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 * 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 <linux/blktrace_api.h>
29
30#include "blk-throttle.h"
31#include "blk.h"
32#include "blk-mq-sched.h"
33#include "blk-rq-qos.h"
34#include "blk-cgroup.h"
35
36static struct kobject *block_depr;
37
38/*
39 * Unique, monotonically increasing sequential number associated with block
40 * devices instances (i.e. incremented each time a device is attached).
41 * Associating uevents with block devices in userspace is difficult and racy:
42 * the uevent netlink socket is lossy, and on slow and overloaded systems has
43 * a very high latency.
44 * Block devices do not have exclusive owners in userspace, any process can set
45 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
46 * can be reused again and again).
47 * A userspace process setting up a block device and watching for its events
48 * cannot thus reliably tell whether an event relates to the device it just set
49 * up or another earlier instance with the same name.
50 * This sequential number allows userspace processes to solve this problem, and
51 * uniquely associate an uevent to the lifetime to a device.
52 */
53static atomic64_t diskseq;
54
55/* for extended dynamic devt allocation, currently only one major is used */
56#define NR_EXT_DEVT (1 << MINORBITS)
57static DEFINE_IDA(ext_devt_ida);
58
59void set_capacity(struct gendisk *disk, sector_t sectors)
60{
61 bdev_set_nr_sectors(disk->part0, sectors);
62}
63EXPORT_SYMBOL(set_capacity);
64
65/*
66 * Set disk capacity and notify if the size is not currently zero and will not
67 * be set to zero. Returns true if a uevent was sent, otherwise false.
68 */
69bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
70{
71 sector_t capacity = get_capacity(disk);
72 char *envp[] = { "RESIZE=1", NULL };
73
74 set_capacity(disk, size);
75
76 /*
77 * Only print a message and send a uevent if the gendisk is user visible
78 * and alive. This avoids spamming the log and udev when setting the
79 * initial capacity during probing.
80 */
81 if (size == capacity ||
82 !disk_live(disk) ||
83 (disk->flags & GENHD_FL_HIDDEN))
84 return false;
85
86 pr_info("%s: detected capacity change from %lld to %lld\n",
87 disk->disk_name, capacity, size);
88
89 /*
90 * Historically we did not send a uevent for changes to/from an empty
91 * device.
92 */
93 if (!capacity || !size)
94 return false;
95 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
96 return true;
97}
98EXPORT_SYMBOL_GPL(set_capacity_and_notify);
99
100static void part_stat_read_all(struct block_device *part,
101 struct disk_stats *stat)
102{
103 int cpu;
104
105 memset(stat, 0, sizeof(struct disk_stats));
106 for_each_possible_cpu(cpu) {
107 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
108 int group;
109
110 for (group = 0; group < NR_STAT_GROUPS; group++) {
111 stat->nsecs[group] += ptr->nsecs[group];
112 stat->sectors[group] += ptr->sectors[group];
113 stat->ios[group] += ptr->ios[group];
114 stat->merges[group] += ptr->merges[group];
115 }
116
117 stat->io_ticks += ptr->io_ticks;
118 }
119}
120
121unsigned int part_in_flight(struct block_device *part)
122{
123 unsigned int inflight = 0;
124 int cpu;
125
126 for_each_possible_cpu(cpu) {
127 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
128 part_stat_local_read_cpu(part, in_flight[1], cpu);
129 }
130 if ((int)inflight < 0)
131 inflight = 0;
132
133 return inflight;
134}
135
136static void part_in_flight_rw(struct block_device *part,
137 unsigned int inflight[2])
138{
139 int cpu;
140
141 inflight[0] = 0;
142 inflight[1] = 0;
143 for_each_possible_cpu(cpu) {
144 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
145 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
146 }
147 if ((int)inflight[0] < 0)
148 inflight[0] = 0;
149 if ((int)inflight[1] < 0)
150 inflight[1] = 0;
151}
152
153/*
154 * Can be deleted altogether. Later.
155 *
156 */
157#define BLKDEV_MAJOR_HASH_SIZE 255
158static struct blk_major_name {
159 struct blk_major_name *next;
160 int major;
161 char name[16];
162#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
163 void (*probe)(dev_t devt);
164#endif
165} *major_names[BLKDEV_MAJOR_HASH_SIZE];
166static DEFINE_MUTEX(major_names_lock);
167static DEFINE_SPINLOCK(major_names_spinlock);
168
169/* index in the above - for now: assume no multimajor ranges */
170static inline int major_to_index(unsigned major)
171{
172 return major % BLKDEV_MAJOR_HASH_SIZE;
173}
174
175#ifdef CONFIG_PROC_FS
176void blkdev_show(struct seq_file *seqf, off_t offset)
177{
178 struct blk_major_name *dp;
179
180 spin_lock(&major_names_spinlock);
181 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
182 if (dp->major == offset)
183 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
184 spin_unlock(&major_names_spinlock);
185}
186#endif /* CONFIG_PROC_FS */
187
188/**
189 * __register_blkdev - register a new block device
190 *
191 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
192 * @major = 0, try to allocate any unused major number.
193 * @name: the name of the new block device as a zero terminated string
194 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
195 * pre-created device node is accessed. When a probe call uses
196 * add_disk() and it fails the driver must cleanup resources. This
197 * interface may soon be removed.
198 *
199 * The @name must be unique within the system.
200 *
201 * The return value depends on the @major input parameter:
202 *
203 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
204 * then the function returns zero on success, or a negative error code
205 * - if any unused major number was requested with @major = 0 parameter
206 * then the return value is the allocated major number in range
207 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
208 *
209 * See Documentation/admin-guide/devices.txt for the list of allocated
210 * major numbers.
211 *
212 * Use register_blkdev instead for any new code.
