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