projects/linux·blob master · refs

1596 lines · 41.1 KiB

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 if (sectors > BLK_DEV_MAX_SECTORS) {
62 pr_warn_once("%s: truncate capacity from %lld to %lld\n",
63 disk->disk_name, sectors,
64 BLK_DEV_MAX_SECTORS);
65 sectors = BLK_DEV_MAX_SECTORS;
66 }
67
68 bdev_set_nr_sectors(disk->part0, sectors);
69}
70EXPORT_SYMBOL(set_capacity);
71
72/*
73 * Set disk capacity and notify if the size is not currently zero and will not
74 * be set to zero. Returns true if a uevent was sent, otherwise false.
75 */
76bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
77{
78 sector_t capacity = get_capacity(disk);
79 char *envp[] = { "RESIZE=1", NULL };
80
81 set_capacity(disk, size);
82
83 /*
84 * Only print a message and send a uevent if the gendisk is user visible
85 * and alive. This avoids spamming the log and udev when setting the
86 * initial capacity during probing.
87 */
88 if (size == capacity ||
89 !disk_live(disk) ||
90 (disk->flags & GENHD_FL_HIDDEN))
91 return false;
92
93 pr_info_ratelimited("%s: detected capacity change from %lld to %lld\n",
94 disk->disk_name, capacity, size);
95
96 /*
97 * Historically we did not send a uevent for changes to/from an empty
98 * device.
99 */
100 if (!capacity || !size)
101 return false;
102 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
103 return true;
104}
105EXPORT_SYMBOL_GPL(set_capacity_and_notify);
106
107static void part_stat_read_all(struct block_device *part,
108 struct disk_stats *stat)
109{
110 int cpu;
111
112 memset(stat, 0, sizeof(struct disk_stats));
113 for_each_possible_cpu(cpu) {
114 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
115 int group;
116
117 for (group = 0; group < NR_STAT_GROUPS; group++) {
118 stat->nsecs[group] += ptr->nsecs[group];
119 stat->sectors[group] += ptr->sectors[group];
120 stat->ios[group] += ptr->ios[group];
121 stat->merges[group] += ptr->merges[group];
122 }
123
124 stat->io_ticks += ptr->io_ticks;
125 }
126}
127
128static void bdev_count_inflight_rw(struct block_device *part,
129 unsigned int inflight[2], bool mq_driver)
130{
131 int write = 0;
132 int read = 0;
133 int cpu;
134
135 if (mq_driver) {
136 blk_mq_in_driver_rw(part, inflight);
137 return;
138 }
139
140 for_each_possible_cpu(cpu) {
141 read += part_stat_local_read_cpu(part, in_flight[READ], cpu);
142 write += part_stat_local_read_cpu(part, in_flight[WRITE], cpu);
143 }
144
145 /*
146 * While iterating all CPUs, some IOs may be issued from a CPU already
147 * traversed and complete on a CPU that has not yet been traversed,
148 * causing the inflight number to be negative.
149 */
150 inflight[READ] = read > 0 ? read : 0;
151 inflight[WRITE] = write > 0 ? write : 0;
152}
153
154/**
155 * bdev_count_inflight - get the number of inflight IOs for a block device.
156 *
157 * @part: the block device.
158 *
159 * Inflight here means started IO accounting, from bdev_start_io_acct() for
160 * bio-based block device, and from blk_account_io_start() for rq-based block
161 * device.
162 */
163unsigned int bdev_count_inflight(struct block_device *part)
164{
165 unsigned int inflight[2] = {0};
166
167 bdev_count_inflight_rw(part, inflight, false);
168
169 return inflight[READ] + inflight[WRITE];
170}
171EXPORT_SYMBOL_GPL(bdev_count_inflight);
172
173/*
174 * Can be deleted altogether. Later.
175 *
176 */
177#define BLKDEV_MAJOR_HASH_SIZE 255
178static struct blk_major_name {
179 struct blk_major_name *next;
180 int major;
181 char name[16];
182#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
183 void (*probe)(dev_t devt);
184#endif
185} *major_names[BLKDEV_MAJOR_HASH_SIZE];
186static DEFINE_MUTEX(major_names_lock);
187static DEFINE_SPINLOCK(major_names_spinlock);
188
189/* index in the above - for now: assume no multimajor ranges */
190static inline int major_to_index(unsigned major)
191{
192 return major % BLKDEV_MAJOR_HASH_SIZE;
193}
194
195#ifdef CONFIG_PROC_FS
196void blkdev_show(struct seq_file *seqf, off_t offset)
197{
198 struct blk_major_name *dp;
199
200 spin_lock(&major_names_spinlock);
201 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
202 if (dp->major == offset)
203 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
204 spin_unlock(&major_names_spinlock);
205}
206#endif /* CONFIG_PROC_FS */
207
208/**
209 * __register_blkdev - register a new block device
210 *
211 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
212 * @major = 0, try to allocate any unused major number.
213 * @name: the name of the new block device as a zero terminated string
214 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
215 * pre-created device node is accessed. When a probe call uses
216 * add_disk() and it fails the driver must cleanup resources. This
217 * interface may soon be removed.
218 *
219 * The @name must be unique within the system.
220 *
221 * The return value depends on the @major input parameter:
222 *
223 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
224 * then the function returns zero on success, or a negative error code
225 * - if any unused major number was requested with @major = 0 parameter
226 * then the return value is the allocated major number in range
227 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
228 *
229 * See Documentation/admin-guide/devices.txt for the list of allocated
230 * major numbers.
231 *
232 * Use register_blkdev instead for any new code.
