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(bdev: 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(kobj: &disk_to_dev(disk)->kobj, action: 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
121static unsigned 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(lock: &major_names_spinlock);
181 for (dp = major_names[major_to_index(major: offset)]; dp; dp = dp->next)
182 if (dp->major == offset)
183 seq_printf(m: seqf, fmt: "%3d %s\n", dp->major, dp->name);
184 spin_unlock(lock: &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(size: 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(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(lock: &major_names_spinlock);
271
272 if (ret < 0) {
273 printk("register_blkdev: cannot get major %u for %s\n",
274 major, name);
275 kfree(objp: p);
276 }
277out:
278 mutex_unlock(lock: &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(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(lock: &major_names_spinlock);
301 mutex_unlock(lock: &major_names_lock);
302 kfree(objp: 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, min: 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, id: 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(bdev: part) && !bdev_nr_sectors(bdev: part))
330 continue;
331 if (!kobject_get_unless_zero(kobj: &part->bd_device.kobj))
332 continue;
333
334 rcu_read_unlock();
335 kobject_uevent(bdev_kobj(part), action);
336 put_device(dev: &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->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
349 return -EINVAL;
350 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
351 return -EINVAL;
352 if (disk->open_partitions)
353 return -EBUSY;
354
355 /*
356 * If the device is opened exclusively by current thread already, it's
357 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
358 * synchronize with other exclusive openers and other partition
359 * scanners.
360 */
361 if (!(mode & BLK_OPEN_EXCL)) {
362 ret = bd_prepare_to_claim(bdev: disk->part0, holder: disk_scan_partitions,
363 NULL);
364 if (ret)
365 return ret;
366 }
367
368 set_bit(GD_NEED_PART_SCAN, addr: &disk->state);
369 file = bdev_file_open_by_dev(dev: disk_devt(disk), mode: mode & ~BLK_OPEN_EXCL,
370 NULL, NULL);
371 if (IS_ERR(ptr: file))
372 ret = PTR_ERR(ptr: file);
373 else
374 fput(file);
375
376 /*
377 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
378 * and this will cause that re-assemble partitioned raid device will
379 * creat partition for underlying disk.
380 */
381 clear_bit(GD_NEED_PART_SCAN, addr: &disk->state);
382 if (!(mode & BLK_OPEN_EXCL))
383 bd_abort_claiming(bdev: disk->part0, holder: disk_scan_partitions);
384 return ret;
385}
386
387/**
388 * device_add_disk - add disk information to kernel list
389 * @parent: parent device for the disk
390 * @disk: per-device partitioning information
391 * @groups: Additional per-device sysfs groups
392 *
393 * This function registers the partitioning information in @disk
394 * with the kernel.
395 */
396int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
397 const struct attribute_group **groups)
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(q: 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(q: disk->queue);
414
415 /* Mark bdev as having a submit_bio, if needed */
416 disk->part0->bd_has_submit_bio = disk->fops->submit_bio != NULL;
417
418 /*
419 * If the driver provides an explicit major number it also must provide
420 * the number of minors numbers supported, and those will be used to
421 * setup the gendisk.
422 * Otherwise just allocate the device numbers for both the whole device
423 * and all partitions from the extended dev_t space.
