Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * gendisk handling
4 *
5 * Portions Copyright (C) 2020 Christoph Hellwig
6 */
7
8#include <linux/module.h>
9#include <linux/ctype.h>
10#include <linux/fs.h>
11#include <linux/kdev_t.h>
12#include <linux/kernel.h>
13#include <linux/blkdev.h>
14#include <linux/backing-dev.h>
15#include <linux/init.h>
16#include <linux/spinlock.h>
17#include <linux/proc_fs.h>
18#include <linux/seq_file.h>
19#include <linux/slab.h>
20#include <linux/kmod.h>
21#include <linux/major.h>
22#include <linux/mutex.h>
23#include <linux/idr.h>
24#include <linux/log2.h>
25#include <linux/pm_runtime.h>
26#include <linux/badblocks.h>
27#include <linux/part_stat.h>
28#include "blk-throttle.h"
29
30#include "blk.h"
31#include "blk-mq-sched.h"
32#include "blk-rq-qos.h"
33#include "blk-cgroup.h"
34
35static struct kobject *block_depr;
36
37/*
38 * Unique, monotonically increasing sequential number associated with block
39 * devices instances (i.e. incremented each time a device is attached).
40 * Associating uevents with block devices in userspace is difficult and racy:
41 * the uevent netlink socket is lossy, and on slow and overloaded systems has
42 * a very high latency.
43 * Block devices do not have exclusive owners in userspace, any process can set
44 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
45 * can be reused again and again).
46 * A userspace process setting up a block device and watching for its events
47 * cannot thus reliably tell whether an event relates to the device it just set
48 * up or another earlier instance with the same name.
49 * This sequential number allows userspace processes to solve this problem, and
50 * uniquely associate an uevent to the lifetime to a device.
51 */
52static atomic64_t diskseq;
53
54/* for extended dynamic devt allocation, currently only one major is used */
55#define NR_EXT_DEVT (1 << MINORBITS)
56static DEFINE_IDA(ext_devt_ida);
57
58void set_capacity(struct gendisk *disk, sector_t sectors)
59{
60 struct block_device *bdev = disk->part0;
61
62 spin_lock(&bdev->bd_size_lock);
63 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
64 bdev->bd_nr_sectors = sectors;
65 spin_unlock(&bdev->bd_size_lock);
66}
67EXPORT_SYMBOL(set_capacity);
68
69/*
70 * Set disk capacity and notify if the size is not currently zero and will not
71 * be set to zero. Returns true if a uevent was sent, otherwise false.
72 */
73bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
74{
75 sector_t capacity = get_capacity(disk);
76 char *envp[] = { "RESIZE=1", NULL };
77
78 set_capacity(disk, size);
79
80 /*
81 * Only print a message and send a uevent if the gendisk is user visible
82 * and alive. This avoids spamming the log and udev when setting the
83 * initial capacity during probing.
84 */
85 if (size == capacity ||
86 !disk_live(disk) ||
87 (disk->flags & GENHD_FL_HIDDEN))
88 return false;
89
90 pr_info("%s: detected capacity change from %lld to %lld\n",
91 disk->disk_name, capacity, size);
92
93 /*
94 * Historically we did not send a uevent for changes to/from an empty
95 * device.
96 */
97 if (!capacity || !size)
98 return false;
99 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
100 return true;
101}
102EXPORT_SYMBOL_GPL(set_capacity_and_notify);
103
104static void part_stat_read_all(struct block_device *part,
105 struct disk_stats *stat)
106{
107 int cpu;
108
109 memset(stat, 0, sizeof(struct disk_stats));
110 for_each_possible_cpu(cpu) {
111 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
112 int group;
113
114 for (group = 0; group < NR_STAT_GROUPS; group++) {
115 stat->nsecs[group] += ptr->nsecs[group];
116 stat->sectors[group] += ptr->sectors[group];
117 stat->ios[group] += ptr->ios[group];
118 stat->merges[group] += ptr->merges[group];
119 }
120
121 stat->io_ticks += ptr->io_ticks;
122 }
123}
124
125static unsigned int part_in_flight(struct block_device *part)
126{
127 unsigned int inflight = 0;
128 int cpu;
129
130 for_each_possible_cpu(cpu) {
131 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
132 part_stat_local_read_cpu(part, in_flight[1], cpu);
133 }
134 if ((int)inflight < 0)
135 inflight = 0;
136
137 return inflight;
138}
139
140static void part_in_flight_rw(struct block_device *part,
141 unsigned int inflight[2])
142{
143 int cpu;
144
145 inflight[0] = 0;
146 inflight[1] = 0;
147 for_each_possible_cpu(cpu) {
148 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
149 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
150 }
151 if ((int)inflight[0] < 0)
152 inflight[0] = 0;
153 if ((int)inflight[1] < 0)
154 inflight[1] = 0;
155}
156
157/*
158 * Can be deleted altogether. Later.
159 *
160 */
161#define BLKDEV_MAJOR_HASH_SIZE 255
162static struct blk_major_name {
163 struct blk_major_name *next;
164 int major;
165 char name[16];
166#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
167 void (*probe)(dev_t devt);
168#endif
169} *major_names[BLKDEV_MAJOR_HASH_SIZE];
170static DEFINE_MUTEX(major_names_lock);
171static DEFINE_SPINLOCK(major_names_spinlock);
172
173/* index in the above - for now: assume no multimajor ranges */
174static inline int major_to_index(unsigned major)
175{
176 return major % BLKDEV_MAJOR_HASH_SIZE;
177}
178
179#ifdef CONFIG_PROC_FS
180void blkdev_show(struct seq_file *seqf, off_t offset)
181{
182 struct blk_major_name *dp;
183
184 spin_lock(&major_names_spinlock);
185 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
186 if (dp->major == offset)
187 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
188 spin_unlock(&major_names_spinlock);
189}
190#endif /* CONFIG_PROC_FS */
191
192/**
193 * __register_blkdev - register a new block device
194 *
195 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
196 * @major = 0, try to allocate any unused major number.
197 * @name: the name of the new block device as a zero terminated string
198 * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
199 * pre-created device node is accessed. When a probe call uses
200 * add_disk() and it fails the driver must cleanup resources. This
201 * interface may soon be removed.
202 *
203 * The @name must be unique within the system.
204 *
205 * The return value depends on the @major input parameter:
206 *
207 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
208 * then the function returns zero on success, or a negative error code
209 * - if any unused major number was requested with @major = 0 parameter
210 * then the return value is the allocated major number in range
211 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
212 *
213 * See Documentation/admin-guide/devices.txt for the list of allocated
214 * major numbers.
215 *
216 * Use register_blkdev instead for any new code.
217 */
218int __register_blkdev(unsigned int major, const char *name,
219 void (*probe)(dev_t devt))
220{
221 struct blk_major_name **n, *p;
222 int index, ret = 0;
223
224 mutex_lock(&major_names_lock);
225
226 /* temporary */
227 if (major == 0) {
228 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
229 if (major_names[index] == NULL)
230 break;
231 }
232
233 if (index == 0) {
234 printk("%s: failed to get major for %s\n",
235 __func__, name);
236 ret = -EBUSY;
237 goto out;
238 }
239 major = index;
240 ret = major;
241 }
242
243 if (major >= BLKDEV_MAJOR_MAX) {
244 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
245 __func__, major, BLKDEV_MAJOR_MAX-1, name);
246
247 ret = -EINVAL;
248 goto out;
249 }
250
251 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
252 if (p == NULL) {
253 ret = -ENOMEM;
254 goto out;
255 }
256
257 p->major = major;
258#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
259 p->probe = probe;
260#endif
261 strlcpy(p->name, name, sizeof(p->name));
262 p->next = NULL;
263 index = major_to_index(major);
264
265 spin_lock(&major_names_spinlock);
266 for (n = &major_names[index]; *n; n = &(*n)->next) {
267 if ((*n)->major == major)
268 break;
269 }
270 if (!*n)
271 *n = p;
272 else
273 ret = -EBUSY;
274 spin_unlock(&major_names_spinlock);
275
276 if (ret < 0) {
277 printk("register_blkdev: cannot get major %u for %s\n",
278 major, name);
279 kfree(p);
280 }
281out:
282 mutex_unlock(&major_names_lock);
283 return ret;
284}
285EXPORT_SYMBOL(__register_blkdev);
286
287void unregister_blkdev(unsigned int major, const char *name)
288{
289 struct blk_major_name **n;
290 struct blk_major_name *p = NULL;
291 int index = major_to_index(major);
292
293 mutex_lock(&major_names_lock);
294 spin_lock(&major_names_spinlock);
295 for (n = &major_names[index]; *n; n = &(*n)->next)
296 if ((*n)->major == major)
297 break;
298 if (!*n || strcmp((*n)->name, name)) {
299 WARN_ON(1);
300 } else {
301 p = *n;
302 *n = p->next;
303 }
304 spin_unlock(&major_names_spinlock);
305 mutex_unlock(&major_names_lock);
306 kfree(p);
307}
308
309EXPORT_SYMBOL(unregister_blkdev);
310
311int blk_alloc_ext_minor(void)
312{
313 int idx;
314
315 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
316 if (idx == -ENOSPC)
317 return -EBUSY;
318 return idx;
319}
320
321void blk_free_ext_minor(unsigned int minor)
322{
323 ida_free(&ext_devt_ida, minor);
324}
325
326static char *bdevt_str(dev_t devt, char *buf)
327{
328 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
329 char tbuf[BDEVT_SIZE];
330 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
331 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
332 } else
333 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
334
335 return buf;
336}
337
338void disk_uevent(struct gendisk *disk, enum kobject_action action)
339{
340 struct block_device *part;
341 unsigned long idx;
342
343 rcu_read_lock();
344 xa_for_each(&disk->part_tbl, idx, part) {
345 if (bdev_is_partition(part) && !bdev_nr_sectors(part))
346 continue;
347 if (!kobject_get_unless_zero(&part->bd_device.kobj))
348 continue;
349
350 rcu_read_unlock();
351 kobject_uevent(bdev_kobj(part), action);
352 put_device(&part->bd_device);
353 rcu_read_lock();
354 }
355 rcu_read_unlock();
356}
357EXPORT_SYMBOL_GPL(disk_uevent);
358
359int disk_scan_partitions(struct gendisk *disk, fmode_t mode, void *owner)
360{
361 struct block_device *bdev;
362
363 if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
364 return -EINVAL;
365 if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
366 return -EINVAL;
367 if (disk->open_partitions)
368 return -EBUSY;
369 /* Someone else has bdev exclusively open? */
370 if (disk->part0->bd_holder && disk->part0->bd_holder != owner)
371 return -EBUSY;
372
373 set_bit(GD_NEED_PART_SCAN, &disk->state);
374 bdev = blkdev_get_by_dev(disk_devt(disk), mode, NULL);
375 if (IS_ERR(bdev))
376 return PTR_ERR(bdev);
377 blkdev_put(bdev, mode);
378 return 0;
379}
380
381/**
382 * device_add_disk - add disk information to kernel list
383 * @parent: parent device for the disk
384 * @disk: per-device partitioning information
385 * @groups: Additional per-device sysfs groups
386 *
387 * This function registers the partitioning information in @disk
388 * with the kernel.
389 */
390int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
391 const struct attribute_group **groups)
392
393{
394 struct device *ddev = disk_to_dev(disk);
395 int ret;
396
397 /* Only makes sense for bio-based to set ->poll_bio */
398 if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
399 return -EINVAL;
400
401 /*
402 * The disk queue should now be all set with enough information about
403 * the device for the elevator code to pick an adequate default
404 * elevator if one is needed, that is, for devices requesting queue
405 * registration.
406 */
407 elevator_init_mq(disk->queue);
408
409 /*
410 * If the driver provides an explicit major number it also must provide
411 * the number of minors numbers supported, and those will be used to
412 * setup the gendisk.
413 * Otherwise just allocate the device numbers for both the whole device
414 * and all partitions from the extended dev_t space.
