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 static 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->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(disk->part0, disk_scan_partitions,
363 					  NULL);
364 		if (ret)
365 			return ret;
366 	}
367 
368 	set_bit(GD_NEED_PART_SCAN, &disk->state);
369 	bdev = blkdev_get_by_dev(disk_devt(disk), mode & ~BLK_OPEN_EXCL, NULL,
370 				 NULL);
371 	if (IS_ERR(bdev))
372 		ret =  PTR_ERR(bdev);
373 	else
374 		blkdev_put(bdev, NULL);
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, &disk->state);
382 	if (!(mode & BLK_OPEN_EXCL))
383 		bd_abort_claiming(disk->part0, 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  */
device_add_disk(struct device * parent,struct gendisk * disk,const struct attribute_group ** groups)396 int __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(disk->queue) && disk->fops->poll_bio)
405 		return -EINVAL;
406 
407 	/*
408 	 * The disk queue should now be all set with enough information about
409 	 * the device for the elevator code to pick an adequate default
410 	 * elevator if one is needed, that is, for devices requesting queue
411 	 * registration.
412 	 */
413 	elevator_init_mq(disk->queue);
414 
415 	/* Mark bdev as having a submit_bio, if needed */
416 	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(ddev, 1);
452 
453 	ddev->parent = parent;
454 	ddev->groups = groups;
455 	dev_set_name(ddev, "%s", disk->disk_name);
456 	if (!(disk->flags & GENHD_FL_HIDDEN))
457 		ddev->devt = MKDEV(disk->major, disk->first_minor);
458 	ret = device_add(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(block_depr, &ddev->kobj,
467 				kobject_name(&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(ddev, true);
477 
478 	disk->part0->bd_holder_dir =
479 		kobject_create_and_add("holders", &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("slaves", &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(disk->bdi, "%u:%u",
496 				   disk->major, disk->first_minor);
497 		if (ret)
498 			goto out_unregister_queue;
499 		bdi_set_owner(disk->bdi, ddev);
500 		ret = sysfs_create_link(&ddev->kobj,
501 					&disk->bdi->dev->kobj, "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, &disk->state);
509 
510 		bdev_add(disk->part0, 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(ddev, 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, &disk->state);
532 	return 0;
533 
534 out_unregister_bdi:
535 	if (!(disk->flags & GENHD_FL_HIDDEN))
536 		bdi_unregister(disk->bdi);
537 out_unregister_queue:
538 	blk_unregister_queue(disk);
539 	rq_qos_exit(disk->queue);
540 out_put_slave_dir:
541 	kobject_put(disk->slave_dir);
542 	disk->slave_dir = NULL;
543 out_put_holder_dir:
544 	kobject_put(disk->part0->bd_holder_dir);
545 out_del_block_link:
546 	sysfs_remove_link(block_depr, dev_name(ddev));
547 	pm_runtime_set_memalloc_noio(ddev, false);
548 out_device_del:
549 	device_del(ddev);
550 out_free_ext_minor:
551 	if (disk->major == BLOCK_EXT_MAJOR)
552 		blk_free_ext_minor(disk->first_minor);
553 out_exit_elevator:
554 	if (disk->queue->elevator)
555 		elevator_exit(disk->queue);
556 	return ret;
557 }
558 EXPORT_SYMBOL(device_add_disk);
559 
blk_report_disk_dead(struct gendisk * disk,bool surprise)560 static void blk_report_disk_dead(struct gendisk *disk, bool surprise)
561 {
562 	struct block_device *bdev;
563 	unsigned long idx;
564 
565 	rcu_read_lock();
566 	xa_for_each(&disk->part_tbl, idx, bdev) {
567 		if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
568 			continue;
569 		rcu_read_unlock();
570 
571 		bdev_mark_dead(bdev, surprise);
572 
573 		put_device(&bdev->bd_device);
574 		rcu_read_lock();
575 	}
576 	rcu_read_unlock();
577 }
578 
__blk_mark_disk_dead(struct gendisk * disk)579 static void __blk_mark_disk_dead(struct gendisk *disk)
580 {
581 	/*
582 	 * Fail any new I/O.
583 	 */
584 	if (test_and_set_bit(GD_DEAD, &disk->state))
585 		return;
586 
587 	if (test_bit(GD_OWNS_QUEUE, &disk->state))
588 		blk_queue_flag_set(QUEUE_FLAG_DYING, disk->queue);
589 
590 	/*
591 	 * Stop buffered writers from dirtying pages that can't be written out.
