xref: /openbmc/linux/block/genhd.c (revision cb8a7d53)
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 
35 static 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  */
52 static atomic64_t diskseq;
53 
54 /* for extended dynamic devt allocation, currently only one major is used */
55 #define NR_EXT_DEVT		(1 << MINORBITS)
56 static DEFINE_IDA(ext_devt_ida);
57 
58 void 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 }
67 EXPORT_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  */
73 bool 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 }
102 EXPORT_SYMBOL_GPL(set_capacity_and_notify);
103 
104 static 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 
125 static 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 
140 static 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
162 static 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];
170 static DEFINE_MUTEX(major_names_lock);
171 static DEFINE_SPINLOCK(major_names_spinlock);
172 
173 /* index in the above - for now: assume no multimajor ranges */
174 static inline int major_to_index(unsigned major)
175 {
176 	return major % BLKDEV_MAJOR_HASH_SIZE;
177 }
178 
179 #ifdef CONFIG_PROC_FS
180 void 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  */
218 int __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 	}
281 out:
282 	mutex_unlock(&major_names_lock);
283 	return ret;
284 }
285 EXPORT_SYMBOL(__register_blkdev);
286 
287 void 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 
309 EXPORT_SYMBOL(unregister_blkdev);
310 
311 int 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 
321 void blk_free_ext_minor(unsigned int minor)
322 {
323 	ida_free(&ext_devt_ida, minor);
324 }
325 
326 static 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 
338 void 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 }
357 EXPORT_SYMBOL_GPL(disk_uevent);
358 
359 int disk_scan_partitions(struct gendisk *disk, fmode_t mode)
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 
370 	set_bit(GD_NEED_PART_SCAN, &disk->state);
371 	bdev = blkdev_get_by_dev(disk_devt(disk), mode, NULL);
372 	if (IS_ERR(bdev))
373 		return PTR_ERR(bdev);
374 	blkdev_put(bdev, mode);
375 	return 0;
376 }
377 
378 /**
379  * device_add_disk - add disk information to kernel list
380  * @parent: parent device for the disk
381  * @disk: per-device partitioning information
382  * @groups: Additional per-device sysfs groups
383  *
384  * This function registers the partitioning information in @disk
385  * with the kernel.
386  */
387 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
388 				 const struct attribute_group **groups)
389 
390 {
391 	struct device *ddev = disk_to_dev(disk);
392 	int ret;
393 
394 	/* Only makes sense for bio-based to set ->poll_bio */
395 	if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
396 		return -EINVAL;
397 
398 	/*
399 	 * The disk queue should now be all set with enough information about
400 	 * the device for the elevator code to pick an adequate default
401 	 * elevator if one is needed, that is, for devices requesting queue
402 	 * registration.
403 	 */
404 	elevator_init_mq(disk->queue);
405 
406 	/*
407 	 * If the driver provides an explicit major number it also must provide
408 	 * the number of minors numbers supported, and those will be used to
409 	 * setup the gendisk.
410 	 * Otherwise just allocate the device numbers for both the whole device
411 	 * and all partitions from the extended dev_t space.
412 	 */
413 	if (disk->major) {
414 		if (WARN_ON(!disk->minors))
415 			return -EINVAL;
416 
417 		if (disk->minors > DISK_MAX_PARTS) {
418 			pr_err("block: can't allocate more than %d partitions\n",
419 				DISK_MAX_PARTS);
420 			disk->minors = DISK_MAX_PARTS;
421 		}
422 		if (disk->first_minor + disk->minors > MINORMASK + 1)
423 			return -EINVAL;
424 	} else {
425 		if (WARN_ON(disk->minors))
426 			return -EINVAL;
427 
428 		ret = blk_alloc_ext_minor();
429 		if (ret < 0)
430 			return ret;
431 		disk->major = BLOCK_EXT_MAJOR;
432 		disk->first_minor = ret;
433 	}
434 
435 	/* delay uevents, until we scanned partition table */
436 	dev_set_uevent_suppress(ddev, 1);
437 
438 	ddev->parent = parent;
439 	ddev->groups = groups;
440 	dev_set_name(ddev, "%s", disk->disk_name);
