attribute *attr, char *buf)
1259{
1260 struct gendisk *disk = dev_to_disk(dev);
1261
1262 return sprintf(buf, "%d\n",
1263 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1264}
1265
1266static ssize_t disk_hidden_show(struct device *dev,
1267 struct device_attribute *attr, char *buf)
1268{
1269 struct gendisk *disk = dev_to_disk(dev);
1270
1271 return sprintf(buf, "%d\n",
1272 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1273}
1274
1275static ssize_t disk_ro_show(struct device *dev,
1276 struct device_attribute *attr, char *buf)
1277{
1278 struct gendisk *disk = dev_to_disk(dev);
1279
1280 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1281}
1282
1283ssize_t part_size_show(struct device *dev,
1284 struct device_attribute *attr, char *buf)
1285{
1286 struct hd_struct *p = dev_to_part(dev);
1287
1288 return sprintf(buf, "%llu\n",
1289 (unsigned long long)part_nr_sects_read(p));
1290}
1291
1292ssize_t part_stat_show(struct device *dev,
1293 struct device_attribute *attr, char *buf)
1294{
1295 struct hd_struct *p = dev_to_part(dev);
1296 struct request_queue *q = part_to_disk(p)->queue;
1297 struct disk_stats stat;
1298 unsigned int inflight;
1299
1300 part_stat_read_all(p, &stat);
1301 if (queue_is_mq(q))
1302 inflight = blk_mq_in_flight(q, p);
1303 else
1304 inflight = part_in_flight(q, p);
1305
1306 return sprintf(buf,
1307 "%8lu %8lu %8llu %8u "
1308 "%8lu %8lu %8llu %8u "
1309 "%8u %8u %8u "
1310 "%8lu %8lu %8llu %8u "
1311 "%8lu %8u"
1312 "\n",
1313 stat.ios[STAT_READ],
1314 stat.merges[STAT_READ],
1315 (unsigned long long)stat.sectors[STAT_READ],
1316 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1317 stat.ios[STAT_WRITE],
1318 stat.merges[STAT_WRITE],
1319 (unsigned long long)stat.sectors[STAT_WRITE],
1320 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1321 inflight,
1322 jiffies_to_msecs(stat.io_ticks),
1323 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1324 stat.nsecs[STAT_WRITE] +
1325 stat.nsecs[STAT_DISCARD] +
1326 stat.nsecs[STAT_FLUSH],
1327 NSEC_PER_MSEC),
1328 stat.ios[STAT_DISCARD],
1329 stat.merges[STAT_DISCARD],
1330 (unsigned long long)stat.sectors[STAT_DISCARD],
1331 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1332 stat.ios[STAT_FLUSH],
1333 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1334}
1335
1336ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1337 char *buf)
1338{
1339 struct hd_struct *p = dev_to_part(dev);
1340 struct request_queue *q = part_to_disk(p)->queue;
1341 unsigned int inflight[2];
1342
1343 if (queue_is_mq(q))
1344 blk_mq_in_flight_rw(q, p, inflight);
1345 else
1346 part_in_flight_rw(q, p, inflight);
1347
1348 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1349}
1350
1351static ssize_t disk_capability_show(struct device *dev,
1352 struct device_attribute *attr, char *buf)
1353{
1354 struct gendisk *disk = dev_to_disk(dev);
1355
1356 return sprintf(buf, "%x\n", disk->flags);
1357}
1358
1359static ssize_t disk_alignment_offset_show(struct device *dev,
1360 struct device_attribute *attr,
1361 char *buf)
1362{
1363 struct gendisk *disk = dev_to_disk(dev);
1364
1365 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1366}
1367
1368static ssize_t disk_discard_alignment_show(struct device *dev,
1369 struct device_attribute *attr,
1370 char *buf)
1371{
1372 struct gendisk *disk = dev_to_disk(dev);
1373
1374 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1375}
1376
1377static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1378static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1379static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1380static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1381static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1382static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1383static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1384static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1385static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1386static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1387static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1388static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1389
1390#ifdef CONFIG_FAIL_MAKE_REQUEST
1391ssize_t part_fail_show(struct device *dev,
1392 struct device_attribute *attr, char *buf)