592 	 */
593 	set_capacity(disk, 0);
594 
595 	/*
596 	 * Prevent new I/O from crossing bio_queue_enter().
597 	 */
598 	blk_queue_start_drain(disk->queue);
599 }
600 
601 /**
602  * blk_mark_disk_dead - mark a disk as dead
603  * @disk: disk to mark as dead
604  *
605  * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
606  * to this disk.
607  */
blk_mark_disk_dead(struct gendisk * disk)608 void blk_mark_disk_dead(struct gendisk *disk)
609 {
610 	__blk_mark_disk_dead(disk);
611 	blk_report_disk_dead(disk, true);
612 }
613 EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
614 
615 /**
616  * del_gendisk - remove the gendisk
617  * @disk: the struct gendisk to remove
618  *
619  * Removes the gendisk and all its associated resources. This deletes the
620  * partitions associated with the gendisk, and unregisters the associated
621  * request_queue.
622  *
623  * This is the counter to the respective __device_add_disk() call.
624  *
625  * The final removal of the struct gendisk happens when its refcount reaches 0
626  * with put_disk(), which should be called after del_gendisk(), if
627  * __device_add_disk() was used.
628  *
629  * Drivers exist which depend on the release of the gendisk to be synchronous,
630  * it should not be deferred.
631  *
632  * Context: can sleep
633  */
del_gendisk(struct gendisk * disk)634 void del_gendisk(struct gendisk *disk)
635 {
636 	struct request_queue *q = disk->queue;
637 	struct block_device *part;
638 	unsigned long idx;
639 
640 	might_sleep();
641 
642 	if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
643 		return;
644 
645 	disk_del_events(disk);
646 
647 	/*
648 	 * Prevent new openers by unlinked the bdev inode.
649 	 */
650 	mutex_lock(&disk->open_mutex);
651 	xa_for_each(&disk->part_tbl, idx, part)
652 		remove_inode_hash(part->bd_inode);
653 	mutex_unlock(&disk->open_mutex);
654 
655 	/*
656 	 * Tell the file system to write back all dirty data and shut down if
657 	 * it hasn't been notified earlier.
658 	 */
659 	if (!test_bit(GD_DEAD, &disk->state))
660 		blk_report_disk_dead(disk, false);
661 	__blk_mark_disk_dead(disk);
662 
663 	/*
664 	 * Drop all partitions now that the disk is marked dead.
665 	 */
666 	mutex_lock(&disk->open_mutex);
667 	xa_for_each_start(&disk->part_tbl, idx, part, 1)
668 		drop_partition(part);
669 	mutex_unlock(&disk->open_mutex);
670 
671 	if (!(disk->flags & GENHD_FL_HIDDEN)) {
672 		sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
673 
674 		/*
675 		 * Unregister bdi before releasing device numbers (as they can
676 		 * get reused and we'd get clashes in sysfs).
677 		 */
678 		bdi_unregister(disk->bdi);
679 	}
680 
681 	blk_unregister_queue(disk);
682 
683 	kobject_put(disk->part0->bd_holder_dir);
684 	kobject_put(disk->slave_dir);
685 	disk->slave_dir = NULL;
686 
687 	part_stat_set_all(disk->part0, 0);
688 	disk->part0->bd_stamp = 0;
689 	sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
690 	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
691 	device_del(disk_to_dev(disk));
692 
693 	blk_mq_freeze_queue_wait(q);
694 
695 	blk_throtl_cancel_bios(disk);
696 
697 	blk_sync_queue(q);
698 	blk_flush_integrity();
699 
700 	if (queue_is_mq(q))
701 		blk_mq_cancel_work_sync(q);
702 
703 	blk_mq_quiesce_queue(q);
704 	if (q->elevator) {
705 		mutex_lock(&q->sysfs_lock);
706 		elevator_exit(q);
707 		mutex_unlock(&q->sysfs_lock);
708 	}
709 	rq_qos_exit(q);
710 	blk_mq_unquiesce_queue(q);
711 
712 	/*
713 	 * If the disk does not own the queue, allow using passthrough requests
714 	 * again.  Else leave the queue frozen to fail all I/O.
715 	 */
716 	if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
717 		blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
718 		__blk_mq_unfreeze_queue(q, true);
719 	} else {
720 		if (queue_is_mq(q))
721 			blk_mq_exit_queue(q);
722 	}
723 }
724 EXPORT_SYMBOL(del_gendisk);
725 
726 /**
727  * invalidate_disk - invalidate the disk
728  * @disk: the struct gendisk to invalidate
729  *
730  * A helper to invalidates the disk. It will clean the disk's associated
731  * buffer/page caches and reset its internal states so that the disk
732  * can be reused by the drivers.