424 */
425 ret = -EINVAL;
426 if (disk->major) {
427 if (WARN_ON(!disk->minors))
428 goto out_exit_elevator;
429
430 if (disk->minors > DISK_MAX_PARTS) {
431 pr_err("block: can't allocate more than %d partitions\n",
432 DISK_MAX_PARTS);
433 disk->minors = DISK_MAX_PARTS;
434 }
435 if (disk->first_minor > MINORMASK ||
436 disk->minors > MINORMASK + 1 ||
437 disk->first_minor + disk->minors > MINORMASK + 1)
438 goto out_exit_elevator;
439 } else {
440 if (WARN_ON(disk->minors))
441 goto out_exit_elevator;
442
443 ret = blk_alloc_ext_minor();
444 if (ret < 0)
445 goto out_exit_elevator;
446 disk->major = BLOCK_EXT_MAJOR;
447 disk->first_minor = ret;
448 }
449
450 /* delay uevents, until we scanned partition table */
451 dev_set_uevent_suppress(dev: ddev, val: 1);
452
453 ddev->parent = parent;
454 ddev->groups = groups;
455 dev_set_name(dev: ddev, name: "%s", disk->disk_name);
456 if (!(disk->flags & GENHD_FL_HIDDEN))
457 ddev->devt = MKDEV(disk->major, disk->first_minor);
458 ret = device_add(dev: ddev);
459 if (ret)
460 goto out_free_ext_minor;
461
462 ret = disk_alloc_events(disk);
463 if (ret)
464 goto out_device_del;
465
466 ret = sysfs_create_link(kobj: block_depr, target: &ddev->kobj,
467 name: kobject_name(kobj: &ddev->kobj));
468 if (ret)
469 goto out_device_del;
470
471 /*
472 * avoid probable deadlock caused by allocating memory with
473 * GFP_KERNEL in runtime_resume callback of its all ancestor
474 * devices
475 */
476 pm_runtime_set_memalloc_noio(dev: ddev, enable: true);
477
478 disk->part0->bd_holder_dir =
479 kobject_create_and_add(name: "holders", parent: &ddev->kobj);
480 if (!disk->part0->bd_holder_dir) {
481 ret = -ENOMEM;
482 goto out_del_block_link;
483 }
484 disk->slave_dir = kobject_create_and_add(name: "slaves", parent: &ddev->kobj);
485 if (!disk->slave_dir) {
486 ret = -ENOMEM;
487 goto out_put_holder_dir;
488 }
489
490 ret = blk_register_queue(disk);
491 if (ret)
492 goto out_put_slave_dir;
493
494 if (!(disk->flags & GENHD_FL_HIDDEN)) {
495 ret = bdi_register(bdi: disk->bdi, fmt: "%u:%u",
496 disk->major, disk->first_minor);
497 if (ret)
498 goto out_unregister_queue;
499 bdi_set_owner(bdi: disk->bdi, owner: ddev);
500 ret = sysfs_create_link(kobj: &ddev->kobj,
501 target: &disk->bdi->dev->kobj, name: "bdi");
502 if (ret)
503 goto out_unregister_bdi;
504
505 /* Make sure the first partition scan will be proceed */
506 if (get_capacity(disk) && !(disk->flags & GENHD_FL_NO_PART) &&
507 !test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
508 set_bit(GD_NEED_PART_SCAN, addr: &disk->state);
509
510 bdev_add(bdev: disk->part0, dev: ddev->devt);
511 if (get_capacity(disk))
512 disk_scan_partitions(disk, BLK_OPEN_READ);
513
514 /*
515 * Announce the disk and partitions after all partitions are
516 * created. (for hidden disks uevents remain suppressed forever)
517 */
518 dev_set_uevent_suppress(dev: ddev, val: 0);
519 disk_uevent(disk, KOBJ_ADD);
520 } else {
521 /*
522 * Even if the block_device for a hidden gendisk is not
523 * registered, it needs to have a valid bd_dev so that the
524 * freeing of the dynamic major works.
525 */
526 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
527 }
528
529 disk_update_readahead(disk);
530 disk_add_events(disk);
531 set_bit(GD_ADDED, addr: &disk->state);
532 return 0;
533
534out_unregister_bdi:
535 if (!(disk->flags & GENHD_FL_HIDDEN))
536 bdi_unregister(bdi: disk->bdi);
537out_unregister_queue:
538 blk_unregister_queue(disk);
539 rq_qos_exit(disk->queue);
540out_put_slave_dir:
541 kobject_put(kobj: disk->slave_dir);
542 disk->slave_dir = NULL;
543out_put_holder_dir:
544 kobject_put(kobj: disk->part0->bd_holder_dir);
545out_del_block_link:
546 sysfs_remove_link(kobj: block_depr, name: dev_name(dev: ddev));
547 pm_runtime_set_memalloc_noio(dev: ddev, enable: false);
548out_device_del:
549 device_del(dev: ddev);
550out_free_ext_minor:
551 if (disk->major == BLOCK_EXT_MAJOR)
552 blk_free_ext_minor(minor: disk->first_minor);
553out_exit_elevator:
554 if (disk->queue->elevator)
555 elevator_exit(q: disk->queue);
556 return ret;
557}
558EXPORT_SYMBOL(device_add_disk);
559
560static void blk_report_disk_dead(struct gendisk *disk, bool surprise)
561{
562 struct block_device *bdev;
563 unsigned long idx;
564
565 /*
566 * On surprise disk removal, bdev_mark_dead() may call into file
567 * systems below. Make it clear that we're expecting to not hold
568 * disk->open_mutex.