1393{
1394 struct hd_struct *p = dev_to_part(dev);
1395
1396 return sprintf(buf, "%d\n", p->make_it_fail);
1397}
1398
1399ssize_t part_fail_store(struct device *dev,
1400 struct device_attribute *attr,
1401 const char *buf, size_t count)
1402{
1403 struct hd_struct *p = dev_to_part(dev);
1404 int i;
1405
1406 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1407 p->make_it_fail = (i == 0) ? 0 : 1;
1408
1409 return count;
1410}
1411
1412static struct device_attribute dev_attr_fail =
1413 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1414#endif /* CONFIG_FAIL_MAKE_REQUEST */
1415
1416#ifdef CONFIG_FAIL_IO_TIMEOUT
1417static struct device_attribute dev_attr_fail_timeout =
1418 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1419#endif
1420
1421static struct attribute *disk_attrs[] = {
1422 &dev_attr_range.attr,
1423 &dev_attr_ext_range.attr,
1424 &dev_attr_removable.attr,
1425 &dev_attr_hidden.attr,
1426 &dev_attr_ro.attr,
1427 &dev_attr_size.attr,
1428 &dev_attr_alignment_offset.attr,
1429 &dev_attr_discard_alignment.attr,
1430 &dev_attr_capability.attr,
1431 &dev_attr_stat.attr,
1432 &dev_attr_inflight.attr,
1433 &dev_attr_badblocks.attr,
1434#ifdef CONFIG_FAIL_MAKE_REQUEST
1435 &dev_attr_fail.attr,
1436#endif
1437#ifdef CONFIG_FAIL_IO_TIMEOUT
1438 &dev_attr_fail_timeout.attr,
1439#endif
1440 NULL
1441};
1442
1443static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1444{
1445 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1446 struct gendisk *disk = dev_to_disk(dev);
1447
1448 if (a == &dev_attr_badblocks.attr && !disk->bb)
1449 return 0;
1450 return a->mode;
1451}
1452
1453static struct attribute_group disk_attr_group = {
1454 .attrs = disk_attrs,
1455 .is_visible = disk_visible,
1456};
1457
1458static const struct attribute_group *disk_attr_groups[] = {
1459 &disk_attr_group,
1460 NULL
1461};
1462
1463/**
1464 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1465 * @disk: disk to replace part_tbl for
1466 * @new_ptbl: new part_tbl to install
1467 *
1468 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1469 * original ptbl is freed using RCU callback.
1470 *
1471 * LOCKING:
1472 * Matching bd_mutex locked or the caller is the only user of @disk.
1473 */
1474static void disk_replace_part_tbl(struct gendisk *disk,
1475 struct disk_part_tbl *new_ptbl)
1476{
1477 struct disk_part_tbl *old_ptbl =
1478 rcu_dereference_protected(disk->part_tbl, 1);
1479
1480 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1481
1482 if (old_ptbl) {
1483 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1484 kfree_rcu(old_ptbl, rcu_head);
1485 }
1486}
1487
1488/**
1489 * disk_expand_part_tbl - expand disk->part_tbl
1490 * @disk: disk to expand part_tbl for
1491 * @partno: expand such that this partno can fit in
1492 *
1493 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1494 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1495 *
1496 * LOCKING:
1497 * Matching bd_mutex locked or the caller is the only user of @disk.
1498 * Might sleep.
1499 *
1500 * RETURNS:
1501 * 0 on success, -errno on failure.
1502 */
1503int disk_expand_part_tbl(struct gendisk *disk, int partno)
1504{
1505 struct disk_part_tbl *old_ptbl =
1506 rcu_dereference_protected(disk->part_tbl, 1);
1507 struct disk_part_tbl *new_ptbl;
1508 int len = old_ptbl ? old_ptbl->len : 0;
1509 int i, target;
1510
1511 /*
1512 * check for int overflow, since we can get here from blkpg_ioctl()
1513 * with a user passed 'partno'.
1514 */
1515 target = partno + 1;
1516 if (target < 0)
1517 return -EINVAL;
1518
1519 /* disk_max_parts() is zero during initialization, ignore if so */
1520 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1521 return -EINVAL;
1522
1523 if (target <= len)
1524 return 0;
1525
1526 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1527 disk->node_id);
1528 if (!new_ptbl)
1529 return -ENOMEM;
1530
1531 new_ptbl->len = target;
1532
1533 for (i = 0; i < len; i++)
1534 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1535
1536 disk_replace_part_tbl(disk, new_ptbl);
1537 return 0;
1538}
1539
1540/**
1541 * disk_release - releases all allocated resources of the gendisk
1542 * @dev: the device representing this disk
1543 *
1544 * This function releases all allocated resources of the gendisk.