733  *
734  * Context: can sleep
735  */
invalidate_disk(struct gendisk * disk)736 void invalidate_disk(struct gendisk *disk)
737 {
738 	struct block_device *bdev = disk->part0;
739 
740 	invalidate_bdev(bdev);
741 	bdev->bd_inode->i_mapping->wb_err = 0;
742 	set_capacity(disk, 0);
743 }
744 EXPORT_SYMBOL(invalidate_disk);
745 
746 /* sysfs access to bad-blocks list. */
disk_badblocks_show(struct device * dev,struct device_attribute * attr,char * page)747 static ssize_t disk_badblocks_show(struct device *dev,
748 					struct device_attribute *attr,
749 					char *page)
750 {
751 	struct gendisk *disk = dev_to_disk(dev);
752 
753 	if (!disk->bb)
754 		return sprintf(page, "\n");
755 
756 	return badblocks_show(disk->bb, page, 0);
757 }
758 
disk_badblocks_store(struct device * dev,struct device_attribute * attr,const char * page,size_t len)759 static ssize_t disk_badblocks_store(struct device *dev,
760 					struct device_attribute *attr,
761 					const char *page, size_t len)
762 {
763 	struct gendisk *disk = dev_to_disk(dev);
764 
765 	if (!disk->bb)
766 		return -ENXIO;
767 
768 	return badblocks_store(disk->bb, page, len, 0);
769 }
770 
771 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
blk_request_module(dev_t devt)772 void blk_request_module(dev_t devt)
773 {
774 	unsigned int major = MAJOR(devt);
775 	struct blk_major_name **n;
776 
777 	mutex_lock(&major_names_lock);
778 	for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
779 		if ((*n)->major == major && (*n)->probe) {
780 			(*n)->probe(devt);
781 			mutex_unlock(&major_names_lock);
782 			return;
783 		}
784 	}
785 	mutex_unlock(&major_names_lock);
786 
787 	if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
788 		/* Make old-style 2.4 aliases work */
789 		request_module("block-major-%d", MAJOR(devt));
790 }
791 #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
792 
793 #ifdef CONFIG_PROC_FS
794 /* iterator */
disk_seqf_start(struct seq_file * seqf,loff_t * pos)795 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
796 {
797 	loff_t skip = *pos;
798 	struct class_dev_iter *iter;
799 	struct device *dev;
800 
801 	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
802 	if (!iter)
803 		return ERR_PTR(-ENOMEM);
804 
805 	seqf->private = iter;
806 	class_dev_iter_init(iter, &block_class, NULL, &disk_type);
807 	do {
808 		dev = class_dev_iter_next(iter);
809 		if (!dev)
810 			return NULL;
811 	} while (skip--);
812 
813 	return dev_to_disk(dev);
814 }
815 
disk_seqf_next(struct seq_file * seqf,void * v,loff_t * pos)816 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
817 {
818 	struct device *dev;
819 
820 	(*pos)++;
821 	dev = class_dev_iter_next(seqf->private);
822 	if (dev)
823 		return dev_to_disk(dev);
824 
825 	return NULL;
826 }
827 
disk_seqf_stop(struct seq_file * seqf,void * v)828 static void disk_seqf_stop(struct seq_file *seqf, void *v)
829 {
830 	struct class_dev_iter *iter = seqf->private;
831 
832 	/* stop is called even after start failed :-( */
833 	if (iter) {
834 		class_dev_iter_exit(iter);
835 		kfree(iter);
836 		seqf->private = NULL;
837 	}
838 }
839 
show_partition_start(struct seq_file * seqf,loff_t * pos)840 static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
841 {
842 	void *p;
843 
844 	p = disk_seqf_start(seqf, pos);
845 	if (!IS_ERR_OR_NULL(p) && !*pos)
846 		seq_puts(seqf, "major minor  #blocks  name\n\n");
847 	return p;
848 }
849 
show_partition(struct seq_file * seqf,void * v)850 static int show_partition(struct seq_file *seqf, void *v)
851 {
852 	struct gendisk *sgp = v;
853 	struct block_device *part;
854 	unsigned long idx;
855 
856 	if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
857 		return 0;
858 
859 	rcu_read_lock();
860 	xa_for_each(&sgp->part_tbl, idx, part) {
861 		if (!bdev_nr_sectors(part))
862 			continue;
863 		seq_printf(seqf, "%4d  %7d %10llu %pg\n",
864 			   MAJOR(part->bd_dev), MINOR(part->bd_dev),
865 			   bdev_nr_sectors(part) >> 1, part);
866 	}
867 	rcu_read_unlock();
868 	return 0;
869 }
870 
871 static const struct seq_operations partitions_op = {