213 */
214int __register_blkdev(unsigned int major, const char *name,
215 void (*probe)(dev_t devt))
216{
217 struct blk_major_name **n, *p;
218 int index, ret = 0;
219
220 mutex_lock(&major_names_lock);
221
222 /* temporary */
223 if (major == 0) {
224 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
225 if (major_names[index] == NULL)
226 break;
227 }
228
229 if (index == 0) {
230 printk("%s: failed to get major for %s\n",
231 __func__, name);
232 ret = -EBUSY;
233 goto out;
234 }
235 major = index;
236 ret = major;
237 }
238
239 if (major >= BLKDEV_MAJOR_MAX) {
240 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
241 __func__, major, BLKDEV_MAJOR_MAX-1, name);
242
243 ret = -EINVAL;
244 goto out;
245 }
246
247 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
248 if (p == NULL) {
249 ret = -ENOMEM;
250 goto out;
251 }
252
253 p->major = major;
254#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
255 p->probe = probe;
256#endif
257 strscpy(p->name, name, sizeof(p->name));
258 p->next = NULL;
259 index = major_to_index(major);
260
261 spin_lock(&major_names_spinlock);
262 for (n = &major_names[index]; *n; n = &(*n)->next) {
263 if ((*n)->major == major)
264 break;
265 }
266 if (!*n)
267 *n = p;
268 else
269 ret = -EBUSY;
270 spin_unlock(&major_names_spinlock);
271
272 if (ret < 0) {
273 printk("register_blkdev: cannot get major %u for %s\n",
274 major, name);
275 kfree(p);
276 }
277out:
278 mutex_unlock(&major_names_lock);
279 return ret;
280}
281EXPORT_SYMBOL(__register_blkdev);
282
283void unregister_blkdev(unsigned int major, const char *name)
284{
285 struct blk_major_name **n;
286 struct blk_major_name *p = NULL;
287 int index = major_to_index(major);
288
289 mutex_lock(&major_names_lock);
290 spin_lock(&major_names_spinlock);
291 for (n = &major_names[index]; *n; n = &(*n)->next)
292 if ((*n)->major == major)
293 break;
294 if (!*n || strcmp((*n)->name, name)) {
295 WARN_ON(1);
296 } else {
297 p = *n;
298 *n = p->next;
299 }
300 spin_unlock(&major_names_spinlock);
301 mutex_unlock(&major_names_lock);
302 kfree(p);
303}
304
305EXPORT_SYMBOL(unregister_blkdev);
306
307int blk_alloc_ext_minor(void)
308{
309 int idx;
310
311 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
312 if (idx == -ENOSPC)
313 return -EBUSY;
314 return idx;
315}
316
317void blk_free_ext_minor(unsigned int minor)
318{
319 ida_free(&ext_devt_ida, minor);
320}
321
322void disk_uevent(struct gendisk *disk, enum kobject_action action)
323{
324 struct block_device *part;
325 unsigned long idx;
326
327 rcu_read_lock();
328 xa_for_each(&disk->part_tbl, idx, part) {
329 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
330 continue;
331 if (!kobject_get_unless_zero(&part->bd_device.kobj))
332 continue;
333
334 rcu_read_unlock();
335 kobject_uevent(bdev_kobj(part), action);
336 put_device(&part->bd_device);
337 rcu_read_lock();
338 }
339 rcu_read_unlock();
340}
341EXPORT_SYMBOL_GPL(disk_uevent);
342
343int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode)
344{
345 struct file *file;
346 int ret = 0;
347
348 if (!disk_has_partscan(disk))
349 return -EINVAL;
350 if (disk->open_partitions)
351 return -EBUSY;
352
353 /*
354 * If the device is opened exclusively by current thread already, it's
355 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
356 * synchronize with other exclusive openers and other partition
357 * scanners.
358 */
359 if (!(mode & BLK_OPEN_EXCL)) {
360 ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions,
361 NULL);
362 if (ret)
363 return ret;
364 }
365
366 set_bit(GD_NEED_PART_SCAN, &disk->state);
367 file = bdev_file_open_by_dev(disk_devt(disk), mode & ~BLK_OPEN_EXCL,
368 NULL, NULL);
369 if (IS_ERR(file))
370 ret = PTR_ERR(file);
371 else
372 fput(file);
373
374 /*
375 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
376 * and this will cause that re-assemble partitioned raid device will
377 * creat partition for underlying disk.
378 */
379 clear_bit(GD_NEED_PART_SCAN, &disk->state);
380 if (!(mode & BLK_OPEN_EXCL))
381 bd_abort_claiming(disk->part0, disk_scan_partitions);
382 return ret;
383}
384
385/**
386 * add_disk_fwnode - add disk information to kernel list with fwnode
387 * @parent: parent device for the disk
388 * @disk: per-device partitioning information
389 * @groups: Additional per-device sysfs groups
390 * @fwnode: attached disk fwnode
391 *
392 * This function registers the partitioning information in @disk
393 * with the kernel. Also attach a fwnode to the disk device.
394 */
395int __must_check add_disk_fwnode(struct device *parent, struct gendisk *disk,
396 const struct attribute_group **groups,
397 struct fwnode_handle *fwnode)
398
399{
400 struct device *ddev = disk_to_dev(disk);
401 int ret;
402
403 /* Only makes sense for bio-based to set ->poll_bio */
404 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
405 return -EINVAL;
406
407 /*
408 * The disk queue should now be all set with enough information about
409 * the device for the elevator code to pick an adequate default
410 * elevator if one is needed, that is, for devices requesting queue
411 * registration.
412 */
413 elevator_init_mq(disk->queue);
414
415 /* Mark bdev as having a submit_bio, if needed */
416 if (disk->fops->submit_bio)
417 bdev_set_flag(disk->part0, BD_HAS_SUBMIT_BIO);
418
419 /*
420 * If the driver provides an explicit major number it also must provide
421 * the number of minors numbers supported, and those will be used to
422 * setup the gendisk.