233 */
234int __register_blkdev(unsigned int major, const char *name,
235 void (*probe)(dev_t devt))
236{
237 struct blk_major_name **n, *p;
238 int index, ret = 0;
239
240 mutex_lock(&major_names_lock);
241
242 /* temporary */
243 if (major == 0) {
244 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
245 if (major_names[index] == NULL)
246 break;
247 }
248
249 if (index == 0) {
250 printk("%s: failed to get major for %s\n",
251 __func__, name);
252 ret = -EBUSY;
253 goto out;
254 }
255 major = index;
256 ret = major;
257 }
258
259 if (major >= BLKDEV_MAJOR_MAX) {
260 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
261 __func__, major, BLKDEV_MAJOR_MAX-1, name);
262
263 ret = -EINVAL;
264 goto out;
265 }
266
267 p = kmalloc_obj(struct blk_major_name);
268 if (p == NULL) {
269 ret = -ENOMEM;
270 goto out;
271 }
272
273 p->major = major;
274#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
275 p->probe = probe;
276#endif
277 strscpy(p->name, name, sizeof(p->name));
278 p->next = NULL;
279 index = major_to_index(major);
280
281 spin_lock(&major_names_spinlock);
282 for (n = &major_names[index]; *n; n = &(*n)->next) {
283 if ((*n)->major == major)
284 break;
285 }
286 if (!*n)
287 *n = p;
288 else
289 ret = -EBUSY;
290 spin_unlock(&major_names_spinlock);
291
292 if (ret < 0) {
293 printk("register_blkdev: cannot get major %u for %s\n",
294 major, name);
295 kfree(p);
296 }
297out:
298 mutex_unlock(&major_names_lock);
299 return ret;
300}
301EXPORT_SYMBOL(__register_blkdev);
302
303void unregister_blkdev(unsigned int major, const char *name)
304{
305 struct blk_major_name **n;
306 struct blk_major_name *p = NULL;
307 int index = major_to_index(major);
308
309 mutex_lock(&major_names_lock);
310 spin_lock(&major_names_spinlock);
311 for (n = &major_names[index]; *n; n = &(*n)->next)
312 if ((*n)->major == major)
313 break;
314 if (!*n || strcmp((*n)->name, name)) {
315 WARN_ON(1);
316 } else {
317 p = *n;
318 *n = p->next;
319 }
320 spin_unlock(&major_names_spinlock);
321 mutex_unlock(&major_names_lock);
322 kfree(p);
323}
324
325EXPORT_SYMBOL(unregister_blkdev);
326
327int blk_alloc_ext_minor(void)
328{
329 int idx;
330
331 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
332 if (idx == -ENOSPC)
333 return -EBUSY;
334 return idx;
335}
336
337void blk_free_ext_minor(unsigned int minor)
338{
339 ida_free(&ext_devt_ida, minor);
340}
341
342void disk_uevent(struct gendisk *disk, enum kobject_action action)
343{
344 struct block_device *part;
345 unsigned long idx;
346
347 rcu_read_lock();
348 xa_for_each(&disk->part_tbl, idx, part) {
349 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
350 continue;
351 if (!kobject_get_unless_zero(&part->bd_device.kobj))
352 continue;
353
354 rcu_read_unlock();
355 kobject_uevent(bdev_kobj(part), action);
356 put_device(&part->bd_device);
357 rcu_read_lock();
358 }
359 rcu_read_unlock();
360}
361EXPORT_SYMBOL_GPL(disk_uevent);
362
363int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode)
364{
365 struct file *file;
366 int ret = 0;
367
368 if (!disk_has_partscan(disk))
369 return -EINVAL;
370 if (disk->open_partitions)
371 return -EBUSY;
372
373 /*
374 * If the device is opened exclusively by current thread already, it's
375 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
376 * synchronize with other exclusive openers and other partition
377 * scanners.
378 */
379 if (!(mode & BLK_OPEN_EXCL)) {
380 ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions,
381 NULL);
382 if (ret)
383 return ret;
384 }
385
386 set_bit(GD_NEED_PART_SCAN, &disk->state);
387 file = bdev_file_open_by_dev(disk_devt(disk), mode & ~BLK_OPEN_EXCL,
388 NULL, NULL);
389 if (IS_ERR(file))
390 ret = PTR_ERR(file);
391 else
392 fput(file);
393
394 /*
395 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
396 * and this will cause that re-assemble partitioned raid device will
397 * creat partition for underlying disk.
398 */
399 clear_bit(GD_NEED_PART_SCAN, &disk->state);
400 if (!(mode & BLK_OPEN_EXCL))
401 bd_abort_claiming(disk->part0, disk_scan_partitions);
402 return ret;
403}
404
405static void add_disk_final(struct gendisk *disk)
406{
407 struct device *ddev = disk_to_dev(disk);
408
409 if (!(disk->flags & GENHD_FL_HIDDEN)) {
410 bdev_add(disk->part0, ddev->devt);
411 if (get_capacity(disk))
412 disk_scan_partitions(disk, BLK_OPEN_READ);
413
414 /*
415 * Announce the disk and partitions after all partitions are
416 * created. (for hidden disks uevents remain suppressed forever)
417 */
418 dev_set_uevent_suppress(ddev, 0);
419 disk_uevent(disk, KOBJ_ADD);
420 }
421
422 blk_apply_bdi_limits(disk->bdi, &disk->queue->limits);
423 disk_add_events(disk);
424 set_bit(GD_ADDED, &disk->state);
425}
426
427static int __add_disk(struct device *parent, struct gendisk *disk,
428 const struct attribute_group **groups,
429 struct fwnode_handle *fwnode)
430
431{
432 struct device *ddev = disk_to_dev(disk);
433 int ret;
434
435 if (WARN_ON_ONCE(bdev_nr_sectors(disk->part0) > BLK_DEV_MAX_SECTORS))
436 return -EINVAL;
437
438 if (queue_is_mq(disk->queue)) {
439 /*
440 * ->submit_bio and ->poll_bio are bypassed for blk-mq drivers.
441 */
442 if (disk->fops->submit_bio || disk->fops->poll_bio)
443 return -EINVAL;
444 } else {
445 if (!disk->fops->submit_bio)
446 return -EINVAL;
447 bdev_set_flag(disk->part0, BD_HAS_SUBMIT_BIO);
448 }
449
450 /*
451 * We do not support partitions with zoned block devices, so do not try
452 * to scan the partitions table.
453 */
454 if (blk_queue_is_zoned(disk->queue))
455 disk->flags |= GENHD_FL_NO_PART;
456
457 /*
458 * If the driver provides an explicit major number it also must provide
459 * the number of minors numbers supported, and those will be used to
460 * setup the gendisk.
