Root/block/genhd.c

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

Archive Download this file



interactive