423 * Otherwise just allocate the device numbers for both the whole device
424 * and all partitions from the extended dev_t space.
425 */
426 ret = -EINVAL;
427 if (disk->major) {
428 if (WARN_ON(!disk->minors))
429 goto out_exit_elevator;
430
431 if (disk->minors > DISK_MAX_PARTS) {
432 pr_err("block: can't allocate more than %d partitions\n",
433 DISK_MAX_PARTS);
434 disk->minors = DISK_MAX_PARTS;
435 }
436 if (disk->first_minor > MINORMASK ||
437 disk->minors > MINORMASK + 1 ||
438 disk->first_minor + disk->minors > MINORMASK + 1)
439 goto out_exit_elevator;
440 } else {
441 if (WARN_ON(disk->minors))
442 goto out_exit_elevator;
443
444 ret = blk_alloc_ext_minor();
445 if (ret < 0)
446 goto out_exit_elevator;
447 disk->major = BLOCK_EXT_MAJOR;
448 disk->first_minor = ret;
449 }
450
451 /* delay uevents, until we scanned partition table */
452 dev_set_uevent_suppress(ddev, 1);
453
454 ddev->parent = parent;
455 ddev->groups = groups;
456 dev_set_name(ddev, "%s", disk->disk_name);
457 if (fwnode)
458 device_set_node(ddev, fwnode);
459 if (!(disk->flags & GENHD_FL_HIDDEN))
460 ddev->devt = MKDEV(disk->major, disk->first_minor);
461 ret = device_add(ddev);
462 if (ret)
463 goto out_free_ext_minor;
464
465 ret = disk_alloc_events(disk);
466 if (ret)
467 goto out_device_del;
468
469 ret = sysfs_create_link(block_depr, &ddev->kobj,
470 kobject_name(&ddev->kobj));
471 if (ret)
472 goto out_device_del;
473
474 /*
475 * avoid probable deadlock caused by allocating memory with
476 * GFP_KERNEL in runtime_resume callback of its all ancestor
477 * devices
478 */
479 pm_runtime_set_memalloc_noio(ddev, true);
480
481 disk->part0->bd_holder_dir =
482 kobject_create_and_add("holders", &ddev->kobj);
483 if (!disk->part0->bd_holder_dir) {
484 ret = -ENOMEM;
485 goto out_del_block_link;
486 }
487 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
488 if (!disk->slave_dir) {
489 ret = -ENOMEM;
490 goto out_put_holder_dir;
491 }
492
493 ret = blk_register_queue(disk);
494 if (ret)
495 goto out_put_slave_dir;
496
497 if (!(disk->flags & GENHD_FL_HIDDEN)) {
498 ret = bdi_register(disk->bdi, "%u:%u",
499 disk->major, disk->first_minor);
500 if (ret)
501 goto out_unregister_queue;
502 bdi_set_owner(disk->bdi, ddev);
503 ret = sysfs_create_link(&ddev->kobj,
504 &disk->bdi->dev->kobj, "bdi");
505 if (ret)
506 goto out_unregister_bdi;
507
508 /* Make sure the first partition scan will be proceed */
509 if (get_capacity(disk) && disk_has_partscan(disk))
510 set_bit(GD_NEED_PART_SCAN, &disk->state);
511
512 bdev_add(disk->part0, ddev->devt);
513 if (get_capacity(disk))
514 disk_scan_partitions(disk, BLK_OPEN_READ);
515
516 /*
517 * Announce the disk and partitions after all partitions are
518 * created. (for hidden disks uevents remain suppressed forever)
519 */
520 dev_set_uevent_suppress(ddev, 0);
521 disk_uevent(disk, KOBJ_ADD);
522 } else {
523 /*
524 * Even if the block_device for a hidden gendisk is not
525 * registered, it needs to have a valid bd_dev so that the
526 * freeing of the dynamic major works.
527 */
528 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
529 }
530
531 blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);
532 disk_add_events(disk);
533 set_bit(GD_ADDED, &disk->state);
534 return 0;
535
536out_unregister_bdi:
537 if (!(disk->flags & GENHD_FL_HIDDEN))
538 bdi_unregister(disk->bdi);
539out_unregister_queue:
540 blk_unregister_queue(disk);
541 rq_qos_exit(disk->queue);
542out_put_slave_dir:
543 kobject_put(disk->slave_dir);
544 disk->slave_dir = NULL;
545out_put_holder_dir:
546 kobject_put(disk->part0->bd_holder_dir);
547out_del_block_link:
548 sysfs_remove_link(block_depr, dev_name(ddev));
549 pm_runtime_set_memalloc_noio(ddev, false);
550out_device_del:
551 device_del(ddev);
552out_free_ext_minor:
553 if (disk->major == BLOCK_EXT_MAJOR)
554 blk_free_ext_minor(disk->first_minor);
555out_exit_elevator:
556 if (disk->queue->elevator)
557 elevator_exit(disk->queue);
558 return ret;
559}
560EXPORT_SYMBOL_GPL(add_disk_fwnode);
561
562/**
563 * device_add_disk - add disk information to kernel list
564 * @parent: parent device for the disk
565 * @disk: per-device partitioning information
566 * @groups: Additional per-device sysfs groups
567 *
568 * This function registers the partitioning information in @disk
569 * with the kernel.
570 */
571int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
572 const struct attribute_group **groups)
573{
574 return add_disk_fwnode(parent, disk, groups, NULL);
575}
576EXPORT_SYMBOL(device_add_disk);
577
578static void blk_report_disk_dead(struct gendisk *disk, bool surprise)
579{
580 struct block_device *bdev;
581 unsigned long idx;
582
583 /*
584 * On surprise disk removal, bdev_mark_dead() may call into file
585 * systems below. Make it clear that we're expecting to not hold
586 * disk->open_mutex.