872 	.start	= show_partition_start,
873 	.next	= disk_seqf_next,
874 	.stop	= disk_seqf_stop,
875 	.show	= show_partition
876 };
877 #endif
878 
genhd_device_init(void)879 static int __init genhd_device_init(void)
880 {
881 	int error;
882 
883 	error = class_register(&block_class);
884 	if (unlikely(error))
885 		return error;
886 	blk_dev_init();
887 
888 	register_blkdev(BLOCK_EXT_MAJOR, "blkext");
889 
890 	/* create top-level block dir */
891 	block_depr = kobject_create_and_add("block", NULL);
892 	return 0;
893 }
894 
895 subsys_initcall(genhd_device_init);
896 
disk_range_show(struct device * dev,struct device_attribute * attr,char * buf)897 static ssize_t disk_range_show(struct device *dev,
898 			       struct device_attribute *attr, char *buf)
899 {
900 	struct gendisk *disk = dev_to_disk(dev);
901 
902 	return sprintf(buf, "%d\n", disk->minors);
903 }
904 
disk_ext_range_show(struct device * dev,struct device_attribute * attr,char * buf)905 static ssize_t disk_ext_range_show(struct device *dev,
906 				   struct device_attribute *attr, char *buf)
907 {
908 	struct gendisk *disk = dev_to_disk(dev);
909 
910 	return sprintf(buf, "%d\n",
911 		(disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
912 }
913 
disk_removable_show(struct device * dev,struct device_attribute * attr,char * buf)914 static ssize_t disk_removable_show(struct device *dev,
915 				   struct device_attribute *attr, char *buf)
916 {
917 	struct gendisk *disk = dev_to_disk(dev);
918 
919 	return sprintf(buf, "%d\n",
920 		       (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
921 }
922 
disk_hidden_show(struct device * dev,struct device_attribute * attr,char * buf)923 static ssize_t disk_hidden_show(struct device *dev,
924 				   struct device_attribute *attr, char *buf)
925 {
926 	struct gendisk *disk = dev_to_disk(dev);
927 
928 	return sprintf(buf, "%d\n",
929 		       (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
930 }
931 
disk_ro_show(struct device * dev,struct device_attribute * attr,char * buf)932 static ssize_t disk_ro_show(struct device *dev,
933 				   struct device_attribute *attr, char *buf)
934 {
935 	struct gendisk *disk = dev_to_disk(dev);
936 
937 	return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
938 }
939 
part_size_show(struct device * dev,struct device_attribute * attr,char * buf)940 ssize_t part_size_show(struct device *dev,
941 		       struct device_attribute *attr, char *buf)
942 {
943 	return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
944 }
945 
part_stat_show(struct device * dev,struct device_attribute * attr,char * buf)946 ssize_t part_stat_show(struct device *dev,
947 		       struct device_attribute *attr, char *buf)
948 {
949 	struct block_device *bdev = dev_to_bdev(dev);
950 	struct request_queue *q = bdev_get_queue(bdev);
951 	struct disk_stats stat;
952 	unsigned int inflight;
953 
954 	if (queue_is_mq(q))
955 		inflight = blk_mq_in_flight(q, bdev);
956 	else
957 		inflight = part_in_flight(bdev);
958 
959 	if (inflight) {
960 		part_stat_lock();
961 		update_io_ticks(bdev, jiffies, true);
962 		part_stat_unlock();
963 	}
964 	part_stat_read_all(bdev, &stat);
965 	return sprintf(buf,
966 		"%8lu %8lu %8llu %8u "
967 		"%8lu %8lu %8llu %8u "
968 		"%8u %8u %8u "
969 		"%8lu %8lu %8llu %8u "
970 		"%8lu %8u"
971 		"\n",
972 		stat.ios[STAT_READ],
973 		stat.merges[STAT_READ],
974 		(unsigned long long)stat.sectors[STAT_READ],
975 		(unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
976 		stat.ios[STAT_WRITE],
977 		stat.merges[STAT_WRITE],
978 		(unsigned long long)stat.sectors[STAT_WRITE],
979 		(unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
980 		inflight,
981 		jiffies_to_msecs(stat.io_ticks),
982 		(unsigned int)div_u64(stat.nsecs[STAT_READ] +
983 				      stat.nsecs[STAT_WRITE] +
984 				      stat.nsecs[STAT_DISCARD] +
985 				      stat.nsecs[STAT_FLUSH],
986 						NSEC_PER_MSEC),
987 		stat.ios[STAT_DISCARD],
988 		stat.merges[STAT_DISCARD],
989 		(unsigned long long)stat.sectors[STAT_DISCARD],
990 		(unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