461 * Otherwise just allocate the device numbers for both the whole device
462 * and all partitions from the extended dev_t space.
463 */
464 ret = -EINVAL;
465 if (disk->major) {
466 if (WARN_ON(!disk->minors))
467 goto out;
468
469 if (disk->minors > DISK_MAX_PARTS) {
470 pr_err("block: can't allocate more than %d partitions\n",
471 DISK_MAX_PARTS);
472 disk->minors = DISK_MAX_PARTS;
473 }
474 if (disk->first_minor > MINORMASK ||
475 disk->minors > MINORMASK + 1 ||
476 disk->first_minor + disk->minors > MINORMASK + 1)
477 goto out;
478 } else {
479 if (WARN_ON(disk->minors))
480 goto out;
481
482 ret = blk_alloc_ext_minor();
483 if (ret < 0)
484 goto out;
485 disk->major = BLOCK_EXT_MAJOR;
486 disk->first_minor = ret;
487 }
488
489 /* delay uevents, until we scanned partition table */
490 dev_set_uevent_suppress(ddev, 1);
491
492 ddev->parent = parent;
493 ddev->groups = groups;
494 dev_set_name(ddev, "%s", disk->disk_name);
495 if (fwnode)
496 device_set_node(ddev, fwnode);
497 if (!(disk->flags & GENHD_FL_HIDDEN))
498 ddev->devt = MKDEV(disk->major, disk->first_minor);
499 ret = device_add(ddev);
500 if (ret)
501 goto out_free_ext_minor;
502
503 ret = disk_alloc_events(disk);
504 if (ret)
505 goto out_device_del;
506
507 ret = sysfs_create_link(block_depr, &ddev->kobj,
508 kobject_name(&ddev->kobj));
509 if (ret)
510 goto out_device_del;
511
512 /*
513 * avoid probable deadlock caused by allocating memory with
514 * GFP_KERNEL in runtime_resume callback of its all ancestor
515 * devices
516 */
517 pm_runtime_set_memalloc_noio(ddev, true);
518
519 disk->part0->bd_holder_dir =
520 kobject_create_and_add("holders", &ddev->kobj);
521 if (!disk->part0->bd_holder_dir) {
522 ret = -ENOMEM;
523 goto out_del_block_link;
524 }
525 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
526 if (!disk->slave_dir) {
527 ret = -ENOMEM;
528 goto out_put_holder_dir;
529 }
530
531 ret = blk_register_queue(disk);
532 if (ret)
533 goto out_put_slave_dir;
534
535 if (!(disk->flags & GENHD_FL_HIDDEN)) {
536 ret = bdi_register(disk->bdi, "%u:%u",
537 disk->major, disk->first_minor);
538 if (ret)
539 goto out_unregister_queue;
540 bdi_set_owner(disk->bdi, ddev);
541 ret = sysfs_create_link(&ddev->kobj,
542 &disk->bdi->dev->kobj, "bdi");
543 if (ret)
544 goto out_unregister_bdi;
545 } else {
546 /*
547 * Even if the block_device for a hidden gendisk is not
548 * registered, it needs to have a valid bd_dev so that the
549 * freeing of the dynamic major works.
550 */
551 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
552 }
553 return 0;
554
555out_unregister_bdi:
556 if (!(disk->flags & GENHD_FL_HIDDEN))
557 bdi_unregister(disk->bdi);
558out_unregister_queue:
559 blk_unregister_queue(disk);
560 rq_qos_exit(disk->queue);
561out_put_slave_dir:
562 kobject_put(disk->slave_dir);
563 disk->slave_dir = NULL;
564out_put_holder_dir:
565 kobject_put(disk->part0->bd_holder_dir);
566out_del_block_link:
567 sysfs_remove_link(block_depr, dev_name(ddev));
568 pm_runtime_set_memalloc_noio(ddev, false);
569out_device_del:
570 device_del(ddev);
571out_free_ext_minor:
572 if (disk->major == BLOCK_EXT_MAJOR)
573 blk_free_ext_minor(disk->first_minor);
574out:
575 return ret;
576}
577
578/**
579 * add_disk_fwnode - add disk information to kernel list with fwnode
580 * @parent: parent device for the disk
581 * @disk: per-device partitioning information
582 * @groups: Additional per-device sysfs groups
583 * @fwnode: attached disk fwnode
584 *
585 * This function registers the partitioning information in @disk
586 * with the kernel. Also attach a fwnode to the disk device.
587 */
588int __must_check add_disk_fwnode(struct device *parent, struct gendisk *disk,
589 const struct attribute_group **groups,
590 struct fwnode_handle *fwnode)
591{
592 struct blk_mq_tag_set *set;
593 unsigned int memflags;
594 int ret;
595
596 if (queue_is_mq(disk->queue)) {
597 set = disk->queue->tag_set;
598 memflags = memalloc_noio_save();
599 down_read(&set->update_nr_hwq_lock);
600 ret = __add_disk(parent, disk, groups, fwnode);
601 up_read(&set->update_nr_hwq_lock);
602 memalloc_noio_restore(memflags);
603 } else {
604 ret = __add_disk(parent, disk, groups, fwnode);
605 }
606
607 /*
608 * add_disk_final() needn't to read `nr_hw_queues`, so move it out
609 * of read lock `set->update_nr_hwq_lock` for avoiding unnecessary
610 * lock dependency on `disk->open_mutex` from scanning partition.
611 */
612 if (!ret)
613 add_disk_final(disk);
614 return ret;
615}
616EXPORT_SYMBOL_GPL(add_disk_fwnode);
617
618/**
619 * device_add_disk - add disk information to kernel list
620 * @parent: parent device for the disk
621 * @disk: per-device partitioning information
622 * @groups: Additional per-device sysfs groups
623 *
624 * This function registers the partitioning information in @disk
625 * with the kernel.
626 */
627int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
628 const struct attribute_group **groups)
629{
630 return add_disk_fwnode(parent, disk, groups, NULL);
631}
632EXPORT_SYMBOL(device_add_disk);
633
634static void blk_report_disk_dead(struct gendisk *disk, bool surprise)
635{
636 struct block_device *bdev;
637 unsigned long idx;
638
639 /*
640 * On surprise disk removal, bdev_mark_dead() may call into file
641 * systems below. Make it clear that we're expecting to not hold
642 * disk->open_mutex.