441 	if (!(disk->flags & GENHD_FL_HIDDEN))
442 		ddev->devt = MKDEV(disk->major, disk->first_minor);
443 	ret = device_add(ddev);
444 	if (ret)
445 		goto out_free_ext_minor;
446 
447 	ret = disk_alloc_events(disk);
448 	if (ret)
449 		goto out_device_del;
450 
451 	if (!sysfs_deprecated) {
452 		ret = sysfs_create_link(block_depr, &ddev->kobj,
453 					kobject_name(&ddev->kobj));
454 		if (ret)
455 			goto out_device_del;
456 	}
457 
458 	/*
459 	 * avoid probable deadlock caused by allocating memory with
460 	 * GFP_KERNEL in runtime_resume callback of its all ancestor
461 	 * devices
462 	 */
463 	pm_runtime_set_memalloc_noio(ddev, true);
464 
465 	ret = blk_integrity_add(disk);
466 	if (ret)
467 		goto out_del_block_link;
468 
469 	disk->part0->bd_holder_dir =
470 		kobject_create_and_add("holders", &ddev->kobj);
471 	if (!disk->part0->bd_holder_dir) {
472 		ret = -ENOMEM;
473 		goto out_del_integrity;
474 	}
475 	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
476 	if (!disk->slave_dir) {
477 		ret = -ENOMEM;
478 		goto out_put_holder_dir;
479 	}
480 
481 	ret = bd_register_pending_holders(disk);
482 	if (ret < 0)
483 		goto out_put_slave_dir;
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);
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 	}
511 
512 	disk_update_readahead(disk);
513 	disk_add_events(disk);
514 	set_bit(GD_ADDED, &disk->state);
515 	return 0;
516 
517 out_unregister_bdi:
518 	if (!(disk->flags & GENHD_FL_HIDDEN))
519 		bdi_unregister(disk->bdi);
520 out_unregister_queue:
521 	blk_unregister_queue(disk);
522 out_put_slave_dir:
523 	kobject_put(disk->slave_dir);
524 out_put_holder_dir:
525 	kobject_put(disk->part0->bd_holder_dir);
526 out_del_integrity:
527 	blk_integrity_del(disk);
528 out_del_block_link:
529 	if (!sysfs_deprecated)
530 		sysfs_remove_link(block_depr, dev_name(ddev));
531 out_device_del:
532 	device_del(ddev);
533 out_free_ext_minor:
534 	if (disk->major == BLOCK_EXT_MAJOR)
535 		blk_free_ext_minor(disk->first_minor);
536 	return ret;
537 }
538 EXPORT_SYMBOL(device_add_disk);
539 
540 /**
541  * blk_mark_disk_dead - mark a disk as dead
542  * @disk: disk to mark as dead
543  *
544  * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
545  * to this disk.
546  */
547 void blk_mark_disk_dead(struct gendisk *disk)
548 {
549 	set_bit(GD_DEAD, &disk->state);
550 	blk_queue_start_drain(disk->queue);
551 }
552 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
553 
554 /**
555  * del_gendisk - remove the gendisk
556  * @disk: the struct gendisk to remove
557  *
558  * Removes the gendisk and all its associated resources. This deletes the
559  * partitions associated with the gendisk, and unregisters the associated
560  * request_queue.
561  *
562  * This is the counter to the respective __device_add_disk() call.
563  *
564  * The final removal of the struct gendisk happens when its refcount reaches 0
565  * with put_disk(), which should be called after del_gendisk(), if
566  * __device_add_disk() was used.
567  *
568  * Drivers exist which depend on the release of the gendisk to be synchronous,
569  * it should not be deferred.
570  *
571  * Context: can sleep
572  */
573 void del_gendisk(struct gendisk *disk)
574 {
575 	struct request_queue *q = disk->queue;
576 
577 	might_sleep();
578 
579 	if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
580 		return;
581 
582 	blk_integrity_del(disk);
583 	disk_del_events(disk);
584 
585 	mutex_lock(&disk->open_mutex);
586 	remove_inode_hash(disk->part0->bd_inode);
587 	blk_drop_partitions(disk);
588 	mutex_unlock(&disk->open_mutex);
589 
590 	fsync_bdev(disk->part0);
591 	__invalidate_device(disk->part0, true);
592 
593 	/*
594 	 * Fail any new I/O.
595 	 */
596 	set_bit(GD_DEAD, &disk->state);
597 	if (test_bit(GD_OWNS_QUEUE, &disk->state))
598 		blk_queue_flag_set(QUEUE_FLAG_DYING, q);
599 	set_capacity(disk, 0);
600 
601 	/*
602 	 * Prevent new I/O from crossing bio_queue_enter().
603 	 */
604 	blk_queue_start_drain(q);
605 
606 	if (!(disk->flags & GENHD_FL_HIDDEN)) {
607 		sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
608 
609 		/*
610 		 * Unregister bdi before releasing device numbers (as they can
611 		 * get reused and we'd get clashes in sysfs).
