blob: 4cd3433c99ac7d64fe05b72a1061cba01059920a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * gendisk handling
3 */
4
Linus Torvalds1da177e2005-04-16 15:20:36 -07005#include <linux/module.h>
6#include <linux/fs.h>
7#include <linux/genhd.h>
Andrew Mortonb446b602007-02-20 13:57:48 -08008#include <linux/kdev_t.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/kernel.h>
10#include <linux/blkdev.h>
11#include <linux/init.h>
12#include <linux/spinlock.h>
13#include <linux/seq_file.h>
14#include <linux/slab.h>
15#include <linux/kmod.h>
16#include <linux/kobj_map.h>
Christoph Hellwig2ef41632005-05-05 16:15:59 -070017#include <linux/buffer_head.h>
Jes Sorensen58383af2006-02-06 14:12:43 -080018#include <linux/mutex.h>
Tejun Heobcce3de2008-08-25 19:47:22 +090019#include <linux/idr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Adrian Bunkff889722008-03-04 11:23:45 +010021#include "blk.h"
22
Kay Sieversedfaa7c2007-05-21 22:08:01 +020023static DEFINE_MUTEX(block_class_lock);
24#ifndef CONFIG_SYSFS_DEPRECATED
25struct kobject *block_depr;
26#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Tejun Heobcce3de2008-08-25 19:47:22 +090028/* for extended dynamic devt allocation, currently only one major is used */
29#define MAX_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
Adrian Bunk1826ead2008-03-04 11:23:46 +010037static struct device_type disk_type;
38
Tejun Heoe71bf0d2008-09-03 09:03:02 +020039/**
40 * disk_get_part - get partition
41 * @disk: disk to look partition from
42 * @partno: partition number
43 *
44 * Look for partition @partno from @disk. If found, increment
45 * reference count and return it.
46 *
47 * CONTEXT:
48 * Don't care.
49 *
50 * RETURNS:
51 * Pointer to the found partition on success, NULL if not found.
52 */
53struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
54{
Tejun Heo540eed52008-08-25 19:56:15 +090055 struct hd_struct *part = NULL;
56 struct disk_part_tbl *ptbl;
Tejun Heoe71bf0d2008-09-03 09:03:02 +020057
Tejun Heo540eed52008-08-25 19:56:15 +090058 if (unlikely(partno < 0))
Tejun Heoe71bf0d2008-09-03 09:03:02 +020059 return NULL;
Tejun Heo540eed52008-08-25 19:56:15 +090060
Tejun Heoe71bf0d2008-09-03 09:03:02 +020061 rcu_read_lock();
Tejun Heo540eed52008-08-25 19:56:15 +090062
63 ptbl = rcu_dereference(disk->part_tbl);
64 if (likely(partno < ptbl->len)) {
65 part = rcu_dereference(ptbl->part[partno]);
66 if (part)
67 get_device(part_to_dev(part));
68 }
69
Tejun Heoe71bf0d2008-09-03 09:03:02 +020070 rcu_read_unlock();
71
72 return part;
73}
74EXPORT_SYMBOL_GPL(disk_get_part);
75
76/**
77 * disk_part_iter_init - initialize partition iterator
78 * @piter: iterator to initialize
79 * @disk: disk to iterate over
80 * @flags: DISK_PITER_* flags
81 *
82 * Initialize @piter so that it iterates over partitions of @disk.
83 *
84 * CONTEXT:
85 * Don't care.
86 */
87void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
88 unsigned int flags)
89{
Tejun Heo540eed52008-08-25 19:56:15 +090090 struct disk_part_tbl *ptbl;
91
92 rcu_read_lock();
93 ptbl = rcu_dereference(disk->part_tbl);
94
Tejun Heoe71bf0d2008-09-03 09:03:02 +020095 piter->disk = disk;
96 piter->part = NULL;
97
98 if (flags & DISK_PITER_REVERSE)
Tejun Heo540eed52008-08-25 19:56:15 +090099 piter->idx = ptbl->len - 1;
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200100 else if (flags & DISK_PITER_INCL_PART0)
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200101 piter->idx = 0;
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200102 else
103 piter->idx = 1;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200104
105 piter->flags = flags;
Tejun Heo540eed52008-08-25 19:56:15 +0900106
107 rcu_read_unlock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200108}
109EXPORT_SYMBOL_GPL(disk_part_iter_init);
110
111/**
112 * disk_part_iter_next - proceed iterator to the next partition and return it
113 * @piter: iterator of interest
114 *
115 * Proceed @piter to the next partition and return it.
116 *
117 * CONTEXT:
118 * Don't care.