1545 *
1546 * The struct gendisk refcount is incremented with get_gendisk() or
1547 * get_disk_and_module(), and its refcount is decremented with
1548 * put_disk_and_module() or put_disk(). Once the refcount reaches 0 this
1549 * function is called.
1550 *
1551 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1552 * assigned. Since the request_queue sits on top of the gendisk for these
1553 * drivers we also call blk_put_queue() for them, and we expect the
1554 * request_queue refcount to reach 0 at this point, and so the request_queue
1555 * will also be freed prior to the disk.
1556 *
1557 * Context: can sleep
1558 */
1559static void disk_release(struct device *dev)
1560{
1561 struct gendisk *disk = dev_to_disk(dev);
1562
1563 might_sleep();
1564
1565 blk_free_devt(dev->devt);
1566 disk_release_events(disk);
1567 kfree(disk->random);
1568 disk_replace_part_tbl(disk, NULL);
1569 hd_free_part(&disk->part0);
1570 if (disk->queue)
1571 blk_put_queue(disk->queue);
1572 kfree(disk);
1573}
1574struct class block_class = {
1575 .name = "block",
1576};
1577
1578static char *block_devnode(struct device *dev, umode_t *mode,
1579 kuid_t *uid, kgid_t *gid)
1580{
1581 struct gendisk *disk = dev_to_disk(dev);
1582
1583 if (disk->fops->devnode)
1584 return disk->fops->devnode(disk, mode);
1585 return NULL;
1586}
1587
1588const struct device_type disk_type = {
1589 .name = "disk",
1590 .groups = disk_attr_groups,
1591 .release = disk_release,
1592 .devnode = block_devnode,
1593};
1594
1595#ifdef CONFIG_PROC_FS
1596/*
1597 * aggregate disk stat collector. Uses the same stats that the sysfs
1598 * entries do, above, but makes them available through one seq_file.
1599 *
1600 * The output looks suspiciously like /proc/partitions with a bunch of
1601 * extra fields.
1602 */
1603static int diskstats_show(struct seq_file *seqf, void *v)
1604{
1605 struct gendisk *gp = v;
1606 struct disk_part_iter piter;
1607 struct hd_struct *hd;
1608 char buf[BDEVNAME_SIZE];
1609 unsigned int inflight;
1610 struct disk_stats stat;
1611
1612 /*
1613 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1614 seq_puts(seqf, "major minor name"
1615 " rio rmerge rsect ruse wio wmerge "
1616 "wsect wuse running use aveq"
1617 "\n\n");
1618 */
1619
1620 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1621 while ((hd = disk_part_iter_next(&piter))) {
1622 part_stat_read_all(hd, &stat);
1623 if (queue_is_mq(gp->queue))
1624 inflight = blk_mq_in_flight(gp->queue, hd);
1625 else
1626 inflight = part_in_flight(gp->queue, hd);
1627
1628 seq_printf(seqf, "%4d %7d %s "
1629 "%lu %lu %lu %u "
1630 "%lu %lu %lu %u "
1631 "%u %u %u "
1632 "%lu %lu %lu %u "
1633 "%lu %u"
1634 "\n",
1635 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1636 disk_name(gp, hd->partno, buf),
1637 stat.ios[STAT_READ],
1638 stat.merges[STAT_READ],
1639 stat.sectors[STAT_READ],
1640 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1641 NSEC_PER_MSEC),
1642 stat.ios[STAT_WRITE],
1643 stat.merges[STAT_WRITE],
1644 stat.sectors[STAT_WRITE],
1645 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1646 NSEC_PER_MSEC),
1647 inflight,
1648 jiffies_to_msecs(stat.io_ticks),
1649 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1650 stat.nsecs[STAT_WRITE] +
1651 stat.nsecs[STAT_DISCARD] +
1652 stat.nsecs[STAT_FLUSH],
1653 NSEC_PER_MSEC),
1654 stat.ios[STAT_DISCARD],
1655 stat.merges[STAT_DISCARD],
1656 stat.sectors[STAT_DISCARD],
