xref: /openbmc/linux/block/genhd.c (revision 5a5625a8)
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 
36 static 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  */
53 static atomic64_t diskseq;
54 
55 /* for extended dynamic devt allocation, currently only one major is used */
56 #define NR_EXT_DEVT		(1 << MINORBITS)
57 static DEFINE_IDA(ext_devt_ida);
58 
set_capacity(struct gendisk * disk,sector_t sectors)59 void set_capacity(struct gendisk *disk, sector_t sectors)
60 {
61 	bdev_set_nr_sectors(disk->part0, sectors);
62 }
63 EXPORT_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  */
set_capacity_and_notify(struct gendisk * disk,sector_t size)69 bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
70 {
71 	sector_t capacity = get_capacity(disk);
72 	char *envp[] = { "RESIZE=1", NULL };
73 
74 	set_capacity(disk, size);
75 
76 	/*
77 	 * Only print a message and send a uevent if the gendisk is user visible
78 	 * and alive.  This avoids spamming the log and udev when setting the
79 	 * initial capacity during probing.
80 	 */
81 	if (size == capacity ||
82 	    !disk_live(disk) ||
83 	    (disk->flags & GENHD_FL_HIDDEN))
84 		return false;
85 
86 	pr_info("%s: detected capacity change from %lld to %lld\n",
87 		disk->disk_name, capacity, size);
88 
89 	/*
90 	 * Historically we did not send a uevent for changes to/from an empty
91 	 * device.
92 	 */
93 	if (!capacity || !size)
94 		return false;
95 	kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
96 	return true;
97 }
98 EXPORT_SYMBOL_GPL(set_capacity_and_notify);
99 
part_stat_read_all(struct block_device * part,struct disk_stats * stat)100 static 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 
part_in_flight(struct block_device * part)121 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 
part_in_flight_rw(struct block_device * part,unsigned int inflight[2])136 static 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
158 static 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];
166 static DEFINE_MUTEX(major_names_lock);
167 static DEFINE_SPINLOCK(major_names_spinlock);
168 
169 /* index in the above - for now: assume no multimajor ranges */
major_to_index(unsigned major)170 static inline int major_to_index(unsigned major)
171 {
172 	return major % BLKDEV_MAJOR_HASH_SIZE;
173 }
174 
175 #ifdef CONFIG_PROC_FS
blkdev_show(struct seq_file * seqf,off_t offset)176 void blkdev_show(struct seq_file *seqf, off_t offset)
177 {
178 	struct blk_major_name *dp;
179 
180 	spin_lock(&major_names_spinlock);
181 	for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
182 		if (dp->major == offset)
183 			seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
184 	spin_unlock(&major_names_spinlock);
185 }
186 #endif /* CONFIG_PROC_FS */
187 
188 /**
189  * __register_blkdev - register a new block device
190  *
191  * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
192  *         @major = 0, try to allocate any unused major number.
193  * @name: the name of the new block device as a zero terminated string
194  * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
195  *	   pre-created device node is accessed. When a probe call uses
196  *	   add_disk() and it fails the driver must cleanup resources. This
197  *	   interface may soon be removed.
198  *
199  * The @name must be unique within the system.
200  *
201  * The return value depends on the @major input parameter:
202  *
203  *  - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
204  *    then the function returns zero on success, or a negative error code
205  *  - if any unused major number was requested with @major = 0 parameter
206  *    then the return value is the allocated major number in range
207  *    [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
208  *
209  * See Documentation/admin-guide/devices.txt for the list of allocated
210  * major numbers.
211  *
212  * Use register_blkdev instead for any new code.
213  */
__register_blkdev(unsigned int major,const char * name,void (* probe)(dev_t devt))214 int __register_blkdev(unsigned int major, const char *name,
215 		void (*probe)(dev_t devt))
216 {
217 	struct blk_major_name **n, *p;
218 	int index, ret = 0;
219 
220 	mutex_lock(&major_names_lock);
221 
222 	/* temporary */
223 	if (major == 0) {
224 		for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
225 			if (major_names[index] == NULL)
226 				break;
227 		}
228 
229 		if (index == 0) {
230 			printk("%s: failed to get major for %s\n",
231 			       __func__, name);
232 			ret = -EBUSY;
233 			goto out;
234 		}
235 		major = index;
236 		ret = major;
237 	}
238 
239 	if (major >= BLKDEV_MAJOR_MAX) {
240 		pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
241 		       __func__, major, BLKDEV_MAJOR_MAX-1, name);
242 
243 		ret = -EINVAL;
244 		goto out;
245 	}
246 
247 	p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
248 	if (p == NULL) {
249 		ret = -ENOMEM;
250 		goto out;
251 	}
252 
253 	p->major = major;
254 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
255 	p->probe = probe;
256 #endif