587 */
588 lockdep_assert_not_held(&disk->open_mutex);
589
590 rcu_read_lock();
591 xa_for_each(&disk->part_tbl, idx, bdev) {
592 if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
593 continue;
594 rcu_read_unlock();
595
596 bdev_mark_dead(bdev, surprise);
597
598 put_device(&bdev->bd_device);
599 rcu_read_lock();
600 }
601 rcu_read_unlock();
602}
603
604static bool __blk_mark_disk_dead(struct gendisk *disk)
605{
606 /*
607 * Fail any new I/O.
608 */
609 if (test_and_set_bit(GD_DEAD, &disk->state))
610 return false;
611
612 if (test_bit(GD_OWNS_QUEUE, &disk->state))
613 blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);
614
615 /*
616 * Stop buffered writers from dirtying pages that can't be written out.
617 */
618 set_capacity(disk, 0);
619
620 /*
621 * Prevent new I/O from crossing bio_queue_enter().
622 */
623 return blk_queue_start_drain(disk->queue);
624}
625
626/**
627 * blk_mark_disk_dead - mark a disk as dead
628 * @disk: disk to mark as dead
629 *
630 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
631 * to this disk.
632 */
633void blk_mark_disk_dead(struct gendisk *disk)
634{
635 __blk_mark_disk_dead(disk);
636 blk_report_disk_dead(disk, true);
637}
638EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
639
640/**
641 * del_gendisk - remove the gendisk
642 * @disk: the struct gendisk to remove
643 *
644 * Removes the gendisk and all its associated resources. This deletes the
645 * partitions associated with the gendisk, and unregisters the associated
646 * request_queue.
647 *
648 * This is the counter to the respective __device_add_disk() call.
649 *
650 * The final removal of the struct gendisk happens when its refcount reaches 0
651 * with put_disk(), which should be called after del_gendisk(), if
652 * __device_add_disk() was used.
653 *
654 * Drivers exist which depend on the release of the gendisk to be synchronous,
655 * it should not be deferred.
656 *
657 * Context: can sleep
658 */
659void del_gendisk(struct gendisk *disk)
660{
661 struct request_queue *q = disk->queue;
662 struct block_device *part;
663 unsigned long idx;
664 bool start_drain, queue_dying;
665
666 might_sleep();
667
668 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
669 return;
670
671 disk_del_events(disk);
672
673 /*
674 * Prevent new openers by unlinked the bdev inode.
675 */
676 mutex_lock(&disk->open_mutex);
677 xa_for_each(&disk->part_tbl, idx, part)
678 bdev_unhash(part);
679 mutex_unlock(&disk->open_mutex);
680
681 /*
682 * Tell the file system to write back all dirty data and shut down if
683 * it hasn't been notified earlier.
684 */
685 if (!test_bit(GD_DEAD, &disk->state))
686 blk_report_disk_dead(disk, false);
687
688 /*
689 * Drop all partitions now that the disk is marked dead.
690 */
691 mutex_lock(&disk->open_mutex);
692 start_drain = __blk_mark_disk_dead(disk);
693 queue_dying = blk_queue_dying(q);
694 if (start_drain)
695 blk_freeze_acquire_lock(q, true, queue_dying);
696 xa_for_each_start(&disk->part_tbl, idx, part, 1)
697 drop_partition(part);
698 mutex_unlock(&disk->open_mutex);
699
700 if (!(disk->flags & GENHD_FL_HIDDEN)) {
701 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
702
703 /*
704 * Unregister bdi before releasing device numbers (as they can
705 * get reused and we'd get clashes in sysfs).
706 */
707 bdi_unregister(disk->bdi);
708 }
709
710 blk_unregister_queue(disk);
711
712 kobject_put(disk->part0->bd_holder_dir);
713 kobject_put(disk->slave_dir);
714 disk->slave_dir = NULL;
715
716 part_stat_set_all(disk->part0, 0);
717 disk->part0->bd_stamp = 0;
718 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
719 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
720 device_del(disk_to_dev(disk));
721
722 blk_mq_freeze_queue_wait(q);
723
724 blk_throtl_cancel_bios(disk);
725
726 blk_sync_queue(q);
727 blk_flush_integrity();
728
729 if (queue_is_mq(q))
730 blk_mq_cancel_work_sync(q);
731
732 blk_mq_quiesce_queue(q);
733 if (q->elevator) {
734 mutex_lock(&q->sysfs_lock);
735 elevator_exit(q);
736 mutex_unlock(&q->sysfs_lock);
737 }
738 rq_qos_exit(q);
739 blk_mq_unquiesce_queue(q);
740
741 /*
742 * If the disk does not own the queue, allow using passthrough requests
743 * again. Else leave the queue frozen to fail all I/O.
744 */
745 if (!test_bit(GD_OWNS_QUEUE, &disk->state))
746 __blk_mq_unfreeze_queue(q, true);
747 else if (queue_is_mq(q))
748 blk_mq_exit_queue(q);
749
750 if (start_drain)
751 blk_unfreeze_release_lock(q, true, queue_dying);
752}
753EXPORT_SYMBOL(del_gendisk);
754
755/**
756 * invalidate_disk - invalidate the disk
757 * @disk: the struct gendisk to invalidate
758 *
759 * A helper to invalidates the disk. It will clean the disk's associated
760 * buffer/page caches and reset its internal states so that the disk
761 * can be reused by the drivers.