991 		stat.ios[STAT_FLUSH],
992 		(unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
993 }
994 
part_inflight_show(struct device * dev,struct device_attribute * attr,char * buf)995 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
996 			   char *buf)
997 {
998 	struct block_device *bdev = dev_to_bdev(dev);
999 	struct request_queue *q = bdev_get_queue(bdev);
1000 	unsigned int inflight[2];
1001 
1002 	if (queue_is_mq(q))
1003 		blk_mq_in_flight_rw(q, bdev, inflight);
1004 	else
1005 		part_in_flight_rw(bdev, inflight);
1006 
1007 	return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1008 }
1009 
disk_capability_show(struct device * dev,struct device_attribute * attr,char * buf)1010 static ssize_t disk_capability_show(struct device *dev,
1011 				    struct device_attribute *attr, char *buf)
1012 {
1013 	dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1014 	return sprintf(buf, "0\n");
1015 }
1016 
disk_alignment_offset_show(struct device * dev,struct device_attribute * attr,char * buf)1017 static ssize_t disk_alignment_offset_show(struct device *dev,
1018 					  struct device_attribute *attr,
1019 					  char *buf)
1020 {
1021 	struct gendisk *disk = dev_to_disk(dev);
1022 
1023 	return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1024 }
1025 
disk_discard_alignment_show(struct device * dev,struct device_attribute * attr,char * buf)1026 static ssize_t disk_discard_alignment_show(struct device *dev,
1027 					   struct device_attribute *attr,
1028 					   char *buf)
1029 {
1030 	struct gendisk *disk = dev_to_disk(dev);
1031 
1032 	return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1033 }
1034 
diskseq_show(struct device * dev,struct device_attribute * attr,char * buf)1035 static ssize_t diskseq_show(struct device *dev,
1036 			    struct device_attribute *attr, char *buf)
1037 {
1038 	struct gendisk *disk = dev_to_disk(dev);
1039 
1040 	return sprintf(buf, "%llu\n", disk->diskseq);
1041 }
1042 
1043 static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1044 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1045 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1046 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1047 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1048 static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1049 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1050 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1051 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1052 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1053 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1054 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1055 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1056 
1057 #ifdef CONFIG_FAIL_MAKE_REQUEST
part_fail_show(struct device * dev,struct device_attribute * attr,char * buf)1058 ssize_t part_fail_show(struct device *dev,
1059 		       struct device_attribute *attr, char *buf)
1060 {
1061 	return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1062 }
1063 
part_fail_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)1064 ssize_t part_fail_store(struct device *dev,
1065 			struct device_attribute *attr,
1066 			const char *buf, size_t count)
1067 {
1068 	int i;
1069 
1070 	if (count > 0 && sscanf(buf, "%d", &i) > 0)
1071 		dev_to_bdev(dev)->bd_make_it_fail = i;
1072 
1073 	return count;
1074 }
1075 
1076 static struct device_attribute dev_attr_fail =
1077 	__ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1078 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1079 
1080 #ifdef CONFIG_FAIL_IO_TIMEOUT
1081 static struct device_attribute dev_attr_fail_timeout =
1082 	__ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1083 #endif
1084 
1085 static struct attribute *disk_attrs[] = {
1086 	&dev_attr_range.attr,
1087 	&dev_attr_ext_range.attr,
1088 	&dev_attr_removable.attr,
1089 	&dev_attr_hidden.attr,
1090 	&dev_attr_ro.attr,
1091 	&dev_attr_size.attr,
1092 	&dev_attr_alignment_offset.attr,
1093 	&dev_attr_discard_alignment.attr,
1094 	&dev_attr_capability.attr,
1095 	&dev_attr_stat.attr,