1657 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1658 NSEC_PER_MSEC),
1659 stat.ios[STAT_FLUSH],
1660 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1661 NSEC_PER_MSEC)
1662 );
1663 }
1664 disk_part_iter_exit(&piter);
1665
1666 return 0;
1667}
1668
1669static const struct seq_operations diskstats_op = {
1670 .start = disk_seqf_start,
1671 .next = disk_seqf_next,
1672 .stop = disk_seqf_stop,
1673 .show = diskstats_show
1674};
1675
1676static int __init proc_genhd_init(void)
1677{
1678 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1679 proc_create_seq("partitions", 0, NULL, &partitions_op);
1680 return 0;
1681}
1682module_init(proc_genhd_init);
1683#endif /* CONFIG_PROC_FS */
1684
1685dev_t blk_lookup_devt(const char *name, int partno)
1686{
1687 dev_t devt = MKDEV(0, 0);
1688 struct class_dev_iter iter;
1689 struct device *dev;
1690
1691 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1692 while ((dev = class_dev_iter_next(&iter))) {
1693 struct gendisk *disk = dev_to_disk(dev);
1694 struct hd_struct *part;
1695
1696 if (strcmp(dev_name(dev), name))
1697 continue;
1698
1699 if (partno < disk->minors) {
1700 /* We need to return the right devno, even
1701 * if the partition doesn't exist yet.
1702 */
1703 devt = MKDEV(MAJOR(dev->devt),
1704 MINOR(dev->devt) + partno);
1705 break;
1706 }
1707 part = disk_get_part(disk, partno);
1708 if (part) {
1709 devt = part_devt(part);
1710 disk_put_part(part);
1711 break;
1712 }
1713 disk_put_part(part);
1714 }
1715 class_dev_iter_exit(&iter);
1716 return devt;
1717}
1718
1719struct gendisk *__alloc_disk_node(int minors, int node_id)
1720{
1721 struct gendisk *disk;
1722 struct disk_part_tbl *ptbl;
1723
1724 if (minors > DISK_MAX_PARTS) {
1725 printk(KERN_ERR
1726 "block: can't allocate more than %d partitions\n",
1727 DISK_MAX_PARTS);
1728 minors = DISK_MAX_PARTS;
1729 }
1730
1731 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1732 if (disk) {
1733 disk->part0.dkstats = alloc_percpu(struct disk_stats);
1734 if (!disk->part0.dkstats) {
1735 kfree(disk);
1736 return NULL;
1737 }
1738 init_rwsem(&disk->lookup_sem);
1739 disk->node_id = node_id;
1740 if (disk_expand_part_tbl(disk, 0)) {
1741 free_percpu(disk->part0.dkstats);
1742 kfree(disk);
1743 return NULL;
1744 }
1745 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1746 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1747
1748 /*
1749 * set_capacity() and get_capacity() currently don't use
1750 * seqcounter to read/update the part0->nr_sects. Still init
1751 * the counter as we can read the sectors in IO submission
1752 * patch using seqence counters.
1753 *
1754 * TODO: Ideally set_capacity() and get_capacity() should be
1755 * converted to make use of bd_mutex and sequence counters.
1756 */
1757 hd_sects_seq_init(&disk->part0);
1758 if (hd_ref_init(&disk->part0)) {
1759 hd_free_part(&disk->part0);
1760 kfree(disk);
1761 return NULL;
1762 }
1763
1764 disk->minors = minors;
1765 rand_initialize_disk(disk);
1766 disk_to_dev(disk)->class = &block_class;
1767 disk_to_dev(disk)->type = &disk_type;
1768 device_initialize(disk_to_dev(disk));
1769 }
1770 return disk;
1771}
1772EXPORT_SYMBOL(__alloc_disk_node);
1773
1774/**
1775 * get_disk_and_module - increments the gendisk and gendisk fops module refcount
1776 * @disk: the struct gendisk to increment the refcount for
1777 *
1778 * This increments the refcount for the struct gendisk, and the gendisk's
1779 * fops module owner.
1780 *
1781 * Context: Any context.