612 		 */
613 		bdi_unregister(disk->bdi);
614 	}
615 
616 	blk_unregister_queue(disk);
617 
618 	kobject_put(disk->part0->bd_holder_dir);
619 	kobject_put(disk->slave_dir);
620 
621 	part_stat_set_all(disk->part0, 0);
622 	disk->part0->bd_stamp = 0;
623 	if (!sysfs_deprecated)
624 		sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
625 	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
626 	device_del(disk_to_dev(disk));
627 
628 	blk_mq_freeze_queue_wait(q);
629 
630 	blk_throtl_cancel_bios(disk->queue);
631 
632 	blk_sync_queue(q);
633 	blk_flush_integrity();
634 	blk_mq_cancel_work_sync(q);
635 
636 	blk_mq_quiesce_queue(q);
637 	if (q->elevator) {
638 		mutex_lock(&q->sysfs_lock);
639 		elevator_exit(q);
640 		mutex_unlock(&q->sysfs_lock);
641 	}
642 	rq_qos_exit(q);
643 	blk_mq_unquiesce_queue(q);
644 
645 	/*
646 	 * If the disk does not own the queue, allow using passthrough requests
647 	 * again.  Else leave the queue frozen to fail all I/O.
648 	 */
649 	if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
650 		blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
651 		__blk_mq_unfreeze_queue(q, true);
652 	} else {
653 		if (queue_is_mq(q))
654 			blk_mq_exit_queue(q);
655 	}
656 }
657 EXPORT_SYMBOL(del_gendisk);
658 
659 /**
660  * invalidate_disk - invalidate the disk
661  * @disk: the struct gendisk to invalidate
662  *
663  * A helper to invalidates the disk. It will clean the disk's associated
664  * buffer/page caches and reset its internal states so that the disk
665  * can be reused by the drivers.
666  *
667  * Context: can sleep
668  */
669 void invalidate_disk(struct gendisk *disk)
670 {
671 	struct block_device *bdev = disk->part0;
672 
673 	invalidate_bdev(bdev);
674 	bdev->bd_inode->i_mapping->wb_err = 0;
675 	set_capacity(disk, 0);
676 }
677 EXPORT_SYMBOL(invalidate_disk);
678 
679 /* sysfs access to bad-blocks list. */
680 static ssize_t disk_badblocks_show(struct device *dev,
681 					struct device_attribute *attr,
682 					char *page)
683 {
684 	struct gendisk *disk = dev_to_disk(dev);
685 
686 	if (!disk->bb)
687 		return sprintf(page, "\n");
688 
689 	return badblocks_show(disk->bb, page, 0);
690 }
691 
692 static ssize_t disk_badblocks_store(struct device *dev,
693 					struct device_attribute *attr,
694 					const char *page, size_t len)
695 {
696 	struct gendisk *disk = dev_to_disk(dev);
697 
698 	if (!disk->bb)
699 		return -ENXIO;
700 
701 	return badblocks_store(disk->bb, page, len, 0);
702 }
703 
704 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
705 void blk_request_module(dev_t devt)
706 {
707 	unsigned int major = MAJOR(devt);
708 	struct blk_major_name **n;
709 
710 	mutex_lock(&major_names_lock);
711 	for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
712 		if ((*n)->major == major && (*n)->probe) {
713 			(*n)->probe(devt);
714 			mutex_unlock(&major_names_lock);
715 			return;
716 		}
717 	}
718 	mutex_unlock(&major_names_lock);
719 
720 	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
721 		/* Make old-style 2.4 aliases work */
722 		request_module("block-major-%d", MAJOR(devt));
723 }
724 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
725 
726 /*
727  * print a full list of all partitions - intended for places where the root
728  * filesystem can't be mounted and thus to give the victim some idea of what
729  * went wrong
730  */
731 void __init printk_all_partitions(void)
732 {
733 	struct class_dev_iter iter;
734 	struct device *dev;
735 
736 	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
737 	while ((dev = class_dev_iter_next(&iter))) {
738 		struct gendisk *disk = dev_to_disk(dev);
739 		struct block_device *part;
740 		char devt_buf[BDEVT_SIZE];
741 		unsigned long idx;
742 
743 		/*
744 		 * Don't show empty devices or things that have been
745 		 * suppressed
746 		 */
747 		if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
748 			continue;
749 
750 		/*
751 		 * Note, unlike /proc/partitions, I am showing the numbers in
752 		 * hex - the same format as the root= option takes.
753 		 */
754 		rcu_read_lock();
755 		xa_for_each(&disk->part_tbl, idx, part) {
756 			if (!bdev_nr_sectors(part))
757 				continue;
758 			printk("%s%s %10llu %pg %s",
759 			       bdev_is_partition(part) ? "  " : "",
760 			       bdevt_str(part->bd_dev, devt_buf),
761 			       bdev_nr_sectors(part) >> 1, part,
762 			       part->bd_meta_info ?