1096 	&dev_attr_inflight.attr,
1097 	&dev_attr_badblocks.attr,
1098 	&dev_attr_events.attr,
1099 	&dev_attr_events_async.attr,
1100 	&dev_attr_events_poll_msecs.attr,
1101 	&dev_attr_diskseq.attr,
1102 #ifdef CONFIG_FAIL_MAKE_REQUEST
1103 	&dev_attr_fail.attr,
1104 #endif
1105 #ifdef CONFIG_FAIL_IO_TIMEOUT
1106 	&dev_attr_fail_timeout.attr,
1107 #endif
1108 	NULL
1109 };
1110 
disk_visible(struct kobject * kobj,struct attribute * a,int n)1111 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1112 {
1113 	struct device *dev = container_of(kobj, typeof(*dev), kobj);
1114 	struct gendisk *disk = dev_to_disk(dev);
1115 
1116 	if (a == &dev_attr_badblocks.attr && !disk->bb)
1117 		return 0;
1118 	return a->mode;
1119 }
1120 
1121 static struct attribute_group disk_attr_group = {
1122 	.attrs = disk_attrs,
1123 	.is_visible = disk_visible,
1124 };
1125 
1126 static const struct attribute_group *disk_attr_groups[] = {
1127 	&disk_attr_group,
1128 #ifdef CONFIG_BLK_DEV_IO_TRACE
1129 	&blk_trace_attr_group,
1130 #endif
1131 #ifdef CONFIG_BLK_DEV_INTEGRITY
1132 	&blk_integrity_attr_group,
1133 #endif
1134 	NULL
1135 };
1136 
1137 /**
1138  * disk_release - releases all allocated resources of the gendisk
1139  * @dev: the device representing this disk
1140  *
1141  * This function releases all allocated resources of the gendisk.
1142  *
1143  * Drivers which used __device_add_disk() have a gendisk with a request_queue
1144  * assigned. Since the request_queue sits on top of the gendisk for these
1145  * drivers we also call blk_put_queue() for them, and we expect the
1146  * request_queue refcount to reach 0 at this point, and so the request_queue
1147  * will also be freed prior to the disk.
1148  *
1149  * Context: can sleep
1150  */
disk_release(struct device * dev)1151 static void disk_release(struct device *dev)
1152 {
1153 	struct gendisk *disk = dev_to_disk(dev);
1154 
1155 	might_sleep();
1156 	WARN_ON_ONCE(disk_live(disk));
1157 
1158 	blk_trace_remove(disk->queue);
1159 
1160 	/*
1161 	 * To undo the all initialization from blk_mq_init_allocated_queue in
1162 	 * case of a probe failure where add_disk is never called we have to
1163 	 * call blk_mq_exit_queue here. We can't do this for the more common
1164 	 * teardown case (yet) as the tagset can be gone by the time the disk
1165 	 * is released once it was added.
1166 	 */
1167 	if (queue_is_mq(disk->queue) &&
1168 	    test_bit(GD_OWNS_QUEUE, &disk->state) &&
1169 	    !test_bit(GD_ADDED, &disk->state))
1170 		blk_mq_exit_queue(disk->queue);
1171 
1172 	blkcg_exit_disk(disk);
1173 
1174 	bioset_exit(&disk->bio_split);
1175 
1176 	disk_release_events(disk);
1177 	kfree(disk->random);
1178 	disk_free_zone_bitmaps(disk);
1179 	xa_destroy(&disk->part_tbl);
1180 
1181 	disk->queue->disk = NULL;
1182 	blk_put_queue(disk->queue);
1183 
1184 	if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1185 		disk->fops->free_disk(disk);
1186 
1187 	iput(disk->part0->bd_inode);	/* frees the disk */
1188 }
1189 
block_uevent(const struct device * dev,struct kobj_uevent_env * env)1190 static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1191 {
1192 	const struct gendisk *disk = dev_to_disk(dev);
1193 
1194 	return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1195 }
1196 
1197 struct class block_class = {
1198 	.name		= "block",
1199 	.dev_uevent	= block_uevent,
1200 };
1201 
block_devnode(const struct device * dev,umode_t * mode,kuid_t * uid,kgid_t * gid)1202 static char *block_devnode(const struct device *dev, umode_t *mode,
1203 			   kuid_t *uid, kgid_t *gid)
1204 {
1205 	struct gendisk *disk = dev_to_disk(dev);
1206 
1207 	if (disk->fops->devnode)
1208 		return disk->fops->devnode(disk, mode);
1209 	return NULL;
1210 }
1211 
1212 const struct device_type disk_type = {
1213 	.name		= "disk",
1214 	.groups		= disk_attr_groups,
1215 	.release	= disk_release,
1216 	.devnode	= block_devnode,
1217 };
1218 
1219 #ifdef CONFIG_PROC_FS
1220 /*
1221  * aggregate disk stat collector.  Uses the same stats that the sysfs
1222  * entries do, above, but makes them available through one seq_file.