119 */
120struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
121{
Tejun Heo540eed52008-08-25 19:56:15 +0900122 struct disk_part_tbl *ptbl;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200123 int inc, end;
124
125 /* put the last partition */
126 disk_put_part(piter->part);
127 piter->part = NULL;
128
Tejun Heo540eed52008-08-25 19:56:15 +0900129 /* get part_tbl */
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200130 rcu_read_lock();
Tejun Heo540eed52008-08-25 19:56:15 +0900131 ptbl = rcu_dereference(piter->disk->part_tbl);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200132
133 /* determine iteration parameters */
134 if (piter->flags & DISK_PITER_REVERSE) {
135 inc = -1;
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200136 if (piter->flags & DISK_PITER_INCL_PART0)
137 end = -1;
138 else
139 end = 0;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200140 } else {
141 inc = 1;
Tejun Heo540eed52008-08-25 19:56:15 +0900142 end = ptbl->len;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200143 }
144
145 /* iterate to the next partition */
146 for (; piter->idx != end; piter->idx += inc) {
147 struct hd_struct *part;
148
Tejun Heo540eed52008-08-25 19:56:15 +0900149 part = rcu_dereference(ptbl->part[piter->idx]);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200150 if (!part)
151 continue;
152 if (!(piter->flags & DISK_PITER_INCL_EMPTY) && !part->nr_sects)
153 continue;
154
Tejun Heoed9e1982008-08-25 19:56:05 +0900155 get_device(part_to_dev(part));
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200156 piter->part = part;
157 piter->idx += inc;
158 break;
159 }
160
161 rcu_read_unlock();
162
163 return piter->part;
164}
165EXPORT_SYMBOL_GPL(disk_part_iter_next);
166
167/**
168 * disk_part_iter_exit - finish up partition iteration
169 * @piter: iter of interest
170 *
171 * Called when iteration is over. Cleans up @piter.
172 *
173 * CONTEXT:
174 * Don't care.
175 */
176void disk_part_iter_exit(struct disk_part_iter *piter)
177{
178 disk_put_part(piter->part);
179 piter->part = NULL;
180}
181EXPORT_SYMBOL_GPL(disk_part_iter_exit);
182
183/**
184 * disk_map_sector_rcu - map sector to partition
185 * @disk: gendisk of interest
186 * @sector: sector to map
187 *
188 * Find out which partition @sector maps to on @disk. This is
189 * primarily used for stats accounting.
190 *
191 * CONTEXT:
192 * RCU read locked. The returned partition pointer is valid only
193 * while preemption is disabled.
194 *
195 * RETURNS:
Tejun Heo074a7ac2008-08-25 19:56:14 +0900196 * Found partition on success, part0 is returned if no partition matches
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200197 */
198struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
199{
Tejun Heo540eed52008-08-25 19:56:15 +0900200 struct disk_part_tbl *ptbl;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200201 int i;
202
Tejun Heo540eed52008-08-25 19:56:15 +0900203 ptbl = rcu_dereference(disk->part_tbl);
204
205 for (i = 1; i < ptbl->len; i++) {
206 struct hd_struct *part = rcu_dereference(ptbl->part[i]);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200207
208 if (part && part->start_sect <= sector &&
209 sector < part->start_sect + part->nr_sects)
210 return part;
211 }
Tejun Heo074a7ac2008-08-25 19:56:14 +0900212 return &disk->part0;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200213}
214EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216/*
217 * Can be deleted altogether. Later.
218 *
219 */
220static struct blk_major_name {
221 struct blk_major_name *next;
222 int major;
223 char name[16];
Joe Korty68eef3b2006-03-31 02:30:32 -0800224} *major_names[BLKDEV_MAJOR_HASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
226/* index in the above - for now: assume no multimajor ranges */
227static inline int major_to_index(int major)
228{
Joe Korty68eef3b2006-03-31 02:30:32 -0800229 return major % BLKDEV_MAJOR_HASH_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230}
231
Joe Korty68eef3b2006-03-31 02:30:32 -0800232#ifdef CONFIG_PROC_FS
Tejun Heocf771cb2008-09-03 09:01:09 +0200233void blkdev_show(struct seq_file *seqf, off_t offset)
Neil Horman7170be52006-01-14 13:20:38 -0800234{
Joe Korty68eef3b2006-03-31 02:30:32 -0800235 struct blk_major_name *dp;
Neil Horman7170be52006-01-14 13:20:38 -0800236
Joe Korty68eef3b2006-03-31 02:30:32 -0800237 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200238 mutex_lock(&block_class_lock);
Joe Korty68eef3b2006-03-31 02:30:32 -0800239 for (dp = major_names[offset]; dp; dp = dp->next)
Tejun Heocf771cb2008-09-03 09:01:09 +0200240 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200241 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
Joe Korty68eef3b2006-03-31 02:30:32 -0800244#endif /* CONFIG_PROC_FS */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246int register_blkdev(unsigned int major, const char *name)
247{
248 struct blk_major_name **n, *p;
249 int index, ret = 0;
250
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200251 mutex_lock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
253 /* temporary */
254 if (major == 0) {
255 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
256 if (major_names[index] == NULL)
257 break;
258 }
259
260 if (index == 0) {
261 printk("register_blkdev: failed to get major for %s\n",
262 name);
263 ret = -EBUSY;
264 goto out;
265 }
266 major = index;
267 ret = major;
268 }
269
270 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
271 if (p == NULL) {
272 ret = -ENOMEM;
273 goto out;
274 }
275
276 p->major = major;
277 strlcpy(p->name, name, sizeof(p->name));
278 p->next = NULL;
279 index = major_to_index(major);
280
281 for (n = &major_names[index]; *n; n = &(*n)->next) {
282 if ((*n)->major == major)
283 break;
284 }
285 if (!*n)
286 *n = p;
287 else
288 ret = -EBUSY;
289
290 if (ret < 0) {
291 printk("register_blkdev: cannot get major %d for %s\n",
292 major, name);
293 kfree(p);
294 }
295out:
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200296 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 return ret;
298}
299
300EXPORT_SYMBOL(register_blkdev);
301
Akinobu Mitaf4480242007-07-17 04:03:47 -0700302void unregister_blkdev(unsigned int major, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
304 struct blk_major_name **n;
305 struct blk_major_name *p = NULL;
306 int index = major_to_index(major);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200308 mutex_lock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 for (n = &major_names[index]; *n; n = &(*n)->next)
310 if ((*n)->major == major)
311 break;
Akinobu Mita294462a2007-07-17 04:03:45 -0700312 if (!*n || strcmp((*n)->name, name)) {
313 WARN_ON(1);
Akinobu Mita294462a2007-07-17 04:03:45 -0700314 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 p = *n;
316 *n = p->next;
317 }
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200318 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320}
321
322EXPORT_SYMBOL(unregister_blkdev);
323
324static struct kobj_map *bdev_map;
325
Tejun Heobcce3de2008-08-25 19:47:22 +0900326/**
Tejun Heo870d6652008-08-25 19:47:25 +0900327 * blk_mangle_minor - scatter minor numbers apart
328 * @minor: minor number to mangle
329 *
330 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
331 * is enabled. Mangling twice gives the original value.