763 					part->bd_meta_info->uuid : "");
764 			if (bdev_is_partition(part))
765 				printk("\n");
766 			else if (dev->parent && dev->parent->driver)
767 				printk(" driver: %s\n",
768 					dev->parent->driver->name);
769 			else
770 				printk(" (driver?)\n");
771 		}
772 		rcu_read_unlock();
773 	}
774 	class_dev_iter_exit(&iter);
775 }
776 
777 #ifdef CONFIG_PROC_FS
778 /* iterator */
779 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
780 {
781 	loff_t skip = *pos;
782 	struct class_dev_iter *iter;
783 	struct device *dev;
784 
785 	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
786 	if (!iter)
787 		return ERR_PTR(-ENOMEM);
788 
789 	seqf->private = iter;
790 	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
791 	do {
792 		dev = class_dev_iter_next(iter);
793 		if (!dev)
794 			return NULL;
795 	} while (skip--);
796 
797 	return dev_to_disk(dev);
798 }
799 
800 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
801 {
802 	struct device *dev;
803 
804 	(*pos)++;
805 	dev = class_dev_iter_next(seqf->private);
806 	if (dev)
807 		return dev_to_disk(dev);
808 
809 	return NULL;
810 }
811 
812 static void disk_seqf_stop(struct seq_file *seqf, void *v)
813 {
814 	struct class_dev_iter *iter = seqf->private;
815 
816 	/* stop is called even after start failed :-( */
817 	if (iter) {
818 		class_dev_iter_exit(iter);
819 		kfree(iter);
820 		seqf->private = NULL;
821 	}
822 }
823 
824 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
825 {
826 	void *p;
827 
828 	p = disk_seqf_start(seqf, pos);
829 	if (!IS_ERR_OR_NULL(p) && !*pos)
830 		seq_puts(seqf, "major minor  #blocks  name\n\n");
831 	return p;
832 }
833 
834 static int show_partition(struct seq_file *seqf, void *v)
835 {
836 	struct gendisk *sgp = v;
837 	struct block_device *part;
838 	unsigned long idx;
839 
840 	if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
841 		return 0;
842 
843 	rcu_read_lock();
844 	xa_for_each(&sgp->part_tbl, idx, part) {
845 		if (!bdev_nr_sectors(part))
846 			continue;
847 		seq_printf(seqf, "%4d  %7d %10llu %pg\n",
848 			   MAJOR(part->bd_dev), MINOR(part->bd_dev),
849 			   bdev_nr_sectors(part) >> 1, part);
850 	}
851 	rcu_read_unlock();
852 	return 0;
853 }
854 
855 static const struct seq_operations partitions_op = {
856 	.start	= show_partition_start,
857 	.next	= disk_seqf_next,
858 	.stop	= disk_seqf_stop,
859 	.show	= show_partition
860 };
861 #endif
862 
863 static int __init genhd_device_init(void)
864 {
865 	int error;
866 
867 	block_class.dev_kobj = sysfs_dev_block_kobj;
868 	error = class_register(&block_class);
869 	if (unlikely(error))
870 		return error;
871 	blk_dev_init();
872 
873 	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
874 
875 	/* create top-level block dir */
876 	if (!sysfs_deprecated)
877 		block_depr = kobject_create_and_add("block", NULL);
878 	return 0;
879 }
880 
881 subsys_initcall(genhd_device_init);
882 
883 static ssize_t disk_range_show(struct device *dev,
884 			       struct device_attribute *attr, char *buf)
885 {
886 	struct gendisk *disk = dev_to_disk(dev);
887 
888 	return sprintf(buf, "%d\n", disk->minors);
889 }
890 
891 static ssize_t disk_ext_range_show(struct device *dev,
892 				   struct device_attribute *attr, char *buf)
893 {
894 	struct gendisk *disk = dev_to_disk(dev);
895 
896 	return sprintf(buf, "%d\n",
897 		(disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
898 }
899 
900 static ssize_t disk_removable_show(struct device *dev,
901 				   struct device_attribute *attr, char *buf)
902 {
903 	struct gendisk *disk = dev_to_disk(dev);
904 
905 	return sprintf(buf, "%d\n",
906 		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
907 }
908 
909 static ssize_t disk_hidden_show(struct device *dev,
910 				   struct device_attribute *attr, char *buf)
911 {
912 	struct gendisk *disk = dev_to_disk(dev);
913 
914 	return sprintf(buf, "%d\n",
915 		       (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
916 }
917 
918 static ssize_t disk_ro_show(struct device *dev,
919 				   struct device_attribute *attr, char *buf)
920 {
921 	struct gendisk *disk = dev_to_disk(dev);
922 
923 	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
924 }
925 
926 ssize_t part_size_show(struct device *dev,
927 		       struct device_attribute *attr, char *buf)
928 {
929 	return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