1782 */
1783struct kobject *get_disk_and_module(struct gendisk *disk)
1784{
1785 struct module *owner;
1786 struct kobject *kobj;
1787
1788 if (!disk->fops)
1789 return NULL;
1790 owner = disk->fops->owner;
1791 if (owner && !try_module_get(owner))
1792 return NULL;
1793 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1794 if (kobj == NULL) {
1795 module_put(owner);
1796 return NULL;
1797 }
1798 return kobj;
1799
1800}
1801EXPORT_SYMBOL(get_disk_and_module);
1802
1803/**
1804 * put_disk - decrements the gendisk refcount
1805 * @disk: the struct gendisk to decrement the refcount for
1806 *
1807 * This decrements the refcount for the struct gendisk. When this reaches 0
1808 * we'll have disk_release() called.
1809 *
1810 * Context: Any context, but the last reference must not be dropped from
1811 * atomic context.
1812 */
1813void put_disk(struct gendisk *disk)
1814{
1815 if (disk)
1816 kobject_put(&disk_to_dev(disk)->kobj);
1817}
1818EXPORT_SYMBOL(put_disk);
1819
1820/**
1821 * put_disk_and_module - decrements the module and gendisk refcount
1822 * @disk: the struct gendisk to decrement the refcount for
1823 *
1824 * This is a counterpart of get_disk_and_module() and thus also of
1825 * get_gendisk().
1826 *
1827 * Context: Any context, but the last reference must not be dropped from
1828 * atomic context.
1829 */
1830void put_disk_and_module(struct gendisk *disk)
1831{
1832 if (disk) {
1833 struct module *owner = disk->fops->owner;
1834
1835 put_disk(disk);
1836 module_put(owner);
1837 }
1838}
1839EXPORT_SYMBOL(put_disk_and_module);
1840
1841static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1842{
1843 char event[] = "DISK_RO=1";
1844 char *envp[] = { event, NULL };
1845
1846 if (!ro)
1847 event[8] = '0';
1848 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1849}
1850
1851void set_device_ro(struct block_device *bdev, int flag)
1852{
1853 bdev->bd_part->policy = flag;
1854}
1855
1856EXPORT_SYMBOL(set_device_ro);
1857
1858void set_disk_ro(struct gendisk *disk, int flag)
1859{
1860 struct disk_part_iter piter;
1861 struct hd_struct *part;
1862
1863 if (disk->part0.policy != flag) {
1864 set_disk_ro_uevent(disk, flag);
1865 disk->part0.policy = flag;
1866 }
1867
1868 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1869 while ((part = disk_part_iter_next(&piter)))
1870 part->policy = flag;
1871 disk_part_iter_exit(&piter);
1872}
1873
1874EXPORT_SYMBOL(set_disk_ro);
1875
1876int bdev_read_only(struct block_device *bdev)
1877{
1878 if (!bdev)
1879 return 0;
1880 return bdev->bd_part->policy;
1881}
1882
1883EXPORT_SYMBOL(bdev_read_only);
1884
1885/*
1886 * Disk events - monitor disk events like media change and eject request.
1887 */
1888struct disk_events {
1889 struct list_head node; /* all disk_event's */
1890 struct gendisk *disk; /* the associated disk */
1891 spinlock_t lock;
1892
1893 struct mutex block_mutex; /* protects blocking */
1894 int block; /* event blocking depth */
1895 unsigned int pending; /* events already sent out */
1896 unsigned int clearing; /* events being cleared */
1897
1898 long poll_msecs; /* interval, -1 for default */
1899 struct delayed_work dwork;
1900};
1901
1902static const char *disk_events_strs[] = {
1903 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1904 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1905};
1906
1907static char *disk_uevents[] = {
1908 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1909 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1910};
1911
1912/* list of all disk_events */
1913static DEFINE_MUTEX(disk_events_mutex);
1914static LIST_HEAD(disk_events);
1915
1916/* disable in-kernel polling by default */
1917static unsigned long disk_events_dfl_poll_msecs;
1918
1919static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1920{
1921 struct disk_events *ev = disk->ev;
1922 long intv_msecs = 0;
1923
1924 /*
1925 * If device-specific poll interval is set, always use it. If
1926 * the default is being used, poll if the POLL flag is set.