332 *
333 * RETURNS:
334 * Mangled value.
335 *
336 * CONTEXT:
337 * Don't care.
338 */
339static int blk_mangle_minor(int minor)
340{
341#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
342 int i;
343
344 for (i = 0; i < MINORBITS / 2; i++) {
345 int low = minor & (1 << i);
346 int high = minor & (1 << (MINORBITS - 1 - i));
347 int distance = MINORBITS - 1 - 2 * i;
348
349 minor ^= low | high; /* clear both bits */
350 low <<= distance; /* swap the positions */
351 high >>= distance;
352 minor |= low | high; /* and set */
353 }
354#endif
355 return minor;
356}
357
358/**
Tejun Heobcce3de2008-08-25 19:47:22 +0900359 * blk_alloc_devt - allocate a dev_t for a partition
360 * @part: partition to allocate dev_t for
361 * @gfp_mask: memory allocation flag
362 * @devt: out parameter for resulting dev_t
363 *
364 * Allocate a dev_t for block device.
365 *
366 * RETURNS:
367 * 0 on success, allocated dev_t is returned in *@devt. -errno on
368 * failure.
369 *
370 * CONTEXT:
371 * Might sleep.
372 */
373int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
374{
375 struct gendisk *disk = part_to_disk(part);
376 int idx, rc;
377
378 /* in consecutive minor range? */
379 if (part->partno < disk->minors) {
380 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
381 return 0;
382 }
383
384 /* allocate ext devt */
385 do {
386 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
387 return -ENOMEM;
388 rc = idr_get_new(&ext_devt_idr, part, &idx);
389 } while (rc == -EAGAIN);
390
391 if (rc)
392 return rc;
393
394 if (idx > MAX_EXT_DEVT) {
395 idr_remove(&ext_devt_idr, idx);
396 return -EBUSY;
397 }
398
Tejun Heo870d6652008-08-25 19:47:25 +0900399 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
Tejun Heobcce3de2008-08-25 19:47:22 +0900400 return 0;
401}
402
403/**
404 * blk_free_devt - free a dev_t
405 * @devt: dev_t to free
406 *
407 * Free @devt which was allocated using blk_alloc_devt().
408 *
409 * CONTEXT:
410 * Might sleep.
411 */
412void blk_free_devt(dev_t devt)
413{
414 might_sleep();
415
416 if (devt == MKDEV(0, 0))
417 return;
418
419 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
420 mutex_lock(&ext_devt_mutex);
Tejun Heo870d6652008-08-25 19:47:25 +0900421 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
Tejun Heobcce3de2008-08-25 19:47:22 +0900422 mutex_unlock(&ext_devt_mutex);
423 }
424}
425
Tejun Heo1f014292008-08-25 19:47:23 +0900426static char *bdevt_str(dev_t devt, char *buf)
427{
428 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
429 char tbuf[BDEVT_SIZE];
430 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
431 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
432 } else
433 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
434
435 return buf;
436}
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438/*
439 * Register device numbers dev..(dev+range-1)
440 * range must be nonzero
441 * The hash chain is sorted on range, so that subranges can override.
442 */
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200443void blk_register_region(dev_t devt, unsigned long range, struct module *module,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 struct kobject *(*probe)(dev_t, int *, void *),
445 int (*lock)(dev_t, void *), void *data)
446{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200447 kobj_map(bdev_map, devt, range, module, probe, lock, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
450EXPORT_SYMBOL(blk_register_region);
451
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200452void blk_unregister_region(dev_t devt, unsigned long range)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200454 kobj_unmap(bdev_map, devt, range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
456
457EXPORT_SYMBOL(blk_unregister_region);
458
Tejun Heocf771cb2008-09-03 09:01:09 +0200459static struct kobject *exact_match(dev_t devt, int *partno, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
461 struct gendisk *p = data;
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200462
Tejun Heoed9e1982008-08-25 19:56:05 +0900463 return &disk_to_dev(p)->kobj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200466static int exact_lock(dev_t devt, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 struct gendisk *p = data;
469
470 if (!get_disk(p))
471 return -1;
472 return 0;
473}
474
475/**
476 * add_disk - add partitioning information to kernel list
477 * @disk: per-device partitioning information
478 *
479 * This function registers the partitioning information in @disk
480 * with the kernel.