1223  *
1224  * The output looks suspiciously like /proc/partitions with a bunch of
1225  * extra fields.
1226  */
diskstats_show(struct seq_file * seqf,void * v)1227 static int diskstats_show(struct seq_file *seqf, void *v)
1228 {
1229 	struct gendisk *gp = v;
1230 	struct block_device *hd;
1231 	unsigned int inflight;
1232 	struct disk_stats stat;
1233 	unsigned long idx;
1234 
1235 	/*
1236 	if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1237 		seq_puts(seqf,	"major minor name"
1238 				"     rio rmerge rsect ruse wio wmerge "
1239 				"wsect wuse running use aveq"
1240 				"\n\n");
1241 	*/
1242 
1243 	rcu_read_lock();
1244 	xa_for_each(&gp->part_tbl, idx, hd) {
1245 		if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1246 			continue;
1247 		if (queue_is_mq(gp->queue))
1248 			inflight = blk_mq_in_flight(gp->queue, hd);
1249 		else
1250 			inflight = part_in_flight(hd);
1251 
1252 		if (inflight) {
1253 			part_stat_lock();
1254 			update_io_ticks(hd, jiffies, true);
1255 			part_stat_unlock();
1256 		}
1257 		part_stat_read_all(hd, &stat);
1258 		seq_printf(seqf, "%4d %7d %pg "
1259 			   "%lu %lu %lu %u "
1260 			   "%lu %lu %lu %u "
1261 			   "%u %u %u "
1262 			   "%lu %lu %lu %u "
1263 			   "%lu %u"
1264 			   "\n",
1265 			   MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1266 			   stat.ios[STAT_READ],
1267 			   stat.merges[STAT_READ],
1268 			   stat.sectors[STAT_READ],
1269 			   (unsigned int)div_u64(stat.nsecs[STAT_READ],
1270 							NSEC_PER_MSEC),
1271 			   stat.ios[STAT_WRITE],
1272 			   stat.merges[STAT_WRITE],
1273 			   stat.sectors[STAT_WRITE],
1274 			   (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1275 							NSEC_PER_MSEC),
1276 			   inflight,
1277 			   jiffies_to_msecs(stat.io_ticks),
1278 			   (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1279 						 stat.nsecs[STAT_WRITE] +
1280 						 stat.nsecs[STAT_DISCARD] +
1281 						 stat.nsecs[STAT_FLUSH],
1282 							NSEC_PER_MSEC),
1283 			   stat.ios[STAT_DISCARD],
1284 			   stat.merges[STAT_DISCARD],
1285 			   stat.sectors[STAT_DISCARD],
1286 			   (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1287 						 NSEC_PER_MSEC),
1288 			   stat.ios[STAT_FLUSH],
1289 			   (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1290 						 NSEC_PER_MSEC)
1291 			);
1292 	}
1293 	rcu_read_unlock();
1294 
1295 	return 0;
1296 }
1297 
1298 static const struct seq_operations diskstats_op = {
1299 	.start	= disk_seqf_start,
1300 	.next	= disk_seqf_next,
1301 	.stop	= disk_seqf_stop,
1302 	.show	= diskstats_show
1303 };
1304 
proc_genhd_init(void)1305 static int __init proc_genhd_init(void)
1306 {
1307 	proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1308 	proc_create_seq("partitions", 0, NULL, &partitions_op);
1309 	return 0;
1310 }
1311 module_init(proc_genhd_init);
1312 #endif /* CONFIG_PROC_FS */
1313 
part_devt(struct gendisk * disk,u8 partno)1314 dev_t part_devt(struct gendisk *disk, u8 partno)
1315 {
1316 	struct block_device *part;
1317 	dev_t devt = 0;
1318 
1319 	rcu_read_lock();
1320 	part = xa_load(&disk->part_tbl, partno);
1321 	if (part)
1322 		devt = part->bd_dev;
1323 	rcu_read_unlock();
1324 
1325 	return devt;
1326 }
1327 
__alloc_disk_node(struct request_queue * q,int node_id,struct lock_class_key * lkclass)1328 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1329 		struct lock_class_key *lkclass)
1330 {
1331 	struct gendisk *disk;
1332 
1333 	disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1334 	if (!disk)
1335 		return NULL;
1336 
1337 	if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1338 		goto out_free_disk;
1339 
1340 	disk->bdi = bdi_alloc(node_id);
1341 	if (!disk->bdi)
1342 		goto out_free_bioset;
1343 