569 */
570 lockdep_assert_not_held(&disk->open_mutex);
571
572 rcu_read_lock();
573 xa_for_each(&disk->part_tbl, idx, bdev) {
574 if (!kobject_get_unless_zero(kobj: &bdev->bd_device.kobj))
575 continue;
576 rcu_read_unlock();
577
578 bdev_mark_dead(bdev, surprise);
579
580 put_device(dev: &bdev->bd_device);
581 rcu_read_lock();
582 }
583 rcu_read_unlock();
584}
585
586static void __blk_mark_disk_dead(struct gendisk *disk)
587{
588 /*
589 * Fail any new I/O.
590 */
591 if (test_and_set_bit(GD_DEAD, addr: &disk->state))
592 return;
593
594 if (test_bit(GD_OWNS_QUEUE, &disk->state))
595 blk_queue_flag_set(QUEUE_FLAG_DYING, q: disk->queue);
596
597 /*
598 * Stop buffered writers from dirtying pages that can't be written out.
599 */
600 set_capacity(disk, 0);
601
602 /*
603 * Prevent new I/O from crossing bio_queue_enter().
604 */
605 blk_queue_start_drain(q: disk->queue);
606}
607
608/**
609 * blk_mark_disk_dead - mark a disk as dead
610 * @disk: disk to mark as dead
611 *
612 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
613 * to this disk.
614 */
615void blk_mark_disk_dead(struct gendisk *disk)
616{
617 __blk_mark_disk_dead(disk);
618 blk_report_disk_dead(disk, surprise: true);
619}
620EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
621
622/**
623 * del_gendisk - remove the gendisk
624 * @disk: the struct gendisk to remove
625 *
626 * Removes the gendisk and all its associated resources. This deletes the
627 * partitions associated with the gendisk, and unregisters the associated
628 * request_queue.
629 *
630 * This is the counter to the respective __device_add_disk() call.
631 *
632 * The final removal of the struct gendisk happens when its refcount reaches 0
633 * with put_disk(), which should be called after del_gendisk(), if
634 * __device_add_disk() was used.
635 *
636 * Drivers exist which depend on the release of the gendisk to be synchronous,
637 * it should not be deferred.
638 *
639 * Context: can sleep
640 */
641void del_gendisk(struct gendisk *disk)
642{
643 struct request_queue *q = disk->queue;
644 struct block_device *part;
645 unsigned long idx;
646
647 might_sleep();
648
649 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
650 return;
651
652 disk_del_events(disk);
653
654 /*
655 * Prevent new openers by unlinked the bdev inode.
656 */
657 mutex_lock(&disk->open_mutex);
658 xa_for_each(&disk->part_tbl, idx, part)
659 remove_inode_hash(inode: part->bd_inode);
660 mutex_unlock(lock: &disk->open_mutex);
661
662 /*
663 * Tell the file system to write back all dirty data and shut down if
664 * it hasn't been notified earlier.
665 */
666 if (!test_bit(GD_DEAD, &disk->state))
667 blk_report_disk_dead(disk, surprise: false);
668 __blk_mark_disk_dead(disk);
669
670 /*
671 * Drop all partitions now that the disk is marked dead.
672 */
673 mutex_lock(&disk->open_mutex);
674 xa_for_each_start(&disk->part_tbl, idx, part, 1)
675 drop_partition(part);
676 mutex_unlock(lock: &disk->open_mutex);
677
678 if (!(disk->flags & GENHD_FL_HIDDEN)) {
679 sysfs_remove_link(kobj: &disk_to_dev(disk)->kobj, name: "bdi");
680
681 /*
682 * Unregister bdi before releasing device numbers (as they can
683 * get reused and we'd get clashes in sysfs).
684 */
685 bdi_unregister(bdi: disk->bdi);
686 }
687
688 blk_unregister_queue(disk);
689
690 kobject_put(kobj: disk->part0->bd_holder_dir);
691 kobject_put(kobj: disk->slave_dir);
692 disk->slave_dir = NULL;
693
694 part_stat_set_all(part: disk->part0, value: 0);
695 disk->part0->bd_stamp = 0;
696 sysfs_remove_link(kobj: block_depr, name: dev_name(disk_to_dev(disk)));
697 pm_runtime_set_memalloc_noio(disk_to_dev(disk), enable: false);
698 device_del(disk_to_dev(disk));
699
700 blk_mq_freeze_queue_wait(q);
701
702 blk_throtl_cancel_bios(disk);
703
704 blk_sync_queue(q);
705 blk_flush_integrity();
706
707 if (queue_is_mq(q))
708 blk_mq_cancel_work_sync(q);
709
710 blk_mq_quiesce_queue(q);
711 if (q->elevator) {
712 mutex_lock(&q->sysfs_lock);
713 elevator_exit(q);
714 mutex_unlock(lock: &q->sysfs_lock);
715 }
716 rq_qos_exit(q);
717 blk_mq_unquiesce_queue(q);
718
719 /*
720 * If the disk does not own the queue, allow using passthrough requests
721 * again. Else leave the queue frozen to fail all I/O.