Tejun Heo3e1a7ff2008-08-25 19:56:17 +0900481 *
482 * FIXME: error handling
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 */
484void add_disk(struct gendisk *disk)
485{
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700486 struct backing_dev_info *bdi;
Tejun Heo3e1a7ff2008-08-25 19:56:17 +0900487 dev_t devt;
Greg Kroah-Hartman6ffeea72008-05-22 17:21:08 -0400488 int retval;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700489
Tejun Heo3e1a7ff2008-08-25 19:56:17 +0900490 /* minors == 0 indicates to use ext devt from part0 and should
491 * be accompanied with EXT_DEVT flag. Make sure all
492 * parameters make sense.
493 */
494 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
495 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 disk->flags |= GENHD_FL_UP;
Tejun Heo3e1a7ff2008-08-25 19:56:17 +0900498
499 retval = blk_alloc_devt(&disk->part0, &devt);
500 if (retval) {
501 WARN_ON(1);
502 return;
503 }
504 disk_to_dev(disk)->devt = devt;
505
506 /* ->major and ->first_minor aren't supposed to be
507 * dereferenced from here on, but set them just in case.
508 */
509 disk->major = MAJOR(devt);
510 disk->first_minor = MINOR(devt);
511
Tejun Heof331c022008-09-03 09:01:48 +0200512 blk_register_region(disk_devt(disk), disk->minors, NULL,
513 exact_match, exact_lock, disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 register_disk(disk);
515 blk_register_queue(disk);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700516
517 bdi = &disk->queue->backing_dev_info;
Tejun Heof331c022008-09-03 09:01:48 +0200518 bdi_register_dev(bdi, disk_devt(disk));
Tejun Heoed9e1982008-08-25 19:56:05 +0900519 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
520 "bdi");
Greg Kroah-Hartman6ffeea72008-05-22 17:21:08 -0400521 WARN_ON(retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
524EXPORT_SYMBOL(add_disk);
525EXPORT_SYMBOL(del_gendisk); /* in partitions/check.c */
526
527void unlink_gendisk(struct gendisk *disk)
528{
Tejun Heoed9e1982008-08-25 19:56:05 +0900529 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700530 bdi_unregister(&disk->queue->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 blk_unregister_queue(disk);
Tejun Heof331c022008-09-03 09:01:48 +0200532 blk_unregister_region(disk_devt(disk), disk->minors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533}
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535/**
536 * get_gendisk - get partitioning information for a given device
Randy Dunlap710027a2008-08-19 20:13:11 +0200537 * @devt: device to get partitioning information for
538 * @part: returned partition index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 *
540 * This function gets the structure containing partitioning
Randy Dunlap710027a2008-08-19 20:13:11 +0200541 * information for the given device @devt.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 */
Tejun Heocf771cb2008-09-03 09:01:09 +0200543struct gendisk *get_gendisk(dev_t devt, int *partno)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544{
Tejun Heobcce3de2008-08-25 19:47:22 +0900545 struct gendisk *disk = NULL;
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200546
Tejun Heobcce3de2008-08-25 19:47:22 +0900547 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
548 struct kobject *kobj;
549
550 kobj = kobj_lookup(bdev_map, devt, partno);
551 if (kobj)
552 disk = dev_to_disk(kobj_to_dev(kobj));
553 } else {
554 struct hd_struct *part;
555
556 mutex_lock(&ext_devt_mutex);
Tejun Heo870d6652008-08-25 19:47:25 +0900557 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
Tejun Heobcce3de2008-08-25 19:47:22 +0900558 if (part && get_disk(part_to_disk(part))) {
559 *partno = part->partno;
560 disk = part_to_disk(part);
561 }
562 mutex_unlock(&ext_devt_mutex);
563 }
564
565 return disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566}
567
Tejun Heof331c022008-09-03 09:01:48 +0200568/**
569 * bdget_disk - do bdget() by gendisk and partition number
570 * @disk: gendisk of interest
571 * @partno: partition number
572 *
573 * Find partition @partno from @disk, do bdget() on it.
574 *
575 * CONTEXT:
576 * Don't care.
577 *
578 * RETURNS:
579 * Resulting block_device on success, NULL on failure.