1344 	/* bdev_alloc() might need the queue, set before the first call */
1345 	disk->queue = q;
1346 
1347 	disk->part0 = bdev_alloc(disk, 0);
1348 	if (!disk->part0)
1349 		goto out_free_bdi;
1350 
1351 	disk->node_id = node_id;
1352 	mutex_init(&disk->open_mutex);
1353 	xa_init(&disk->part_tbl);
1354 	if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1355 		goto out_destroy_part_tbl;
1356 
1357 	if (blkcg_init_disk(disk))
1358 		goto out_erase_part0;
1359 
1360 	rand_initialize_disk(disk);
1361 	disk_to_dev(disk)->class = &block_class;
1362 	disk_to_dev(disk)->type = &disk_type;
1363 	device_initialize(disk_to_dev(disk));
1364 	inc_diskseq(disk);
1365 	q->disk = disk;
1366 	lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1367 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1368 	INIT_LIST_HEAD(&disk->slave_bdevs);
1369 #endif
1370 	return disk;
1371 
1372 out_erase_part0:
1373 	xa_erase(&disk->part_tbl, 0);
1374 out_destroy_part_tbl:
1375 	xa_destroy(&disk->part_tbl);
1376 	disk->part0->bd_disk = NULL;
1377 	iput(disk->part0->bd_inode);
1378 out_free_bdi:
1379 	bdi_put(disk->bdi);
1380 out_free_bioset:
1381 	bioset_exit(&disk->bio_split);
1382 out_free_disk:
1383 	kfree(disk);
1384 	return NULL;
1385 }
1386 
__blk_alloc_disk(int node,struct lock_class_key * lkclass)1387 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1388 {
1389 	struct request_queue *q;
1390 	struct gendisk *disk;
1391 
1392 	q = blk_alloc_queue(node);
1393 	if (!q)
1394 		return NULL;
1395 
1396 	disk = __alloc_disk_node(q, node, lkclass);
1397 	if (!disk) {
1398 		blk_put_queue(q);
1399 		return NULL;
1400 	}
1401 	set_bit(GD_OWNS_QUEUE, &disk->state);
1402 	return disk;
1403 }
1404 EXPORT_SYMBOL(__blk_alloc_disk);
1405 
1406 /**
1407  * put_disk - decrements the gendisk refcount
1408  * @disk: the struct gendisk to decrement the refcount for
1409  *
1410  * This decrements the refcount for the struct gendisk. When this reaches 0
1411  * we'll have disk_release() called.
1412  *
1413  * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1414  * when handling probe errors (that is before add_disk() is called).
1415  *
1416  * Context: Any context, but the last reference must not be dropped from
1417  *          atomic context.
1418  */
put_disk(struct gendisk * disk)1419 void put_disk(struct gendisk *disk)
1420 {
1421 	if (disk)
1422 		put_device(disk_to_dev(disk));
1423 }
1424 EXPORT_SYMBOL(put_disk);
1425 
set_disk_ro_uevent(struct gendisk * gd,int ro)1426 static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1427 {
1428 	char event[] = "DISK_RO=1";
1429 	char *envp[] = { event, NULL };
1430 
1431 	if (!ro)
1432 		event[8] = '0';
1433 	kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1434 }
1435 
1436 /**
1437  * set_disk_ro - set a gendisk read-only
1438  * @disk:	gendisk to operate on
1439  * @read_only:	%true to set the disk read-only, %false set the disk read/write
1440  *
1441  * This function is used to indicate whether a given disk device should have its
1442  * read-only flag set. set_disk_ro() is typically used by device drivers to
1443  * indicate whether the underlying physical device is write-protected.
1444  */
set_disk_ro(struct gendisk * disk,bool read_only)1445 void set_disk_ro(struct gendisk *disk, bool read_only)
1446 {
1447 	if (read_only) {
1448 		if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1449 			return;
1450 	} else {
1451 		if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1452 			return;
1453 	}
1454 	set_disk_ro_uevent(disk, read_only);
1455 }
1456 EXPORT_SYMBOL(set_disk_ro);
1457 
inc_diskseq(struct gendisk * disk)1458 void inc_diskseq(struct gendisk *disk)
1459 {
1460 	disk->diskseq = atomic64_inc_return(&diskseq);
1461 }
1462