1927 */
1928 if (ev->poll_msecs >= 0)
1929 intv_msecs = ev->poll_msecs;
1930 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1931 intv_msecs = disk_events_dfl_poll_msecs;
1932
1933 return msecs_to_jiffies(intv_msecs);
1934}
1935
1936/**
1937 * disk_block_events - block and flush disk event checking
1938 * @disk: disk to block events for
1939 *
1940 * On return from this function, it is guaranteed that event checking
1941 * isn't in progress and won't happen until unblocked by
1942 * disk_unblock_events(). Events blocking is counted and the actual
1943 * unblocking happens after the matching number of unblocks are done.
1944 *
1945 * Note that this intentionally does not block event checking from
1946 * disk_clear_events().
1947 *
1948 * CONTEXT:
1949 * Might sleep.
1950 */
1951void disk_block_events(struct gendisk *disk)
1952{
1953 struct disk_events *ev = disk->ev;
1954 unsigned long flags;
1955 bool cancel;
1956
1957 if (!ev)
1958 return;
1959
1960 /*
1961 * Outer mutex ensures that the first blocker completes canceling
1962 * the event work before further blockers are allowed to finish.
1963 */
1964 mutex_lock(&ev->block_mutex);
1965
1966 spin_lock_irqsave(&ev->lock, flags);
1967 cancel = !ev->block++;
1968 spin_unlock_irqrestore(&ev->lock, flags);
1969
1970 if (cancel)
1971 cancel_delayed_work_sync(&disk->ev->dwork);
1972
1973 mutex_unlock(&ev->block_mutex);
1974}
1975
1976static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1977{
1978 struct disk_events *ev = disk->ev;
1979 unsigned long intv;
1980 unsigned long flags;
1981
1982 spin_lock_irqsave(&ev->lock, flags);
1983
1984 if (WARN_ON_ONCE(ev->block <= 0))
1985 goto out_unlock;
1986
1987 if (--ev->block)
1988 goto out_unlock;
1989
1990 intv = disk_events_poll_jiffies(disk);
1991 if (check_now)
1992 queue_delayed_work(system_freezable_power_efficient_wq,
1993 &ev->dwork, 0);
1994 else if (intv)
1995 queue_delayed_work(system_freezable_power_efficient_wq,
1996 &ev->dwork, intv);
1997out_unlock:
1998 spin_unlock_irqrestore(&ev->lock, flags);
1999}
2000
2001/**
2002 * disk_unblock_events - unblock disk event checking
2003 * @disk: disk to unblock events for
2004 *
2005 * Undo disk_block_events(). When the block count reaches zero, it
2006 * starts events polling if configured.
2007 *
2008 * CONTEXT:
2009 * Don't care. Safe to call from irq context.
2010 */
2011void disk_unblock_events(struct gendisk *disk)
2012{
2013 if (disk->ev)
2014 __disk_unblock_events(disk, false);
2015}
2016
2017/**
2018 * disk_flush_events - schedule immediate event checking and flushing
2019 * @disk: disk to check and flush events for
2020 * @mask: events to flush
2021 *
2022 * Schedule immediate event checking on @disk if not blocked. Events in
2023 * @mask are scheduled to be cleared from the driver. Note that this
2024 * doesn't clear the events from @disk->ev.
2025 *
2026 * CONTEXT:
2027 * If @mask is non-zero must be called with bdev->bd_mutex held.
2028 */
2029void disk_flush_events(struct gendisk *disk, unsigned int mask)
2030{
2031 struct disk_events *ev = disk->ev;
2032
2033 if (!ev)
2034 return;
2035
2036 spin_lock_irq(&ev->lock);
2037 ev->clearing |= mask;
2038 if (!ev->block)
2039 mod_delayed_work(system_freezable_power_efficient_wq,
2040 &ev->dwork, 0);
2041 spin_unlock_irq(&ev->lock);
2042}
2043
2044/**
2045 * disk_clear_events - synchronously check, clear and return pending events
2046 * @disk: disk to fetch and clear events from
2047 * @mask: mask of events to be fetched and cleared
2048 *
2049 * Disk events are synchronously checked and pending events in @mask
2050 * are cleared and returned. This ignores the block count.
2051 *
2052 * CONTEXT:
2053 * Might sleep.