580 */
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200581struct block_device *bdget_disk(struct gendisk *disk, int partno)
Tejun Heof331c022008-09-03 09:01:48 +0200582{
Tejun Heo548b10e2008-08-29 09:01:47 +0200583 struct hd_struct *part;
584 struct block_device *bdev = NULL;
Tejun Heof331c022008-09-03 09:01:48 +0200585
Tejun Heo548b10e2008-08-29 09:01:47 +0200586 part = disk_get_part(disk, partno);
Tejun Heo2bbedcb2008-08-29 11:41:51 +0200587 if (part)
Tejun Heo548b10e2008-08-29 09:01:47 +0200588 bdev = bdget(part_devt(part));
589 disk_put_part(part);
Tejun Heof331c022008-09-03 09:01:48 +0200590
Tejun Heo548b10e2008-08-29 09:01:47 +0200591 return bdev;
Tejun Heof331c022008-09-03 09:01:48 +0200592}
593EXPORT_SYMBOL(bdget_disk);
594
Dave Gilbertdd2a3452007-05-09 02:33:24 -0700595/*
596 * print a full list of all partitions - intended for places where the root
597 * filesystem can't be mounted and thus to give the victim some idea of what
598 * went wrong
599 */
600void __init printk_all_partitions(void)
601{
Tejun Heodef4e382008-09-03 08:57:12 +0200602 struct class_dev_iter iter;
603 struct device *dev;
604
605 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
606 while ((dev = class_dev_iter_next(&iter))) {
607 struct gendisk *disk = dev_to_disk(dev);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200608 struct disk_part_iter piter;
609 struct hd_struct *part;
Tejun Heo1f014292008-08-25 19:47:23 +0900610 char name_buf[BDEVNAME_SIZE];
611 char devt_buf[BDEVT_SIZE];
Tejun Heodef4e382008-09-03 08:57:12 +0200612
613 /*
614 * Don't show empty devices or things that have been
615 * surpressed
616 */
617 if (get_capacity(disk) == 0 ||
618 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
619 continue;
620
621 /*
622 * Note, unlike /proc/partitions, I am showing the
623 * numbers in hex - the same format as the root=
624 * option takes.
625 */
Tejun Heo074a7ac2008-08-25 19:56:14 +0900626 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
627 while ((part = disk_part_iter_next(&piter))) {
628 bool is_part0 = part == &disk->part0;
Tejun Heodef4e382008-09-03 08:57:12 +0200629
Tejun Heo074a7ac2008-08-25 19:56:14 +0900630 printk("%s%s %10llu %s", is_part0 ? "" : " ",
Tejun Heo1f014292008-08-25 19:47:23 +0900631 bdevt_str(part_devt(part), devt_buf),
Tejun Heof331c022008-09-03 09:01:48 +0200632 (unsigned long long)part->nr_sects >> 1,
Tejun Heo1f014292008-08-25 19:47:23 +0900633 disk_name(disk, part->partno, name_buf));
Tejun Heo074a7ac2008-08-25 19:56:14 +0900634 if (is_part0) {
635 if (disk->driverfs_dev != NULL &&
636 disk->driverfs_dev->driver != NULL)
637 printk(" driver: %s\n",
638 disk->driverfs_dev->driver->name);
639 else
640 printk(" (driver?)\n");
641 } else
642 printk("\n");
643 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200644 disk_part_iter_exit(&piter);
Tejun Heodef4e382008-09-03 08:57:12 +0200645 }
646 class_dev_iter_exit(&iter);
Dave Gilbertdd2a3452007-05-09 02:33:24 -0700647}
648
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649#ifdef CONFIG_PROC_FS
650/* iterator */
Tejun Heodef4e382008-09-03 08:57:12 +0200651static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
Greg Kroah-Hartman68c4d4a2008-05-22 17:21:08 -0400652{
Tejun Heodef4e382008-09-03 08:57:12 +0200653 loff_t skip = *pos;
654 struct class_dev_iter *iter;
655 struct device *dev;
Greg Kroah-Hartman68c4d4a2008-05-22 17:21:08 -0400656
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200657 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
Tejun Heodef4e382008-09-03 08:57:12 +0200658 if (!iter)
659 return ERR_PTR(-ENOMEM);
660
661 seqf->private = iter;
662 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
663 do {
664 dev = class_dev_iter_next(iter);
665 if (!dev)
666 return NULL;
667 } while (skip--);
668
669 return dev_to_disk(dev);
Greg Kroah-Hartman68c4d4a2008-05-22 17:21:08 -0400670}
671
Tejun Heodef4e382008-09-03 08:57:12 +0200672static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200674 struct device *dev;
Greg Kroah-Hartman66c64af2008-05-22 17:21:08 -0400675
Tejun Heodef4e382008-09-03 08:57:12 +0200676 (*pos)++;
677 dev = class_dev_iter_next(seqf->private);
Tejun Heo2ac3cee2008-09-03 08:53:37 +0200678 if (dev)
Greg Kroah-Hartman68c4d4a2008-05-22 17:21:08 -0400679 return dev_to_disk(dev);
Tejun Heo2ac3cee2008-09-03 08:53:37 +0200680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 return NULL;
682}
683
Tejun Heodef4e382008-09-03 08:57:12 +0200684static void disk_seqf_stop(struct seq_file *seqf, void *v)
Greg Kroah-Hartman27f30252008-05-22 17:21:08 -0400685{
Tejun Heodef4e382008-09-03 08:57:12 +0200686 struct class_dev_iter *iter = seqf->private;
Greg Kroah-Hartman27f30252008-05-22 17:21:08 -0400687
Tejun Heodef4e382008-09-03 08:57:12 +0200688 /* stop is called even after start failed :-( */
689 if (iter) {
690 class_dev_iter_exit(iter);
691 kfree(iter);
Kay Sievers5c0ef6d2008-08-16 14:30:30 +0200692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
Tejun Heodef4e382008-09-03 08:57:12 +0200695static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Tejun Heodef4e382008-09-03 08:57:12 +0200697 static void *p;
698
699 p = disk_seqf_start(seqf, pos);
Tejun Heo243294d2008-09-04 09:17:31 +0200700 if (!IS_ERR(p) && p && !*pos)
Tejun Heodef4e382008-09-03 08:57:12 +0200701 seq_puts(seqf, "major minor #blocks name\n\n");
702 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703}
704