930 }
931 
932 ssize_t part_stat_show(struct device *dev,
933 		       struct device_attribute *attr, char *buf)
934 {
935 	struct block_device *bdev = dev_to_bdev(dev);
936 	struct request_queue *q = bdev_get_queue(bdev);
937 	struct disk_stats stat;
938 	unsigned int inflight;
939 
940 	if (queue_is_mq(q))
941 		inflight = blk_mq_in_flight(q, bdev);
942 	else
943 		inflight = part_in_flight(bdev);
944 
945 	if (inflight) {
946 		part_stat_lock();
947 		update_io_ticks(bdev, jiffies, true);
948 		part_stat_unlock();
949 	}
950 	part_stat_read_all(bdev, &stat);
951 	return sprintf(buf,
952 		"%8lu %8lu %8llu %8u "
953 		"%8lu %8lu %8llu %8u "
954 		"%8u %8u %8u "
955 		"%8lu %8lu %8llu %8u "
956 		"%8lu %8u"
957 		"\n",
958 		stat.ios[STAT_READ],
959 		stat.merges[STAT_READ],
960 		(unsigned long long)stat.sectors[STAT_READ],
961 		(unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
962 		stat.ios[STAT_WRITE],
963 		stat.merges[STAT_WRITE],
964 		(unsigned long long)stat.sectors[STAT_WRITE],
965 		(unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
966 		inflight,
967 		jiffies_to_msecs(stat.io_ticks),
968 		(unsigned int)div_u64(stat.nsecs[STAT_READ] +
969 				      stat.nsecs[STAT_WRITE] +
970 				      stat.nsecs[STAT_DISCARD] +
971 				      stat.nsecs[STAT_FLUSH],
972 						NSEC_PER_MSEC),
973 		stat.ios[STAT_DISCARD],
974 		stat.merges[STAT_DISCARD],
975 		(unsigned long long)stat.sectors[STAT_DISCARD],
976 		(unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
977 		stat.ios[STAT_FLUSH],
978 		(unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
979 }
980 
981 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
982 			   char *buf)
983 {
984 	struct block_device *bdev = dev_to_bdev(dev);
985 	struct request_queue *q = bdev_get_queue(bdev);
986 	unsigned int inflight[2];
987 
988 	if (queue_is_mq(q))
989 		blk_mq_in_flight_rw(q, bdev, inflight);
990 	else
991 		part_in_flight_rw(bdev, inflight);
992 
993 	return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
994 }
995 
996 static ssize_t disk_capability_show(struct device *dev,
997 				    struct device_attribute *attr, char *buf)
998 {
999 	struct gendisk *disk = dev_to_disk(dev);
1000 
1001 	return sprintf(buf, "%x\n", disk->flags);
1002 }
1003 
1004 static ssize_t disk_alignment_offset_show(struct device *dev,
1005 					  struct device_attribute *attr,
1006 					  char *buf)
1007 {
1008 	struct gendisk *disk = dev_to_disk(dev);
1009 
1010 	return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1011 }
1012 
1013 static ssize_t disk_discard_alignment_show(struct device *dev,
1014 					   struct device_attribute *attr,
1015 					   char *buf)
1016 {
1017 	struct gendisk *disk = dev_to_disk(dev);
1018 
1019 	return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1020 }
1021 
1022 static ssize_t diskseq_show(struct device *dev,
1023 			    struct device_attribute *attr, char *buf)
1024 {
1025 	struct gendisk *disk = dev_to_disk(dev);
1026 
1027 	return sprintf(buf, "%llu\n", disk->diskseq);
1028 }
1029 
1030 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1031 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1032 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1033 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1034 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1035 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1036 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1037 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1038 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1039 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1040 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1041 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1042 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1043 
1044 #ifdef CONFIG_FAIL_MAKE_REQUEST
1045 ssize_t part_fail_show(struct device *dev,
1046 		       struct device_attribute *attr, char *buf)
1047 {
1048 	return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1049 }
1050 
1051 ssize_t part_fail_store(struct device *dev,
1052 			struct device_attribute *attr,
1053 			const char *buf, size_t count)
1054 {
1055 	int i;
1056 
1057 	if (count > 0 && sscanf(buf, "%d", &i) > 0)
1058 		dev_to_bdev(dev)->bd_make_it_fail = i;
1059 
1060 	return count;
1061 }
1062 
1063 static struct device_attribute dev_attr_fail =