2054 */
2055unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
2056{
2057 struct disk_events *ev = disk->ev;
2058 unsigned int pending;
2059 unsigned int clearing = mask;
2060
2061 if (!ev)
2062 return 0;
2063
2064 disk_block_events(disk);
2065
2066 /*
2067 * store the union of mask and ev->clearing on the stack so that the
2068 * race with disk_flush_events does not cause ambiguity (ev->clearing
2069 * can still be modified even if events are blocked).
2070 */
2071 spin_lock_irq(&ev->lock);
2072 clearing |= ev->clearing;
2073 ev->clearing = 0;
2074 spin_unlock_irq(&ev->lock);
2075
2076 disk_check_events(ev, &clearing);
2077 /*
2078 * if ev->clearing is not 0, the disk_flush_events got called in the
2079 * middle of this function, so we want to run the workfn without delay.
2080 */
2081 __disk_unblock_events(disk, ev->clearing ? true : false);
2082
2083 /* then, fetch and clear pending events */
2084 spin_lock_irq(&ev->lock);
2085 pending = ev->pending & mask;
2086 ev->pending &= ~mask;
2087 spin_unlock_irq(&ev->lock);
2088 WARN_ON_ONCE(clearing & mask);
2089
2090 return pending;
2091}
2092
2093/*
2094 * Separate this part out so that a different pointer for clearing_ptr can be
2095 * passed in for disk_clear_events.
2096 */
2097static void disk_events_workfn(struct work_struct *work)
2098{
2099 struct delayed_work *dwork = to_delayed_work(work);
2100 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
2101
2102 disk_check_events(ev, &ev->clearing);
2103}
2104
2105static void disk_check_events(struct disk_events *ev,
2106 unsigned int *clearing_ptr)
2107{
2108 struct gendisk *disk = ev->disk;
2109 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
2110 unsigned int clearing = *clearing_ptr;
2111 unsigned int events;
2112 unsigned long intv;
2113 int nr_events = 0, i;
2114
2115 /* check events */
2116 events = disk->fops->check_events(disk, clearing);
2117
2118 /* accumulate pending events and schedule next poll if necessary */
2119 spin_lock_irq(&ev->lock);
2120
2121 events &= ~ev->pending;
2122 ev->pending |= events;
2123 *clearing_ptr &= ~clearing;
2124
2125 intv = disk_events_poll_jiffies(disk);
2126 if (!ev->block && intv)
2127 queue_delayed_work(system_freezable_power_efficient_wq,
2128 &ev->dwork, intv);
2129
2130 spin_unlock_irq(&ev->lock);
2131
2132 /*
2133 * Tell userland about new events. Only the events listed in
2134 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
2135 * is set. Otherwise, events are processed internally but never
2136 * get reported to userland.
2137 */
2138 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
2139 if ((events & disk->events & (1 << i)) &&
2140 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2141 envp[nr_events++] = disk_uevents[i];
2142
2143 if (nr_events)
2144 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
2145}
2146
2147/*
2148 * A disk events enabled device has the following sysfs nodes under
2149 * its /sys/block/X/ directory.
2150 *
2151 * events : list of all supported events
2152 * events_async : list of events which can be detected w/o polling
2153 * (always empty, only for backwards compatibility)
2154 * events_poll_msecs : polling interval, 0: disable, -1: system default
2155 */
2156static ssize_t __disk_events_show(unsigned int events, char *buf)
2157{
2158 const char *delim = "";
2159 ssize_t pos = 0;
2160 int i;
2161
2162 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
2163 if (events & (1 << i)) {
2164 pos += sprintf(buf + pos, "%s%s",
2165 delim, disk_events_strs[i]);
2166 delim = " ";
2167 }
2168 if (pos)
2169 pos += sprintf(buf + pos, "\n");
2170 return pos;
2171}
2172
2173static ssize_t disk_events_show(struct device *dev,
2174 struct device_attribute *attr, char *buf)
2175{
2176 struct gendisk *disk = dev_to_disk(dev);
2177
2178 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
2179 return 0;
2180
2181 return __disk_events_show(disk->events, buf);
2182}
2183
2184static ssize_t disk_events_async_show(struct device *dev,
2185 struct device_attribute *attr, char *buf)
2186{
2187 return 0;
2188}
2189