415 */
416 ret = -EINVAL;
417 if (disk->major) {
418 if (WARN_ON(!disk->minors))
419 goto out_exit_elevator;
420
421 if (disk->minors > DISK_MAX_PARTS) {
422 pr_err("block: can't allocate more than %d partitions\n",
423 DISK_MAX_PARTS);
424 disk->minors = DISK_MAX_PARTS;
425 }
426 if (disk->first_minor + disk->minors > MINORMASK + 1)
427 goto out_exit_elevator;
428 } else {
429 if (WARN_ON(disk->minors))
430 goto out_exit_elevator;
431
432 ret = blk_alloc_ext_minor();
433 if (ret < 0)
434 goto out_exit_elevator;
435 disk->major = BLOCK_EXT_MAJOR;
436 disk->first_minor = ret;
437 }
438
439 /* delay uevents, until we scanned partition table */
440 dev_set_uevent_suppress(ddev, 1);
441
442 ddev->parent = parent;
443 ddev->groups = groups;
444 dev_set_name(ddev, "%s", disk->disk_name);
445 if (!(disk->flags & GENHD_FL_HIDDEN))
446 ddev->devt = MKDEV(disk->major, disk->first_minor);
447 ret = device_add(ddev);
448 if (ret)
449 goto out_free_ext_minor;
450
451 ret = disk_alloc_events(disk);
452 if (ret)
453 goto out_device_del;
454
455 if (!sysfs_deprecated) {
456 ret = sysfs_create_link(block_depr, &ddev->kobj,
457 kobject_name(&ddev->kobj));
458 if (ret)
459 goto out_device_del;
460 }
461
462 /*
463 * avoid probable deadlock caused by allocating memory with
464 * GFP_KERNEL in runtime_resume callback of its all ancestor
465 * devices
466 */
467 pm_runtime_set_memalloc_noio(ddev, true);
468
469 ret = blk_integrity_add(disk);
470 if (ret)
471 goto out_del_block_link;
472
473 disk->part0->bd_holder_dir =
474 kobject_create_and_add("holders", &ddev->kobj);
475 if (!disk->part0->bd_holder_dir) {
476 ret = -ENOMEM;
477 goto out_del_integrity;
478 }
479 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
480 if (!disk->slave_dir) {
481 ret = -ENOMEM;
482 goto out_put_holder_dir;
483 }
484
485 ret = blk_register_queue(disk);
486 if (ret)
487 goto out_put_slave_dir;
488
489 if (!(disk->flags & GENHD_FL_HIDDEN)) {
490 ret = bdi_register(disk->bdi, "%u:%u",
491 disk->major, disk->first_minor);
492 if (ret)
493 goto out_unregister_queue;
494 bdi_set_owner(disk->bdi, ddev);
495 ret = sysfs_create_link(&ddev->kobj,
496 &disk->bdi->dev->kobj, "bdi");
497 if (ret)
498 goto out_unregister_bdi;
499
500 bdev_add(disk->part0, ddev->devt);
501 if (get_capacity(disk))
502 disk_scan_partitions(disk, FMODE_READ, NULL);
503
504 /*
505 * Announce the disk and partitions after all partitions are
506 * created. (for hidden disks uevents remain suppressed forever)
507 */
508 dev_set_uevent_suppress(ddev, 0);
509 disk_uevent(disk, KOBJ_ADD);
510 } else {
511 /*
512 * Even if the block_device for a hidden gendisk is not
513 * registered, it needs to have a valid bd_dev so that the
514 * freeing of the dynamic major works.
515 */
516 disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
517 }
518
519 disk_update_readahead(disk);
520 disk_add_events(disk);
521 set_bit(GD_ADDED, &disk->state);
522 return 0;
523
524out_unregister_bdi:
525 if (!(disk->flags & GENHD_FL_HIDDEN))
526 bdi_unregister(disk->bdi);
527out_unregister_queue:
528 blk_unregister_queue(disk);
529 rq_qos_exit(disk->queue);
530out_put_slave_dir:
531 kobject_put(disk->slave_dir);
532 disk->slave_dir = NULL;
533out_put_holder_dir:
534 kobject_put(disk->part0->bd_holder_dir);
535out_del_integrity:
536 blk_integrity_del(disk);
537out_del_block_link:
538 if (!sysfs_deprecated)
539 sysfs_remove_link(block_depr, dev_name(ddev));
540out_device_del:
541 device_del(ddev);
542out_free_ext_minor:
543 if (disk->major == BLOCK_EXT_MAJOR)
544 blk_free_ext_minor(disk->first_minor);
545out_exit_elevator:
546 if (disk->queue->elevator)
547 elevator_exit(disk->queue);
548 return ret;
549}
550EXPORT_SYMBOL(device_add_disk);
551
552/**
553 * blk_mark_disk_dead - mark a disk as dead
554 * @disk: disk to mark as dead
555 *
556 * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
557 * to this disk.
558 */
559void blk_mark_disk_dead(struct gendisk *disk)
560{
561 set_bit(GD_DEAD, &disk->state);
562 blk_queue_start_drain(disk->queue);
563
564 /*
565 * Stop buffered writers from dirtying pages that can't be written out.
566 */
567 set_capacity_and_notify(disk, 0);
568}
569EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
570
571/**
572 * del_gendisk - remove the gendisk
573 * @disk: the struct gendisk to remove
574 *
575 * Removes the gendisk and all its associated resources. This deletes the
576 * partitions associated with the gendisk, and unregisters the associated
577 * request_queue.
578 *
579 * This is the counter to the respective __device_add_disk() call.
580 *
581 * The final removal of the struct gendisk happens when its refcount reaches 0
582 * with put_disk(), which should be called after del_gendisk(), if
583 * __device_add_disk() was used.
584 *
585 * Drivers exist which depend on the release of the gendisk to be synchronous,
586 * it should not be deferred.
587 *
588 * Context: can sleep
589 */
590void del_gendisk(struct gendisk *disk)
591{
592 struct request_queue *q = disk->queue;
593
594 might_sleep();
595
596 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
597 return;
598
599 blk_integrity_del(disk);
600 disk_del_events(disk);
601
602 mutex_lock(&disk->open_mutex);
603 remove_inode_hash(disk->part0->bd_inode);
604 blk_drop_partitions(disk);
605 mutex_unlock(&disk->open_mutex);
606
607 fsync_bdev(disk->part0);
608 __invalidate_device(disk->part0, true);
609
610 /*
611 * Fail any new I/O.
612 */
613 set_bit(GD_DEAD, &disk->state);
614 if (test_bit(GD_OWNS_QUEUE, &disk->state))
615 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
616 set_capacity(disk, 0);
617
618 /*
619 * Prevent new I/O from crossing bio_queue_enter().
620 */
621 blk_queue_start_drain(q);
622
623 if (!(disk->flags & GENHD_FL_HIDDEN)) {
624 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
625
626 /*
627 * Unregister bdi before releasing device numbers (as they can
628 * get reused and we'd get clashes in sysfs).
629 */
630 bdi_unregister(disk->bdi);
631 }
632
633 blk_unregister_queue(disk);
634
635 kobject_put(disk->part0->bd_holder_dir);
636 kobject_put(disk->slave_dir);
637 disk->slave_dir = NULL;
638
639 part_stat_set_all(disk->part0, 0);
640 disk->part0->bd_stamp = 0;
641 if (!sysfs_deprecated)
642 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
643 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
644 device_del(disk_to_dev(disk));
645
646 blk_mq_freeze_queue_wait(q);
647
648 blk_throtl_cancel_bios(disk);
649
650 blk_sync_queue(q);
651 blk_flush_integrity();
652
653 if (queue_is_mq(q))
654 blk_mq_cancel_work_sync(q);
655
656 blk_mq_quiesce_queue(q);
657 if (q->elevator) {
658 mutex_lock(&q->sysfs_lock);
659 elevator_exit(q);
660 mutex_unlock(&q->sysfs_lock);
661 }
662 rq_qos_exit(q);
663 blk_mq_unquiesce_queue(q);
664
665 /*
666 * If the disk does not own the queue, allow using passthrough requests
667 * again. Else leave the queue frozen to fail all I/O.
668 */
669 if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
670 blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
671 __blk_mq_unfreeze_queue(q, true);
672 } else {
673 if (queue_is_mq(q))
674 blk_mq_exit_queue(q);
675 }
676}
677EXPORT_SYMBOL(del_gendisk);
678
679/**
680 * invalidate_disk - invalidate the disk
681 * @disk: the struct gendisk to invalidate
682 *
683 * A helper to invalidates the disk. It will clean the disk's associated
684 * buffer/page caches and reset its internal states so that the disk
685 * can be reused by the drivers.
686 *
687 * Context: can sleep
688 */
689void invalidate_disk(struct gendisk *disk)
690{
691 struct block_device *bdev = disk->part0;
692
693 invalidate_bdev(bdev);
694 bdev->bd_inode->i_mapping->wb_err = 0;
695 set_capacity(disk, 0);
696}
697EXPORT_SYMBOL(invalidate_disk);
698
699/* sysfs access to bad-blocks list. */
700static ssize_t disk_badblocks_show(struct device *dev,
701 struct device_attribute *attr,
702 char *page)
703{
704 struct gendisk *disk = dev_to_disk(dev);
705
706 if (!disk->bb)
707 return sprintf(page, "\n");
708
709 return badblocks_show(disk->bb, page, 0);
710}
711
712static ssize_t disk_badblocks_store(struct device *dev,
713 struct device_attribute *attr,
714 const char *page, size_t len)
715{
716 struct gendisk *disk = dev_to_disk(dev);
717
718 if (!disk->bb)
719 return -ENXIO;
720
721 return badblocks_store(disk->bb, page, len, 0);
722}
723
724#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
725void blk_request_module(dev_t devt)
726{
727 unsigned int major = MAJOR(devt);
728 struct blk_major_name **n;
729
730 mutex_lock(&major_names_lock);
731 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
732 if ((*n)->major == major && (*n)->probe) {
733 (*n)->probe(devt);
734 mutex_unlock(&major_names_lock);
735 return;
736 }
737 }
738 mutex_unlock(&major_names_lock);
739
740 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
741 /* Make old-style 2.4 aliases work */
742 request_module("block-major-%d", MAJOR(devt));
743}
744#endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
745
746/*
747 * print a full list of all partitions - intended for places where the root
748 * filesystem can't be mounted and thus to give the victim some idea of what
749 * went wrong
750 */
751void __init printk_all_partitions(void)
752{
753 struct class_dev_iter iter;
754 struct device *dev;
755
756 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
757 while ((dev = class_dev_iter_next(&iter))) {
758 struct gendisk *disk = dev_to_disk(dev);
759 struct block_device *part;
760 char devt_buf[BDEVT_SIZE];
761 unsigned long idx;
762
763 /*
764 * Don't show empty devices or things that have been
765 * suppressed
766 */
767 if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
768 continue;
769
770 /*
771 * Note, unlike /proc/partitions, I am showing the numbers in
772 * hex - the same format as the root= option takes.
773 */
774 rcu_read_lock();
775 xa_for_each(&disk->part_tbl, idx, part) {
776 if (!bdev_nr_sectors(part))
777 continue;
778 printk("%s%s %10llu %pg %s",
779 bdev_is_partition(part) ? " " : "",
780 bdevt_str(part->bd_dev, devt_buf),
781 bdev_nr_sectors(part) >> 1, part,
782 part->bd_meta_info ?