1064 	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1065 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1066 
1067 #ifdef CONFIG_FAIL_IO_TIMEOUT
1068 static struct device_attribute dev_attr_fail_timeout =
1069 	__ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1070 #endif
1071 
1072 static struct attribute *disk_attrs[] = {
1073 	&dev_attr_range.attr,
1074 	&dev_attr_ext_range.attr,
1075 	&dev_attr_removable.attr,
1076 	&dev_attr_hidden.attr,
1077 	&dev_attr_ro.attr,
1078 	&dev_attr_size.attr,
1079 	&dev_attr_alignment_offset.attr,
1080 	&dev_attr_discard_alignment.attr,
1081 	&dev_attr_capability.attr,
1082 	&dev_attr_stat.attr,
1083 	&dev_attr_inflight.attr,
1084 	&dev_attr_badblocks.attr,
1085 	&dev_attr_events.attr,
1086 	&dev_attr_events_async.attr,
1087 	&dev_attr_events_poll_msecs.attr,
1088 	&dev_attr_diskseq.attr,
1089 #ifdef CONFIG_FAIL_MAKE_REQUEST
1090 	&dev_attr_fail.attr,
1091 #endif
1092 #ifdef CONFIG_FAIL_IO_TIMEOUT
1093 	&dev_attr_fail_timeout.attr,
1094 #endif
1095 	NULL
1096 };
1097 
1098 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1099 {
1100 	struct device *dev = container_of(kobj, typeof(*dev), kobj);
1101 	struct gendisk *disk = dev_to_disk(dev);
1102 
1103 	if (a == &dev_attr_badblocks.attr && !disk->bb)
1104 		return 0;
1105 	return a->mode;
1106 }
1107 
1108 static struct attribute_group disk_attr_group = {
1109 	.attrs = disk_attrs,
1110 	.is_visible = disk_visible,
1111 };
1112 
1113 static const struct attribute_group *disk_attr_groups[] = {
1114 	&disk_attr_group,
1115 #ifdef CONFIG_BLK_DEV_IO_TRACE
1116 	&blk_trace_attr_group,
1117 #endif
1118 	NULL
1119 };
1120 
1121 /**
1122  * disk_release - releases all allocated resources of the gendisk
1123  * @dev: the device representing this disk
1124  *
1125  * This function releases all allocated resources of the gendisk.
1126  *
1127  * Drivers which used __device_add_disk() have a gendisk with a request_queue
1128  * assigned. Since the request_queue sits on top of the gendisk for these
1129  * drivers we also call blk_put_queue() for them, and we expect the
1130  * request_queue refcount to reach 0 at this point, and so the request_queue
1131  * will also be freed prior to the disk.
1132  *
1133  * Context: can sleep
1134  */
1135 static void disk_release(struct device *dev)
1136 {
1137 	struct gendisk *disk = dev_to_disk(dev);
1138 
1139 	might_sleep();
1140 	WARN_ON_ONCE(disk_live(disk));
1141 
1142 	/*
1143 	 * To undo the all initialization from blk_mq_init_allocated_queue in
1144 	 * case of a probe failure where add_disk is never called we have to
1145 	 * call blk_mq_exit_queue here. We can't do this for the more common
1146 	 * teardown case (yet) as the tagset can be gone by the time the disk
1147 	 * is released once it was added.
1148 	 */
1149 	if (queue_is_mq(disk->queue) &&
1150 	    test_bit(GD_OWNS_QUEUE, &disk->state) &&
1151 	    !test_bit(GD_ADDED, &disk->state))
1152 		blk_mq_exit_queue(disk->queue);
1153 
1154 	blkcg_exit_queue(disk->queue);
1155 	bioset_exit(&disk->bio_split);
1156 
1157 	disk_release_events(disk);
1158 	kfree(disk->random);
1159 	disk_free_zone_bitmaps(disk);
1160 	xa_destroy(&disk->part_tbl);
1161 
1162 	disk->queue->disk = NULL;
1163 	blk_put_queue(disk->queue);
1164 
1165 	if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1166 		disk->fops->free_disk(disk);
1167 
1168 	iput(disk->part0->bd_inode);	/* frees the disk */
1169 }
1170 
1171 static int block_uevent(struct device *dev, struct kobj_uevent_env *env)
1172 {
1173 	struct gendisk *disk = dev_to_disk(dev);
1174 
1175 	return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1176 }
1177 
1178 struct class block_class = {
1179 	.name		= "block",
1180 	.dev_uevent	= block_uevent,
1181 };
1182 
1183 static char *block_devnode(struct device *dev, umode_t *mode,
1184 			   kuid_t *uid, kgid_t *gid)
1185 {
1186 	struct gendisk *disk = dev_to_disk(dev);
1187 
1188 	if (disk->fops->devnode)
1189 		return disk->fops->devnode(disk, mode);
1190 	return NULL;
1191 }
1192 
1193 const struct device_type disk_type = {
1194 	.name		= "disk",
1195 	.groups		= disk_attr_groups,
1196 	.release	= disk_release,
1197 	.devnode	= block_devnode,
1198 };
1199 
1200 #ifdef CONFIG_PROC_FS
1201 /*
1202  * aggregate disk stat collector.  Uses the same stats that the sysfs
1203  * entries do, above, but makes them available through one seq_file.