643 */
644 lockdep_assert_not_held(&disk->open_mutex);
645
646 rcu_read_lock();
647 xa_for_each(&disk->part_tbl, idx, bdev) {
648 if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
649 continue;
650 rcu_read_unlock();
651
652 bdev_mark_dead(bdev, surprise);
653
654 put_device(&bdev->bd_device);
655 rcu_read_lock();
656 }
657 rcu_read_unlock();
658}
659
660static bool __blk_mark_disk_dead(struct gendisk *disk)
661{
662 /*
663 * Fail any new I/O.
664 */
665 if (test_and_set_bit(GD_DEAD, &disk->state))
666 return false;
667
668 if (test_bit(GD_OWNS_QUEUE, &disk->state))
669 blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);
670
671 /*
672 * Stop buffered writers from dirtying pages that can't be written out.
673 */
674 set_capacity(disk, 0);
675
676 /*
677 * Prevent new I/O from crossing bio_queue_enter().
678 */
679 return blk_queue_start_drain(disk->queue);
680}
681
682/**
683 * blk_mark_disk_dead - mark a disk as dead
684 * @disk: disk to mark as dead
685 *
686 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
687 * to this disk.
688 */
689void blk_mark_disk_dead(struct gendisk *disk)
690{
691 blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);
692 __blk_mark_disk_dead(disk);
693 blk_report_disk_dead(disk, true);
694}
695EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
696
697static void __del_gendisk(struct gendisk *disk)
698{
699 struct request_queue *q = disk->queue;
700 struct block_device *part;
701 unsigned long idx;
702 bool start_drain;
703
704 might_sleep();
705
706 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
707 return;
708
709 disk_del_events(disk);
710
711 /*
712 * Prevent new openers by unlinked the bdev inode.
713 */
714 mutex_lock(&disk->open_mutex);
715 xa_for_each(&disk->part_tbl, idx, part)
716 bdev_unhash(part);
717 mutex_unlock(&disk->open_mutex);
718
719 /*
720 * Tell the file system to write back all dirty data and shut down if
721 * it hasn't been notified earlier.
722 */
723 if (!test_bit(GD_DEAD, &disk->state))
724 blk_report_disk_dead(disk, false);
725
726 /*
727 * Drop all partitions now that the disk is marked dead.
728 */
729 mutex_lock(&disk->open_mutex);
730 start_drain = __blk_mark_disk_dead(disk);
731 if (start_drain)
732 blk_freeze_acquire_lock(q);
733 xa_for_each_start(&disk->part_tbl, idx, part, 1)
734 drop_partition(part);
735 mutex_unlock(&disk->open_mutex);
736
737 if (!(disk->flags & GENHD_FL_HIDDEN)) {
738 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
739
740 /*
741 * Unregister bdi before releasing device numbers (as they can
742 * get reused and we'd get clashes in sysfs).
743 */
744 bdi_unregister(disk->bdi);
745 }
746
747 blk_unregister_queue(disk);
748
749 kobject_put(disk->part0->bd_holder_dir);
750 kobject_put(disk->slave_dir);
751 disk->slave_dir = NULL;
752
753 part_stat_set_all(disk->part0, 0);
754 disk->part0->bd_stamp = 0;
755 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
756 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
757 device_del(disk_to_dev(disk));
758
759 blk_mq_freeze_queue_wait(q);
760
761 blk_throtl_cancel_bios(disk);
762
763 blk_sync_queue(q);
764 blk_flush_integrity();
765
766 if (queue_is_mq(q))
767 blk_mq_cancel_work_sync(q);
768
769 rq_qos_exit(q);
770
771 /*
772 * If the disk does not own the queue, allow using passthrough requests
773 * again. Else leave the queue frozen to fail all I/O.
774 */
775 if (!test_bit(GD_OWNS_QUEUE, &disk->state))
776 __blk_mq_unfreeze_queue(q, true);
777 else if (queue_is_mq(q))
778 blk_mq_exit_queue(q);
779
780 if (start_drain)
781 blk_unfreeze_release_lock(q);
782}
783
784static void disable_elv_switch(struct request_queue *q)
785{
786 struct blk_mq_tag_set *set = q->tag_set;
787 WARN_ON_ONCE(!queue_is_mq(q));
788
789 down_write(&set->update_nr_hwq_lock);
790 blk_queue_flag_set(QUEUE_FLAG_NO_ELV_SWITCH, q);
791 up_write(&set->update_nr_hwq_lock);
792}
793
794/**
795 * del_gendisk - remove the gendisk
796 * @disk: the struct gendisk to remove
797 *
798 * Removes the gendisk and all its associated resources. This deletes the
799 * partitions associated with the gendisk, and unregisters the associated
800 * request_queue.
801 *
802 * This is the counter to the respective device_add_disk() call.
803 *
804 * The final removal of the struct gendisk happens when its refcount reaches 0
805 * with put_disk(), which should be called after del_gendisk(), if
806 * device_add_disk() was used.
807 *
808 * Drivers exist which depend on the release of the gendisk to be synchronous,
809 * it should not be deferred.
810 *
811 * Context: can sleep
812 */
813void del_gendisk(struct gendisk *disk)
814{
815 struct blk_mq_tag_set *set;
816 unsigned int memflags;
817
818 if (!queue_is_mq(disk->queue)) {
819 __del_gendisk(disk);
820 } else {
821 set = disk->queue->tag_set;
822
823 disable_elv_switch(disk->queue);
824
825 memflags = memalloc_noio_save();
826 down_read(&set->update_nr_hwq_lock);
827 __del_gendisk(disk);
828 up_read(&set->update_nr_hwq_lock);
829 memalloc_noio_restore(memflags);
830 }
831}
832EXPORT_SYMBOL(del_gendisk);
833
834/**
835 * invalidate_disk - invalidate the disk
836 * @disk: the struct gendisk to invalidate
837 *
838 * A helper to invalidates the disk. It will clean the disk's associated
839 * buffer/page caches and reset its internal states so that the disk
840 * can be reused by the drivers.