722 */
723 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
724 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
725 __blk_mq_unfreeze_queue(q, force_atomic: true);
726 } else {
727 if (queue_is_mq(q))
728 blk_mq_exit_queue(q);
729 }
730}
731EXPORT_SYMBOL(del_gendisk);
732
733/**
734 * invalidate_disk - invalidate the disk
735 * @disk: the struct gendisk to invalidate
736 *
737 * A helper to invalidates the disk. It will clean the disk's associated
738 * buffer/page caches and reset its internal states so that the disk
739 * can be reused by the drivers.
740 *
741 * Context: can sleep
742 */
743void invalidate_disk(struct gendisk *disk)
744{
745 struct block_device *bdev = disk->part0;
746
747 invalidate_bdev(bdev);
748 bdev->bd_inode->i_mapping->wb_err = 0;
749 set_capacity(disk, 0);
750}
751EXPORT_SYMBOL(invalidate_disk);
752
753/* sysfs access to bad-blocks list. */
754static ssize_t disk_badblocks_show(struct device *dev,
755 struct device_attribute *attr,
756 char *page)
757{
758 struct gendisk *disk = dev_to_disk(dev);
759
760 if (!disk->bb)
761 return sprintf(buf: page, fmt: "\n");
762
763 return badblocks_show(bb: disk->bb, page, unack: 0);
764}
765
766static ssize_t disk_badblocks_store(struct device *dev,
767 struct device_attribute *attr,
768 const char *page, size_t len)
769{
770 struct gendisk *disk = dev_to_disk(dev);
771
772 if (!disk->bb)
773 return -ENXIO;
774
775 return badblocks_store(bb: disk->bb, page, len, unack: 0);
776}
777
778#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
779void blk_request_module(dev_t devt)
780{
781 unsigned int major = MAJOR(devt);
782 struct blk_major_name **n;
783
784 mutex_lock(&major_names_lock);
785 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
786 if ((*n)->major == major && (*n)->probe) {
787 (*n)->probe(devt);
788 mutex_unlock(lock: &major_names_lock);
789 return;
790 }
791 }
792 mutex_unlock(lock: &major_names_lock);
793
794 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
795 /* Make old-style 2.4 aliases work */
796 request_module("block-major-%d", MAJOR(devt));
797}
798#endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
799
800#ifdef CONFIG_PROC_FS
801/* iterator */
802static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
803{
804 loff_t skip = *pos;
805 struct class_dev_iter *iter;
806 struct device *dev;
807
808 iter = kmalloc(size: sizeof(*iter), GFP_KERNEL);
809 if (!iter)
810 return ERR_PTR(error: -ENOMEM);
811
812 seqf->private = iter;
813 class_dev_iter_init(iter, class: &block_class, NULL, type: &disk_type);
814 do {
815 dev = class_dev_iter_next(iter);
816 if (!dev)
817 return NULL;
818 } while (skip--);
819
820 return dev_to_disk(dev);
821}
822
823static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
824{
825 struct device *dev;
826
827 (*pos)++;
828 dev = class_dev_iter_next(iter: seqf->private);
829 if (dev)
830 return dev_to_disk(dev);
831
832 return NULL;
833}
834
835static void disk_seqf_stop(struct seq_file *seqf, void *v)
836{
837 struct class_dev_iter *iter = seqf->private;
838
839 /* stop is called even after start failed :-( */
840 if (iter) {
841 class_dev_iter_exit(iter);
842 kfree(objp: iter);
843 seqf->private = NULL;
844 }
845}
846
847static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
848{
849 void *p;
850
851 p = disk_seqf_start(seqf, pos);
852 if (!IS_ERR_OR_NULL(ptr: p) && !*pos)
853 seq_puts(m: seqf, s: "major minor #blocks name\n\n");
854 return p;
855}
856
857static int show_partition(struct seq_file *seqf, void *v)
858{
859 struct gendisk *sgp = v;
860 struct block_device *part;
861 unsigned long idx;
862
863 if (!get_capacity(disk: sgp) || (sgp->flags & GENHD_FL_HIDDEN))
864 return 0;
865
866 rcu_read_lock();
867 xa_for_each(&sgp->part_tbl, idx, part) {
868 if (!bdev_nr_sectors(bdev: part))
869 continue;
870 seq_printf(m: seqf, fmt: "%4d %7d %10llu %pg\n",