257 	strscpy(p->name, name, sizeof(p->name));
258 	p->next = NULL;
259 	index = major_to_index(major);
260 
261 	spin_lock(&major_names_spinlock);
262 	for (n = &major_names[index]; *n; n = &(*n)->next) {
263 		if ((*n)->major == major)
264 			break;
265 	}
266 	if (!*n)
267 		*n = p;
268 	else
269 		ret = -EBUSY;
270 	spin_unlock(&major_names_spinlock);
271 
272 	if (ret < 0) {
273 		printk("register_blkdev: cannot get major %u for %s\n",
274 		       major, name);
275 		kfree(p);
276 	}
277 out:
278 	mutex_unlock(&major_names_lock);
279 	return ret;
280 }
281 EXPORT_SYMBOL(__register_blkdev);
282 
unregister_blkdev(unsigned int major,const char * name)283 void unregister_blkdev(unsigned int major, const char *name)
284 {
285 	struct blk_major_name **n;
286 	struct blk_major_name *p = NULL;
287 	int index = major_to_index(major);
288 
289 	mutex_lock(&major_names_lock);
290 	spin_lock(&major_names_spinlock);
291 	for (n = &major_names[index]; *n; n = &(*n)->next)
292 		if ((*n)->major == major)
293 			break;
294 	if (!*n || strcmp((*n)->name, name)) {
295 		WARN_ON(1);
296 	} else {
297 		p = *n;
298 		*n = p->next;
299 	}
300 	spin_unlock(&major_names_spinlock);
301 	mutex_unlock(&major_names_lock);
302 	kfree(p);
303 }
304 
305 EXPORT_SYMBOL(unregister_blkdev);
306 
blk_alloc_ext_minor(void)307 int blk_alloc_ext_minor(void)
308 {
309 	int idx;
310 
311 	idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
312 	if (idx == -ENOSPC)
313 		return -EBUSY;
314 	return idx;
315 }
316 
blk_free_ext_minor(unsigned int minor)317 void blk_free_ext_minor(unsigned int minor)
318 {
319 	ida_free(&ext_devt_ida, minor);
320 }
321 
disk_uevent(struct gendisk * disk,enum kobject_action action)322 void disk_uevent(struct gendisk *disk, enum kobject_action action)
323 {
324 	struct block_device *part;
325 	unsigned long idx;
326 
327 	rcu_read_lock();
328 	xa_for_each(&disk->part_tbl, idx, part) {
329 		if (bdev_is_partition(part) && !bdev_nr_sectors(part))
330 			continue;
331 		if (!kobject_get_unless_zero(&part->bd_device.kobj))
332 			continue;
333 
334 		rcu_read_unlock();
335 		kobject_uevent(bdev_kobj(part), action);
336 		put_device(&part->bd_device);
337 		rcu_read_lock();
338 	}
339 	rcu_read_unlock();
340 }
341 EXPORT_SYMBOL_GPL(disk_uevent);
342 
disk_scan_partitions(struct gendisk * disk,blk_mode_t mode)343 int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode)
344 {
345 	struct block_device *bdev;
346 	int ret = 0;
347 
348 	if (!disk_has_partscan(disk))
349 		return -EINVAL;
350 	if (disk->open_partitions)
351 		return -EBUSY;
352 
353 	/*
354 	 * If the device is opened exclusively by current thread already, it's
355 	 * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
356 	 * synchronize with other exclusive openers and other partition
357 	 * scanners.
358 	 */
359 	if (!(mode & BLK_OPEN_EXCL)) {
360 		ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions,
361 					  NULL);
362 		if (ret)
363 			return ret;
364 	}
365 
366 	set_bit(GD_NEED_PART_SCAN, &disk->state);
367 	bdev = blkdev_get_by_dev(disk_devt(disk), mode & ~BLK_OPEN_EXCL, NULL,
368 				 NULL);
369 	if (IS_ERR(bdev))
370 		ret =  PTR_ERR(bdev);
371 	else
372 		blkdev_put(bdev, NULL);
373 
374 	/*
375 	 * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
376 	 * and this will cause that re-assemble partitioned raid device will
377 	 * creat partition for underlying disk.
378 	 */
379 	clear_bit(GD_NEED_PART_SCAN, &disk->state);
380 	if (!(mode & BLK_OPEN_EXCL))
381 		bd_abort_claiming(disk->part0, disk_scan_partitions);
382 	return ret;
383 }
384 
385 /**
386  * device_add_disk - add disk information to kernel list
387  * @parent: parent device for the disk
388  * @disk: per-device partitioning information
389  * @groups: Additional per-device sysfs groups
390  *
391  * This function registers the partitioning information in @disk
392  * with the kernel.
393  */
device_add_disk(struct device * parent,struct gendisk * disk,const struct attribute_group ** groups)394 int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
395 				 const struct attribute_group **groups)
396 
397 {
398 	struct device *ddev = disk_to_dev(disk);
399 	int ret;
400 
401 	/* Only makes sense for bio-based to set ->poll_bio */
402 	if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
403 		return -EINVAL;
404 
405 	/*
406 	 * The disk queue should now be all set with enough information about
407 	 * the device for the elevator code to pick an adequate default
408 	 * elevator if one is needed, that is, for devices requesting queue
409 	 * registration.
410 	 */
411 	elevator_init_mq(disk->queue);
412 
413 	/* Mark bdev as having a submit_bio, if needed */
414 	disk->part0->bd_has_submit_bio = disk->fops->submit_bio != NULL;
415 
416 	/*
417 	 * If the driver provides an explicit major number it also must provide
418 	 * the number of minors numbers supported, and those will be used to
419 	 * setup the gendisk.
420 	 * Otherwise just allocate the device numbers for both the whole device
421 	 * and all partitions from the extended dev_t space.