Tejun Heocf771cb2008-09-03 09:01:09 +0200705static int show_partition(struct seq_file *seqf, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
707 struct gendisk *sgp = v;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200708 struct disk_part_iter piter;
709 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 char buf[BDEVNAME_SIZE];
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 /* Don't show non-partitionable removeable devices or empty devices */
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200713 if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
Tejun Heof331c022008-09-03 09:01:48 +0200714 (sgp->flags & GENHD_FL_REMOVABLE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 return 0;
716 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
717 return 0;
718
719 /* show the full disk and all non-0 size partitions of it */
Tejun Heo074a7ac2008-08-25 19:56:14 +0900720 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200721 while ((part = disk_part_iter_next(&piter)))
Tejun Heo1f014292008-08-25 19:47:23 +0900722 seq_printf(seqf, "%4d %7d %10llu %s\n",
Tejun Heof331c022008-09-03 09:01:48 +0200723 MAJOR(part_devt(part)), MINOR(part_devt(part)),
724 (unsigned long long)part->nr_sects >> 1,
725 disk_name(sgp, part->partno, buf));
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200726 disk_part_iter_exit(&piter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
728 return 0;
729}
730
Jan Engelhardt12f32bb2008-01-29 20:57:51 +0100731const struct seq_operations partitions_op = {
Tejun Heodef4e382008-09-03 08:57:12 +0200732 .start = show_partition_start,
733 .next = disk_seqf_next,
734 .stop = disk_seqf_stop,
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200735 .show = show_partition
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736};
737#endif
738
739
Tejun Heocf771cb2008-09-03 09:01:09 +0200740static struct kobject *base_probe(dev_t devt, int *partno, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200742 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 /* Make old-style 2.4 aliases work */
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200744 request_module("block-major-%d", MAJOR(devt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return NULL;
746}
747
748static int __init genhd_device_init(void)
749{
Dan Williamse105b8b2008-04-21 10:51:07 -0700750 int error;
751
752 block_class.dev_kobj = sysfs_dev_block_kobj;
753 error = class_register(&block_class);
Roland McGrathee27a552008-03-11 17:13:15 -0700754 if (unlikely(error))
755 return error;
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200756 bdev_map = kobj_map_init(base_probe, &block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 blk_dev_init();
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200758
759#ifndef CONFIG_SYSFS_DEPRECATED
760 /* create top-level block dir */
761 block_depr = kobject_create_and_add("block", NULL);
762#endif
Greg Kroah-Hartman830d3cf2007-11-06 10:36:58 -0800763 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764}
765
766subsys_initcall(genhd_device_init);
767
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200768static ssize_t disk_range_show(struct device *dev,
769 struct device_attribute *attr, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200771 struct gendisk *disk = dev_to_disk(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200773 return sprintf(buf, "%d\n", disk->minors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774}
775
Tejun Heo1f014292008-08-25 19:47:23 +0900776static ssize_t disk_ext_range_show(struct device *dev,
777 struct device_attribute *attr, char *buf)
778{
779 struct gendisk *disk = dev_to_disk(dev);
780
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200781 return sprintf(buf, "%d\n", disk_max_parts(disk));
Tejun Heo1f014292008-08-25 19:47:23 +0900782}
783
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200784static ssize_t disk_removable_show(struct device *dev,
785 struct device_attribute *attr, char *buf)
Kay Sieversa7fd6702005-10-01 14:49:43 +0200786{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200787 struct gendisk *disk = dev_to_disk(dev);
Kay Sieversa7fd6702005-10-01 14:49:43 +0200788
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200789 return sprintf(buf, "%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200791}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792
Kay Sievers1c9ce522008-06-13 09:41:00 +0200793static ssize_t disk_ro_show(struct device *dev,
794 struct device_attribute *attr, char *buf)
795{
796 struct gendisk *disk = dev_to_disk(dev);
797
Tejun Heob7db9952008-08-25 19:56:10 +0900798 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
Kay Sievers1c9ce522008-06-13 09:41:00 +0200799}
800
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200801static ssize_t disk_capability_show(struct device *dev,
802 struct device_attribute *attr, char *buf)
Kristen Carlson Accardi86ce18d2007-05-23 13:57:38 -0700803{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200804 struct gendisk *disk = dev_to_disk(dev);
805
806 return sprintf(buf, "%x\n", disk->flags);
Kristen Carlson Accardi86ce18d2007-05-23 13:57:38 -0700807}
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200808
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200809static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