783 part->bd_meta_info->uuid : "");
784 if (bdev_is_partition(part))
785 printk("\n");
786 else if (dev->parent && dev->parent->driver)
787 printk(" driver: %s\n",
788 dev->parent->driver->name);
789 else
790 printk(" (driver?)\n");
791 }
792 rcu_read_unlock();
793 }
794 class_dev_iter_exit(&iter);
795}
796
797#ifdef CONFIG_PROC_FS
798/* iterator */
799static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
800{
801 loff_t skip = *pos;
802 struct class_dev_iter *iter;
803 struct device *dev;
804
805 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
806 if (!iter)
807 return ERR_PTR(-ENOMEM);
808
809 seqf->private = iter;
810 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
811 do {
812 dev = class_dev_iter_next(iter);
813 if (!dev)
814 return NULL;
815 } while (skip--);
816
817 return dev_to_disk(dev);
818}
819
820static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
821{
822 struct device *dev;
823
824 (*pos)++;
825 dev = class_dev_iter_next(seqf->private);
826 if (dev)
827 return dev_to_disk(dev);
828
829 return NULL;
830}
831
832static void disk_seqf_stop(struct seq_file *seqf, void *v)
833{
834 struct class_dev_iter *iter = seqf->private;
835
836 /* stop is called even after start failed :-( */
837 if (iter) {
838 class_dev_iter_exit(iter);
839 kfree(iter);
840 seqf->private = NULL;
841 }
842}
843
844static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
845{
846 void *p;
847
848 p = disk_seqf_start(seqf, pos);
849 if (!IS_ERR_OR_NULL(p) && !*pos)
850 seq_puts(seqf, "major minor #blocks name\n\n");
851 return p;
852}
853
854static int show_partition(struct seq_file *seqf, void *v)
855{
856 struct gendisk *sgp = v;
857 struct block_device *part;
858 unsigned long idx;
859
860 if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
861 return 0;
862
863 rcu_read_lock();
864 xa_for_each(&sgp->part_tbl, idx, part) {
865 if (!bdev_nr_sectors(part))
866 continue;
867 seq_printf(seqf, "%4d %7d %10llu %pg\n",
868 MAJOR(part->bd_dev), MINOR(part->bd_dev),
869 bdev_nr_sectors(part) >> 1, part);
870 }
871 rcu_read_unlock();
872 return 0;
873}
874
875static const struct seq_operations partitions_op = {
876 .start = show_partition_start,
877 .next = disk_seqf_next,
878 .stop = disk_seqf_stop,
879 .show = show_partition
880};
881#endif
882
883static int __init genhd_device_init(void)
884{
885 int error;
886
887 block_class.dev_kobj = sysfs_dev_block_kobj;
888 error = class_register(&block_class);
889 if (unlikely(error))
890 return error;
891 blk_dev_init();
892
893 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
894
895 /* create top-level block dir */
896 if (!sysfs_deprecated)
897 block_depr = kobject_create_and_add("block", NULL);
898 return 0;
899}
900
901subsys_initcall(genhd_device_init);
902
903static ssize_t disk_range_show(struct device *dev,
904 struct device_attribute *attr, char *buf)
905{
906 struct gendisk *disk = dev_to_disk(dev);
907
908 return sprintf(buf, "%d\n", disk->minors);
909}
910
911static ssize_t disk_ext_range_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_NO_PART) ? 1 : DISK_MAX_PARTS);
918}
919
920static ssize_t disk_removable_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_REMOVABLE ? 1 : 0));
927}
928
929static ssize_t disk_hidden_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",
935 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
936}
937
938static ssize_t disk_ro_show(struct device *dev,
939 struct device_attribute *attr, char *buf)
940{
941 struct gendisk *disk = dev_to_disk(dev);
942
943 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
944}
945
946ssize_t part_size_show(struct device *dev,
947 struct device_attribute *attr, char *buf)
948{
949 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
950}
951
952ssize_t part_stat_show(struct device *dev,
953 struct device_attribute *attr, char *buf)
954{
955 struct block_device *bdev = dev_to_bdev(dev);
956 struct request_queue *q = bdev_get_queue(bdev);
957 struct disk_stats stat;
958 unsigned int inflight;
959
960 if (queue_is_mq(q))
961 inflight = blk_mq_in_flight(q, bdev);
962 else
963 inflight = part_in_flight(bdev);
964
965 if (inflight) {
966 part_stat_lock();
967 update_io_ticks(bdev, jiffies, true);
968 part_stat_unlock();
969 }
970 part_stat_read_all(bdev, &stat);
971 return sprintf(buf,
972 "%8lu %8lu %8llu %8u "
973 "%8lu %8lu %8llu %8u "
974 "%8u %8u %8u "
975 "%8lu %8lu %8llu %8u "
976 "%8lu %8u"
977 "\n",
978 stat.ios[STAT_READ],
979 stat.merges[STAT_READ],
980 (unsigned long long)stat.sectors[STAT_READ],
981 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
982 stat.ios[STAT_WRITE],
983 stat.merges[STAT_WRITE],
984 (unsigned long long)stat.sectors[STAT_WRITE],
985 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
986 inflight,
987 jiffies_to_msecs(stat.io_ticks),
988 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
989 stat.nsecs[STAT_WRITE] +
990 stat.nsecs[STAT_DISCARD] +
991 stat.nsecs[STAT_FLUSH],
992 NSEC_PER_MSEC),
993 stat.ios[STAT_DISCARD],
994 stat.merges[STAT_DISCARD],
995 (unsigned long long)stat.sectors[STAT_DISCARD],
996 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
997 stat.ios[STAT_FLUSH],
998 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
999}
1000
1001ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1002 char *buf)
1003{
1004 struct block_device *bdev = dev_to_bdev(dev);
1005 struct request_queue *q = bdev_get_queue(bdev);
1006 unsigned int inflight[2];
1007
1008 if (queue_is_mq(q))
1009 blk_mq_in_flight_rw(q, bdev, inflight);
1010 else
1011 part_in_flight_rw(bdev, inflight);
1012
1013 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1014}
1015
1016static ssize_t disk_capability_show(struct device *dev,
1017 struct device_attribute *attr, char *buf)
1018{
1019 struct gendisk *disk = dev_to_disk(dev);
1020
1021 return sprintf(buf, "%x\n", disk->flags);
1022}
1023
1024static ssize_t disk_alignment_offset_show(struct device *dev,
1025 struct device_attribute *attr,
1026 char *buf)
1027{
1028 struct gendisk *disk = dev_to_disk(dev);
1029
1030 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1031}
1032
1033static ssize_t disk_discard_alignment_show(struct device *dev,
1034 struct device_attribute *attr,
1035 char *buf)
1036{
1037 struct gendisk *disk = dev_to_disk(dev);
1038
1039 return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1040}
1041
1042static ssize_t diskseq_show(struct device *dev,
1043 struct device_attribute *attr, char *buf)
1044{
1045 struct gendisk *disk = dev_to_disk(dev);
1046
1047 return sprintf(buf, "%llu\n", disk->diskseq);
1048}
1049
1050static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1051static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1052static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1053static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1054static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1055static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1056static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1057static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1058static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1059static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1060static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1061static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1062static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1063
1064#ifdef CONFIG_FAIL_MAKE_REQUEST
1065ssize_t part_fail_show(struct device *dev,
1066 struct device_attribute *attr, char *buf)
1067{
1068 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1069}
1070
1071ssize_t part_fail_store(struct device *dev,
1072 struct device_attribute *attr,
1073 const char *buf, size_t count)
1074{
1075 int i;
1076
1077 if (count > 0 && sscanf(buf, "%d", &i) > 0)
1078 dev_to_bdev(dev)->bd_make_it_fail = i;
1079
1080 return count;
1081}
1082
1083static struct device_attribute dev_attr_fail =
1084 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1085#endif /* CONFIG_FAIL_MAKE_REQUEST */
1086
1087#ifdef CONFIG_FAIL_IO_TIMEOUT
1088static struct device_attribute dev_attr_fail_timeout =
1089 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1090#endif
1091
1092static struct attribute *disk_attrs[] = {
1093 &dev_attr_range.attr,
1094 &dev_attr_ext_range.attr,
1095 &dev_attr_removable.attr,
1096 &dev_attr_hidden.attr,
1097 &dev_attr_ro.attr,
1098 &dev_attr_size.attr,
1099 &dev_attr_alignment_offset.attr,
1100 &dev_attr_discard_alignment.attr,
1101 &dev_attr_capability.attr,
1102 &dev_attr_stat.attr,
1103 &dev_attr_inflight.attr,
1104 &dev_attr_badblocks.attr,
1105 &dev_attr_events.attr,
1106 &dev_attr_events_async.attr,
1107 &dev_attr_events_poll_msecs.attr,
1108 &dev_attr_diskseq.attr,
1109#ifdef CONFIG_FAIL_MAKE_REQUEST
1110 &dev_attr_fail.attr,
1111#endif
1112#ifdef CONFIG_FAIL_IO_TIMEOUT
1113 &dev_attr_fail_timeout.attr,
1114#endif
1115 NULL
1116};
1117
1118static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1119{
1120 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1121 struct gendisk *disk = dev_to_disk(dev);
1122
1123 if (a == &dev_attr_badblocks.attr && !disk->bb)
1124 return 0;
1125 return a->mode;
1126}
1127
1128static struct attribute_group disk_attr_group = {
1129 .attrs = disk_attrs,
1130 .is_visible = disk_visible,
1131};
1132
1133static const struct attribute_group *disk_attr_groups[] = {
1134 &disk_attr_group,
1135#ifdef CONFIG_BLK_DEV_IO_TRACE
1136 &blk_trace_attr_group,
1137#endif
1138 NULL
1139};
1140
1141/**
1142 * disk_release - releases all allocated resources of the gendisk
1143 * @dev: the device representing this disk
1144 *
1145 * This function releases all allocated resources of the gendisk.
1146 *
1147 * Drivers which used __device_add_disk() have a gendisk with a request_queue
1148 * assigned. Since the request_queue sits on top of the gendisk for these
1149 * drivers we also call blk_put_queue() for them, and we expect the
1150 * request_queue refcount to reach 0 at this point, and so the request_queue
1151 * will also be freed prior to the disk.
1152 *
1153 * Context: can sleep
1154 */
1155static void disk_release(struct device *dev)
1156{
1157 struct gendisk *disk = dev_to_disk(dev);
1158
1159 might_sleep();
1160 WARN_ON_ONCE(disk_live(disk));
1161
1162 /*
1163 * To undo the all initialization from blk_mq_init_allocated_queue in
1164 * case of a probe failure where add_disk is never called we have to
1165 * call blk_mq_exit_queue here. We can't do this for the more common
1166 * teardown case (yet) as the tagset can be gone by the time the disk
1167 * is released once it was added.
1168 */
1169 if (queue_is_mq(disk->queue) &&
1170 test_bit(GD_OWNS_QUEUE, &disk->state) &&
1171 !test_bit(GD_ADDED, &disk->state))
1172 blk_mq_exit_queue(disk->queue);
1173
1174 blkcg_exit_disk(disk);
1175
1176 bioset_exit(&disk->bio_split);
1177
1178 disk_release_events(disk);
1179 kfree(disk->random);
1180 disk_free_zone_bitmaps(disk);
1181 xa_destroy(&disk->part_tbl);
1182
1183 disk->queue->disk = NULL;
1184 blk_put_queue(disk->queue);
1185
1186 if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1187 disk->fops->free_disk(disk);
1188
1189 iput(disk->part0->bd_inode); /* frees the disk */
1190}
1191
1192static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1193{
1194 const struct gendisk *disk = dev_to_disk(dev);
1195
1196 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1197}
1198
1199struct class block_class = {
1200 .name = "block",
1201 .dev_uevent = block_uevent,
1202};
1203
1204static char *block_devnode(struct device *dev, umode_t *mode,
1205 kuid_t *uid, kgid_t *gid)
1206{
1207 struct gendisk *disk = dev_to_disk(dev);
1208
1209 if (disk->fops->devnode)
1210 return disk->fops->devnode(disk, mode);
1211 return NULL;
1212}
1213
1214const struct device_type disk_type = {
1215 .name = "disk",
1216 .groups = disk_attr_groups,
1217 .release = disk_release,
1218 .devnode = block_devnode,
1219};
1220
1221#ifdef CONFIG_PROC_FS
1222/*
1223 * aggregate disk stat collector. Uses the same stats that the sysfs
1224 * entries do, above, but makes them available through one seq_file.
1225 *
1226 * The output looks suspiciously like /proc/partitions with a bunch of
1227 * extra fields.
1228 */
1229static int diskstats_show(struct seq_file *seqf, void *v)
1230{
1231 struct gendisk *gp = v;
1232 struct block_device *hd;
1233 unsigned int inflight;
1234 struct disk_stats stat;
1235 unsigned long idx;
1236
1237 /*
1238 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1239 seq_puts(seqf, "major minor name"
1240 " rio rmerge rsect ruse wio wmerge "
1241 "wsect wuse running use aveq"
1242 "\n\n");
1243 */
1244
1245 rcu_read_lock();
1246 xa_for_each(&gp->part_tbl, idx, hd) {
1247 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1248 continue;
1249 if (queue_is_mq(gp->queue))
1250 inflight = blk_mq_in_flight(gp->queue, hd);
1251 else
1252 inflight = part_in_flight(hd);
1253
1254 if (inflight) {
1255 part_stat_lock();
1256 update_io_ticks(hd, jiffies, true);
1257 part_stat_unlock();
1258 }
1259 part_stat_read_all(hd, &stat);
1260 seq_printf(seqf, "%4d %7d %pg "
1261 "%lu %lu %lu %u "
1262 "%lu %lu %lu %u "
1263 "%u %u %u "
1264 "%lu %lu %lu %u "
1265 "%lu %u"
1266 "\n",
1267 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1268 stat.ios[STAT_READ],
1269 stat.merges[STAT_READ],
1270 stat.sectors[STAT_READ],
1271 (unsigned int)div_u64(stat.nsecs[STAT_READ],
1272 NSEC_PER_MSEC),
1273 stat.ios[STAT_WRITE],
1274 stat.merges[STAT_WRITE],
1275 stat.sectors[STAT_WRITE],
1276 (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1277 NSEC_PER_MSEC),
1278 inflight,
1279 jiffies_to_msecs(stat.io_ticks),
1280 (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1281 stat.nsecs[STAT_WRITE] +
1282 stat.nsecs[STAT_DISCARD] +
1283 stat.nsecs[STAT_FLUSH],
1284 NSEC_PER_MSEC),
1285 stat.ios[STAT_DISCARD],
1286 stat.merges[STAT_DISCARD],
1287 stat.sectors[STAT_DISCARD],
1288 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1289 NSEC_PER_MSEC),
1290 stat.ios[STAT_FLUSH],
1291 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1292 NSEC_PER_MSEC)
1293 );
1294 }
1295 rcu_read_unlock();
1296
1297 return 0;
1298}
1299
1300static const struct seq_operations diskstats_op = {
1301 .start = disk_seqf_start,
1302 .next = disk_seqf_next,
1303 .stop = disk_seqf_stop,
1304 .show = diskstats_show
1305};
1306
1307static int __init proc_genhd_init(void)
1308{
1309 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1310 proc_create_seq("partitions", 0, NULL, &partitions_op);
1311 return 0;
1312}
1313module_init(proc_genhd_init);
1314#endif /* CONFIG_PROC_FS */
1315
1316dev_t part_devt(struct gendisk *disk, u8 partno)
1317{
1318 struct block_device *part;
1319 dev_t devt = 0;
1320
1321 rcu_read_lock();
1322 part = xa_load(&disk->part_tbl, partno);
1323 if (part)
1324 devt = part->bd_dev;
1325 rcu_read_unlock();
1326
1327 return devt;
1328}
1329
1330dev_t blk_lookup_devt(const char *name, int partno)
1331{
1332 dev_t devt = MKDEV(0, 0);
1333 struct class_dev_iter iter;
1334 struct device *dev;
1335
1336 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1337 while ((dev = class_dev_iter_next(&iter))) {
1338 struct gendisk *disk = dev_to_disk(dev);
1339
1340 if (strcmp(dev_name(dev), name))
1341 continue;
1342
1343 if (partno < disk->minors) {
1344 /* We need to return the right devno, even
1345 * if the partition doesn't exist yet.