422 	 */
423 	ret = -EINVAL;
424 	if (disk->major) {
425 		if (WARN_ON(!disk->minors))
426 			goto out_exit_elevator;
427 
428 		if (disk->minors > DISK_MAX_PARTS) {
429 			pr_err("block: can't allocate more than %d partitions\n",
430 				DISK_MAX_PARTS);
431 			disk->minors = DISK_MAX_PARTS;
432 		}
433 		if (disk->first_minor > MINORMASK ||
434 		    disk->minors > MINORMASK + 1 ||
435 		    disk->first_minor + disk->minors > MINORMASK + 1)
436 			goto out_exit_elevator;
437 	} else {
438 		if (WARN_ON(disk->minors))
439 			goto out_exit_elevator;
440 
441 		ret = blk_alloc_ext_minor();
442 		if (ret < 0)
443 			goto out_exit_elevator;
444 		disk->major = BLOCK_EXT_MAJOR;
445 		disk->first_minor = ret;
446 	}
447 
448 	/* delay uevents, until we scanned partition table */
449 	dev_set_uevent_suppress(ddev, 1);
450 
451 	ddev->parent = parent;
452 	ddev->groups = groups;
453 	dev_set_name(ddev, "%s", disk->disk_name);
454 	if (!(disk->flags & GENHD_FL_HIDDEN))
455 		ddev->devt = MKDEV(disk->major, disk->first_minor);
456 	ret = device_add(ddev);
457 	if (ret)
458 		goto out_free_ext_minor;
459 
460 	ret = disk_alloc_events(disk);
461 	if (ret)
462 		goto out_device_del;
463 
464 	ret = sysfs_create_link(block_depr, &ddev->kobj,
465 				kobject_name(&ddev->kobj));
466 	if (ret)
467 		goto out_device_del;
468 
469 	/*
470 	 * avoid probable deadlock caused by allocating memory with
471 	 * GFP_KERNEL in runtime_resume callback of its all ancestor
472 	 * devices
473 	 */
474 	pm_runtime_set_memalloc_noio(ddev, true);
475 
476 	disk->part0->bd_holder_dir =
477 		kobject_create_and_add("holders", &ddev->kobj);
478 	if (!disk->part0->bd_holder_dir) {
479 		ret = -ENOMEM;
480 		goto out_del_block_link;
481 	}
482 	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
483 	if (!disk->slave_dir) {
484 		ret = -ENOMEM;
485 		goto out_put_holder_dir;
486 	}
487 
488 	ret = blk_register_queue(disk);
489 	if (ret)
490 		goto out_put_slave_dir;
491 
492 	if (!(disk->flags & GENHD_FL_HIDDEN)) {
493 		ret = bdi_register(disk->bdi, "%u:%u",
494 				   disk->major, disk->first_minor);
495 		if (ret)
496 			goto out_unregister_queue;
497 		bdi_set_owner(disk->bdi, ddev);
498 		ret = sysfs_create_link(&ddev->kobj,
499 					&disk->bdi->dev->kobj, "bdi");
500 		if (ret)
501 			goto out_unregister_bdi;
502 
503 		/* Make sure the first partition scan will be proceed */
504 		if (get_capacity(disk) && disk_has_partscan(disk))
505 			set_bit(GD_NEED_PART_SCAN, &disk->state);
506 
507 		bdev_add(disk->part0, ddev->devt);
508 		if (get_capacity(disk))
509 			disk_scan_partitions(disk, BLK_OPEN_READ);
510 
511 		/*
512 		 * Announce the disk and partitions after all partitions are
513 		 * created. (for hidden disks uevents remain suppressed forever)
514 		 */
515 		dev_set_uevent_suppress(ddev, 0);
516 		disk_uevent(disk, KOBJ_ADD);
517 	} else {
518 		/*
519 		 * Even if the block_device for a hidden gendisk is not
520 		 * registered, it needs to have a valid bd_dev so that the
521 		 * freeing of the dynamic major works.
522 		 */
523 		disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
524 	}
525 
526 	disk_update_readahead(disk);
527 	disk_add_events(disk);
528 	set_bit(GD_ADDED, &disk->state);
529 	return 0;
530 
531 out_unregister_bdi:
532 	if (!(disk->flags & GENHD_FL_HIDDEN))
533 		bdi_unregister(disk->bdi);
534 out_unregister_queue:
535 	blk_unregister_queue(disk);
536 	rq_qos_exit(disk->queue);
537 out_put_slave_dir:
538 	kobject_put(disk->slave_dir);
539 	disk->slave_dir = NULL;
540 out_put_holder_dir:
541 	kobject_put(disk->part0->bd_holder_dir);
542 out_del_block_link:
543 	sysfs_remove_link(block_depr, dev_name(ddev));
544 	pm_runtime_set_memalloc_noio(ddev, false);
545 out_device_del:
546 	device_del(ddev);
547 out_free_ext_minor:
548 	if (disk->major == BLOCK_EXT_MAJOR)
549 		blk_free_ext_minor(disk->first_minor);
550 out_exit_elevator:
551 	if (disk->queue->elevator)
552 		elevator_exit(disk->queue);
553 	return ret;
554 }
555 EXPORT_SYMBOL(device_add_disk);
556 
blk_report_disk_dead(struct gendisk * disk,bool surprise)557 static void blk_report_disk_dead(struct gendisk *disk, bool surprise)
558 {
559 	struct block_device *bdev;
560 	unsigned long idx;
561 
562 	rcu_read_lock();
563 	xa_for_each(&disk->part_tbl, idx, bdev) {
564 		if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
565 			continue;
566 		rcu_read_unlock();
567 
568 		bdev_mark_dead(bdev, surprise);
569 
570 		put_device(&bdev->bd_device);
571 		rcu_read_lock();
572 	}
573 	rcu_read_unlock();
574 }
575 
__blk_mark_disk_dead(struct gendisk * disk)576 static void __blk_mark_disk_dead(struct gendisk *disk)
577 {
578 	/*
579 	 * Fail any new I/O.
580 	 */
581 	if (test_and_set_bit(GD_DEAD, &disk->state))
582 		return;
583 
584 	if (test_bit(GD_OWNS_QUEUE, &disk->state))
585 		blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);
586 
587 	/*
588 	 * Stop buffered writers from dirtying pages that can't be written out.
589 	 */
590 	set_capacity(disk, 0);
591 
592 	/*
593 	 * Prevent new I/O from crossing bio_queue_enter().
594 	 */
595 	blk_queue_start_drain(disk->queue);
596 }
597 
598 /**
599  * blk_mark_disk_dead - mark a disk as dead
600  * @disk: disk to mark as dead
601  *
602  * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
603  * to this disk.
604  */
blk_mark_disk_dead(struct gendisk * disk)605 void blk_mark_disk_dead(struct gendisk *disk)
606 {
607 	__blk_mark_disk_dead(disk);
608 	blk_report_disk_dead(disk, true);
609 }
610 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
611 
612 /**
613  * del_gendisk - remove the gendisk
614  * @disk: the struct gendisk to remove
615  *
616  * Removes the gendisk and all its associated resources. This deletes the
617  * partitions associated with the gendisk, and unregisters the associated
618  * request_queue.
619  *
620  * This is the counter to the respective __device_add_disk() call.
621  *
622  * The final removal of the struct gendisk happens when its refcount reaches 0
623  * with put_disk(), which should be called after del_gendisk(), if
624  * __device_add_disk() was used.
625  *
626  * Drivers exist which depend on the release of the gendisk to be synchronous,
627  * it should not be deferred.