Tejun Heo1f014292008-08-25 19:47:23 +0900810static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200811static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
Kay Sievers1c9ce522008-06-13 09:41:00 +0200812static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
Tejun Heoe5610522008-08-25 19:56:09 +0900813static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200814static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
Tejun Heo074a7ac2008-08-25 19:56:14 +0900815static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800816#ifdef CONFIG_FAIL_MAKE_REQUEST
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200817static struct device_attribute dev_attr_fail =
Tejun Heoeddb2e22008-08-25 19:56:13 +0900818 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800819#endif
Jens Axboe581d4e22008-09-14 05:56:33 -0700820#ifdef CONFIG_FAIL_IO_TIMEOUT
821static struct device_attribute dev_attr_fail_timeout =
822 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
823 part_timeout_store);
824#endif
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200825
826static struct attribute *disk_attrs[] = {
827 &dev_attr_range.attr,
Tejun Heo1f014292008-08-25 19:47:23 +0900828 &dev_attr_ext_range.attr,
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200829 &dev_attr_removable.attr,
Kay Sievers1c9ce522008-06-13 09:41:00 +0200830 &dev_attr_ro.attr,
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200831 &dev_attr_size.attr,
832 &dev_attr_capability.attr,
833 &dev_attr_stat.attr,
834#ifdef CONFIG_FAIL_MAKE_REQUEST
835 &dev_attr_fail.attr,
836#endif
Jens Axboe581d4e22008-09-14 05:56:33 -0700837#ifdef CONFIG_FAIL_IO_TIMEOUT
838 &dev_attr_fail_timeout.attr,
839#endif
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200840 NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841};
842
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200843static struct attribute_group disk_attr_group = {
844 .attrs = disk_attrs,
845};
846
847static struct attribute_group *disk_attr_groups[] = {
848 &disk_attr_group,
849 NULL
850};
851
Tejun Heo540eed52008-08-25 19:56:15 +0900852static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
853{
854 struct disk_part_tbl *ptbl =
855 container_of(head, struct disk_part_tbl, rcu_head);
856
857 kfree(ptbl);
858}
859
860/**
861 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
862 * @disk: disk to replace part_tbl for
863 * @new_ptbl: new part_tbl to install
864 *
865 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
866 * original ptbl is freed using RCU callback.
867 *
868 * LOCKING:
869 * Matching bd_mutx locked.
870 */
871static void disk_replace_part_tbl(struct gendisk *disk,
872 struct disk_part_tbl *new_ptbl)
873{
874 struct disk_part_tbl *old_ptbl = disk->part_tbl;
875
876 rcu_assign_pointer(disk->part_tbl, new_ptbl);
877 if (old_ptbl)
878 call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
879}
880
881/**
882 * disk_expand_part_tbl - expand disk->part_tbl
883 * @disk: disk to expand part_tbl for
884 * @partno: expand such that this partno can fit in
885 *
886 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
887 * uses RCU to allow unlocked dereferencing for stats and other stuff.
888 *
889 * LOCKING:
890 * Matching bd_mutex locked, might sleep.
891 *
892 * RETURNS:
893 * 0 on success, -errno on failure.
894 */
895int disk_expand_part_tbl(struct gendisk *disk, int partno)
896{
897 struct disk_part_tbl *old_ptbl = disk->part_tbl;
898 struct disk_part_tbl *new_ptbl;
899 int len = old_ptbl ? old_ptbl->len : 0;
900 int target = partno + 1;
901 size_t size;
902 int i;
903
904 /* disk_max_parts() is zero during initialization, ignore if so */
905 if (disk_max_parts(disk) && target > disk_max_parts(disk))
906 return -EINVAL;
907
908 if (target <= len)
909 return 0;
910
911 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
912 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
913 if (!new_ptbl)
914 return -ENOMEM;
915
916 INIT_RCU_HEAD(&new_ptbl->rcu_head);
917 new_ptbl->len = target;
918
919 for (i = 0; i < len; i++)
920 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
921
922 disk_replace_part_tbl(disk, new_ptbl);
923 return 0;
924}
925
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200926static void disk_release(struct device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200928 struct gendisk *disk = dev_to_disk(dev);
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 kfree(disk->random);
Tejun Heo540eed52008-08-25 19:56:15 +0900931 disk_replace_part_tbl(disk, NULL);
Tejun Heo074a7ac2008-08-25 19:56:14 +0900932 free_part_stats(&disk->part0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 kfree(disk);
934}
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200935struct class block_class = {
936 .name = "block",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937};
938
Adrian Bunk1826ead2008-03-04 11:23:46 +0100939static struct device_type disk_type = {
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200940 .name = "disk",
941 .groups = disk_attr_groups,
942 .release = disk_release,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943};
944
Randy Dunlapa6e2ba82008-05-23 09:44:11 -0700945#ifdef CONFIG_PROC_FS
Tejun Heocf771cb2008-09-03 09:01:09 +0200946/*
947 * aggregate disk stat collector. Uses the same stats that the sysfs
948 * entries do, above, but makes them available through one seq_file.
949 *
950 * The output looks suspiciously like /proc/partitions with a bunch of
951 * extra fields.