1204  *
1205  * The output looks suspiciously like /proc/partitions with a bunch of
1206  * extra fields.
1207  */
1208 static int diskstats_show(struct seq_file *seqf, void *v)
1209 {
1210 	struct gendisk *gp = v;
1211 	struct block_device *hd;
1212 	unsigned int inflight;
1213 	struct disk_stats stat;
1214 	unsigned long idx;
1215 
1216 	/*
1217 	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1218 		seq_puts(seqf,	"major minor name"
1219 				"     rio rmerge rsect ruse wio wmerge "
1220 				"wsect wuse running use aveq"
1221 				"\n\n");
1222 	*/
1223 
1224 	rcu_read_lock();
1225 	xa_for_each(&gp->part_tbl, idx, hd) {
1226 		if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1227 			continue;
1228 		if (queue_is_mq(gp->queue))
1229 			inflight = blk_mq_in_flight(gp->queue, hd);
1230 		else
1231 			inflight = part_in_flight(hd);
1232 
1233 		if (inflight) {
1234 			part_stat_lock();
1235 			update_io_ticks(hd, jiffies, true);
1236 			part_stat_unlock();
1237 		}
1238 		part_stat_read_all(hd, &stat);
1239 		seq_printf(seqf, "%4d %7d %pg "
1240 			   "%lu %lu %lu %u "
1241 			   "%lu %lu %lu %u "
1242 			   "%u %u %u "
1243 			   "%lu %lu %lu %u "
1244 			   "%lu %u"
1245 			   "\n",
1246 			   MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1247 			   stat.ios[STAT_READ],
1248 			   stat.merges[STAT_READ],
1249 			   stat.sectors[STAT_READ],
1250 			   (unsigned int)div_u64(stat.nsecs[STAT_READ],
1251 							NSEC_PER_MSEC),
1252 			   stat.ios[STAT_WRITE],
1253 			   stat.merges[STAT_WRITE],
1254 			   stat.sectors[STAT_WRITE],
1255 			   (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1256 							NSEC_PER_MSEC),
1257 			   inflight,
1258 			   jiffies_to_msecs(stat.io_ticks),
1259 			   (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1260 						 stat.nsecs[STAT_WRITE] +
1261 						 stat.nsecs[STAT_DISCARD] +
1262 						 stat.nsecs[STAT_FLUSH],
1263 							NSEC_PER_MSEC),
1264 			   stat.ios[STAT_DISCARD],
1265 			   stat.merges[STAT_DISCARD],
1266 			   stat.sectors[STAT_DISCARD],
1267 			   (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1268 						 NSEC_PER_MSEC),
1269 			   stat.ios[STAT_FLUSH],
1270 			   (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1271 						 NSEC_PER_MSEC)
1272 			);
1273 	}
1274 	rcu_read_unlock();
1275 
1276 	return 0;
1277 }
1278 
1279 static const struct seq_operations diskstats_op = {
1280 	.start	= disk_seqf_start,
1281 	.next	= disk_seqf_next,
1282 	.stop	= disk_seqf_stop,
1283 	.show	= diskstats_show
1284 };
1285 
1286 static int __init proc_genhd_init(void)
1287 {
1288 	proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1289 	proc_create_seq("partitions", 0, NULL, &partitions_op);
1290 	return 0;
1291 }
1292 module_init(proc_genhd_init);
1293 #endif /* CONFIG_PROC_FS */
1294 
1295 dev_t part_devt(struct gendisk *disk, u8 partno)
1296 {
1297 	struct block_device *part;
1298 	dev_t devt = 0;
1299 
1300 	rcu_read_lock();
1301 	part = xa_load(&disk->part_tbl, partno);
1302 	if (part)
1303 		devt = part->bd_dev;
1304 	rcu_read_unlock();
1305 
1306 	return devt;
1307 }
1308 
1309 dev_t blk_lookup_devt(const char *name, int partno)
1310 {
1311 	dev_t devt = MKDEV(0, 0);
1312 	struct class_dev_iter iter;
1313 	struct device *dev;
1314 
1315 	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1316 	while ((dev = class_dev_iter_next(&iter))) {
1317 		struct gendisk *disk = dev_to_disk(dev);
1318 
1319 		if (strcmp(dev_name(dev), name))
1320 			continue;
1321 
1322 		if (partno < disk->minors) {
1323 			/* We need to return the right devno, even
1324 			 * if the partition doesn't exist yet.