1346 */
1347 devt = MKDEV(MAJOR(dev->devt),
1348 MINOR(dev->devt) + partno);
1349 } else {
1350 devt = part_devt(disk, partno);
1351 if (devt)
1352 break;
1353 }
1354 }
1355 class_dev_iter_exit(&iter);
1356 return devt;
1357}
1358
1359struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1360 struct lock_class_key *lkclass)
1361{
1362 struct gendisk *disk;
1363
1364 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1365 if (!disk)
1366 return NULL;
1367
1368 if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1369 goto out_free_disk;
1370
1371 disk->bdi = bdi_alloc(node_id);
1372 if (!disk->bdi)
1373 goto out_free_bioset;
1374
1375 /* bdev_alloc() might need the queue, set before the first call */
1376 disk->queue = q;
1377
1378 disk->part0 = bdev_alloc(disk, 0);
1379 if (!disk->part0)
1380 goto out_free_bdi;
1381
1382 disk->node_id = node_id;
1383 mutex_init(&disk->open_mutex);
1384 xa_init(&disk->part_tbl);
1385 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1386 goto out_destroy_part_tbl;
1387
1388 if (blkcg_init_disk(disk))
1389 goto out_erase_part0;
1390
1391 rand_initialize_disk(disk);
1392 disk_to_dev(disk)->class = &block_class;
1393 disk_to_dev(disk)->type = &disk_type;
1394 device_initialize(disk_to_dev(disk));
1395 inc_diskseq(disk);
1396 q->disk = disk;
1397 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1398#ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1399 INIT_LIST_HEAD(&disk->slave_bdevs);
1400#endif
1401 return disk;
1402
1403out_erase_part0:
1404 xa_erase(&disk->part_tbl, 0);
1405out_destroy_part_tbl:
1406 xa_destroy(&disk->part_tbl);
1407 disk->part0->bd_disk = NULL;
1408 iput(disk->part0->bd_inode);
1409out_free_bdi:
1410 bdi_put(disk->bdi);
1411out_free_bioset:
1412 bioset_exit(&disk->bio_split);
1413out_free_disk:
1414 kfree(disk);
1415 return NULL;
1416}
1417
1418struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1419{
1420 struct request_queue *q;
1421 struct gendisk *disk;
1422
1423 q = blk_alloc_queue(node);
1424 if (!q)
1425 return NULL;
1426
1427 disk = __alloc_disk_node(q, node, lkclass);
1428 if (!disk) {
1429 blk_put_queue(q);
1430 return NULL;
1431 }
1432 set_bit(GD_OWNS_QUEUE, &disk->state);
1433 return disk;
1434}
1435EXPORT_SYMBOL(__blk_alloc_disk);
1436
1437/**
1438 * put_disk - decrements the gendisk refcount
1439 * @disk: the struct gendisk to decrement the refcount for
1440 *
1441 * This decrements the refcount for the struct gendisk. When this reaches 0
1442 * we'll have disk_release() called.
1443 *
1444 * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1445 * when handling probe errors (that is before add_disk() is called).
1446 *
1447 * Context: Any context, but the last reference must not be dropped from
1448 * atomic context.
1449 */
1450void put_disk(struct gendisk *disk)
1451{
1452 if (disk)
1453 put_device(disk_to_dev(disk));
1454}
1455EXPORT_SYMBOL(put_disk);
1456
1457static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1458{
1459 char event[] = "DISK_RO=1";
1460 char *envp[] = { event, NULL };
1461
1462 if (!ro)
1463 event[8] = '0';
1464 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1465}
1466
1467/**
1468 * set_disk_ro - set a gendisk read-only
1469 * @disk: gendisk to operate on
1470 * @read_only: %true to set the disk read-only, %false set the disk read/write
1471 *
1472 * This function is used to indicate whether a given disk device should have its
1473 * read-only flag set. set_disk_ro() is typically used by device drivers to
1474 * indicate whether the underlying physical device is write-protected.
1475 */
1476void set_disk_ro(struct gendisk *disk, bool read_only)
1477{
1478 if (read_only) {
1479 if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1480 return;
1481 } else {
1482 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1483 return;
1484 }
1485 set_disk_ro_uevent(disk, read_only);
1486}
1487EXPORT_SYMBOL(set_disk_ro);
1488
1489void inc_diskseq(struct gendisk *disk)
1490{
1491 disk->diskseq = atomic64_inc_return(&diskseq);
1492}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * gendisk handling
4 */
5
6#include <linux/module.h>
7#include <linux/fs.h>
8#include <linux/genhd.h>
9#include <linux/kdev_t.h>
10#include <linux/kernel.h>
11#include <linux/blkdev.h>
12#include <linux/backing-dev.h>
13#include <linux/init.h>
14#include <linux/spinlock.h>
15#include <linux/proc_fs.h>
16#include <linux/seq_file.h>
17#include <linux/slab.h>
18#include <linux/kmod.h>
19#include <linux/kobj_map.h>
20#include <linux/mutex.h>
21#include <linux/idr.h>
22#include <linux/log2.h>
23#include <linux/pm_runtime.h>
24#include <linux/badblocks.h>
25
26#include "blk.h"
27
28static DEFINE_MUTEX(block_class_lock);
29struct kobject *block_depr;
30
31/* for extended dynamic devt allocation, currently only one major is used */
32#define NR_EXT_DEVT (1 << MINORBITS)
33
34/* For extended devt allocation. ext_devt_lock prevents look up
35 * results from going away underneath its user.
36 */
37static DEFINE_SPINLOCK(ext_devt_lock);
38static DEFINE_IDR(ext_devt_idr);
39
40static const struct device_type disk_type;
41
42static void disk_check_events(struct disk_events *ev,
43 unsigned int *clearing_ptr);
44static void disk_alloc_events(struct gendisk *disk);
45static void disk_add_events(struct gendisk *disk);
46static void disk_del_events(struct gendisk *disk);
47static void disk_release_events(struct gendisk *disk);
48
49void part_inc_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
50{
51 if (queue_is_mq(q))
52 return;
53
54 part_stat_local_inc(part, in_flight[rw]);
55 if (part->partno)
56 part_stat_local_inc(&part_to_disk(part)->part0, in_flight[rw]);
57}
58
59void part_dec_in_flight(struct request_queue *q, struct hd_struct *part, int rw)
60{
61 if (queue_is_mq(q))
62 return;
63
64 part_stat_local_dec(part, in_flight[rw]);
65 if (part->partno)
66 part_stat_local_dec(&part_to_disk(part)->part0, in_flight[rw]);
67}
68
69unsigned int part_in_flight(struct request_queue *q, struct hd_struct *part)
70{
71 int cpu;
72 unsigned int inflight;
73
74 if (queue_is_mq(q)) {
75 return blk_mq_in_flight(q, part);
76 }
77
78 inflight = 0;
79 for_each_possible_cpu(cpu) {
80 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
81 part_stat_local_read_cpu(part, in_flight[1], cpu);
82 }
83 if ((int)inflight < 0)
84 inflight = 0;
85
86 return inflight;
87}
88
89void part_in_flight_rw(struct request_queue *q, struct hd_struct *part,
90 unsigned int inflight[2])
91{
92 int cpu;
93
94 if (queue_is_mq(q)) {
95 blk_mq_in_flight_rw(q, part, inflight);
96 return;
97 }
98
99 inflight[0] = 0;
100 inflight[1] = 0;
101 for_each_possible_cpu(cpu) {
102 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
103 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
104 }
105 if ((int)inflight[0] < 0)
106 inflight[0] = 0;
107 if ((int)inflight[1] < 0)
108 inflight[1] = 0;
109}
110
111struct hd_struct *__disk_get_part(struct gendisk *disk, int partno)
112{
113 struct disk_part_tbl *ptbl = rcu_dereference(disk->part_tbl);
114
115 if (unlikely(partno < 0 || partno >= ptbl->len))
116 return NULL;
117 return rcu_dereference(ptbl->part[partno]);
118}
119
120/**
121 * disk_get_part - get partition
122 * @disk: disk to look partition from
123 * @partno: partition number
124 *
125 * Look for partition @partno from @disk. If found, increment
126 * reference count and return it.
127 *
128 * CONTEXT:
129 * Don't care.
130 *
131 * RETURNS:
132 * Pointer to the found partition on success, NULL if not found.
133 */
134struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
135{
136 struct hd_struct *part;
137
138 rcu_read_lock();
139 part = __disk_get_part(disk, partno);
140 if (part)
141 get_device(part_to_dev(part));
142 rcu_read_unlock();
143
144 return part;
145}
146EXPORT_SYMBOL_GPL(disk_get_part);
147
148/**
149 * disk_part_iter_init - initialize partition iterator
150 * @piter: iterator to initialize
151 * @disk: disk to iterate over
152 * @flags: DISK_PITER_* flags
153 *
154 * Initialize @piter so that it iterates over partitions of @disk.
155 *
156 * CONTEXT:
157 * Don't care.
158 */
159void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
160 unsigned int flags)
161{
162 struct disk_part_tbl *ptbl;
163
164 rcu_read_lock();
165 ptbl = rcu_dereference(disk->part_tbl);
166
167 piter->disk = disk;
168 piter->part = NULL;
169
170 if (flags & DISK_PITER_REVERSE)
171 piter->idx = ptbl->len - 1;
172 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0))
173 piter->idx = 0;
174 else
175 piter->idx = 1;
176
177 piter->flags = flags;
178
179 rcu_read_unlock();
180}
181EXPORT_SYMBOL_GPL(disk_part_iter_init);
182
183/**
184 * disk_part_iter_next - proceed iterator to the next partition and return it
185 * @piter: iterator of interest
186 *
187 * Proceed @piter to the next partition and return it.
188 *
189 * CONTEXT:
190 * Don't care.
191 */
192struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
193{
194 struct disk_part_tbl *ptbl;
195 int inc, end;
196
197 /* put the last partition */
198 disk_put_part(piter->part);
199 piter->part = NULL;
200
201 /* get part_tbl */
202 rcu_read_lock();
203 ptbl = rcu_dereference(piter->disk->part_tbl);
204
205 /* determine iteration parameters */
206 if (piter->flags & DISK_PITER_REVERSE) {
207 inc = -1;
208 if (piter->flags & (DISK_PITER_INCL_PART0 |
209 DISK_PITER_INCL_EMPTY_PART0))
210 end = -1;
211 else
212 end = 0;
213 } else {
214 inc = 1;
215 end = ptbl->len;
216 }
217
218 /* iterate to the next partition */
219 for (; piter->idx != end; piter->idx += inc) {
220 struct hd_struct *part;
221
222 part = rcu_dereference(ptbl->part[piter->idx]);
223 if (!part)
224 continue;
225 if (!part_nr_sects_read(part) &&
226 !(piter->flags & DISK_PITER_INCL_EMPTY) &&
227 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 &&
228 piter->idx == 0))
229 continue;
230
231 get_device(part_to_dev(part));
232 piter->part = part;
233 piter->idx += inc;
234 break;
235 }
236
237 rcu_read_unlock();
238
239 return piter->part;
240}
241EXPORT_SYMBOL_GPL(disk_part_iter_next);
242
243/**
244 * disk_part_iter_exit - finish up partition iteration
245 * @piter: iter of interest
246 *
247 * Called when iteration is over. Cleans up @piter.
248 *
249 * CONTEXT:
250 * Don't care.
251 */
252void disk_part_iter_exit(struct disk_part_iter *piter)
253{
254 disk_put_part(piter->part);
255 piter->part = NULL;
256}
257EXPORT_SYMBOL_GPL(disk_part_iter_exit);
258
259static inline int sector_in_part(struct hd_struct *part, sector_t sector)
260{
261 return part->start_sect <= sector &&
262 sector < part->start_sect + part_nr_sects_read(part);
263}
264
265/**
266 * disk_map_sector_rcu - map sector to partition
267 * @disk: gendisk of interest
268 * @sector: sector to map
269 *
270 * Find out which partition @sector maps to on @disk. This is
271 * primarily used for stats accounting.
272 *
273 * CONTEXT:
274 * RCU read locked. The returned partition pointer is valid only
275 * while preemption is disabled.
276 *
277 * RETURNS:
278 * Found partition on success, part0 is returned if no partition matches
279 */
280struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
281{
282 struct disk_part_tbl *ptbl;
283 struct hd_struct *part;
284 int i;
285
286 ptbl = rcu_dereference(disk->part_tbl);
287
288 part = rcu_dereference(ptbl->last_lookup);
289 if (part && sector_in_part(part, sector))
290 return part;
291
292 for (i = 1; i < ptbl->len; i++) {
293 part = rcu_dereference(ptbl->part[i]);
294
295 if (part && sector_in_part(part, sector)) {
296 rcu_assign_pointer(ptbl->last_lookup, part);
297 return part;
298 }
299 }
300 return &disk->part0;
301}
302EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
303
304/*
305 * Can be deleted altogether. Later.
306 *
307 */
308#define BLKDEV_MAJOR_HASH_SIZE 255
309static struct blk_major_name {
310 struct blk_major_name *next;
311 int major;
312 char name[16];
313} *major_names[BLKDEV_MAJOR_HASH_SIZE];
314
315/* index in the above - for now: assume no multimajor ranges */
316static inline int major_to_index(unsigned major)
317{
318 return major % BLKDEV_MAJOR_HASH_SIZE;
319}
320
321#ifdef CONFIG_PROC_FS
322void blkdev_show(struct seq_file *seqf, off_t offset)
323{
324 struct blk_major_name *dp;
325
326 mutex_lock(&block_class_lock);
327 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
328 if (dp->major == offset)
329 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
330 mutex_unlock(&block_class_lock);
331}
332#endif /* CONFIG_PROC_FS */
333
334/**
335 * register_blkdev - register a new block device
336 *
337 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
338 * @major = 0, try to allocate any unused major number.