871 MAJOR(part->bd_dev), MINOR(part->bd_dev),
872 bdev_nr_sectors(bdev: part) >> 1, part);
873 }
874 rcu_read_unlock();
875 return 0;
876}
877
878static const struct seq_operations partitions_op = {
879 .start = show_partition_start,
880 .next = disk_seqf_next,
881 .stop = disk_seqf_stop,
882 .show = show_partition
883};
884#endif
885
886static int __init genhd_device_init(void)
887{
888 int error;
889
890 error = class_register(class: &block_class);
891 if (unlikely(error))
892 return error;
893 blk_dev_init();
894
895 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
896
897 /* create top-level block dir */
898 block_depr = kobject_create_and_add(name: "block", NULL);
899 return 0;
900}
901
902subsys_initcall(genhd_device_init);
903
904static ssize_t disk_range_show(struct device *dev,
905 struct device_attribute *attr, char *buf)
906{
907 struct gendisk *disk = dev_to_disk(dev);
908
909 return sprintf(buf, fmt: "%d\n", disk->minors);
910}
911
912static ssize_t disk_ext_range_show(struct device *dev,
913 struct device_attribute *attr, char *buf)
914{
915 struct gendisk *disk = dev_to_disk(dev);
916
917 return sprintf(buf, fmt: "%d\n",
918 (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
919}
920
921static ssize_t disk_removable_show(struct device *dev,
922 struct device_attribute *attr, char *buf)
923{
924 struct gendisk *disk = dev_to_disk(dev);
925
926 return sprintf(buf, fmt: "%d\n",
927 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
928}
929
930static ssize_t disk_hidden_show(struct device *dev,
931 struct device_attribute *attr, char *buf)
932{
933 struct gendisk *disk = dev_to_disk(dev);
934
935 return sprintf(buf, fmt: "%d\n",
936 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
937}
938
939static ssize_t disk_ro_show(struct device *dev,
940 struct device_attribute *attr, char *buf)
941{
942 struct gendisk *disk = dev_to_disk(dev);
943
944 return sprintf(buf, fmt: "%d\n", get_disk_ro(disk) ? 1 : 0);
945}
946
947ssize_t part_size_show(struct device *dev,
948 struct device_attribute *attr, char *buf)
949{
950 return sprintf(buf, fmt: "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
951}
952
953ssize_t part_stat_show(struct device *dev,
954 struct device_attribute *attr, char *buf)
955{
956 struct block_device *bdev = dev_to_bdev(dev);
957 struct request_queue *q = bdev_get_queue(bdev);
958 struct disk_stats stat;
959 unsigned int inflight;
960
961 if (queue_is_mq(q))
962 inflight = blk_mq_in_flight(q, part: bdev);
963 else
964 inflight = part_in_flight(part: bdev);
965
966 if (inflight) {
967 part_stat_lock();
968 update_io_ticks(part: bdev, now: jiffies, end: true);
969 part_stat_unlock();
970 }
971 part_stat_read_all(part: bdev, stat: &stat);
972 return sprintf(buf,
973 fmt: "%8lu %8lu %8llu %8u "
974 "%8lu %8lu %8llu %8u "
975 "%8u %8u %8u "
976 "%8lu %8lu %8llu %8u "
977 "%8lu %8u"
978 "\n",
979 stat.ios[STAT_READ],
980 stat.merges[STAT_READ],
981 (unsigned long long)stat.sectors[STAT_READ],
982 (unsigned int)div_u64(dividend: stat.nsecs[STAT_READ], NSEC_PER_MSEC),
983 stat.ios[STAT_WRITE],
984 stat.merges[STAT_WRITE],
985 (unsigned long long)stat.sectors[STAT_WRITE],
986 (unsigned int)div_u64(dividend: stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
987 inflight,
988 jiffies_to_msecs(j: stat.io_ticks),
989 (unsigned int)div_u64(dividend: stat.nsecs[STAT_READ] +
990 stat.nsecs[STAT_WRITE] +
991 stat.nsecs[STAT_DISCARD] +
992 stat.nsecs[STAT_FLUSH],
993 NSEC_PER_MSEC),
994 stat.ios[STAT_DISCARD],
995 stat.merges[STAT_DISCARD],
996 (unsigned long long)stat.sectors[STAT_DISCARD],
997 (unsigned int)div_u64(dividend: stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
998 stat.ios[STAT_FLUSH],
999 (unsigned int)div_u64(dividend: stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1000}
1001
1002ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1003 char *buf)