841 *
842 * Context: can sleep
843 */
844void invalidate_disk(struct gendisk *disk)
845{
846 struct block_device *bdev = disk->part0;
847
848 invalidate_bdev(bdev);
849 bdev->bd_mapping->wb_err = 0;
850 set_capacity(disk, 0);
851}
852EXPORT_SYMBOL(invalidate_disk);
853
854/* sysfs access to bad-blocks list. */
855static ssize_t disk_badblocks_show(struct device *dev,
856 struct device_attribute *attr,
857 char *page)
858{
859 struct gendisk *disk = dev_to_disk(dev);
860
861 if (!disk->bb)
862 return sysfs_emit(page, "\n");
863
864 return badblocks_show(disk->bb, page, 0);
865}
866
867static ssize_t disk_badblocks_store(struct device *dev,
868 struct device_attribute *attr,
869 const char *page, size_t len)
870{
871 struct gendisk *disk = dev_to_disk(dev);
872
873 if (!disk->bb)
874 return -ENXIO;
875
876 return badblocks_store(disk->bb, page, len, 0);
877}
878
879#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
880static bool blk_probe_dev(dev_t devt)
881{
882 unsigned int major = MAJOR(devt);
883 struct blk_major_name **n;
884
885 mutex_lock(&major_names_lock);
886 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
887 if ((*n)->major == major && (*n)->probe) {
888 (*n)->probe(devt);
889 mutex_unlock(&major_names_lock);
890 return true;
891 }
892 }
893 mutex_unlock(&major_names_lock);
894 return false;
895}
896
897void blk_request_module(dev_t devt)
898{
899 int error;
900
901 if (blk_probe_dev(devt))
902 return;
903
904 error = request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt));
905 /* Make old-style 2.4 aliases work */
906 if (error > 0)
907 error = request_module("block-major-%d", MAJOR(devt));
908 if (!error)
909 blk_probe_dev(devt);
910}
911#endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
912
913#ifdef CONFIG_PROC_FS
914/* iterator */
915static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
916{
917 loff_t skip = *pos;
918 struct class_dev_iter *iter;
919 struct device *dev;
920
921 iter = kmalloc_obj(*iter);
922 if (!iter)
923 return ERR_PTR(-ENOMEM);
924
925 seqf->private = iter;
926 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
927 do {
928 dev = class_dev_iter_next(iter);
929 if (!dev)
930 return NULL;
931 } while (skip--);
932
933 return dev_to_disk(dev);
934}
935
936static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
937{
938 struct device *dev;
939
940 (*pos)++;
941 dev = class_dev_iter_next(seqf->private);
942 if (dev)
943 return dev_to_disk(dev);
944
945 return NULL;
946}
947
948static void disk_seqf_stop(struct seq_file *seqf, void *v)
949{
950 struct class_dev_iter *iter = seqf->private;
951
952 /* stop is called even after start failed :-( */
953 if (iter) {
954 class_dev_iter_exit(iter);
955 kfree(iter);
956 seqf->private = NULL;
957 }
958}
959
960static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
961{
962 void *p;
963
964 p = disk_seqf_start(seqf, pos);
965 if (!IS_ERR_OR_NULL(p) && !*pos)
966 seq_puts(seqf, "major minor #blocks name\n\n");
967 return p;
968}
969
970static int show_partition(struct seq_file *seqf, void *v)
971{
972 struct gendisk *sgp = v;
973 struct block_device *part;
974 unsigned long idx;
975
976 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
977 return 0;
978
979 rcu_read_lock();
980 xa_for_each(&sgp->part_tbl, idx, part) {
981 if (!bdev_nr_sectors(part))
982 continue;
983 seq_printf(seqf, "%4d %7d %10llu %pg\n",
984 MAJOR(part->bd_dev), MINOR(part->bd_dev),
985 bdev_nr_sectors(part) >> 1, part);
986 }
987 rcu_read_unlock();
988 return 0;
989}
990
991static const struct seq_operations partitions_op = {
992 .start = show_partition_start,
993 .next = disk_seqf_next,
994 .stop = disk_seqf_stop,
995 .show = show_partition
996};
997#endif
998
999static int __init genhd_device_init(void)
1000{
1001 int error;
1002
1003 error = class_register(&block_class);
1004 if (unlikely(error))
1005 return error;
1006 blk_dev_init();
1007
1008 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1009
1010 /* create top-level block dir */
1011 block_depr = kobject_create_and_add("block", NULL);
1012 return 0;
1013}
1014
1015subsys_initcall(genhd_device_init);
1016
1017static ssize_t disk_range_show(struct device *dev,
1018 struct device_attribute *attr, char *buf)
1019{
1020 struct gendisk *disk = dev_to_disk(dev);
1021
1022 return sysfs_emit(buf, "%d\n", disk->minors);
1023}
1024
1025static ssize_t disk_ext_range_show(struct device *dev,
1026 struct device_attribute *attr, char *buf)
1027{
1028 struct gendisk *disk = dev_to_disk(dev);
1029
1030 return sysfs_emit(buf, "%d\n",
1031 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
1032}
1033
1034static ssize_t disk_removable_show(struct device *dev,
1035 struct device_attribute *attr, char *buf)
1036{
1037 struct gendisk *disk = dev_to_disk(dev);
1038
1039 return sysfs_emit(buf, "%d\n",
1040 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1041}
1042
1043static ssize_t disk_hidden_show(struct device *dev,
1044 struct device_attribute *attr, char *buf)
1045{
1046 struct gendisk *disk = dev_to_disk(dev);
1047
1048 return sysfs_emit(buf, "%d\n",
1049 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1050}
1051
1052static ssize_t disk_ro_show(struct device *dev,
1053 struct device_attribute *attr, char *buf)
1054{
1055 struct gendisk *disk = dev_to_disk(dev);
1056
1057 return sysfs_emit(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1058}
1059
1060ssize_t part_size_show(struct device *dev,
1061 struct device_attribute *attr, char *buf)
1062{
1063 return sysfs_emit(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
1064}
1065
1066ssize_t part_stat_show(struct device *dev,
1067 struct device_attribute *attr, char *buf)
1068{
1069 struct block_device *bdev = dev_to_bdev(dev);
1070 struct disk_stats stat;
1071 unsigned int inflight;
1072
1073 inflight = bdev_count_inflight(bdev);
1074 if (inflight) {
1075 part_stat_lock();
1076 update_io_ticks(bdev, jiffies, true);
1077 part_stat_unlock();
1078 }
1079 part_stat_read_all(bdev, &stat);
1080 return sysfs_emit(buf,
1081 "%8lu %8lu %8llu %8u "
1082 "%8lu %8lu %8llu %8u "
1083 "%8u %8u %8u "
1084 "%8lu %8lu %8llu %8u "
1085 "%8lu %8u"
1086 "\n",
1087 stat.ios[STAT_READ],
1088 stat.merges[STAT_READ],
1089 (unsigned long long)stat.sectors[STAT_READ],
1090 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
1091 stat.ios[STAT_WRITE],
1092 stat.merges[STAT_WRITE],
1093 (unsigned long long)stat.sectors[STAT_WRITE],
1094 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
1095 inflight,
1096 jiffies_to_msecs(stat.io_ticks),
1097 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1098 stat.nsecs[STAT_WRITE] +
1099 stat.nsecs[STAT_DISCARD] +
1100 stat.nsecs[STAT_FLUSH],
1101 NSEC_PER_MSEC),
1102 stat.ios[STAT_DISCARD],
1103 stat.merges[STAT_DISCARD],
1104 (unsigned long long)stat.sectors[STAT_DISCARD],
1105 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1106 stat.ios[STAT_FLUSH],
1107 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1108}
1109
1110/*
1111 * Show the number of IOs issued to driver.