628  *
629  * Context: can sleep
630  */
del_gendisk(struct gendisk * disk)631 void del_gendisk(struct gendisk *disk)
632 {
633 	struct request_queue *q = disk->queue;
634 	struct block_device *part;
635 	unsigned long idx;
636 
637 	might_sleep();
638 
639 	if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
640 		return;
641 
642 	disk_del_events(disk);
643 
644 	/*
645 	 * Prevent new openers by unlinked the bdev inode.
646 	 */
647 	mutex_lock(&disk->open_mutex);
648 	xa_for_each(&disk->part_tbl, idx, part)
649 		remove_inode_hash(part->bd_inode);
650 	mutex_unlock(&disk->open_mutex);
651 
652 	/*
653 	 * Tell the file system to write back all dirty data and shut down if
654 	 * it hasn't been notified earlier.
655 	 */
656 	if (!test_bit(GD_DEAD, &disk->state))
657 		blk_report_disk_dead(disk, false);
658 
659 	/*
660 	 * Drop all partitions now that the disk is marked dead.
661 	 */
662 	mutex_lock(&disk->open_mutex);
663 	__blk_mark_disk_dead(disk);
664 	xa_for_each_start(&disk->part_tbl, idx, part, 1)
665 		drop_partition(part);
666 	mutex_unlock(&disk->open_mutex);
667 
668 	if (!(disk->flags & GENHD_FL_HIDDEN)) {
669 		sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
670 
671 		/*
672 		 * Unregister bdi before releasing device numbers (as they can
673 		 * get reused and we'd get clashes in sysfs).
674 		 */
675 		bdi_unregister(disk->bdi);
676 	}
677 
678 	blk_unregister_queue(disk);
679 
680 	kobject_put(disk->part0->bd_holder_dir);
681 	kobject_put(disk->slave_dir);
682 	disk->slave_dir = NULL;
683 
684 	part_stat_set_all(disk->part0, 0);
685 	disk->part0->bd_stamp = 0;
686 	sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
687 	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
688 	device_del(disk_to_dev(disk));
689 
690 	blk_mq_freeze_queue_wait(q);
691 
692 	blk_throtl_cancel_bios(disk);
693 
694 	blk_sync_queue(q);
695 	blk_flush_integrity();
696 
697 	if (queue_is_mq(q))
698 		blk_mq_cancel_work_sync(q);
699 
700 	blk_mq_quiesce_queue(q);
701 	if (q->elevator) {
702 		mutex_lock(&q->sysfs_lock);
703 		elevator_exit(q);
704 		mutex_unlock(&q->sysfs_lock);
705 	}
706 	rq_qos_exit(q);
707 	blk_mq_unquiesce_queue(q);
708 
709 	/*
710 	 * If the disk does not own the queue, allow using passthrough requests
711 	 * again.  Else leave the queue frozen to fail all I/O.
712 	 */
713 	if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
714 		blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
715 		__blk_mq_unfreeze_queue(q, true);
716 	} else {
717 		if (queue_is_mq(q))
718 			blk_mq_exit_queue(q);
719 	}
720 }
721 EXPORT_SYMBOL(del_gendisk);
722 
723 /**
724  * invalidate_disk - invalidate the disk
725  * @disk: the struct gendisk to invalidate
726  *
727  * A helper to invalidates the disk. It will clean the disk's associated
728  * buffer/page caches and reset its internal states so that the disk
729  * can be reused by the drivers.
730  *
731  * Context: can sleep
732  */
invalidate_disk(struct gendisk * disk)733 void invalidate_disk(struct gendisk *disk)
734 {
735 	struct block_device *bdev = disk->part0;
736 
737 	invalidate_bdev(bdev);
738 	bdev->bd_inode->i_mapping->wb_err = 0;
739 	set_capacity(disk, 0);
740 }
741 EXPORT_SYMBOL(invalidate_disk);
742 
743 /* sysfs access to bad-blocks list. */
disk_badblocks_show(struct device * dev,struct device_attribute * attr,char * page)744 static ssize_t disk_badblocks_show(struct device *dev,
745 					struct device_attribute *attr,
746 					char *page)
747 {
748 	struct gendisk *disk = dev_to_disk(dev);
749 
750 	if (!disk->bb)
751 		return sprintf(page, "\n");
752 
753 	return badblocks_show(disk->bb, page, 0);
754 }
755 
disk_badblocks_store(struct device * dev,struct device_attribute * attr,const char * page,size_t len)756 static ssize_t disk_badblocks_store(struct device *dev,
757 					struct device_attribute *attr,
758 					const char *page, size_t len)
759 {
760 	struct gendisk *disk = dev_to_disk(dev);
761 
762 	if (!disk->bb)
763 		return -ENXIO;
764 
765 	return badblocks_store(disk->bb, page, len, 0);
766 }
767 
768 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
blk_request_module(dev_t devt)769 void blk_request_module(dev_t devt)
770 {
771 	unsigned int major = MAJOR(devt);
772 	struct blk_major_name **n;
773 
774 	mutex_lock(&major_names_lock);
775 	for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
776 		if ((*n)->major == major && (*n)->probe) {
777 			(*n)->probe(devt);
778 			mutex_unlock(&major_names_lock);
779 			return;
780 		}
781 	}
782 	mutex_unlock(&major_names_lock);
783 
784 	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
785 		/* Make old-style 2.4 aliases work */
786 		request_module("block-major-%d", MAJOR(devt));
787 }
788 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
789 
790 #ifdef CONFIG_PROC_FS
791 /* iterator */
disk_seqf_start(struct seq_file * seqf,loff_t * pos)792 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
793 {
794 	loff_t skip = *pos;
795 	struct class_dev_iter *iter;
796 	struct device *dev;
797 
798 	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
799 	if (!iter)
800 		return ERR_PTR(-ENOMEM);
801 
802 	seqf->private = iter;
803 	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
804 	do {
805 		dev = class_dev_iter_next(iter);
806 		if (!dev)
807 			return NULL;
808 	} while (skip--);
809 
810 	return dev_to_disk(dev);
811 }
812 
disk_seqf_next(struct seq_file * seqf,void * v,loff_t * pos)813 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
814 {
815 	struct device *dev;
816 
817 	(*pos)++;
818 	dev = class_dev_iter_next(seqf->private);
819 	if (dev)