339 * @name: the name of the new block device as a zero terminated string
340 *
341 * The @name must be unique within the system.
342 *
343 * The return value depends on the @major input parameter:
344 *
345 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
346 * then the function returns zero on success, or a negative error code
347 * - if any unused major number was requested with @major = 0 parameter
348 * then the return value is the allocated major number in range
349 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
350 *
351 * See Documentation/admin-guide/devices.txt for the list of allocated
352 * major numbers.
353 */
354int register_blkdev(unsigned int major, const char *name)
355{
356 struct blk_major_name **n, *p;
357 int index, ret = 0;
358
359 mutex_lock(&block_class_lock);
360
361 /* temporary */
362 if (major == 0) {
363 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
364 if (major_names[index] == NULL)
365 break;
366 }
367
368 if (index == 0) {
369 printk("%s: failed to get major for %s\n",
370 __func__, name);
371 ret = -EBUSY;
372 goto out;
373 }
374 major = index;
375 ret = major;
376 }
377
378 if (major >= BLKDEV_MAJOR_MAX) {
379 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
380 __func__, major, BLKDEV_MAJOR_MAX-1, name);
381
382 ret = -EINVAL;
383 goto out;
384 }
385
386 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
387 if (p == NULL) {
388 ret = -ENOMEM;
389 goto out;
390 }
391
392 p->major = major;
393 strlcpy(p->name, name, sizeof(p->name));
394 p->next = NULL;
395 index = major_to_index(major);
396
397 for (n = &major_names[index]; *n; n = &(*n)->next) {
398 if ((*n)->major == major)
399 break;
400 }
401 if (!*n)
402 *n = p;
403 else
404 ret = -EBUSY;
405
406 if (ret < 0) {
407 printk("register_blkdev: cannot get major %u for %s\n",
408 major, name);
409 kfree(p);
410 }
411out:
412 mutex_unlock(&block_class_lock);
413 return ret;
414}
415
416EXPORT_SYMBOL(register_blkdev);
417
418void unregister_blkdev(unsigned int major, const char *name)
419{
420 struct blk_major_name **n;
421 struct blk_major_name *p = NULL;
422 int index = major_to_index(major);
423
424 mutex_lock(&block_class_lock);
425 for (n = &major_names[index]; *n; n = &(*n)->next)
426 if ((*n)->major == major)
427 break;
428 if (!*n || strcmp((*n)->name, name)) {
429 WARN_ON(1);
430 } else {
431 p = *n;
432 *n = p->next;
433 }
434 mutex_unlock(&block_class_lock);
435 kfree(p);
436}
437
438EXPORT_SYMBOL(unregister_blkdev);
439
440static struct kobj_map *bdev_map;
441
442/**
443 * blk_mangle_minor - scatter minor numbers apart
444 * @minor: minor number to mangle
445 *
446 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
447 * is enabled. Mangling twice gives the original value.
448 *
449 * RETURNS:
450 * Mangled value.
451 *
452 * CONTEXT:
453 * Don't care.
454 */
455static int blk_mangle_minor(int minor)
456{
457#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
458 int i;
459
460 for (i = 0; i < MINORBITS / 2; i++) {
461 int low = minor & (1 << i);
462 int high = minor & (1 << (MINORBITS - 1 - i));
463 int distance = MINORBITS - 1 - 2 * i;
464
465 minor ^= low | high; /* clear both bits */
466 low <<= distance; /* swap the positions */
467 high >>= distance;
468 minor |= low | high; /* and set */
469 }
470#endif
471 return minor;
472}
473
474/**
475 * blk_alloc_devt - allocate a dev_t for a partition
476 * @part: partition to allocate dev_t for
477 * @devt: out parameter for resulting dev_t
478 *
479 * Allocate a dev_t for block device.
480 *
481 * RETURNS:
482 * 0 on success, allocated dev_t is returned in *@devt. -errno on
483 * failure.
484 *
485 * CONTEXT:
486 * Might sleep.
487 */
488int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
489{
490 struct gendisk *disk = part_to_disk(part);
491 int idx;
492
493 /* in consecutive minor range? */
494 if (part->partno < disk->minors) {
495 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
496 return 0;
497 }
498
499 /* allocate ext devt */
500 idr_preload(GFP_KERNEL);
501
502 spin_lock_bh(&ext_devt_lock);
503 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT);
504 spin_unlock_bh(&ext_devt_lock);
505
506 idr_preload_end();
507 if (idx < 0)
508 return idx == -ENOSPC ? -EBUSY : idx;
509
510 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
511 return 0;
512}
513
514/**
515 * blk_free_devt - free a dev_t
516 * @devt: dev_t to free
517 *
518 * Free @devt which was allocated using blk_alloc_devt().
519 *
520 * CONTEXT:
521 * Might sleep.
522 */
523void blk_free_devt(dev_t devt)
524{
525 if (devt == MKDEV(0, 0))
526 return;
527
528 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
529 spin_lock_bh(&ext_devt_lock);
530 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
531 spin_unlock_bh(&ext_devt_lock);
532 }
533}
534
535/*
536 * We invalidate devt by assigning NULL pointer for devt in idr.
537 */
538void blk_invalidate_devt(dev_t devt)
539{
540 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
541 spin_lock_bh(&ext_devt_lock);
542 idr_replace(&ext_devt_idr, NULL, blk_mangle_minor(MINOR(devt)));
543 spin_unlock_bh(&ext_devt_lock);
544 }
545}
546
547static char *bdevt_str(dev_t devt, char *buf)
548{
549 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
550 char tbuf[BDEVT_SIZE];
551 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
552 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
553 } else
554 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
555
556 return buf;
557}
558
559/*
560 * Register device numbers dev..(dev+range-1)
561 * range must be nonzero
562 * The hash chain is sorted on range, so that subranges can override.
563 */
564void blk_register_region(dev_t devt, unsigned long range, struct module *module,
565 struct kobject *(*probe)(dev_t, int *, void *),
566 int (*lock)(dev_t, void *), void *data)
567{
568 kobj_map(bdev_map, devt, range, module, probe, lock, data);
569}
570
571EXPORT_SYMBOL(blk_register_region);
572
573void blk_unregister_region(dev_t devt, unsigned long range)
574{
575 kobj_unmap(bdev_map, devt, range);
576}
577
578EXPORT_SYMBOL(blk_unregister_region);
579
580static struct kobject *exact_match(dev_t devt, int *partno, void *data)
581{
582 struct gendisk *p = data;
583
584 return &disk_to_dev(p)->kobj;
585}
586
587static int exact_lock(dev_t devt, void *data)
588{
589 struct gendisk *p = data;
590
591 if (!get_disk_and_module(p))
592 return -1;
593 return 0;
594}
595
596static void register_disk(struct device *parent, struct gendisk *disk,
597 const struct attribute_group **groups)
598{
599 struct device *ddev = disk_to_dev(disk);
600 struct block_device *bdev;
601 struct disk_part_iter piter;
602 struct hd_struct *part;
603 int err;
604
605 ddev->parent = parent;
606
607 dev_set_name(ddev, "%s", disk->disk_name);
608
609 /* delay uevents, until we scanned partition table */
610 dev_set_uevent_suppress(ddev, 1);
611
612 if (groups) {
613 WARN_ON(ddev->groups);
614 ddev->groups = groups;
615 }
616 if (device_add(ddev))
617 return;
618 if (!sysfs_deprecated) {
619 err = sysfs_create_link(block_depr, &ddev->kobj,
620 kobject_name(&ddev->kobj));
621 if (err) {
622 device_del(ddev);
623 return;
624 }
625 }
626
627 /*
628 * avoid probable deadlock caused by allocating memory with
629 * GFP_KERNEL in runtime_resume callback of its all ancestor
630 * devices
631 */
632 pm_runtime_set_memalloc_noio(ddev, true);
633
634 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
635 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
636
637 if (disk->flags & GENHD_FL_HIDDEN) {
638 dev_set_uevent_suppress(ddev, 0);
639 return;
640 }
641
642 /* No minors to use for partitions */
643 if (!disk_part_scan_enabled(disk))
644 goto exit;
645
646 /* No such device (e.g., media were just removed) */
647 if (!get_capacity(disk))
648 goto exit;
649
650 bdev = bdget_disk(disk, 0);
651 if (!bdev)
652 goto exit;
653
654 bdev->bd_invalidated = 1;
655 err = blkdev_get(bdev, FMODE_READ, NULL);
656 if (err < 0)
657 goto exit;
658 blkdev_put(bdev, FMODE_READ);
659
660exit:
661 /* announce disk after possible partitions are created */
662 dev_set_uevent_suppress(ddev, 0);
663 kobject_uevent(&ddev->kobj, KOBJ_ADD);
664
665 /* announce possible partitions */
666 disk_part_iter_init(&piter, disk, 0);
667 while ((part = disk_part_iter_next(&piter)))
668 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
669 disk_part_iter_exit(&piter);
670
671 if (disk->queue->backing_dev_info->dev) {
672 err = sysfs_create_link(&ddev->kobj,
673 &disk->queue->backing_dev_info->dev->kobj,
674 "bdi");
675 WARN_ON(err);
676 }
677}
678
679/**
680 * __device_add_disk - add disk information to kernel list
681 * @parent: parent device for the disk
682 * @disk: per-device partitioning information
683 * @groups: Additional per-device sysfs groups
684 * @register_queue: register the queue if set to true
685 *
686 * This function registers the partitioning information in @disk
687 * with the kernel.
688 *
689 * FIXME: error handling
690 */
691static void __device_add_disk(struct device *parent, struct gendisk *disk,
692 const struct attribute_group **groups,
693 bool register_queue)
694{
695 dev_t devt;
696 int retval;
697
698 /*
699 * The disk queue should now be all set with enough information about
700 * the device for the elevator code to pick an adequate default
701 * elevator if one is needed, that is, for devices requesting queue
702 * registration.
703 */
704 if (register_queue)
705 elevator_init_mq(disk->queue);
706
707 /* minors == 0 indicates to use ext devt from part0 and should
708 * be accompanied with EXT_DEVT flag. Make sure all
709 * parameters make sense.
710 */
711 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
712 WARN_ON(!disk->minors &&
713 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN)));
714
715 disk->flags |= GENHD_FL_UP;
716
717 retval = blk_alloc_devt(&disk->part0, &devt);
718 if (retval) {
719 WARN_ON(1);
720 return;
721 }
722 disk->major = MAJOR(devt);
723 disk->first_minor = MINOR(devt);
724
725 disk_alloc_events(disk);
726
727 if (disk->flags & GENHD_FL_HIDDEN) {
728 /*
729 * Don't let hidden disks show up in /proc/partitions,
730 * and don't bother scanning for partitions either.
731 */
732 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
733 disk->flags |= GENHD_FL_NO_PART_SCAN;
734 } else {
735 int ret;
736
737 /* Register BDI before referencing it from bdev */
738 disk_to_dev(disk)->devt = devt;
739 ret = bdi_register_owner(disk->queue->backing_dev_info,
740 disk_to_dev(disk));
741 WARN_ON(ret);
742 blk_register_region(disk_devt(disk), disk->minors, NULL,
743 exact_match, exact_lock, disk);
744 }
745 register_disk(parent, disk, groups);
746 if (register_queue)
747 blk_register_queue(disk);
748
749 /*
750 * Take an extra ref on queue which will be put on disk_release()
751 * so that it sticks around as long as @disk is there.
752 */
753 WARN_ON_ONCE(!blk_get_queue(disk->queue));
754
755 disk_add_events(disk);
756 blk_integrity_add(disk);
757}
758
759void device_add_disk(struct device *parent, struct gendisk *disk,
760 const struct attribute_group **groups)
761
762{
763 __device_add_disk(parent, disk, groups, true);
764}
765EXPORT_SYMBOL(device_add_disk);
766
767void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
768{
769 __device_add_disk(parent, disk, NULL, false);
770}
771EXPORT_SYMBOL(device_add_disk_no_queue_reg);
772
773void del_gendisk(struct gendisk *disk)
774{
775 struct disk_part_iter piter;
776 struct hd_struct *part;
777
778 blk_integrity_del(disk);
779 disk_del_events(disk);
780
781 /*
782 * Block lookups of the disk until all bdevs are unhashed and the
783 * disk is marked as dead (GENHD_FL_UP cleared).
784 */
785 down_write(&disk->lookup_sem);
786 /* invalidate stuff */
787 disk_part_iter_init(&piter, disk,
788 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
789 while ((part = disk_part_iter_next(&piter))) {
790 invalidate_partition(disk, part->partno);
791 bdev_unhash_inode(part_devt(part));
792 delete_partition(disk, part->partno);
793 }
794 disk_part_iter_exit(&piter);
795
796 invalidate_partition(disk, 0);
797 bdev_unhash_inode(disk_devt(disk));
798 set_capacity(disk, 0);
799 disk->flags &= ~GENHD_FL_UP;
800 up_write(&disk->lookup_sem);
801
802 if (!(disk->flags & GENHD_FL_HIDDEN))
803 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
804 if (disk->queue) {
805 /*
806 * Unregister bdi before releasing device numbers (as they can
807 * get reused and we'd get clashes in sysfs).