1112 * For bio-based device, started from bdev_start_io_acct();
1113 * For rq-based device, started from blk_mq_start_request();
1114 */
1115ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1116 char *buf)
1117{
1118 struct block_device *bdev = dev_to_bdev(dev);
1119 struct request_queue *q = bdev_get_queue(bdev);
1120 unsigned int inflight[2] = {0};
1121
1122 bdev_count_inflight_rw(bdev, inflight, queue_is_mq(q));
1123
1124 return sysfs_emit(buf, "%8u %8u\n", inflight[READ], inflight[WRITE]);
1125}
1126
1127static ssize_t disk_capability_show(struct device *dev,
1128 struct device_attribute *attr, char *buf)
1129{
1130 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1131 return sysfs_emit(buf, "0\n");
1132}
1133
1134static ssize_t disk_alignment_offset_show(struct device *dev,
1135 struct device_attribute *attr,
1136 char *buf)
1137{
1138 struct gendisk *disk = dev_to_disk(dev);
1139
1140 return sysfs_emit(buf, "%d\n", bdev_alignment_offset(disk->part0));
1141}
1142
1143static ssize_t disk_discard_alignment_show(struct device *dev,
1144 struct device_attribute *attr,
1145 char *buf)
1146{
1147 struct gendisk *disk = dev_to_disk(dev);
1148
1149 return sysfs_emit(buf, "%d\n", bdev_alignment_offset(disk->part0));
1150}
1151
1152static ssize_t diskseq_show(struct device *dev,
1153 struct device_attribute *attr, char *buf)
1154{
1155 struct gendisk *disk = dev_to_disk(dev);
1156
1157 return sysfs_emit(buf, "%llu\n", disk->diskseq);
1158}
1159
1160static ssize_t partscan_show(struct device *dev,
1161 struct device_attribute *attr, char *buf)
1162{
1163 return sysfs_emit(buf, "%u\n", disk_has_partscan(dev_to_disk(dev)));
1164}
1165
1166static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1167static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1168static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1169static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1170static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1171static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1172static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1173static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1174static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1175static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1176static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1177static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1178static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1179static DEVICE_ATTR(partscan, 0444, partscan_show, NULL);
1180
1181#ifdef CONFIG_FAIL_MAKE_REQUEST
1182ssize_t part_fail_show(struct device *dev,
1183 struct device_attribute *attr, char *buf)
1184{
1185 return sysfs_emit(buf, "%d\n",
1186 bdev_test_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL));
1187}
1188
1189ssize_t part_fail_store(struct device *dev,
1190 struct device_attribute *attr,
1191 const char *buf, size_t count)
1192{
1193 int i;
1194
1195 if (count > 0 && sscanf(buf, "%d", &i) > 0) {
1196 if (i)
1197 bdev_set_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL);
1198 else
1199 bdev_clear_flag(dev_to_bdev(dev), BD_MAKE_IT_FAIL);
1200 }
1201 return count;
1202}
1203
1204static struct device_attribute dev_attr_fail =
1205 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1206#endif /* CONFIG_FAIL_MAKE_REQUEST */
1207
1208#ifdef CONFIG_FAIL_IO_TIMEOUT
1209static struct device_attribute dev_attr_fail_timeout =
1210 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1211#endif
1212
1213static struct attribute *disk_attrs[] = {
1214 &dev_attr_range.attr,
1215 &dev_attr_ext_range.attr,
1216 &dev_attr_removable.attr,
1217 &dev_attr_hidden.attr,
1218 &dev_attr_ro.attr,
1219 &dev_attr_size.attr,
1220 &dev_attr_alignment_offset.attr,
1221 &dev_attr_discard_alignment.attr,
1222 &dev_attr_capability.attr,
1223 &dev_attr_stat.attr,
1224 &dev_attr_inflight.attr,
1225 &dev_attr_badblocks.attr,
1226 &dev_attr_events.attr,
1227 &dev_attr_events_async.attr,
1228 &dev_attr_events_poll_msecs.attr,
1229 &dev_attr_diskseq.attr,
1230 &dev_attr_partscan.attr,
1231#ifdef CONFIG_FAIL_MAKE_REQUEST
1232 &dev_attr_fail.attr,
1233#endif
1234#ifdef CONFIG_FAIL_IO_TIMEOUT
1235 &dev_attr_fail_timeout.attr,
1236#endif
1237 NULL
1238};
1239
1240static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1241{
1242 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1243 struct gendisk *disk = dev_to_disk(dev);
1244
1245 if (a == &dev_attr_badblocks.attr && !disk->bb)
1246 return 0;
1247 return a->mode;
1248}
1249
1250static struct attribute_group disk_attr_group = {
1251 .attrs = disk_attrs,
1252 .is_visible = disk_visible,
1253};
1254
1255static const struct attribute_group *disk_attr_groups[] = {
1256 &disk_attr_group,
1257#ifdef CONFIG_BLK_DEV_IO_TRACE
1258 &blk_trace_attr_group,
1259#endif
1260#ifdef CONFIG_BLK_DEV_INTEGRITY
1261 &blk_integrity_attr_group,
1262#endif
1263 NULL
1264};
1265
1266/**
1267 * disk_release - releases all allocated resources of the gendisk
1268 * @dev: the device representing this disk
1269 *
1270 * This function releases all allocated resources of the gendisk.