1004{
1005 struct block_device *bdev = dev_to_bdev(dev);
1006 struct request_queue *q = bdev_get_queue(bdev);
1007 unsigned int inflight[2];
1008
1009 if (queue_is_mq(q))
1010 blk_mq_in_flight_rw(q, part: bdev, inflight);
1011 else
1012 part_in_flight_rw(part: bdev, inflight);
1013
1014 return sprintf(buf, fmt: "%8u %8u\n", inflight[0], inflight[1]);
1015}
1016
1017static ssize_t disk_capability_show(struct device *dev,
1018 struct device_attribute *attr, char *buf)
1019{
1020 dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1021 return sprintf(buf, fmt: "0\n");
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, fmt: "%d\n", bdev_alignment_offset(bdev: 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, fmt: "%d\n", bdev_alignment_offset(bdev: 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, fmt: "%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, fmt: "%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#ifdef CONFIG_BLK_DEV_INTEGRITY
1139 &blk_integrity_attr_group,
1140#endif
1141 NULL
1142};
1143
1144/**
1145 * disk_release - releases all allocated resources of the gendisk
1146 * @dev: the device representing this disk
1147 *
1148 * This function releases all allocated resources of the gendisk.
1149 *
1150 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1151 * assigned. Since the request_queue sits on top of the gendisk for these
1152 * drivers we also call blk_put_queue() for them, and we expect the
1153 * request_queue refcount to reach 0 at this point, and so the request_queue
1154 * will also be freed prior to the disk.
1155 *
1156 * Context: can sleep
1157 */
1158static void disk_release(struct device *dev)
1159{
1160 struct gendisk *disk = dev_to_disk(dev);
1161
1162 might_sleep();
1163 WARN_ON_ONCE(disk_live(disk));
1164
1165 blk_trace_remove(q: disk->queue);
1166
1167 /*
1168 * To undo the all initialization from blk_mq_init_allocated_queue in
1169 * case of a probe failure where add_disk is never called we have to
1170 * call blk_mq_exit_queue here. We can't do this for the more common
1171 * teardown case (yet) as the tagset can be gone by the time the disk
1172 * is released once it was added.
1173 */
1174 if (queue_is_mq(q: disk->queue) &&
1175 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1176 !test_bit(GD_ADDED, &disk->state))
1177 blk_mq_exit_queue(q: disk->queue);
1178
1179 blkcg_exit_disk(disk);
1180
1181 bioset_exit(&disk->bio_split);
1182
1183 disk_release_events(disk);
1184 kfree(objp: disk->random);
1185 disk_free_zone_bitmaps(disk);
1186 xa_destroy(&disk->part_tbl);
1187
1188 disk->queue->disk = NULL;
1189 blk_put_queue(disk->queue);
1190
1191 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1192 disk->fops->free_disk(disk);
1193
1194 iput(disk->part0->bd_inode); /* frees the disk */
1195}
1196
1197static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1198{
1199 const struct gendisk *disk = dev_to_disk(dev);
1200
1201 return add_uevent_var(env, format: "DISKSEQ=%llu", disk->diskseq);
1202}
1203
1204const struct class block_class = {
1205 .name = "block",
1206 .dev_uevent = block_uevent,
1207};
1208
1209static char *block_devnode(const struct device *dev, umode_t *mode,
1210 kuid_t *uid, kgid_t *gid)
1211{
1212 struct gendisk *disk = dev_to_disk(dev);
1213
1214 if (disk->fops->devnode)
1215 return disk->fops->devnode(disk, mode);
1216 return NULL;
1217}
1218
1219const struct device_type disk_type = {
1220 .name = "disk",
1221 .groups = disk_attr_groups,
1222 .release = disk_release,
1223 .devnode = block_devnode,
1224};
1225
1226#ifdef CONFIG_PROC_FS
1227/*
1228 * aggregate disk stat collector. Uses the same stats that the sysfs
1229 * entries do, above, but makes them available through one seq_file.
1230 *
1231 * The output looks suspiciously like /proc/partitions with a bunch of
1232 * extra fields.