808 */
809 if (!(disk->flags & GENHD_FL_HIDDEN))
810 bdi_unregister(disk->queue->backing_dev_info);
811 blk_unregister_queue(disk);
812 } else {
813 WARN_ON(1);
814 }
815
816 if (!(disk->flags & GENHD_FL_HIDDEN))
817 blk_unregister_region(disk_devt(disk), disk->minors);
818 /*
819 * Remove gendisk pointer from idr so that it cannot be looked up
820 * while RCU period before freeing gendisk is running to prevent
821 * use-after-free issues. Note that the device number stays
822 * "in-use" until we really free the gendisk.
823 */
824 blk_invalidate_devt(disk_devt(disk));
825
826 kobject_put(disk->part0.holder_dir);
827 kobject_put(disk->slave_dir);
828
829 part_stat_set_all(&disk->part0, 0);
830 disk->part0.stamp = 0;
831 if (!sysfs_deprecated)
832 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
833 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
834 device_del(disk_to_dev(disk));
835}
836EXPORT_SYMBOL(del_gendisk);
837
838/* sysfs access to bad-blocks list. */
839static ssize_t disk_badblocks_show(struct device *dev,
840 struct device_attribute *attr,
841 char *page)
842{
843 struct gendisk *disk = dev_to_disk(dev);
844
845 if (!disk->bb)
846 return sprintf(page, "\n");
847
848 return badblocks_show(disk->bb, page, 0);
849}
850
851static ssize_t disk_badblocks_store(struct device *dev,
852 struct device_attribute *attr,
853 const char *page, size_t len)
854{
855 struct gendisk *disk = dev_to_disk(dev);
856
857 if (!disk->bb)
858 return -ENXIO;
859
860 return badblocks_store(disk->bb, page, len, 0);
861}
862
863/**
864 * get_gendisk - get partitioning information for a given device
865 * @devt: device to get partitioning information for
866 * @partno: returned partition index
867 *
868 * This function gets the structure containing partitioning
869 * information for the given device @devt.
870 */
871struct gendisk *get_gendisk(dev_t devt, int *partno)
872{
873 struct gendisk *disk = NULL;
874
875 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
876 struct kobject *kobj;
877
878 kobj = kobj_lookup(bdev_map, devt, partno);
879 if (kobj)
880 disk = dev_to_disk(kobj_to_dev(kobj));
881 } else {
882 struct hd_struct *part;
883
884 spin_lock_bh(&ext_devt_lock);
885 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
886 if (part && get_disk_and_module(part_to_disk(part))) {
887 *partno = part->partno;
888 disk = part_to_disk(part);
889 }
890 spin_unlock_bh(&ext_devt_lock);
891 }
892
893 if (!disk)
894 return NULL;
895
896 /*
897 * Synchronize with del_gendisk() to not return disk that is being
898 * destroyed.
899 */
900 down_read(&disk->lookup_sem);
901 if (unlikely((disk->flags & GENHD_FL_HIDDEN) ||
902 !(disk->flags & GENHD_FL_UP))) {
903 up_read(&disk->lookup_sem);
904 put_disk_and_module(disk);
905 disk = NULL;
906 } else {
907 up_read(&disk->lookup_sem);
908 }
909 return disk;
910}
911EXPORT_SYMBOL(get_gendisk);
912
913/**
914 * bdget_disk - do bdget() by gendisk and partition number
915 * @disk: gendisk of interest
916 * @partno: partition number
917 *
918 * Find partition @partno from @disk, do bdget() on it.
919 *
920 * CONTEXT:
921 * Don't care.
922 *
923 * RETURNS:
924 * Resulting block_device on success, NULL on failure.
925 */
926struct block_device *bdget_disk(struct gendisk *disk, int partno)
927{
928 struct hd_struct *part;
929 struct block_device *bdev = NULL;
930
931 part = disk_get_part(disk, partno);
932 if (part)
933 bdev = bdget(part_devt(part));
934 disk_put_part(part);
935
936 return bdev;
937}
938EXPORT_SYMBOL(bdget_disk);
939
940/*
941 * print a full list of all partitions - intended for places where the root
942 * filesystem can't be mounted and thus to give the victim some idea of what
943 * went wrong
944 */
945void __init printk_all_partitions(void)
946{
947 struct class_dev_iter iter;
948 struct device *dev;
949
950 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
951 while ((dev = class_dev_iter_next(&iter))) {
952 struct gendisk *disk = dev_to_disk(dev);
953 struct disk_part_iter piter;
954 struct hd_struct *part;
955 char name_buf[BDEVNAME_SIZE];
956 char devt_buf[BDEVT_SIZE];
957
958 /*
959 * Don't show empty devices or things that have been
960 * suppressed
961 */
962 if (get_capacity(disk) == 0 ||
963 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
964 continue;
965
966 /*
967 * Note, unlike /proc/partitions, I am showing the
968 * numbers in hex - the same format as the root=
969 * option takes.
970 */
971 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
972 while ((part = disk_part_iter_next(&piter))) {
973 bool is_part0 = part == &disk->part0;
974
975 printk("%s%s %10llu %s %s", is_part0 ? "" : " ",
976 bdevt_str(part_devt(part), devt_buf),
977 (unsigned long long)part_nr_sects_read(part) >> 1
978 , disk_name(disk, part->partno, name_buf),
979 part->info ? part->info->uuid : "");
980 if (is_part0) {
981 if (dev->parent && dev->parent->driver)
982 printk(" driver: %s\n",
983 dev->parent->driver->name);
984 else
985 printk(" (driver?)\n");
986 } else
987 printk("\n");
988 }
989 disk_part_iter_exit(&piter);
990 }
991 class_dev_iter_exit(&iter);
992}
993
994#ifdef CONFIG_PROC_FS
995/* iterator */
996static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
997{
998 loff_t skip = *pos;
999 struct class_dev_iter *iter;
1000 struct device *dev;
1001
1002 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1003 if (!iter)
1004 return ERR_PTR(-ENOMEM);
1005
1006 seqf->private = iter;
1007 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
1008 do {
1009 dev = class_dev_iter_next(iter);
1010 if (!dev)
1011 return NULL;
1012 } while (skip--);
1013
1014 return dev_to_disk(dev);
1015}
1016
1017static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
1018{
1019 struct device *dev;
1020
1021 (*pos)++;
1022 dev = class_dev_iter_next(seqf->private);
1023 if (dev)
1024 return dev_to_disk(dev);
1025
1026 return NULL;
1027}
1028
1029static void disk_seqf_stop(struct seq_file *seqf, void *v)
1030{
1031 struct class_dev_iter *iter = seqf->private;
1032
1033 /* stop is called even after start failed :-( */
1034 if (iter) {
1035 class_dev_iter_exit(iter);
1036 kfree(iter);
1037 seqf->private = NULL;
1038 }
1039}
1040
1041static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
1042{
1043 void *p;
1044
1045 p = disk_seqf_start(seqf, pos);
1046 if (!IS_ERR_OR_NULL(p) && !*pos)
1047 seq_puts(seqf, "major minor #blocks name\n\n");
1048 return p;
1049}
1050
1051static int show_partition(struct seq_file *seqf, void *v)
1052{
1053 struct gendisk *sgp = v;
1054 struct disk_part_iter piter;
1055 struct hd_struct *part;
1056 char buf[BDEVNAME_SIZE];
1057
1058 /* Don't show non-partitionable removeable devices or empty devices */
1059 if (!get_capacity(sgp) || (!disk_max_parts(sgp) &&
1060 (sgp->flags & GENHD_FL_REMOVABLE)))
1061 return 0;
1062 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
1063 return 0;
1064
1065 /* show the full disk and all non-0 size partitions of it */
1066 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
1067 while ((part = disk_part_iter_next(&piter)))
1068 seq_printf(seqf, "%4d %7d %10llu %s\n",
1069 MAJOR(part_devt(part)), MINOR(part_devt(part)),
1070 (unsigned long long)part_nr_sects_read(part) >> 1,
1071 disk_name(sgp, part->partno, buf));
1072 disk_part_iter_exit(&piter);
1073
1074 return 0;
1075}
1076
1077static const struct seq_operations partitions_op = {
1078 .start = show_partition_start,
1079 .next = disk_seqf_next,
1080 .stop = disk_seqf_stop,
1081 .show = show_partition
1082};
1083#endif
1084
1085
1086static struct kobject *base_probe(dev_t devt, int *partno, void *data)
1087{
1088 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
1089 /* Make old-style 2.4 aliases work */
1090 request_module("block-major-%d", MAJOR(devt));
1091 return NULL;
1092}
1093
1094static int __init genhd_device_init(void)
1095{
1096 int error;
1097
1098 block_class.dev_kobj = sysfs_dev_block_kobj;
1099 error = class_register(&block_class);
1100 if (unlikely(error))
1101 return error;
1102 bdev_map = kobj_map_init(base_probe, &block_class_lock);
1103 blk_dev_init();
1104
1105 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
1106
1107 /* create top-level block dir */
1108 if (!sysfs_deprecated)
1109 block_depr = kobject_create_and_add("block", NULL);
1110 return 0;
1111}
1112
1113subsys_initcall(genhd_device_init);
1114
1115static ssize_t disk_range_show(struct device *dev,
1116 struct device_attribute *attr, char *buf)
1117{
1118 struct gendisk *disk = dev_to_disk(dev);
1119
1120 return sprintf(buf, "%d\n", disk->minors);
1121}
1122
1123static ssize_t disk_ext_range_show(struct device *dev,
1124 struct device_attribute *attr, char *buf)
1125{
1126 struct gendisk *disk = dev_to_disk(dev);
1127
1128 return sprintf(buf, "%d\n", disk_max_parts(disk));
1129}
1130
1131static ssize_t disk_removable_show(struct device *dev,
1132 struct device_attribute *attr, char *buf)
1133{
1134 struct gendisk *disk = dev_to_disk(dev);
1135
1136 return sprintf(buf, "%d\n",
1137 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
1138}
1139
1140static ssize_t disk_hidden_show(struct device *dev,
1141 struct device_attribute *attr, char *buf)
1142{
1143 struct gendisk *disk = dev_to_disk(dev);
1144
1145 return sprintf(buf, "%d\n",
1146 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
1147}
1148
1149static ssize_t disk_ro_show(struct device *dev,
1150 struct device_attribute *attr, char *buf)
1151{
1152 struct gendisk *disk = dev_to_disk(dev);
1153
1154 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
1155}
1156
1157static ssize_t disk_capability_show(struct device *dev,
1158 struct device_attribute *attr, char *buf)
1159{
1160 struct gendisk *disk = dev_to_disk(dev);
1161
1162 return sprintf(buf, "%x\n", disk->flags);
1163}
1164
1165static ssize_t disk_alignment_offset_show(struct device *dev,
1166 struct device_attribute *attr,
1167 char *buf)
1168{
1169 struct gendisk *disk = dev_to_disk(dev);
1170
1171 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue));
1172}
1173
1174static ssize_t disk_discard_alignment_show(struct device *dev,
1175 struct device_attribute *attr,
1176 char *buf)
1177{
1178 struct gendisk *disk = dev_to_disk(dev);
1179
1180 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue));
1181}
1182
1183static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1184static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1185static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1186static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1187static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1188static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1189static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1190static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1191static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1192static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1193static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1194static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1195#ifdef CONFIG_FAIL_MAKE_REQUEST
1196static struct device_attribute dev_attr_fail =
1197 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1198#endif
1199#ifdef CONFIG_FAIL_IO_TIMEOUT
1200static struct device_attribute dev_attr_fail_timeout =
1201 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1202#endif
1203
1204static struct attribute *disk_attrs[] = {
1205 &dev_attr_range.attr,
1206 &dev_attr_ext_range.attr,
1207 &dev_attr_removable.attr,
1208 &dev_attr_hidden.attr,
1209 &dev_attr_ro.attr,
1210 &dev_attr_size.attr,
1211 &dev_attr_alignment_offset.attr,
1212 &dev_attr_discard_alignment.attr,
1213 &dev_attr_capability.attr,
1214 &dev_attr_stat.attr,
1215 &dev_attr_inflight.attr,
1216 &dev_attr_badblocks.attr,
1217#ifdef CONFIG_FAIL_MAKE_REQUEST
1218 &dev_attr_fail.attr,
1219#endif
1220#ifdef CONFIG_FAIL_IO_TIMEOUT
1221 &dev_attr_fail_timeout.attr,
1222#endif
1223 NULL
1224};
1225
1226static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1227{
1228 struct device *dev = container_of(kobj, typeof(*dev), kobj);
1229 struct gendisk *disk = dev_to_disk(dev);
1230
1231 if (a == &dev_attr_badblocks.attr && !disk->bb)
1232 return 0;
1233 return a->mode;
1234}
1235
1236static struct attribute_group disk_attr_group = {
1237 .attrs = disk_attrs,
1238 .is_visible = disk_visible,
1239};
1240
1241static const struct attribute_group *disk_attr_groups[] = {
1242 &disk_attr_group,
1243 NULL
1244};
1245
1246/**
1247 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
1248 * @disk: disk to replace part_tbl for
1249 * @new_ptbl: new part_tbl to install
1250 *
1251 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
1252 * original ptbl is freed using RCU callback.