820 		return dev_to_disk(dev);
821 
822 	return NULL;
823 }
824 
disk_seqf_stop(struct seq_file * seqf,void * v)825 static void disk_seqf_stop(struct seq_file *seqf, void *v)
826 {
827 	struct class_dev_iter *iter = seqf->private;
828 
829 	/* stop is called even after start failed :-( */
830 	if (iter) {
831 		class_dev_iter_exit(iter);
832 		kfree(iter);
833 		seqf->private = NULL;
834 	}
835 }
836 
show_partition_start(struct seq_file * seqf,loff_t * pos)837 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
838 {
839 	void *p;
840 
841 	p = disk_seqf_start(seqf, pos);
842 	if (!IS_ERR_OR_NULL(p) && !*pos)
843 		seq_puts(seqf, "major minor  #blocks  name\n\n");
844 	return p;
845 }
846 
show_partition(struct seq_file * seqf,void * v)847 static int show_partition(struct seq_file *seqf, void *v)
848 {
849 	struct gendisk *sgp = v;
850 	struct block_device *part;
851 	unsigned long idx;
852 
853 	if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
854 		return 0;
855 
856 	rcu_read_lock();
857 	xa_for_each(&sgp->part_tbl, idx, part) {
858 		if (!bdev_nr_sectors(part))
859 			continue;
860 		seq_printf(seqf, "%4d  %7d %10llu %pg\n",
861 			   MAJOR(part->bd_dev), MINOR(part->bd_dev),
862 			   bdev_nr_sectors(part) >> 1, part);
863 	}
864 	rcu_read_unlock();
865 	return 0;
866 }
867 
868 static const struct seq_operations partitions_op = {
869 	.start	= show_partition_start,
870 	.next	= disk_seqf_next,
871 	.stop	= disk_seqf_stop,
872 	.show	= show_partition
873 };
874 #endif
875 
genhd_device_init(void)876 static int __init genhd_device_init(void)
877 {
878 	int error;
879 
880 	error = class_register(&block_class);
881 	if (unlikely(error))
882 		return error;
883 	blk_dev_init();
884 
885 	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
886 
887 	/* create top-level block dir */
888 	block_depr = kobject_create_and_add("block", NULL);
889 	return 0;
890 }
891 
892 subsys_initcall(genhd_device_init);
893 
disk_range_show(struct device * dev,struct device_attribute * attr,char * buf)894 static ssize_t disk_range_show(struct device *dev,
895 			       struct device_attribute *attr, char *buf)
896 {
897 	struct gendisk *disk = dev_to_disk(dev);
898 
899 	return sprintf(buf, "%d\n", disk->minors);
900 }
901 
disk_ext_range_show(struct device * dev,struct device_attribute * attr,char * buf)902 static ssize_t disk_ext_range_show(struct device *dev,
903 				   struct device_attribute *attr, char *buf)
904 {
905 	struct gendisk *disk = dev_to_disk(dev);
906 
907 	return sprintf(buf, "%d\n",
908 		(disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
909 }
910 
disk_removable_show(struct device * dev,struct device_attribute * attr,char * buf)911 static ssize_t disk_removable_show(struct device *dev,
912 				   struct device_attribute *attr, char *buf)
913 {
914 	struct gendisk *disk = dev_to_disk(dev);
915 
916 	return sprintf(buf, "%d\n",
917 		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
918 }
919 
disk_hidden_show(struct device * dev,struct device_attribute * attr,char * buf)920 static ssize_t disk_hidden_show(struct device *dev,
921 				   struct device_attribute *attr, char *buf)
922 {
923 	struct gendisk *disk = dev_to_disk(dev);
924 
925 	return sprintf(buf, "%d\n",
926 		       (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
927 }
928 
disk_ro_show(struct device * dev,struct device_attribute * attr,char * buf)929 static ssize_t disk_ro_show(struct device *dev,
930 				   struct device_attribute *attr, char *buf)
931 {
932 	struct gendisk *disk = dev_to_disk(dev);
933 
934 	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
935 }
936 
part_size_show(struct device * dev,struct device_attribute * attr,char * buf)937 ssize_t part_size_show(struct device *dev,
938 		       struct device_attribute *attr, char *buf)
939 {
940 	return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
941 }
942 
part_stat_show(struct device * dev,struct device_attribute * attr,char * buf)943 ssize_t part_stat_show(struct device *dev,
944 		       struct device_attribute *attr, char *buf)
945 {
946 	struct block_device *bdev = dev_to_bdev(dev);
947 	struct request_queue *q = bdev_get_queue(bdev);
948 	struct disk_stats stat;
949 	unsigned int inflight;
950 
951 	if (queue_is_mq(q))
952 		inflight = blk_mq_in_flight(q, bdev);
953 	else
954 		inflight = part_in_flight(bdev);
955 
956 	if (inflight) {
957 		part_stat_lock();
958 		update_io_ticks(bdev, jiffies, true);
959 		part_stat_unlock();
960 	}
961 	part_stat_read_all(bdev, &stat);
962 	return sprintf(buf,
963 		"%8lu %8lu %8llu %8u "
964 		"%8lu %8lu %8llu %8u "
965 		"%8u %8u %8u "
966 		"%8lu %8lu %8llu %8u "
967 		"%8lu %8u"
968 		"\n",
969 		stat.ios[STAT_READ],
970 		stat.merges[STAT_READ],
971 		(unsigned long long)stat.sectors[STAT_READ],
972 		(unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
973 		stat.ios[STAT_WRITE],
974 		stat.merges[STAT_WRITE],
975 		(unsigned long long)stat.sectors[STAT_WRITE],
976 		(unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
977 		inflight,
978 		jiffies_to_msecs(stat.io_ticks),
979 		(unsigned int)div_u64(stat.nsecs[STAT_READ] +
980 				      stat.nsecs[STAT_WRITE] +
981 				      stat.nsecs[STAT_DISCARD] +
982 				      stat.nsecs[STAT_FLUSH],
983 						NSEC_PER_MSEC),
984 		stat.ios[STAT_DISCARD],
985 		stat.merges[STAT_DISCARD],
986 		(unsigned long long)stat.sectors[STAT_DISCARD],
987 		(unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
988 		stat.ios[STAT_FLUSH],
989 		(unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
990 }
991 