1325 			 */
1326 			devt = MKDEV(MAJOR(dev->devt),
1327 				     MINOR(dev->devt) + partno);
1328 		} else {
1329 			devt = part_devt(disk, partno);
1330 			if (devt)
1331 				break;
1332 		}
1333 	}
1334 	class_dev_iter_exit(&iter);
1335 	return devt;
1336 }
1337 
1338 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1339 		struct lock_class_key *lkclass)
1340 {
1341 	struct gendisk *disk;
1342 
1343 	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1344 	if (!disk)
1345 		return NULL;
1346 
1347 	if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1348 		goto out_free_disk;
1349 
1350 	disk->bdi = bdi_alloc(node_id);
1351 	if (!disk->bdi)
1352 		goto out_free_bioset;
1353 
1354 	/* bdev_alloc() might need the queue, set before the first call */
1355 	disk->queue = q;
1356 
1357 	disk->part0 = bdev_alloc(disk, 0);
1358 	if (!disk->part0)
1359 		goto out_free_bdi;
1360 
1361 	disk->node_id = node_id;
1362 	mutex_init(&disk->open_mutex);
1363 	xa_init(&disk->part_tbl);
1364 	if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1365 		goto out_destroy_part_tbl;
1366 
1367 	if (blkcg_init_queue(q))
1368 		goto out_erase_part0;
1369 
1370 	rand_initialize_disk(disk);
1371 	disk_to_dev(disk)->class = &block_class;
1372 	disk_to_dev(disk)->type = &disk_type;
1373 	device_initialize(disk_to_dev(disk));
1374 	inc_diskseq(disk);
1375 	q->disk = disk;
1376 	lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1377 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1378 	INIT_LIST_HEAD(&disk->slave_bdevs);
1379 #endif
1380 	return disk;
1381 
1382 out_erase_part0:
1383 	xa_erase(&disk->part_tbl, 0);
1384 out_destroy_part_tbl:
1385 	xa_destroy(&disk->part_tbl);
1386 	disk->part0->bd_disk = NULL;
1387 	iput(disk->part0->bd_inode);
1388 out_free_bdi:
1389 	bdi_put(disk->bdi);
1390 out_free_bioset:
1391 	bioset_exit(&disk->bio_split);
1392 out_free_disk:
1393 	kfree(disk);
1394 	return NULL;
1395 }
1396 
1397 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1398 {
1399 	struct request_queue *q;
1400 	struct gendisk *disk;
1401 
1402 	q = blk_alloc_queue(node, false);
1403 	if (!q)
1404 		return NULL;
1405 
1406 	disk = __alloc_disk_node(q, node, lkclass);
1407 	if (!disk) {
1408 		blk_put_queue(q);
1409 		return NULL;
1410 	}
1411 	set_bit(GD_OWNS_QUEUE, &disk->state);
1412 	return disk;
1413 }
1414 EXPORT_SYMBOL(__blk_alloc_disk);
1415 
1416 /**
1417  * put_disk - decrements the gendisk refcount
1418  * @disk: the struct gendisk to decrement the refcount for
1419  *
1420  * This decrements the refcount for the struct gendisk. When this reaches 0
1421  * we'll have disk_release() called.
1422  *
1423  * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1424  * when handling probe errors (that is before add_disk() is called).
1425  *
1426  * Context: Any context, but the last reference must not be dropped from
1427  *          atomic context.
1428  */
1429 void put_disk(struct gendisk *disk)
1430 {
1431 	if (disk)
1432 		put_device(disk_to_dev(disk));
1433 }
1434 EXPORT_SYMBOL(put_disk);
1435 
1436 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1437 {
1438 	char event[] = "DISK_RO=1";
1439 	char *envp[] = { event, NULL };
1440 
1441 	if (!ro)
1442 		event[8] = '0';
1443 	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1444 }
1445 
1446 /**
1447  * set_disk_ro - set a gendisk read-only
1448  * @disk:	gendisk to operate on
1449  * @read_only:	%true to set the disk read-only, %false set the disk read/write
1450  *
1451  * This function is used to indicate whether a given disk device should have its
1452  * read-only flag set. set_disk_ro() is typically used by device drivers to
1453  * indicate whether the underlying physical device is write-protected.
1454  */
1455 void set_disk_ro(struct gendisk *disk, bool read_only)
1456 {
1457 	if (read_only) {
1458 		if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1459 			return;
1460 	} else {
1461 		if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1462 			return;
1463 	}
1464 	set_disk_ro_uevent(disk, read_only);
1465 }
1466 EXPORT_SYMBOL(set_disk_ro);
1467 
1468 void inc_diskseq(struct gendisk *disk)
1469 {
1470 	disk->diskseq = atomic64_inc_return(&diskseq);
1471 }
1472