1271 *
1272 * Drivers which used device_add_disk() have a gendisk with a request_queue
1273 * assigned. Since the request_queue sits on top of the gendisk for these
1274 * drivers we also call blk_put_queue() for them, and we expect the
1275 * request_queue refcount to reach 0 at this point, and so the request_queue
1276 * will also be freed prior to the disk.
1277 *
1278 * Context: can sleep
1279 */
1280static void disk_release(struct device *dev)
1281{
1282 struct gendisk *disk = dev_to_disk(dev);
1283
1284 might_sleep();
1285 WARN_ON_ONCE(disk_live(disk));
1286
1287 blk_trace_remove(disk->queue);
1288
1289 /*
1290 * To undo the all initialization from blk_mq_init_allocated_queue in
1291 * case of a probe failure where add_disk is never called we have to
1292 * call blk_mq_exit_queue here, after stopping the timer and work items
1293 * that I/O issued before add_disk may have left pending. We can't do
1294 * this for the more common teardown case (yet) as the tagset can be
1295 * gone by the time the disk is released once it was added.
1296 */
1297 if (queue_is_mq(disk->queue) &&
1298 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1299 !test_bit(GD_ADDED, &disk->state)) {
1300 blk_sync_queue(disk->queue);
1301 blk_mq_cancel_work_sync(disk->queue);
1302 blk_mq_exit_queue(disk->queue);
1303 }
1304
1305 blkcg_exit_disk(disk);
1306
1307 bioset_exit(&disk->bio_split);
1308
1309 disk_release_events(disk);
1310 kfree(disk->random);
1311 disk_release_zone_resources(disk);
1312 xa_destroy(&disk->part_tbl);
1313
1314 kobject_put(&disk->queue_kobj);
1315 disk->queue->disk = NULL;
1316 blk_put_queue(disk->queue);
1317
1318 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1319 disk->fops->free_disk(disk);
1320
1321 bdev_drop(disk->part0); /* frees the disk */
1322}
1323
1324static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1325{
1326 const struct gendisk *disk = dev_to_disk(dev);
1327
1328 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1329}
1330
1331const struct class block_class = {
1332 .name = "block",
1333 .dev_uevent = block_uevent,
1334};
1335
1336static char *block_devnode(const struct device *dev, umode_t *mode,
1337 kuid_t *uid, kgid_t *gid)
1338{
1339 struct gendisk *disk = dev_to_disk(dev);
1340
1341 if (disk->fops->devnode)
1342 return disk->fops->devnode(disk, mode);
1343 return NULL;
1344}
1345
1346const struct device_type disk_type = {
1347 .name = "disk",
1348 .groups = disk_attr_groups,
1349 .release = disk_release,
1350 .devnode = block_devnode,
1351};
1352
1353#ifdef CONFIG_PROC_FS
1354/*
1355 * aggregate disk stat collector. Uses the same stats that the sysfs
1356 * entries do, above, but makes them available through one seq_file.
1357 *
1358 * The output looks suspiciously like /proc/partitions with a bunch of
1359 * extra fields.
1360 */
1361static int diskstats_show(struct seq_file *seqf, void *v)
1362{
1363 struct gendisk *gp = v;
1364 struct block_device *hd;
1365 unsigned int inflight;
1366 struct disk_stats stat;
1367 unsigned long idx;
1368
1369 /*
1370 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1371 seq_puts(seqf, "major minor name"
1372 " rio rmerge rsect ruse wio wmerge "
1373 "wsect wuse running use aveq"
1374 "\n\n");
1375 */
1376
1377 rcu_read_lock();
1378 xa_for_each(&gp->part_tbl, idx, hd) {
1379 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1380 continue;
1381
1382 inflight = bdev_count_inflight(hd);
1383 if (inflight) {
1384 part_stat_lock();
1385 update_io_ticks(hd, jiffies, true);
1386 part_stat_unlock();
1387 }
1388 part_stat_read_all(hd, &stat);
1389 seq_put_decimal_ull_width(seqf, "", MAJOR(hd->bd_dev), 4);
1390 seq_put_decimal_ull_width(seqf, " ", MINOR(hd->bd_dev), 7);
1391 seq_printf(seqf, " %pg", hd);
1392 seq_put_decimal_ull(seqf, " ", stat.ios[STAT_READ]);
1393 seq_put_decimal_ull(seqf, " ", stat.merges[STAT_READ]);
1394 seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_READ]);
1395 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_READ],
1396 NSEC_PER_MSEC));
1397 seq_put_decimal_ull(seqf, " ", stat.ios[STAT_WRITE]);
1398 seq_put_decimal_ull(seqf, " ", stat.merges[STAT_WRITE]);
1399 seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_WRITE]);
1400 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1401 NSEC_PER_MSEC));
1402 seq_put_decimal_ull(seqf, " ", inflight);
1403 seq_put_decimal_ull(seqf, " ", jiffies_to_msecs(stat.io_ticks));
1404 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1405 stat.nsecs[STAT_WRITE] +
1406 stat.nsecs[STAT_DISCARD] +
1407 stat.nsecs[STAT_FLUSH],
1408 NSEC_PER_MSEC));
1409 seq_put_decimal_ull(seqf, " ", stat.ios[STAT_DISCARD]);
1410 seq_put_decimal_ull(seqf, " ", stat.merges[STAT_DISCARD]);
1411 seq_put_decimal_ull(seqf, " ", stat.sectors[STAT_DISCARD]);
1412 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1413 NSEC_PER_MSEC));
1414 seq_put_decimal_ull(seqf, " ", stat.ios[STAT_FLUSH]);
1415 seq_put_decimal_ull(seqf, " ", (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1416 NSEC_PER_MSEC));