part_inflight_show(struct device * dev,struct device_attribute * attr,char * buf)992 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
993 			   char *buf)
994 {
995 	struct block_device *bdev = dev_to_bdev(dev);
996 	struct request_queue *q = bdev_get_queue(bdev);
997 	unsigned int inflight[2];
998 
999 	if (queue_is_mq(q))
1000 		blk_mq_in_flight_rw(q, bdev, inflight);
1001 	else
1002 		part_in_flight_rw(bdev, inflight);
1003 
1004 	return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1005 }
1006 
disk_capability_show(struct device * dev,struct device_attribute * attr,char * buf)1007 static ssize_t disk_capability_show(struct device *dev,
1008 				    struct device_attribute *attr, char *buf)
1009 {
1010 	dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1011 	return sprintf(buf, "0\n");
1012 }
1013 
disk_alignment_offset_show(struct device * dev,struct device_attribute * attr,char * buf)1014 static ssize_t disk_alignment_offset_show(struct device *dev,
1015 					  struct device_attribute *attr,
1016 					  char *buf)
1017 {
1018 	struct gendisk *disk = dev_to_disk(dev);
1019 
1020 	return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1021 }
1022 
disk_discard_alignment_show(struct device * dev,struct device_attribute * attr,char * buf)1023 static ssize_t disk_discard_alignment_show(struct device *dev,
1024 					   struct device_attribute *attr,
1025 					   char *buf)
1026 {
1027 	struct gendisk *disk = dev_to_disk(dev);
1028 
1029 	return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1030 }
1031 
diskseq_show(struct device * dev,struct device_attribute * attr,char * buf)1032 static ssize_t diskseq_show(struct device *dev,
1033 			    struct device_attribute *attr, char *buf)
1034 {
1035 	struct gendisk *disk = dev_to_disk(dev);
1036 
1037 	return sprintf(buf, "%llu\n", disk->diskseq);
1038 }
1039 
partscan_show(struct device * dev,struct device_attribute * attr,char * buf)1040 static ssize_t partscan_show(struct device *dev,
1041 		struct device_attribute *attr, char *buf)
1042 {
1043 	return sprintf(buf, "%u\n", disk_has_partscan(dev_to_disk(dev)));
1044 }
1045 
1046 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1047 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1048 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1049 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1050 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1051 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1052 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1053 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1054 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1055 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1056 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1057 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1058 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1059 static DEVICE_ATTR(partscan, 0444, partscan_show, NULL);
1060 
1061 #ifdef CONFIG_FAIL_MAKE_REQUEST
part_fail_show(struct device * dev,struct device_attribute * attr,char * buf)1062 ssize_t part_fail_show(struct device *dev,
1063 		       struct device_attribute *attr, char *buf)
1064 {
1065 	return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1066 }
1067 
part_fail_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)1068 ssize_t part_fail_store(struct device *dev,
1069 			struct device_attribute *attr,
1070 			const char *buf, size_t count)
1071 {
1072 	int i;
1073 
1074 	if (count > 0 && sscanf(buf, "%d", &i) > 0)
1075 		dev_to_bdev(dev)->bd_make_it_fail = i;
1076 
1077 	return count;
1078 }
1079 
1080 static struct device_attribute dev_attr_fail =
1081 	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1082 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1083 
1084 #ifdef CONFIG_FAIL_IO_TIMEOUT
1085 static struct device_attribute dev_attr_fail_timeout =
1086 	__ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1087 #endif
1088 
1089 static struct attribute *disk_attrs[] = {
1090 	&dev_attr_range.attr,
1091 	&dev_attr_ext_range.attr,
1092 	&dev_attr_removable.attr,
1093 	&dev_attr_hidden.attr,
1094 	&dev_attr_ro.attr,
1095 	&dev_attr_size.attr,
1096 	&dev_attr_alignment_offset.attr,
1097 	&dev_attr_discard_alignment.attr,
1098 	&dev_attr_capability.attr,
1099 	&dev_attr_stat.attr,
1100 	&dev_attr_inflight.attr,
1101 	&dev_attr_badblocks.attr,
1102 	&dev_attr_events.attr,
1103 	&dev_attr_events_async.attr,
1104 	&dev_attr_events_poll_msecs.attr,
1105 	&dev_attr_diskseq.attr,
1106 	&dev_attr_partscan.attr,
1107 #ifdef CONFIG_FAIL_MAKE_REQUEST
1108 	&dev_attr_fail.attr,
1109 #endif
1110 #ifdef CONFIG_FAIL_IO_TIMEOUT
1111 	&dev_attr_fail_timeout.attr,
1112 #endif
1113 	NULL
1114 };
1115 
disk_visible(struct kobject * kobj,struct attribute * a,int n)1116 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1117 {
1118 	struct device *dev = container_of(kobj, typeof(*dev), kobj);
1119 	struct gendisk *disk = dev_to_disk(dev);
1120 
1121 	if (a == &dev_attr_badblocks.attr && !disk->bb)
1122 		return 0;
1123 	return a->mode;
1124 }
1125 
1126 static struct attribute_group disk_attr_group = {
1127 	.attrs = disk_attrs,
1128 	.is_visible = disk_visible,
1129 };
1130 
1131 static const struct attribute_group *disk_attr_groups[] = {
1132 	&disk_attr_group,
1133 #ifdef CONFIG_BLK_DEV_IO_TRACE
1134 	&blk_trace_attr_group,
1135 #endif
1136 #ifdef CONFIG_BLK_DEV_INTEGRITY
1137 	&blk_integrity_attr_group,
1138 #endif
1139 	NULL
1140 };
1141 
1142 /**
1143  * disk_release - releases all allocated resources of the gendisk
1144  * @dev: the device representing this disk
1145  *
1146  * This function releases all allocated resources of the gendisk.