2190static ssize_t disk_events_poll_msecs_show(struct device *dev,
2191 struct device_attribute *attr,
2192 char *buf)
2193{
2194 struct gendisk *disk = dev_to_disk(dev);
2195
2196 if (!disk->ev)
2197 return sprintf(buf, "-1\n");
2198
2199 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
2200}
2201
2202static ssize_t disk_events_poll_msecs_store(struct device *dev,
2203 struct device_attribute *attr,
2204 const char *buf, size_t count)
2205{
2206 struct gendisk *disk = dev_to_disk(dev);
2207 long intv;
2208
2209 if (!count || !sscanf(buf, "%ld", &intv))
2210 return -EINVAL;
2211
2212 if (intv < 0 && intv != -1)
2213 return -EINVAL;
2214
2215 if (!disk->ev)
2216 return -ENODEV;
2217
2218 disk_block_events(disk);
2219 disk->ev->poll_msecs = intv;
2220 __disk_unblock_events(disk, true);
2221
2222 return count;
2223}
2224
2225static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
2226static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
2227static const DEVICE_ATTR(events_poll_msecs, 0644,
2228 disk_events_poll_msecs_show,
2229 disk_events_poll_msecs_store);
2230
2231static const struct attribute *disk_events_attrs[] = {
2232 &dev_attr_events.attr,
2233 &dev_attr_events_async.attr,
2234 &dev_attr_events_poll_msecs.attr,
2235 NULL,
2236};
2237
2238/*
2239 * The default polling interval can be specified by the kernel
2240 * parameter block.events_dfl_poll_msecs which defaults to 0
2241 * (disable). This can also be modified runtime by writing to
2242 * /sys/module/block/parameters/events_dfl_poll_msecs.
2243 */
2244static int disk_events_set_dfl_poll_msecs(const char *val,
2245 const struct kernel_param *kp)
2246{
2247 struct disk_events *ev;
2248 int ret;
2249
2250 ret = param_set_ulong(val, kp);
2251 if (ret < 0)
2252 return ret;
2253
2254 mutex_lock(&disk_events_mutex);
2255
2256 list_for_each_entry(ev, &disk_events, node)
2257 disk_flush_events(ev->disk, 0);
2258
2259 mutex_unlock(&disk_events_mutex);
2260
2261 return 0;
2262}
2263
2264static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2265 .set = disk_events_set_dfl_poll_msecs,
2266 .get = param_get_ulong,
2267};
2268
2269#undef MODULE_PARAM_PREFIX
2270#define MODULE_PARAM_PREFIX "block."
2271
2272module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2273 &disk_events_dfl_poll_msecs, 0644);
2274
2275/*
2276 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2277 */
2278static void disk_alloc_events(struct gendisk *disk)
2279{
2280 struct disk_events *ev;
2281
2282 if (!disk->fops->check_events || !disk->events)
2283 return;
2284
2285 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2286 if (!ev) {
2287 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2288 return;
2289 }
2290
2291 INIT_LIST_HEAD(&ev->node);
2292 ev->disk = disk;
2293 spin_lock_init(&ev->lock);
2294 mutex_init(&ev->block_mutex);
2295 ev->block = 1;
2296 ev->poll_msecs = -1;
2297 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2298
2299 disk->ev = ev;
2300}
2301
2302static void disk_add_events(struct gendisk *disk)
2303{
2304 /* FIXME: error handling */
2305 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2306 pr_warn("%s: failed to create sysfs files for events\n",
2307 disk->disk_name);
2308
2309 if (!disk->ev)
2310 return;
2311
2312 mutex_lock(&disk_events_mutex);
2313 list_add_tail(&disk->ev->node, &disk_events);
2314 mutex_unlock(&disk_events_mutex);
2315
2316 /*
2317 * Block count is initialized to 1 and the following initial
2318 * unblock kicks it into action.
2319 */
2320 __disk_unblock_events(disk, true);
2321}
2322
2323static void disk_del_events(struct gendisk *disk)
2324{
2325 if (disk->ev) {
2326 disk_block_events(disk);
2327
2328 mutex_lock(&disk_events_mutex);
2329 list_del_init(&disk->ev->node);
2330 mutex_unlock(&disk_events_mutex);
2331 }
2332
2333 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2334}
2335
2336static void disk_release_events(struct gendisk *disk)
2337{
2338 /* the block count should be 1 from disk_del_events() */
2339 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
2340 kfree(disk->ev);
2341}