1417 seq_putc(seqf, '\n');
1418 }
1419 rcu_read_unlock();
1420
1421 return 0;
1422}
1423
1424static const struct seq_operations diskstats_op = {
1425 .start = disk_seqf_start,
1426 .next = disk_seqf_next,
1427 .stop = disk_seqf_stop,
1428 .show = diskstats_show
1429};
1430
1431static int __init proc_genhd_init(void)
1432{
1433 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1434 proc_create_seq("partitions", 0, NULL, &partitions_op);
1435 return 0;
1436}
1437module_init(proc_genhd_init);
1438#endif /* CONFIG_PROC_FS */
1439
1440dev_t part_devt(struct gendisk *disk, u8 partno)
1441{
1442 struct block_device *part;
1443 dev_t devt = 0;
1444
1445 rcu_read_lock();
1446 part = xa_load(&disk->part_tbl, partno);
1447 if (part)
1448 devt = part->bd_dev;
1449 rcu_read_unlock();
1450
1451 return devt;
1452}
1453
1454struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1455 struct lock_class_key *lkclass)
1456{
1457 struct gendisk *disk;
1458
1459 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1460 if (!disk)
1461 return NULL;
1462
1463 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1464 goto out_free_disk;
1465
1466 disk->bdi = bdi_alloc(node_id);
1467 if (!disk->bdi)
1468 goto out_free_bioset;
1469
1470 /* bdev_alloc() might need the queue, set before the first call */
1471 disk->queue = q;
1472
1473 disk->part0 = bdev_alloc(disk, 0);
1474 if (!disk->part0)
1475 goto out_free_bdi;
1476
1477 disk->node_id = node_id;
1478 mutex_init(&disk->open_mutex);
1479 xa_init(&disk->part_tbl);
1480 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1481 goto out_destroy_part_tbl;
1482
1483 if (blkcg_init_disk(disk))
1484 goto out_erase_part0;
1485
1486 disk_init_zone_resources(disk);
1487 rand_initialize_disk(disk);
1488 disk_to_dev(disk)->class = &block_class;
1489 disk_to_dev(disk)->type = &disk_type;
1490 device_initialize(disk_to_dev(disk));
1491 inc_diskseq(disk);
1492 q->disk = disk;
1493 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1494#ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1495 INIT_LIST_HEAD(&disk->slave_bdevs);
1496#endif
1497#ifdef CONFIG_BLK_ERROR_INJECTION
1498 mutex_init(&disk->error_injection_lock);
1499 INIT_LIST_HEAD(&disk->error_injection_list);
1500#endif
1501 mutex_init(&disk->rqos_state_mutex);
1502 kobject_init(&disk->queue_kobj, &blk_queue_ktype);
1503 return disk;
1504
1505out_erase_part0:
1506 xa_erase(&disk->part_tbl, 0);
1507out_destroy_part_tbl:
1508 xa_destroy(&disk->part_tbl);
1509 disk->part0->bd_disk = NULL;
1510 bdev_drop(disk->part0);
1511out_free_bdi:
1512 bdi_put(disk->bdi);
1513out_free_bioset:
1514 bioset_exit(&disk->bio_split);
1515out_free_disk:
1516 kfree(disk);
1517 return NULL;
1518}
1519
1520struct gendisk *__blk_alloc_disk(struct queue_limits *lim, int node,
1521 struct lock_class_key *lkclass)
1522{
1523 struct queue_limits default_lim = { };
1524 struct request_queue *q;
1525 struct gendisk *disk;
1526
1527 q = blk_alloc_queue(lim ? lim : &default_lim, node);
1528 if (IS_ERR(q))
1529 return ERR_CAST(q);
1530
1531 disk = __alloc_disk_node(q, node, lkclass);
1532 if (!disk) {
1533 blk_put_queue(q);
1534 return ERR_PTR(-ENOMEM);
1535 }
1536 set_bit(GD_OWNS_QUEUE, &disk->state);
1537 return disk;
1538}
1539EXPORT_SYMBOL(__blk_alloc_disk);
1540
1541/**
1542 * put_disk - decrements the gendisk refcount
1543 * @disk: the struct gendisk to decrement the refcount for
1544 *
1545 * This decrements the refcount for the struct gendisk. When this reaches 0
1546 * we'll have disk_release() called.
1547 *
1548 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1549 * when handling probe errors (that is before add_disk() is called).
1550 *
1551 * Context: Any context, but the last reference must not be dropped from
1552 * atomic context.
1553 */
1554void put_disk(struct gendisk *disk)
1555{
1556 if (disk)
1557 put_device(disk_to_dev(disk));
1558}
1559EXPORT_SYMBOL(put_disk);
1560
1561static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1562{
1563 char event[] = "DISK_RO=1";
1564 char *envp[] = { event, NULL };
1565
1566 if (!ro)
1567 event[8] = '0';
1568 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1569}
1570
1571/**
1572 * set_disk_ro - set a gendisk read-only
1573 * @disk: gendisk to operate on
1574 * @read_only: %true to set the disk read-only, %false set the disk read/write
1575 *
1576 * This function is used to indicate whether a given disk device should have its
1577 * read-only flag set. set_disk_ro() is typically used by device drivers to
1578 * indicate whether the underlying physical device is write-protected.
1579 */
1580void set_disk_ro(struct gendisk *disk, bool read_only)
1581{
1582 if (read_only) {
1583 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1584 return;
1585 } else {
1586 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1587 return;
1588 }
1589 set_disk_ro_uevent(disk, read_only);
1590}
1591EXPORT_SYMBOL(set_disk_ro);
1592
1593void inc_diskseq(struct gendisk *disk)
1594{
1595 disk->diskseq = atomic64_inc_return(&diskseq);
1596}