1147  *
1148  * Drivers which used __device_add_disk() have a gendisk with a request_queue
1149  * assigned. Since the request_queue sits on top of the gendisk for these
1150  * drivers we also call blk_put_queue() for them, and we expect the
1151  * request_queue refcount to reach 0 at this point, and so the request_queue
1152  * will also be freed prior to the disk.
1153  *
1154  * Context: can sleep
1155  */
disk_release(struct device * dev)1156 static void disk_release(struct device *dev)
1157 {
1158 	struct gendisk *disk = dev_to_disk(dev);
1159 
1160 	might_sleep();
1161 	WARN_ON_ONCE(disk_live(disk));
1162 
1163 	blk_trace_remove(disk->queue);
1164 
1165 	/*
1166 	 * To undo the all initialization from blk_mq_init_allocated_queue in
1167 	 * case of a probe failure where add_disk is never called we have to
1168 	 * call blk_mq_exit_queue here. We can't do this for the more common
1169 	 * teardown case (yet) as the tagset can be gone by the time the disk
1170 	 * is released once it was added.
1171 	 */
1172 	if (queue_is_mq(disk->queue) &&
1173 	    test_bit(GD_OWNS_QUEUE, &disk->state) &&
1174 	    !test_bit(GD_ADDED, &disk->state))
1175 		blk_mq_exit_queue(disk->queue);
1176 
1177 	blkcg_exit_disk(disk);
1178 
1179 	bioset_exit(&disk->bio_split);
1180 
1181 	disk_release_events(disk);
1182 	kfree(disk->random);
1183 	disk_free_zone_bitmaps(disk);
1184 	xa_destroy(&disk->part_tbl);
1185 
1186 	disk->queue->disk = NULL;
1187 	blk_put_queue(disk->queue);
1188 
1189 	if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1190 		disk->fops->free_disk(disk);
1191 
1192 	iput(disk->part0->bd_inode);	/* frees the disk */
1193 }
1194 
block_uevent(const struct device * dev,struct kobj_uevent_env * env)1195 static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1196 {
1197 	const struct gendisk *disk = dev_to_disk(dev);
1198 
1199 	return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1200 }
1201 
1202 struct class block_class = {
1203 	.name		= "block",
1204 	.dev_uevent	= block_uevent,
1205 };
1206 
block_devnode(const struct device * dev,umode_t * mode,kuid_t * uid,kgid_t * gid)1207 static char *block_devnode(const struct device *dev, umode_t *mode,
1208 			   kuid_t *uid, kgid_t *gid)
1209 {
1210 	struct gendisk *disk = dev_to_disk(dev);
1211 
1212 	if (disk->fops->devnode)
1213 		return disk->fops->devnode(disk, mode);
1214 	return NULL;
1215 }
1216 
1217 const struct device_type disk_type = {
1218 	.name		= "disk",
1219 	.groups		= disk_attr_groups,
1220 	.release	= disk_release,
1221 	.devnode	= block_devnode,
1222 };
1223 
1224 #ifdef CONFIG_PROC_FS
1225 /*
1226  * aggregate disk stat collector.  Uses the same stats that the sysfs
1227  * entries do, above, but makes them available through one seq_file.
1228  *
1229  * The output looks suspiciously like /proc/partitions with a bunch of
1230  * extra fields.
1231  */
diskstats_show(struct seq_file * seqf,void * v)1232 static int diskstats_show(struct seq_file *seqf, void *v)
1233 {
1234 	struct gendisk *gp = v;
1235 	struct block_device *hd;
1236 	unsigned int inflight;
1237 	struct disk_stats stat;
1238 	unsigned long idx;
1239 
1240 	/*
1241 	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1242 		seq_puts(seqf,	"major minor name"
1243 				"     rio rmerge rsect ruse wio wmerge "
1244 				"wsect wuse running use aveq"
1245 				"\n\n");
1246 	*/
1247 
1248 	rcu_read_lock();
1249 	xa_for_each(&gp->part_tbl, idx, hd) {
1250 		if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1251 			continue;
1252 		if (queue_is_mq(gp->queue))
1253 			inflight = blk_mq_in_flight(gp->queue, hd);
1254 		else
1255 			inflight = part_in_flight(hd);
1256 
1257 		if (inflight) {
1258 			part_stat_lock();
1259 			update_io_ticks(hd, jiffies, true);
1260 			part_stat_unlock();
1261 		}
1262 		part_stat_read_all(hd, &stat);
1263 		seq_printf(seqf, "%4d %7d %pg "
1264 			   "%lu %lu %lu %u "
1265 			   "%lu %lu %lu %u "
1266 			   "%u %u %u "
1267 			   "%lu %lu %lu %u "
1268 			   "%lu %u"
1269 			   "\n",
1270 			   MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1271 			   stat.ios[STAT_READ],
1272 			   stat.merges[STAT_READ],
1273 			   stat.sectors[STAT_READ],
1274 			   (unsigned int)div_u64(stat.nsecs[STAT_READ],
1275 							NSEC_PER_MSEC),
1276 			   stat.ios[STAT_WRITE],
1277 			   stat.merges[STAT_WRITE],
1278 			   stat.sectors[STAT_WRITE],
1279 			   (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1280 							NSEC_PER_MSEC),
1281 			   inflight,
1282 			   jiffies_to_msecs(stat.io_ticks),
1283 			   (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1284 						 stat.nsecs[STAT_WRITE] +
1285 						 stat.nsecs[STAT_DISCARD] +
1286 						 stat.nsecs[STAT_FLUSH],
1287 							NSEC_PER_MSEC),
1288 			   stat.ios[STAT_DISCARD],
1289 			   stat.merges[STAT_DISCARD],
1290 			   stat.sectors[STAT_DISCARD],
1291 			   (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1292 						 NSEC_PER_MSEC),
1293 			   stat.ios[STAT_FLUSH],
1294 			   (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1295 						 NSEC_PER_MSEC)
1296 			);
1297 	}
1298 	rcu_read_unlock();