762 *
763 * Context: can sleep
764 */
765void invalidate_disk(struct gendisk *disk)
766{
767 struct block_device *bdev = disk->part0;
768
769 invalidate_bdev(bdev);
770 bdev->bd_mapping->wb_err = 0;
771 set_capacity(disk, 0);
772}
773EXPORT_SYMBOL(invalidate_disk);
774
775/* sysfs access to bad-blocks list. */
776static ssize_t disk_badblocks_show(struct device *dev,
777 struct device_attribute *attr,
778 char *page)
779{
780 struct gendisk *disk = dev_to_disk(dev);
781
782 if (!disk->bb)
783 return sysfs_emit(page, "\n");
784
785 return badblocks_show(disk->bb, page, 0);
786}
787
788static ssize_t disk_badblocks_store(struct device *dev,
789 struct device_attribute *attr,
790 const char *page, size_t len)
791{
792 struct gendisk *disk = dev_to_disk(dev);
793
794 if (!disk->bb)
795 return -ENXIO;
796
797 return badblocks_store(disk->bb, page, len, 0);
798}
799
800#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
801static bool blk_probe_dev(dev_t devt)
802{
803 unsigned int major = MAJOR(devt);
804 struct blk_major_name **n;
805
806 mutex_lock(&major_names_lock);
807 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
808 if ((*n)->major == major && (*n)->probe) {
809 (*n)->probe(devt);
810 mutex_unlock(&major_names_lock);
811 return true;
812 }
813 }
814 mutex_unlock(&major_names_lock);
815 return false;
816}
817
818void blk_request_module(dev_t devt)
819{
820 int error;
821
822 if (blk_probe_dev(devt))
823 return;
824
825 error = request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt));
826 /* Make old-style 2.4 aliases work */
827 if (error > 0)
828 error = request_module("block-major-%d", MAJOR(devt));
829 if (!error)
830 blk_probe_dev(devt);
831}
832#endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
833
834#ifdef CONFIG_PROC_FS
835/* iterator */
836static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
837{
838 loff_t skip = *pos;
839 struct class_dev_iter *iter;
840 struct device *dev;
841
842 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
843 if (!iter)
844 return ERR_PTR(-ENOMEM);
845
846 seqf->private = iter;
847 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
848 do {
849 dev = class_dev_iter_next(iter);
850 if (!dev)
851 return NULL;
852 } while (skip--);
853
854 return dev_to_disk(dev);
855}
856
857static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
858{
859 struct device *dev;
860
861 (*pos)++;
862 dev = class_dev_iter_next(seqf->private);
863 if (dev)
864 return dev_to_disk(dev);
865
866 return NULL;
867}
868
869static void disk_seqf_stop(struct seq_file *seqf, void *v)
870{
871 struct class_dev_iter *iter = seqf->private;
872
873 /* stop is called even after start failed :-( */
874 if (iter) {
875 class_dev_iter_exit(iter);
876 kfree(iter);
877 seqf->private = NULL;
878 }
879}
880
881static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
882{
883 void *p;
884
885 p = disk_seqf_start(seqf, pos);
886 if (!IS_ERR_OR_NULL(p) && !*pos)
887 seq_puts(seqf, "major minor #blocks name\n\n");
888 return p;
889}
890
891static int show_partition(struct seq_file *seqf, void *v)
892{
893 struct gendisk *sgp = v;
894 struct block_device *part;
895 unsigned long idx;
896
897 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
898 return 0;
899
900 rcu_read_lock();
901 xa_for_each(&sgp->part_tbl, idx, part) {
902 if (!bdev_nr_sectors(part))
903 continue;
904 seq_printf(seqf, "%4d %7d %10llu %pg\n",
905 MAJOR(part->bd_dev), MINOR(part->bd_dev),
906 bdev_nr_sectors(part) >> 1, part);
907 }
908 rcu_read_unlock();
909 return 0;
910}
911
912static const struct seq_operations partitions_op = {
913 .start = show_partition_start,
914 .next = disk_seqf_next,
915 .stop = disk_seqf_stop,
916 .show = show_partition
917};
918#endif
919
920static int __init genhd_device_init(void)
921{
922 int error;
923
924 error = class_register(&block_class);
925 if (unlikely(error))
926 return error;
927 blk_dev_init();
928
929 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
930
931 /* create top-level block dir */
932 block_depr = kobject_create_and_add("block", NULL);
933 return 0;
934}
935
936subsys_initcall(genhd_device_init);
937
938static ssize_t disk_range_show(struct device *dev,
939 struct device_attribute *attr, char *buf)
940{
941 struct gendisk *disk = dev_to_disk(dev);
942
943 return sysfs_emit(buf, "%d\n", disk->minors);
944}
945
946static ssize_t disk_ext_range_show(struct device *dev,
947 struct device_attribute *attr, char *buf)
948{
949 struct gendisk *disk = dev_to_disk(dev);
950
951 return sysfs_emit(buf, "%d\n",
952 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
953}
954
955static ssize_t disk_removable_show(struct device *dev,
956 struct device_attribute *attr, char *buf)
957{
958 struct gendisk *disk = dev_to_disk(dev);
959
960 return sysfs_emit(buf, "%d\n",
961 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
962}
963
964static ssize_t disk_hidden_show(struct device *dev,
965 struct device_attribute *attr, char *buf)
966{
967 struct gendisk *disk = dev_to_disk(dev);
968
969 return sysfs_emit(buf, "%d\n",
970 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
971}
972
973static ssize_t disk_ro_show(struct device *dev,
974 struct device_attribute *attr, char *buf)
975{
976 struct gendisk *disk = dev_to_disk(dev);
977
978 return sysfs_emit(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
979}
980
981ssize_t part_size_show(struct device *dev,
982 struct device_attribute *attr, char *buf)
983{
984 return sysfs_emit(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
985}
986
987ssize_t part_stat_show(struct device *dev,
988 struct device_attribute *attr, char *buf)
989{
990 struct block_device *bdev = dev_to_bdev(dev);
991 struct disk_stats stat;
992 unsigned int inflight;
993
994 inflight = part_in_flight(bdev);
995 if (inflight) {
996 part_stat_lock();
997 update_io_ticks(bdev, jiffies, true);
998 part_stat_unlock();
999 }
1000 part_stat_read_all(bdev, &stat);
1001 return sysfs_emit(buf,