1233 */
1234static int diskstats_show(struct seq_file *seqf, void *v)
1235{
1236 struct gendisk *gp = v;
1237 struct block_device *hd;
1238 unsigned int inflight;
1239 struct disk_stats stat;
1240 unsigned long idx;
1241
1242 /*
1243 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1244 seq_puts(seqf, "major minor name"
1245 " rio rmerge rsect ruse wio wmerge "
1246 "wsect wuse running use aveq"
1247 "\n\n");
1248 */
1249
1250 rcu_read_lock();
1251 xa_for_each(&gp->part_tbl, idx, hd) {
1252 if (bdev_is_partition(bdev: hd) && !bdev_nr_sectors(bdev: hd))
1253 continue;
1254 if (queue_is_mq(q: gp->queue))
1255 inflight = blk_mq_in_flight(q: gp->queue, part: hd);
1256 else
1257 inflight = part_in_flight(part: hd);
1258
1259 if (inflight) {
1260 part_stat_lock();
1261 update_io_ticks(part: hd, now: jiffies, end: true);
1262 part_stat_unlock();
1263 }
1264 part_stat_read_all(part: hd, stat: &stat);
1265 seq_printf(m: seqf, fmt: "%4d %7d %pg "
1266 "%lu %lu %lu %u "
1267 "%lu %lu %lu %u "
1268 "%u %u %u "
1269 "%lu %lu %lu %u "
1270 "%lu %u"
1271 "\n",
1272 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1273 stat.ios[STAT_READ],
1274 stat.merges[STAT_READ],
1275 stat.sectors[STAT_READ],
1276 (unsigned int)div_u64(dividend: stat.nsecs[STAT_READ],
1277 NSEC_PER_MSEC),
1278 stat.ios[STAT_WRITE],
1279 stat.merges[STAT_WRITE],
1280 stat.sectors[STAT_WRITE],
1281 (unsigned int)div_u64(dividend: stat.nsecs[STAT_WRITE],
1282 NSEC_PER_MSEC),
1283 inflight,
1284 jiffies_to_msecs(j: stat.io_ticks),
1285 (unsigned int)div_u64(dividend: stat.nsecs[STAT_READ] +
1286 stat.nsecs[STAT_WRITE] +
1287 stat.nsecs[STAT_DISCARD] +
1288 stat.nsecs[STAT_FLUSH],
1289 NSEC_PER_MSEC),
1290 stat.ios[STAT_DISCARD],
1291 stat.merges[STAT_DISCARD],
1292 stat.sectors[STAT_DISCARD],
1293 (unsigned int)div_u64(dividend: stat.nsecs[STAT_DISCARD],
1294 NSEC_PER_MSEC),
1295 stat.ios[STAT_FLUSH],
1296 (unsigned int)div_u64(dividend: stat.nsecs[STAT_FLUSH],
1297 NSEC_PER_MSEC)
1298 );
1299 }
1300 rcu_read_unlock();
1301
1302 return 0;
1303}
1304
1305static const struct seq_operations diskstats_op = {
1306 .start = disk_seqf_start,
1307 .next = disk_seqf_next,
1308 .stop = disk_seqf_stop,
1309 .show = diskstats_show
1310};
1311
1312static int __init proc_genhd_init(void)
1313{
1314 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1315 proc_create_seq("partitions", 0, NULL, &partitions_op);
1316 return 0;
1317}
1318module_init(proc_genhd_init);
1319#endif /* CONFIG_PROC_FS */
1320
1321dev_t part_devt(struct gendisk *disk, u8 partno)
1322{
1323 struct block_device *part;
1324 dev_t devt = 0;
1325
1326 rcu_read_lock();
1327 part = xa_load(&disk->part_tbl, index: partno);
1328 if (part)
1329 devt = part->bd_dev;
1330 rcu_read_unlock();
1331
1332 return devt;
1333}
1334
1335struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1336 struct lock_class_key *lkclass)
1337{
1338 struct gendisk *disk;
1339
1340 disk = kzalloc_node(size: sizeof(struct gendisk), GFP_KERNEL, node: node_id);
1341 if (!disk)
1342 return NULL;
1343
1344 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, flags: 0))
1345 goto out_free_disk;
1346
1347 disk->bdi = bdi_alloc(node_id);
1348 if (!disk->bdi)
1349 goto out_free_bioset;
1350
1351 /* bdev_alloc() might need the queue, set before the first call */
1352 disk->queue = q;
1353
1354 disk->part0 = bdev_alloc(disk, partno: 0);
1355 if (!disk->part0)
1356 goto out_free_bdi;
1357
1358 disk->node_id = node_id;
1359 mutex_init(&disk->open_mutex);