1299 
1300 	return 0;
1301 }
1302 
1303 static const struct seq_operations diskstats_op = {
1304 	.start	= disk_seqf_start,
1305 	.next	= disk_seqf_next,
1306 	.stop	= disk_seqf_stop,
1307 	.show	= diskstats_show
1308 };
1309 
proc_genhd_init(void)1310 static int __init proc_genhd_init(void)
1311 {
1312 	proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1313 	proc_create_seq("partitions", 0, NULL, &partitions_op);
1314 	return 0;
1315 }
1316 module_init(proc_genhd_init);
1317 #endif /* CONFIG_PROC_FS */
1318 
part_devt(struct gendisk * disk,u8 partno)1319 dev_t part_devt(struct gendisk *disk, u8 partno)
1320 {
1321 	struct block_device *part;
1322 	dev_t devt = 0;
1323 
1324 	rcu_read_lock();
1325 	part = xa_load(&disk->part_tbl, partno);
1326 	if (part)
1327 		devt = part->bd_dev;
1328 	rcu_read_unlock();
1329 
1330 	return devt;
1331 }
1332 
__alloc_disk_node(struct request_queue * q,int node_id,struct lock_class_key * lkclass)1333 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1334 		struct lock_class_key *lkclass)
1335 {
1336 	struct gendisk *disk;
1337 
1338 	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1339 	if (!disk)
1340 		return NULL;
1341 
1342 	if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1343 		goto out_free_disk;
1344 
1345 	disk->bdi = bdi_alloc(node_id);
1346 	if (!disk->bdi)
1347 		goto out_free_bioset;
1348 
1349 	/* bdev_alloc() might need the queue, set before the first call */
1350 	disk->queue = q;
1351 
1352 	disk->part0 = bdev_alloc(disk, 0);
1353 	if (!disk->part0)
1354 		goto out_free_bdi;
1355 
1356 	disk->node_id = node_id;
1357 	mutex_init(&disk->open_mutex);
1358 	xa_init(&disk->part_tbl);
1359 	if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1360 		goto out_destroy_part_tbl;
1361 
1362 	if (blkcg_init_disk(disk))
1363 		goto out_erase_part0;
1364 
1365 	rand_initialize_disk(disk);
1366 	disk_to_dev(disk)->class = &block_class;
1367 	disk_to_dev(disk)->type = &disk_type;
1368 	device_initialize(disk_to_dev(disk));
1369 	inc_diskseq(disk);
1370 	q->disk = disk;
1371 	lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1372 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1373 	INIT_LIST_HEAD(&disk->slave_bdevs);
1374 #endif
1375 	return disk;
1376 
1377 out_erase_part0:
1378 	xa_erase(&disk->part_tbl, 0);
1379 out_destroy_part_tbl:
1380 	xa_destroy(&disk->part_tbl);
1381 	disk->part0->bd_disk = NULL;
1382 	iput(disk->part0->bd_inode);
1383 out_free_bdi:
1384 	bdi_put(disk->bdi);
1385 out_free_bioset:
1386 	bioset_exit(&disk->bio_split);
1387 out_free_disk:
1388 	kfree(disk);
1389 	return NULL;
1390 }
1391 
__blk_alloc_disk(int node,struct lock_class_key * lkclass)1392 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1393 {
1394 	struct request_queue *q;
1395 	struct gendisk *disk;
1396 
1397 	q = blk_alloc_queue(node);
1398 	if (!q)
1399 		return NULL;
1400 
1401 	disk = __alloc_disk_node(q, node, lkclass);
1402 	if (!disk) {
1403 		blk_put_queue(q);
1404 		return NULL;
1405 	}
1406 	set_bit(GD_OWNS_QUEUE, &disk->state);
1407 	return disk;
1408 }
1409 EXPORT_SYMBOL(__blk_alloc_disk);
1410 
1411 /**
1412  * put_disk - decrements the gendisk refcount
1413  * @disk: the struct gendisk to decrement the refcount for
1414  *
1415  * This decrements the refcount for the struct gendisk. When this reaches 0
1416  * we'll have disk_release() called.
1417  *
1418  * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1419  * when handling probe errors (that is before add_disk() is called).
1420  *
1421  * Context: Any context, but the last reference must not be dropped from
1422  *          atomic context.
1423  */
put_disk(struct gendisk * disk)1424 void put_disk(struct gendisk *disk)
1425 {
1426 	if (disk)
1427 		put_device(disk_to_dev(disk));
1428 }
1429 EXPORT_SYMBOL(put_disk);
1430 
set_disk_ro_uevent(struct gendisk * gd,int ro)1431 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1432 {
1433 	char event[] = "DISK_RO=1";
1434 	char *envp[] = { event, NULL };
1435 
1436 	if (!ro)
1437 		event[8] = '0';
1438 	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1439 }
1440 
1441 /**
1442  * set_disk_ro - set a gendisk read-only
1443  * @disk:	gendisk to operate on
1444  * @read_only:	%true to set the disk read-only, %false set the disk read/write
1445  *
1446  * This function is used to indicate whether a given disk device should have its
1447  * read-only flag set. set_disk_ro() is typically used by device drivers to
1448  * indicate whether the underlying physical device is write-protected.
1449  */
set_disk_ro(struct gendisk * disk,bool read_only)1450 void set_disk_ro(struct gendisk *disk, bool read_only)
1451 {
1452 	if (read_only) {
1453 		if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1454 			return;
1455 	} else {
1456 		if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1457 			return;
1458 	}
1459 	set_disk_ro_uevent(disk, read_only);
1460 }
1461 EXPORT_SYMBOL(set_disk_ro);
1462 
inc_diskseq(struct gendisk * disk)1463 void inc_diskseq(struct gendisk *disk)
1464 {
1465 	disk->diskseq = atomic64_inc_return(&diskseq);
1466 }
1467