1002 "%8lu %8lu %8llu %8u "
1003 "%8lu %8lu %8llu %8u "
1004 "%8u %8u %8u "
1005 "%8lu %8lu %8llu %8u "
1006 "%8lu %8u"
1007 "\n",
1008 stat.ios[STAT_READ],
1009 stat.merges[STAT_READ],
1010 (unsigned long long)stat.sectors[STAT_READ],
1011 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1012 stat.ios[STAT_WRITE],
1013 stat.merges[STAT_WRITE],
1014 (unsigned long long)stat.sectors[STAT_WRITE],
1015 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1016 inflight,
1017 jiffies_to_msecs(stat.io_ticks),
1018 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1019 stat.nsecs[STAT_WRITE] +
1020 stat.nsecs[STAT_DISCARD] +
1021 stat.nsecs[STAT_FLUSH],
1022 NSEC_PER_MSEC),
1023 stat.ios[STAT_DISCARD],
1024 stat.merges[STAT_DISCARD],
1025 (unsigned long long)stat.sectors[STAT_DISCARD],
1026 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1027 stat.ios[STAT_FLUSH],
1028 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1029}
1030
1031ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1032 char *buf)
1033{
1034 struct block_device *bdev = dev_to_bdev(dev);
1035 struct request_queue *q = bdev_get_queue(bdev);
1036 unsigned int inflight[2];
1037
1038 if (queue_is_mq(q))
1039 blk_mq_in_flight_rw(q, bdev, inflight);
1040 else
1041 part_in_flight_rw(bdev, inflight);
1042
1043 return sysfs_emit(buf, "%8u %8u\n", inflight[0], inflight[1]);
1044}
1045
1046static ssize_t disk_capability_show(struct device *dev,
1047 struct device_attribute *attr, char *buf)
1048{
1049 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1050 return sysfs_emit(buf, "0\n");
1051}
1052
1053static ssize_t disk_alignment_offset_show(struct device *dev,
1054 struct device_attribute *attr,
1055 char *buf)
1056{
1057 struct gendisk *disk = dev_to_disk(dev);
1058
1059 return sysfs_emit(buf, "%d\n", bdev_alignment_offset(disk->part0));
1060}
1061
1062static ssize_t disk_discard_alignment_show(struct device *dev,
1063 struct device_attribute *attr,
1064 char *buf)
1065{
1066 struct gendisk *disk = dev_to_disk(dev);
1067
1068 return sysfs_emit(buf, "%d\n", bdev_alignment_offset(disk->part0));
1069}
1070
1071static ssize_t diskseq_show(struct device *dev,
1072 struct device_attribute *attr, char *buf)
1073{
1074 struct gendisk *disk = dev_to_disk(dev);
1075
1076 return sysfs_emit(buf, "%llu\n", disk->diskseq);
1077}
1078
1079static ssize_t partscan_show(struct device *dev,
1080 struct device_attribute *attr, char *buf)
1081{
1082 return sysfs_emit(buf, "%u\n", disk_has_partscan(dev_to_disk(dev)));
1083}
1084
1085static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1086static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1087static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1088static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1089static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1090static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1091static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1092static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1093static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1094static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1095static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1096static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1097static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1098static DEVICE_ATTR(partscan, 0444, partscan_show, NULL);
1099
1100#ifdef CONFIG_FAIL_MAKE_REQUEST
1101ssize_t part_fail_show(struct device *dev,
1102 struct device_attribute *attr, char *buf)
1103{
1104 return sysfs_emit(buf, "%d\n",
1105 bdev_test_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL));
1106}
1107
1108ssize_t part_fail_store(struct device *dev,
1109 struct device_attribute *attr,
1110 const char *buf, size_t count)
1111{
1112 int i;
1113
1114 if (count > 0 && sscanf(buf, "%d", &i) > 0) {
1115 if (i)
1116 bdev_set_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL);
1117 else
1118 bdev_clear_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL);
1119 }
1120 return count;
1121}
1122
1123static struct device_attribute dev_attr_fail =
1124 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1125#endif /* CONFIG_FAIL_MAKE_REQUEST */
1126
1127#ifdef CONFIG_FAIL_IO_TIMEOUT
1128static struct device_attribute dev_attr_fail_timeout =
1129 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1130#endif
1131
1132static struct attribute *disk_attrs[] = {
1133 &dev_attr_range.attr,
1134 &dev_attr_ext_range.attr,
1135 &dev_attr_removable.attr,
1136 &dev_attr_hidden.attr,
1137 &dev_attr_ro.attr,
1138 &dev_attr_size.attr,
1139 &dev_attr_alignment_offset.attr,
1140 &dev_attr_discard_alignment.attr,
1141 &dev_attr_capability.attr,
1142 &dev_attr_stat.attr,
1143 &dev_attr_inflight.attr,
1144 &dev_attr_badblocks.attr,
1145 &dev_attr_events.attr,
1146 &dev_attr_events_async.attr,
1147 &dev_attr_events_poll_msecs.attr,
1148 &dev_attr_diskseq.attr,
1149 &dev_attr_partscan.attr,
1150#ifdef CONFIG_FAIL_MAKE_REQUEST
1151 &dev_attr_fail.attr,
1152#endif
1153#ifdef CONFIG_FAIL_IO_TIMEOUT
1154 &dev_attr_fail_timeout.attr,
1155#endif
1156 NULL
1157};
1158
1159static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1160{
1161 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1162 struct gendisk *disk = dev_to_disk(dev);
1163
1164 if (a == &dev_attr_badblocks.attr && !disk->bb)
1165 return 0;
1166 return a->mode;
1167}
1168
1169static struct attribute_group disk_attr_group = {
1170 .attrs = disk_attrs,
1171 .is_visible = disk_visible,
1172};
1173
1174static const struct attribute_group *disk_attr_groups[] = {
1175 &disk_attr_group,
1176#ifdef CONFIG_BLK_DEV_IO_TRACE
1177 &blk_trace_attr_group,
1178#endif
1179#ifdef CONFIG_BLK_DEV_INTEGRITY
1180 &blk_integrity_attr_group,
1181#endif
1182 NULL
1183};
1184
1185/**
1186 * disk_release - releases all allocated resources of the gendisk
1187 * @dev: the device representing this disk
1188 *
1189 * This function releases all allocated resources of the gendisk.