1360 xa_init(xa: &disk->part_tbl);
1361 if (xa_insert(xa: &disk->part_tbl, index: 0, entry: disk->part0, GFP_KERNEL))
1362 goto out_destroy_part_tbl;
1363
1364 if (blkcg_init_disk(disk))
1365 goto out_erase_part0;
1366
1367 rand_initialize_disk(disk);
1368 disk_to_dev(disk)->class = &block_class;
1369 disk_to_dev(disk)->type = &disk_type;
1370 device_initialize(disk_to_dev(disk));
1371 inc_diskseq(disk);
1372 q->disk = disk;
1373 lockdep_init_map(lock: &disk->lockdep_map, name: "(bio completion)", key: lkclass, subclass: 0);
1374#ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1375 INIT_LIST_HEAD(list: &disk->slave_bdevs);
1376#endif
1377 return disk;
1378
1379out_erase_part0:
1380 xa_erase(&disk->part_tbl, index: 0);
1381out_destroy_part_tbl:
1382 xa_destroy(&disk->part_tbl);
1383 disk->part0->bd_disk = NULL;
1384 iput(disk->part0->bd_inode);
1385out_free_bdi:
1386 bdi_put(bdi: disk->bdi);
1387out_free_bioset:
1388 bioset_exit(&disk->bio_split);
1389out_free_disk:
1390 kfree(objp: disk);
1391 return NULL;
1392}
1393
1394struct gendisk *__blk_alloc_disk(struct queue_limits *lim, int node,
1395 struct lock_class_key *lkclass)
1396{
1397 struct queue_limits default_lim = { };
1398 struct request_queue *q;
1399 struct gendisk *disk;
1400
1401 q = blk_alloc_queue(lim: lim ? lim : &default_lim, node_id: node);
1402 if (IS_ERR(ptr: q))
1403 return ERR_CAST(ptr: q);
1404
1405 disk = __alloc_disk_node(q, node_id: node, lkclass);
1406 if (!disk) {
1407 blk_put_queue(q);
1408 return ERR_PTR(error: -ENOMEM);
1409 }
1410 set_bit(GD_OWNS_QUEUE, addr: &disk->state);
1411 return disk;
1412}
1413EXPORT_SYMBOL(__blk_alloc_disk);
1414
1415/**
1416 * put_disk - decrements the gendisk refcount
1417 * @disk: the struct gendisk to decrement the refcount for
1418 *
1419 * This decrements the refcount for the struct gendisk. When this reaches 0
1420 * we'll have disk_release() called.
1421 *
1422 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1423 * when handling probe errors (that is before add_disk() is called).
1424 *
1425 * Context: Any context, but the last reference must not be dropped from
1426 * atomic context.
1427 */
1428void put_disk(struct gendisk *disk)
1429{
1430 if (disk)
1431 put_device(disk_to_dev(disk));
1432}
1433EXPORT_SYMBOL(put_disk);
1434
1435static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1436{
1437 char event[] = "DISK_RO=1";
1438 char *envp[] = { event, NULL };
1439
1440 if (!ro)
1441 event[8] = '0';
1442 kobject_uevent_env(kobj: &disk_to_dev(gd)->kobj, action: KOBJ_CHANGE, envp);
1443}
1444
1445/**
1446 * set_disk_ro - set a gendisk read-only
1447 * @disk: gendisk to operate on
1448 * @read_only: %true to set the disk read-only, %false set the disk read/write
1449 *
1450 * This function is used to indicate whether a given disk device should have its
1451 * read-only flag set. set_disk_ro() is typically used by device drivers to
1452 * indicate whether the underlying physical device is write-protected.
1453 */
1454void set_disk_ro(struct gendisk *disk, bool read_only)
1455{
1456 if (read_only) {
1457 if (test_and_set_bit(GD_READ_ONLY, addr: &disk->state))
1458 return;
1459 } else {
1460 if (!test_and_clear_bit(GD_READ_ONLY, addr: &disk->state))
1461 return;
1462 }
1463 set_disk_ro_uevent(gd: disk, ro: read_only);
1464}
1465EXPORT_SYMBOL(set_disk_ro);
1466
1467void inc_diskseq(struct gendisk *disk)
1468{
1469 disk->diskseq = atomic64_inc_return(v: &diskseq);
1470}
1471

source code of linux/block/genhd.c