1253 *
1254 * LOCKING:
1255 * Matching bd_mutex locked or the caller is the only user of @disk.
1256 */
1257static void disk_replace_part_tbl(struct gendisk *disk,
1258 struct disk_part_tbl *new_ptbl)
1259{
1260 struct disk_part_tbl *old_ptbl =
1261 rcu_dereference_protected(disk->part_tbl, 1);
1262
1263 rcu_assign_pointer(disk->part_tbl, new_ptbl);
1264
1265 if (old_ptbl) {
1266 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
1267 kfree_rcu(old_ptbl, rcu_head);
1268 }
1269}
1270
1271/**
1272 * disk_expand_part_tbl - expand disk->part_tbl
1273 * @disk: disk to expand part_tbl for
1274 * @partno: expand such that this partno can fit in
1275 *
1276 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
1277 * uses RCU to allow unlocked dereferencing for stats and other stuff.
1278 *
1279 * LOCKING:
1280 * Matching bd_mutex locked or the caller is the only user of @disk.
1281 * Might sleep.
1282 *
1283 * RETURNS:
1284 * 0 on success, -errno on failure.
1285 */
1286int disk_expand_part_tbl(struct gendisk *disk, int partno)
1287{
1288 struct disk_part_tbl *old_ptbl =
1289 rcu_dereference_protected(disk->part_tbl, 1);
1290 struct disk_part_tbl *new_ptbl;
1291 int len = old_ptbl ? old_ptbl->len : 0;
1292 int i, target;
1293
1294 /*
1295 * check for int overflow, since we can get here from blkpg_ioctl()
1296 * with a user passed 'partno'.
1297 */
1298 target = partno + 1;
1299 if (target < 0)
1300 return -EINVAL;
1301
1302 /* disk_max_parts() is zero during initialization, ignore if so */
1303 if (disk_max_parts(disk) && target > disk_max_parts(disk))
1304 return -EINVAL;
1305
1306 if (target <= len)
1307 return 0;
1308
1309 new_ptbl = kzalloc_node(struct_size(new_ptbl, part, target), GFP_KERNEL,
1310 disk->node_id);
1311 if (!new_ptbl)
1312 return -ENOMEM;
1313
1314 new_ptbl->len = target;
1315
1316 for (i = 0; i < len; i++)
1317 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
1318
1319 disk_replace_part_tbl(disk, new_ptbl);
1320 return 0;
1321}
1322
1323static void disk_release(struct device *dev)
1324{
1325 struct gendisk *disk = dev_to_disk(dev);
1326
1327 blk_free_devt(dev->devt);
1328 disk_release_events(disk);
1329 kfree(disk->random);
1330 disk_replace_part_tbl(disk, NULL);
1331 hd_free_part(&disk->part0);
1332 if (disk->queue)
1333 blk_put_queue(disk->queue);
1334 kfree(disk);
1335}
1336struct class block_class = {
1337 .name = "block",
1338};
1339
1340static char *block_devnode(struct device *dev, umode_t *mode,
1341 kuid_t *uid, kgid_t *gid)
1342{
1343 struct gendisk *disk = dev_to_disk(dev);
1344
1345 if (disk->devnode)
1346 return disk->devnode(disk, mode);
1347 return NULL;
1348}
1349
1350static const struct device_type disk_type = {
1351 .name = "disk",
1352 .groups = disk_attr_groups,
1353 .release = disk_release,
1354 .devnode = block_devnode,
1355};
1356
1357#ifdef CONFIG_PROC_FS
1358/*
1359 * aggregate disk stat collector. Uses the same stats that the sysfs
1360 * entries do, above, but makes them available through one seq_file.
1361 *
1362 * The output looks suspiciously like /proc/partitions with a bunch of
1363 * extra fields.
1364 */
1365static int diskstats_show(struct seq_file *seqf, void *v)
1366{
1367 struct gendisk *gp = v;
1368 struct disk_part_iter piter;
1369 struct hd_struct *hd;
1370 char buf[BDEVNAME_SIZE];
1371 unsigned int inflight;
1372
1373 /*
1374 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1375 seq_puts(seqf, "major minor name"
1376 " rio rmerge rsect ruse wio wmerge "
1377 "wsect wuse running use aveq"
1378 "\n\n");
1379 */
1380
1381 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
1382 while ((hd = disk_part_iter_next(&piter))) {
1383 inflight = part_in_flight(gp->queue, hd);
1384 seq_printf(seqf, "%4d %7d %s "
1385 "%lu %lu %lu %u "
1386 "%lu %lu %lu %u "
1387 "%u %u %u "
1388 "%lu %lu %lu %u\n",
1389 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1390 disk_name(gp, hd->partno, buf),
1391 part_stat_read(hd, ios[STAT_READ]),
1392 part_stat_read(hd, merges[STAT_READ]),
1393 part_stat_read(hd, sectors[STAT_READ]),
1394 (unsigned int)part_stat_read_msecs(hd, STAT_READ),
1395 part_stat_read(hd, ios[STAT_WRITE]),
1396 part_stat_read(hd, merges[STAT_WRITE]),
1397 part_stat_read(hd, sectors[STAT_WRITE]),
1398 (unsigned int)part_stat_read_msecs(hd, STAT_WRITE),
1399 inflight,
1400 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1401 jiffies_to_msecs(part_stat_read(hd, time_in_queue)),
1402 part_stat_read(hd, ios[STAT_DISCARD]),
1403 part_stat_read(hd, merges[STAT_DISCARD]),
1404 part_stat_read(hd, sectors[STAT_DISCARD]),
1405 (unsigned int)part_stat_read_msecs(hd, STAT_DISCARD)
1406 );
1407 }
1408 disk_part_iter_exit(&piter);
1409
1410 return 0;
1411}
1412
1413static const struct seq_operations diskstats_op = {
1414 .start = disk_seqf_start,
1415 .next = disk_seqf_next,
1416 .stop = disk_seqf_stop,
1417 .show = diskstats_show
1418};
1419
1420static int __init proc_genhd_init(void)
1421{
1422 proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1423 proc_create_seq("partitions", 0, NULL, &partitions_op);
1424 return 0;
1425}
1426module_init(proc_genhd_init);
1427#endif /* CONFIG_PROC_FS */
1428
1429dev_t blk_lookup_devt(const char *name, int partno)
1430{
1431 dev_t devt = MKDEV(0, 0);
1432 struct class_dev_iter iter;
1433 struct device *dev;
1434
1435 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1436 while ((dev = class_dev_iter_next(&iter))) {
1437 struct gendisk *disk = dev_to_disk(dev);
1438 struct hd_struct *part;
1439
1440 if (strcmp(dev_name(dev), name))
1441 continue;
1442
1443 if (partno < disk->minors) {
1444 /* We need to return the right devno, even
1445 * if the partition doesn't exist yet.
1446 */
1447 devt = MKDEV(MAJOR(dev->devt),
1448 MINOR(dev->devt) + partno);
1449 break;
1450 }
1451 part = disk_get_part(disk, partno);
1452 if (part) {
1453 devt = part_devt(part);
1454 disk_put_part(part);
1455 break;
1456 }
1457 disk_put_part(part);
1458 }
1459 class_dev_iter_exit(&iter);
1460 return devt;
1461}
1462EXPORT_SYMBOL(blk_lookup_devt);
1463
1464struct gendisk *__alloc_disk_node(int minors, int node_id)
1465{
1466 struct gendisk *disk;
1467 struct disk_part_tbl *ptbl;
1468
1469 if (minors > DISK_MAX_PARTS) {
1470 printk(KERN_ERR
1471 "block: can't allocate more than %d partitions\n",
1472 DISK_MAX_PARTS);
1473 minors = DISK_MAX_PARTS;
1474 }
1475
1476 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1477 if (disk) {
1478 if (!init_part_stats(&disk->part0)) {
1479 kfree(disk);
1480 return NULL;
1481 }
1482 init_rwsem(&disk->lookup_sem);
1483 disk->node_id = node_id;
1484 if (disk_expand_part_tbl(disk, 0)) {
1485 free_part_stats(&disk->part0);
1486 kfree(disk);
1487 return NULL;
1488 }
1489 ptbl = rcu_dereference_protected(disk->part_tbl, 1);
1490 rcu_assign_pointer(ptbl->part[0], &disk->part0);
1491
1492 /*
1493 * set_capacity() and get_capacity() currently don't use
1494 * seqcounter to read/update the part0->nr_sects. Still init
1495 * the counter as we can read the sectors in IO submission
1496 * patch using seqence counters.
1497 *
1498 * TODO: Ideally set_capacity() and get_capacity() should be
1499 * converted to make use of bd_mutex and sequence counters.
1500 */
1501 seqcount_init(&disk->part0.nr_sects_seq);
1502 if (hd_ref_init(&disk->part0)) {
1503 hd_free_part(&disk->part0);
1504 kfree(disk);
1505 return NULL;
1506 }
1507
1508 disk->minors = minors;
1509 rand_initialize_disk(disk);
1510 disk_to_dev(disk)->class = &block_class;
1511 disk_to_dev(disk)->type = &disk_type;
1512 device_initialize(disk_to_dev(disk));
1513 }
1514 return disk;
1515}
1516EXPORT_SYMBOL(__alloc_disk_node);
1517
1518struct kobject *get_disk_and_module(struct gendisk *disk)
1519{
1520 struct module *owner;
1521 struct kobject *kobj;
1522
1523 if (!disk->fops)
1524 return NULL;
1525 owner = disk->fops->owner;
1526 if (owner && !try_module_get(owner))
1527 return NULL;
1528 kobj = kobject_get_unless_zero(&disk_to_dev(disk)->kobj);
1529 if (kobj == NULL) {
1530 module_put(owner);
1531 return NULL;
1532 }
1533 return kobj;
1534
1535}
1536EXPORT_SYMBOL(get_disk_and_module);
1537
1538void put_disk(struct gendisk *disk)
1539{
1540 if (disk)
1541 kobject_put(&disk_to_dev(disk)->kobj);
1542}
1543EXPORT_SYMBOL(put_disk);
1544
1545/*
1546 * This is a counterpart of get_disk_and_module() and thus also of
1547 * get_gendisk().
1548 */
1549void put_disk_and_module(struct gendisk *disk)
1550{
1551 if (disk) {
1552 struct module *owner = disk->fops->owner;
1553
1554 put_disk(disk);
1555 module_put(owner);
1556 }
1557}
1558EXPORT_SYMBOL(put_disk_and_module);
1559
1560static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1561{
1562 char event[] = "DISK_RO=1";
1563 char *envp[] = { event, NULL };
1564
1565 if (!ro)
1566 event[8] = '0';
1567 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1568}
1569
1570void set_device_ro(struct block_device *bdev, int flag)
1571{
1572 bdev->bd_part->policy = flag;
1573}
1574
1575EXPORT_SYMBOL(set_device_ro);
1576
1577void set_disk_ro(struct gendisk *disk, int flag)
1578{
1579 struct disk_part_iter piter;
1580 struct hd_struct *part;
1581
1582 if (disk->part0.policy != flag) {
1583 set_disk_ro_uevent(disk, flag);
1584 disk->part0.policy = flag;
1585 }
1586
1587 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
1588 while ((part = disk_part_iter_next(&piter)))
1589 part->policy = flag;
1590 disk_part_iter_exit(&piter);
1591}
1592
1593EXPORT_SYMBOL(set_disk_ro);
1594
1595int bdev_read_only(struct block_device *bdev)
1596{
1597 if (!bdev)
1598 return 0;
1599 return bdev->bd_part->policy;
1600}
1601
1602EXPORT_SYMBOL(bdev_read_only);
1603
1604int invalidate_partition(struct gendisk *disk, int partno)
1605{
1606 int res = 0;
1607 struct block_device *bdev = bdget_disk(disk, partno);
1608 if (bdev) {
1609 fsync_bdev(bdev);
1610 res = __invalidate_device(bdev, true);
1611 bdput(bdev);
1612 }
1613 return res;
1614}
1615
1616EXPORT_SYMBOL(invalidate_partition);
1617
1618/*
1619 * Disk events - monitor disk events like media change and eject request.
1620 */
1621struct disk_events {
1622 struct list_head node; /* all disk_event's */
1623 struct gendisk *disk; /* the associated disk */
1624 spinlock_t lock;
1625
1626 struct mutex block_mutex; /* protects blocking */
1627 int block; /* event blocking depth */
1628 unsigned int pending; /* events already sent out */
1629 unsigned int clearing; /* events being cleared */
1630
1631 long poll_msecs; /* interval, -1 for default */
1632 struct delayed_work dwork;
1633};
1634
1635static const char *disk_events_strs[] = {
1636 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change",
1637 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request",
1638};
1639
1640static char *disk_uevents[] = {
1641 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1",
1642 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1",
1643};
1644
1645/* list of all disk_events */
1646static DEFINE_MUTEX(disk_events_mutex);
1647static LIST_HEAD(disk_events);
1648
1649/* disable in-kernel polling by default */
1650static unsigned long disk_events_dfl_poll_msecs;
1651
1652static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
1653{
1654 struct disk_events *ev = disk->ev;
1655 long intv_msecs = 0;
1656
1657 /*
1658 * If device-specific poll interval is set, always use it. If
1659 * the default is being used, poll if the POLL flag is set.