1190 *
1191 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1192 * assigned. Since the request_queue sits on top of the gendisk for these
1193 * drivers we also call blk_put_queue() for them, and we expect the
1194 * request_queue refcount to reach 0 at this point, and so the request_queue
1195 * will also be freed prior to the disk.
1196 *
1197 * Context: can sleep
1198 */
1199static void disk_release(struct device *dev)
1200{
1201 struct gendisk *disk = dev_to_disk(dev);
1202
1203 might_sleep();
1204 WARN_ON_ONCE(disk_live(disk));
1205
1206 blk_trace_remove(disk->queue);
1207
1208 /*
1209 * To undo the all initialization from blk_mq_init_allocated_queue in
1210 * case of a probe failure where add_disk is never called we have to
1211 * call blk_mq_exit_queue here. We can't do this for the more common
1212 * teardown case (yet) as the tagset can be gone by the time the disk
1213 * is released once it was added.
1214 */
1215 if (queue_is_mq(disk->queue) &&
1216 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1217 !test_bit(GD_ADDED, &disk->state))
1218 blk_mq_exit_queue(disk->queue);
1219
1220 blkcg_exit_disk(disk);
1221
1222 bioset_exit(&disk->bio_split);
1223
1224 disk_release_events(disk);
1225 kfree(disk->random);
1226 disk_free_zone_resources(disk);
1227 xa_destroy(&disk->part_tbl);
1228
1229 disk->queue->disk = NULL;
1230 blk_put_queue(disk->queue);
1231
1232 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1233 disk->fops->free_disk(disk);
1234
1235 bdev_drop(disk->part0); /* frees the disk */
1236}
1237
1238static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1239{
1240 const struct gendisk *disk = dev_to_disk(dev);
1241
1242 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1243}
1244
1245const struct class block_class = {
1246 .name = "block",
1247 .dev_uevent = block_uevent,
1248};
1249
1250static char *block_devnode(const struct device *dev, umode_t *mode,
1251 kuid_t *uid, kgid_t *gid)
1252{
1253 struct gendisk *disk = dev_to_disk(dev);
1254
1255 if (disk->fops->devnode)
1256 return disk->fops->devnode(disk, mode);
1257 return NULL;
1258}
1259
1260const struct device_type disk_type = {
1261 .name = "disk",
1262 .groups = disk_attr_groups,
1263 .release = disk_release,
1264 .devnode = block_devnode,
1265};
1266
1267#ifdef CONFIG_PROC_FS
1268/*
1269 * aggregate disk stat collector. Uses the same stats that the sysfs
1270 * entries do, above, but makes them available through one seq_file.
1271 *
1272 * The output looks suspiciously like /proc/partitions with a bunch of
1273 * extra fields.
1274 */
1275static int diskstats_show(struct seq_file *seqf, void *v)
1276{
1277 struct gendisk *gp = v;
1278 struct block_device *hd;
1279 unsigned int inflight;
1280 struct disk_stats stat;
1281 unsigned long idx;
1282
1283 /*
1284 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1285 seq_puts(seqf, "major minor name"
1286 " rio rmerge rsect ruse wio wmerge "
1287 "wsect wuse running use aveq"
1288 "\n\n");
1289 */
1290
1291 rcu_read_lock();
1292 xa_for_each(&gp->part_tbl, idx, hd) {
1293 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1294 continue;
1295
1296 inflight = part_in_flight(hd);
1297 if (inflight) {
1298 part_stat_lock();
1299 update_io_ticks(hd, jiffies, true);
1300 part_stat_unlock();
1301 }
1302 part_stat_read_all(hd, &stat);
1303 seq_put_decimal_ull_width(seqf, "", MAJOR(hd->bd_dev), 4);
1304 seq_put_decimal_ull_width(seqf, " ", MINOR(hd->bd_dev), 7);
1305 seq_printf(seqf, " %pg", hd);
1306 seq_put_decimal_ull(seqf, " ", stat.ios[STAT_READ]);
1307 seq_put_decimal_ull(seqf, " ", stat.merges[STAT_READ]);
1308 seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_READ]);
1309 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_READ],
1310 NSEC_PER_MSEC));
1311 seq_put_decimal_ull(seqf, " ", stat.ios[STAT_WRITE]);
1312 seq_put_decimal_ull(seqf, " ", stat.merges[STAT_WRITE]);
1313 seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_WRITE]);
1314 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1315 NSEC_PER_MSEC));
1316 seq_put_decimal_ull(seqf, " ", inflight);
1317 seq_put_decimal_ull(seqf, " ", jiffies_to_msecs(stat.io_ticks));
1318 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1319 stat.nsecs[STAT_WRITE] +
1320 stat.nsecs[STAT_DISCARD] +
1321 stat.nsecs[STAT_FLUSH],
1322 NSEC_PER_MSEC));
1323 seq_put_decimal_ull(seqf, " ", stat.ios[STAT_DISCARD]);
1324 seq_put_decimal_ull(seqf, " ", stat.merges[STAT_DISCARD]);
1325 seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_DISCARD]);
1326 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1327 NSEC_PER_MSEC));
1328 seq_put_decimal_ull(seqf, " ", stat.ios[STAT_FLUSH]);
1329 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1330 NSEC_PER_MSEC));
1331 seq_putc(seqf, '\n');
1332 }
1333 rcu_read_unlock();
1334
1335 return 0;
1336}
1337
1338static const struct seq_operations diskstats_op = {
1339 .start = disk_seqf_start,
1340 .next = disk_seqf_next,
1341 .stop = disk_seqf_stop,
1342 .show = diskstats_show
1343};
1344
1345static int __init proc_genhd_init(void)
1346{
1347 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1348 proc_create_seq("partitions", 0, NULL, &partitions_op);
1349 return 0;
1350}
1351module_init(proc_genhd_init);
1352#endif /* CONFIG_PROC_FS */
1353
1354dev_t part_devt(struct gendisk *disk, u8 partno)
1355{
1356 struct block_device *part;
1357 dev_t devt = 0;
1358
1359 rcu_read_lock();
1360 part = xa_load(&disk->part_tbl, partno);
1361 if (part)
1362 devt = part->bd_dev;
1363 rcu_read_unlock();
1364
1365 return devt;
1366}
1367
1368struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1369 struct lock_class_key *lkclass)
1370{
1371 struct gendisk *disk;
1372
1373 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1374 if (!disk)
1375 return NULL;
1376
1377 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1378 goto out_free_disk;
1379
1380 disk->bdi = bdi_alloc(node_id);
1381 if (!disk->bdi)
1382 goto out_free_bioset;
1383
1384 /* bdev_alloc() might need the queue, set before the first call */
1385 disk->queue = q;
1386
1387 disk->part0 = bdev_alloc(disk, 0);
1388 if (!disk->part0)
1389 goto out_free_bdi;
1390
1391 disk->node_id = node_id;
1392 mutex_init(&disk->open_mutex);
1393 xa_init(&disk->part_tbl);
1394 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1395 goto out_destroy_part_tbl;
1396
1397 if (blkcg_init_disk(disk))
1398 goto out_erase_part0;
1399
1400 disk_init_zone_resources(disk);
1401 rand_initialize_disk(disk);
1402 disk_to_dev(disk)->class = &block_class;
1403 disk_to_dev(disk)->type = &disk_type;
1404 device_initialize(disk_to_dev(disk));
1405 inc_diskseq(disk);
1406 q->disk = disk;
1407 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1408#ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1409 INIT_LIST_HEAD(&disk->slave_bdevs);
1410#endif
1411 return disk;
1412
1413out_erase_part0:
1414 xa_erase(&disk->part_tbl, 0);
1415out_destroy_part_tbl:
1416 xa_destroy(&disk->part_tbl);
1417 disk->part0->bd_disk = NULL;
1418 bdev_drop(disk->part0);
1419out_free_bdi:
1420 bdi_put(disk->bdi);
1421out_free_bioset:
1422 bioset_exit(&disk->bio_split);
1423out_free_disk:
1424 kfree(disk);
1425 return NULL;
1426}
1427
1428struct gendisk *__blk_alloc_disk(struct queue_limits *lim, int node,
1429 struct lock_class_key *lkclass)
1430{
1431 struct queue_limits default_lim = { };
1432 struct request_queue *q;
1433 struct gendisk *disk;
1434
1435 q = blk_alloc_queue(lim ? lim : &default_lim, node);
1436 if (IS_ERR(q))
1437 return ERR_CAST(q);
1438
1439 disk = __alloc_disk_node(q, node, lkclass);
1440 if (!disk) {
1441 blk_put_queue(q);
1442 return ERR_PTR(-ENOMEM);
1443 }
1444 set_bit(GD_OWNS_QUEUE, &disk->state);
1445 return disk;
1446}
1447EXPORT_SYMBOL(__blk_alloc_disk);
1448
1449/**
1450 * put_disk - decrements the gendisk refcount
1451 * @disk: the struct gendisk to decrement the refcount for
1452 *
1453 * This decrements the refcount for the struct gendisk. When this reaches 0
1454 * we'll have disk_release() called.
1455 *
1456 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1457 * when handling probe errors (that is before add_disk() is called).
1458 *
1459 * Context: Any context, but the last reference must not be dropped from
1460 * atomic context.
1461 */
1462void put_disk(struct gendisk *disk)
1463{
1464 if (disk)
1465 put_device(disk_to_dev(disk));
1466}
1467EXPORT_SYMBOL(put_disk);
1468
1469static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1470{
1471 char event[] = "DISK_RO=1";
1472 char *envp[] = { event, NULL };
1473
1474 if (!ro)
1475 event[8] = '0';
1476 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1477}
1478
1479/**
1480 * set_disk_ro - set a gendisk read-only
1481 * @disk: gendisk to operate on
1482 * @read_only: %true to set the disk read-only, %false set the disk read/write
1483 *
1484 * This function is used to indicate whether a given disk device should have its
1485 * read-only flag set. set_disk_ro() is typically used by device drivers to
1486 * indicate whether the underlying physical device is write-protected.
1487 */
1488void set_disk_ro(struct gendisk *disk, bool read_only)
1489{
1490 if (read_only) {
1491 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1492 return;
1493 } else {
1494 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1495 return;
1496 }
1497 set_disk_ro_uevent(disk, read_only);
1498}
1499EXPORT_SYMBOL(set_disk_ro);
1500
1501void inc_diskseq(struct gendisk *disk)
1502{
1503 disk->diskseq = atomic64_inc_return(&diskseq);
1504}