1660 */
1661 if (ev->poll_msecs >= 0)
1662 intv_msecs = ev->poll_msecs;
1663 else if (disk->event_flags & DISK_EVENT_FLAG_POLL)
1664 intv_msecs = disk_events_dfl_poll_msecs;
1665
1666 return msecs_to_jiffies(intv_msecs);
1667}
1668
1669/**
1670 * disk_block_events - block and flush disk event checking
1671 * @disk: disk to block events for
1672 *
1673 * On return from this function, it is guaranteed that event checking
1674 * isn't in progress and won't happen until unblocked by
1675 * disk_unblock_events(). Events blocking is counted and the actual
1676 * unblocking happens after the matching number of unblocks are done.
1677 *
1678 * Note that this intentionally does not block event checking from
1679 * disk_clear_events().
1680 *
1681 * CONTEXT:
1682 * Might sleep.
1683 */
1684void disk_block_events(struct gendisk *disk)
1685{
1686 struct disk_events *ev = disk->ev;
1687 unsigned long flags;
1688 bool cancel;
1689
1690 if (!ev)
1691 return;
1692
1693 /*
1694 * Outer mutex ensures that the first blocker completes canceling
1695 * the event work before further blockers are allowed to finish.
1696 */
1697 mutex_lock(&ev->block_mutex);
1698
1699 spin_lock_irqsave(&ev->lock, flags);
1700 cancel = !ev->block++;
1701 spin_unlock_irqrestore(&ev->lock, flags);
1702
1703 if (cancel)
1704 cancel_delayed_work_sync(&disk->ev->dwork);
1705
1706 mutex_unlock(&ev->block_mutex);
1707}
1708
1709static void __disk_unblock_events(struct gendisk *disk, bool check_now)
1710{
1711 struct disk_events *ev = disk->ev;
1712 unsigned long intv;
1713 unsigned long flags;
1714
1715 spin_lock_irqsave(&ev->lock, flags);
1716
1717 if (WARN_ON_ONCE(ev->block <= 0))
1718 goto out_unlock;
1719
1720 if (--ev->block)
1721 goto out_unlock;
1722
1723 intv = disk_events_poll_jiffies(disk);
1724 if (check_now)
1725 queue_delayed_work(system_freezable_power_efficient_wq,
1726 &ev->dwork, 0);
1727 else if (intv)
1728 queue_delayed_work(system_freezable_power_efficient_wq,
1729 &ev->dwork, intv);
1730out_unlock:
1731 spin_unlock_irqrestore(&ev->lock, flags);
1732}
1733
1734/**
1735 * disk_unblock_events - unblock disk event checking
1736 * @disk: disk to unblock events for
1737 *
1738 * Undo disk_block_events(). When the block count reaches zero, it
1739 * starts events polling if configured.
1740 *
1741 * CONTEXT:
1742 * Don't care. Safe to call from irq context.
1743 */
1744void disk_unblock_events(struct gendisk *disk)
1745{
1746 if (disk->ev)
1747 __disk_unblock_events(disk, false);
1748}
1749
1750/**
1751 * disk_flush_events - schedule immediate event checking and flushing
1752 * @disk: disk to check and flush events for
1753 * @mask: events to flush
1754 *
1755 * Schedule immediate event checking on @disk if not blocked. Events in
1756 * @mask are scheduled to be cleared from the driver. Note that this
1757 * doesn't clear the events from @disk->ev.
1758 *
1759 * CONTEXT:
1760 * If @mask is non-zero must be called with bdev->bd_mutex held.
1761 */
1762void disk_flush_events(struct gendisk *disk, unsigned int mask)
1763{
1764 struct disk_events *ev = disk->ev;
1765
1766 if (!ev)
1767 return;
1768
1769 spin_lock_irq(&ev->lock);
1770 ev->clearing |= mask;
1771 if (!ev->block)
1772 mod_delayed_work(system_freezable_power_efficient_wq,
1773 &ev->dwork, 0);
1774 spin_unlock_irq(&ev->lock);
1775}
1776
1777/**
1778 * disk_clear_events - synchronously check, clear and return pending events
1779 * @disk: disk to fetch and clear events from
1780 * @mask: mask of events to be fetched and cleared
1781 *
1782 * Disk events are synchronously checked and pending events in @mask
1783 * are cleared and returned. This ignores the block count.
1784 *
1785 * CONTEXT:
1786 * Might sleep.
1787 */
1788unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
1789{
1790 const struct block_device_operations *bdops = disk->fops;
1791 struct disk_events *ev = disk->ev;
1792 unsigned int pending;
1793 unsigned int clearing = mask;
1794
1795 if (!ev) {
1796 /* for drivers still using the old ->media_changed method */
1797 if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
1798 bdops->media_changed && bdops->media_changed(disk))
1799 return DISK_EVENT_MEDIA_CHANGE;
1800 return 0;
1801 }
1802
1803 disk_block_events(disk);
1804
1805 /*
1806 * store the union of mask and ev->clearing on the stack so that the
1807 * race with disk_flush_events does not cause ambiguity (ev->clearing
1808 * can still be modified even if events are blocked).
1809 */
1810 spin_lock_irq(&ev->lock);
1811 clearing |= ev->clearing;
1812 ev->clearing = 0;
1813 spin_unlock_irq(&ev->lock);
1814
1815 disk_check_events(ev, &clearing);
1816 /*
1817 * if ev->clearing is not 0, the disk_flush_events got called in the
1818 * middle of this function, so we want to run the workfn without delay.
1819 */
1820 __disk_unblock_events(disk, ev->clearing ? true : false);
1821
1822 /* then, fetch and clear pending events */
1823 spin_lock_irq(&ev->lock);
1824 pending = ev->pending & mask;
1825 ev->pending &= ~mask;
1826 spin_unlock_irq(&ev->lock);
1827 WARN_ON_ONCE(clearing & mask);
1828
1829 return pending;
1830}
1831
1832/*
1833 * Separate this part out so that a different pointer for clearing_ptr can be
1834 * passed in for disk_clear_events.
1835 */
1836static void disk_events_workfn(struct work_struct *work)
1837{
1838 struct delayed_work *dwork = to_delayed_work(work);
1839 struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
1840
1841 disk_check_events(ev, &ev->clearing);
1842}
1843
1844static void disk_check_events(struct disk_events *ev,
1845 unsigned int *clearing_ptr)
1846{
1847 struct gendisk *disk = ev->disk;
1848 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
1849 unsigned int clearing = *clearing_ptr;
1850 unsigned int events;
1851 unsigned long intv;
1852 int nr_events = 0, i;
1853
1854 /* check events */
1855 events = disk->fops->check_events(disk, clearing);
1856
1857 /* accumulate pending events and schedule next poll if necessary */
1858 spin_lock_irq(&ev->lock);
1859
1860 events &= ~ev->pending;
1861 ev->pending |= events;
1862 *clearing_ptr &= ~clearing;
1863
1864 intv = disk_events_poll_jiffies(disk);
1865 if (!ev->block && intv)
1866 queue_delayed_work(system_freezable_power_efficient_wq,
1867 &ev->dwork, intv);
1868
1869 spin_unlock_irq(&ev->lock);
1870
1871 /*
1872 * Tell userland about new events. Only the events listed in
1873 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT
1874 * is set. Otherwise, events are processed internally but never
1875 * get reported to userland.
1876 */
1877 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
1878 if ((events & disk->events & (1 << i)) &&
1879 (disk->event_flags & DISK_EVENT_FLAG_UEVENT))
1880 envp[nr_events++] = disk_uevents[i];
1881
1882 if (nr_events)
1883 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
1884}
1885
1886/*
1887 * A disk events enabled device has the following sysfs nodes under
1888 * its /sys/block/X/ directory.
1889 *
1890 * events : list of all supported events
1891 * events_async : list of events which can be detected w/o polling
1892 * (always empty, only for backwards compatibility)
1893 * events_poll_msecs : polling interval, 0: disable, -1: system default
1894 */
1895static ssize_t __disk_events_show(unsigned int events, char *buf)
1896{
1897 const char *delim = "";
1898 ssize_t pos = 0;
1899 int i;
1900
1901 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
1902 if (events & (1 << i)) {
1903 pos += sprintf(buf + pos, "%s%s",
1904 delim, disk_events_strs[i]);
1905 delim = " ";
1906 }
1907 if (pos)
1908 pos += sprintf(buf + pos, "\n");
1909 return pos;
1910}
1911
1912static ssize_t disk_events_show(struct device *dev,
1913 struct device_attribute *attr, char *buf)
1914{
1915 struct gendisk *disk = dev_to_disk(dev);
1916
1917 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT))
1918 return 0;
1919
1920 return __disk_events_show(disk->events, buf);
1921}
1922
1923static ssize_t disk_events_async_show(struct device *dev,
1924 struct device_attribute *attr, char *buf)
1925{
1926 return 0;
1927}
1928
1929static ssize_t disk_events_poll_msecs_show(struct device *dev,
1930 struct device_attribute *attr,
1931 char *buf)
1932{
1933 struct gendisk *disk = dev_to_disk(dev);
1934
1935 if (!disk->ev)
1936 return sprintf(buf, "-1\n");
1937
1938 return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
1939}
1940
1941static ssize_t disk_events_poll_msecs_store(struct device *dev,
1942 struct device_attribute *attr,
1943 const char *buf, size_t count)
1944{
1945 struct gendisk *disk = dev_to_disk(dev);
1946 long intv;
1947
1948 if (!count || !sscanf(buf, "%ld", &intv))
1949 return -EINVAL;
1950
1951 if (intv < 0 && intv != -1)
1952 return -EINVAL;
1953
1954 if (!disk->ev)
1955 return -ENODEV;
1956
1957 disk_block_events(disk);
1958 disk->ev->poll_msecs = intv;
1959 __disk_unblock_events(disk, true);
1960
1961 return count;
1962}
1963
1964static const DEVICE_ATTR(events, 0444, disk_events_show, NULL);
1965static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL);
1966static const DEVICE_ATTR(events_poll_msecs, 0644,
1967 disk_events_poll_msecs_show,
1968 disk_events_poll_msecs_store);
1969
1970static const struct attribute *disk_events_attrs[] = {
1971 &dev_attr_events.attr,
1972 &dev_attr_events_async.attr,
1973 &dev_attr_events_poll_msecs.attr,
1974 NULL,
1975};
1976
1977/*
1978 * The default polling interval can be specified by the kernel
1979 * parameter block.events_dfl_poll_msecs which defaults to 0
1980 * (disable). This can also be modified runtime by writing to
1981 * /sys/module/block/parameters/events_dfl_poll_msecs.
1982 */
1983static int disk_events_set_dfl_poll_msecs(const char *val,
1984 const struct kernel_param *kp)
1985{
1986 struct disk_events *ev;
1987 int ret;
1988
1989 ret = param_set_ulong(val, kp);
1990 if (ret < 0)
1991 return ret;
1992
1993 mutex_lock(&disk_events_mutex);
1994
1995 list_for_each_entry(ev, &disk_events, node)
1996 disk_flush_events(ev->disk, 0);
1997
1998 mutex_unlock(&disk_events_mutex);
1999
2000 return 0;
2001}
2002
2003static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
2004 .set = disk_events_set_dfl_poll_msecs,
2005 .get = param_get_ulong,
2006};
2007
2008#undef MODULE_PARAM_PREFIX
2009#define MODULE_PARAM_PREFIX "block."
2010
2011module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
2012 &disk_events_dfl_poll_msecs, 0644);
2013
2014/*
2015 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events.
2016 */
2017static void disk_alloc_events(struct gendisk *disk)
2018{
2019 struct disk_events *ev;
2020
2021 if (!disk->fops->check_events || !disk->events)
2022 return;
2023
2024 ev = kzalloc(sizeof(*ev), GFP_KERNEL);
2025 if (!ev) {
2026 pr_warn("%s: failed to initialize events\n", disk->disk_name);
2027 return;
2028 }
2029
2030 INIT_LIST_HEAD(&ev->node);
2031 ev->disk = disk;
2032 spin_lock_init(&ev->lock);
2033 mutex_init(&ev->block_mutex);
2034 ev->block = 1;
2035 ev->poll_msecs = -1;
2036 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
2037
2038 disk->ev = ev;
2039}
2040
2041static void disk_add_events(struct gendisk *disk)
2042{
2043 /* FIXME: error handling */
2044 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0)
2045 pr_warn("%s: failed to create sysfs files for events\n",
2046 disk->disk_name);
2047
2048 if (!disk->ev)
2049 return;
2050
2051 mutex_lock(&disk_events_mutex);
2052 list_add_tail(&disk->ev->node, &disk_events);
2053 mutex_unlock(&disk_events_mutex);
2054
2055 /*
2056 * Block count is initialized to 1 and the following initial
2057 * unblock kicks it into action.
2058 */
2059 __disk_unblock_events(disk, true);
2060}
2061
2062static void disk_del_events(struct gendisk *disk)
2063{
2064 if (disk->ev) {
2065 disk_block_events(disk);
2066
2067 mutex_lock(&disk_events_mutex);
2068 list_del_init(&disk->ev->node);
2069 mutex_unlock(&disk_events_mutex);
2070 }
2071
2072 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
2073}
2074
2075static void disk_release_events(struct gendisk *disk)
2076{
2077 /* the block count should be 1 from disk_del_events() */
2078 WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
2079 kfree(disk->ev);
2080}