952 */
953static int diskstats_show(struct seq_file *seqf, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
955 struct gendisk *gp = v;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200956 struct disk_part_iter piter;
957 struct hd_struct *hd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 char buf[BDEVNAME_SIZE];
Tejun Heoc9959052008-08-25 19:47:21 +0900959 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
961 /*
Tejun Heoed9e1982008-08-25 19:56:05 +0900962 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
Tejun Heocf771cb2008-09-03 09:01:09 +0200963 seq_puts(seqf, "major minor name"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 " rio rmerge rsect ruse wio wmerge "
965 "wsect wuse running use aveq"
966 "\n\n");
967 */
968
Tejun Heo074a7ac2008-08-25 19:56:14 +0900969 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_PART0);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200970 while ((hd = disk_part_iter_next(&piter))) {
Tejun Heo074a7ac2008-08-25 19:56:14 +0900971 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +0900972 part_round_stats(cpu, hd);
Tejun Heo074a7ac2008-08-25 19:56:14 +0900973 part_stat_unlock();
Tejun Heo1f014292008-08-25 19:47:23 +0900974 seq_printf(seqf, "%4d %7d %s %lu %lu %llu "
Jerome Marchand28f39d52008-02-08 11:04:56 +0100975 "%u %lu %lu %llu %u %u %u %u\n",
Tejun Heof331c022008-09-03 09:01:48 +0200976 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
977 disk_name(gp, hd->partno, buf),
Jerome Marchand28f39d52008-02-08 11:04:56 +0100978 part_stat_read(hd, ios[0]),
979 part_stat_read(hd, merges[0]),
980 (unsigned long long)part_stat_read(hd, sectors[0]),
981 jiffies_to_msecs(part_stat_read(hd, ticks[0])),
982 part_stat_read(hd, ios[1]),
983 part_stat_read(hd, merges[1]),
984 (unsigned long long)part_stat_read(hd, sectors[1]),
985 jiffies_to_msecs(part_stat_read(hd, ticks[1])),
986 hd->in_flight,
987 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
988 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
989 );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200991 disk_part_iter_exit(&piter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
993 return 0;
994}
995
Jan Engelhardt12f32bb2008-01-29 20:57:51 +0100996const struct seq_operations diskstats_op = {
Tejun Heodef4e382008-09-03 08:57:12 +0200997 .start = disk_seqf_start,
998 .next = disk_seqf_next,
999 .stop = disk_seqf_stop,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 .show = diskstats_show
1001};
Randy Dunlapa6e2ba82008-05-23 09:44:11 -07001002#endif /* CONFIG_PROC_FS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001004static void media_change_notify_thread(struct work_struct *work)
1005{
1006 struct gendisk *gd = container_of(work, struct gendisk, async_notify);
1007 char event[] = "MEDIA_CHANGE=1";
1008 char *envp[] = { event, NULL };
1009
1010 /*
1011 * set enviroment vars to indicate which event this is for
1012 * so that user space will know to go check the media status.
1013 */
Tejun Heoed9e1982008-08-25 19:56:05 +09001014 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001015 put_device(gd->driverfs_dev);
1016}
1017
Adrian Bunk1826ead2008-03-04 11:23:46 +01001018#if 0
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001019void genhd_media_change_notify(struct gendisk *disk)
1020{
1021 get_device(disk->driverfs_dev);
1022 schedule_work(&disk->async_notify);
1023}
1024EXPORT_SYMBOL_GPL(genhd_media_change_notify);
Adrian Bunk1826ead2008-03-04 11:23:46 +01001025#endif /* 0 */
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001026
Tejun Heocf771cb2008-09-03 09:01:09 +02001027dev_t blk_lookup_devt(const char *name, int partno)
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001028{
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001029 dev_t devt = MKDEV(0, 0);
Tejun Heodef4e382008-09-03 08:57:12 +02001030 struct class_dev_iter iter;
1031 struct device *dev;
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001032
Tejun Heodef4e382008-09-03 08:57:12 +02001033 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1034 while ((dev = class_dev_iter_next(&iter))) {
1035 struct gendisk *disk = dev_to_disk(dev);
Tejun Heo548b10e2008-08-29 09:01:47 +02001036 struct hd_struct *part;
Tejun Heodef4e382008-09-03 08:57:12 +02001037
Tejun Heof331c022008-09-03 09:01:48 +02001038 if (strcmp(dev->bus_id, name))
1039 continue;
Tejun Heof331c022008-09-03 09:01:48 +02001040
Tejun Heo548b10e2008-08-29 09:01:47 +02001041 part = disk_get_part(disk, partno);
Tejun Heo2bbedcb2008-08-29 11:41:51 +02001042 if (part) {
Tejun Heof331c022008-09-03 09:01:48 +02001043 devt = part_devt(part);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001044 disk_put_part(part);
Tejun Heo548b10e2008-08-29 09:01:47 +02001045 break;
Tejun Heodef4e382008-09-03 08:57:12 +02001046 }
Tejun Heo548b10e2008-08-29 09:01:47 +02001047 disk_put_part(part);
Kay Sievers5c0ef6d2008-08-16 14:30:30 +02001048 }
Tejun Heodef4e382008-09-03 08:57:12 +02001049 class_dev_iter_exit(&iter);
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001050 return devt;
1051}
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001052EXPORT_SYMBOL(blk_lookup_devt);
1053
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054struct gendisk *alloc_disk(int minors)
1055{
Christoph Lameter19460892005-06-23 00:08:19 -07001056 return alloc_disk_node(minors, -1);
1057}
Tejun Heo689d6fa2008-08-25 19:56:16 +09001058EXPORT_SYMBOL(alloc_disk);
Christoph Lameter19460892005-06-23 00:08:19 -07001059
1060struct gendisk *alloc_disk_node(int minors, int node_id)
1061{
1062 struct gendisk *disk;
1063
Christoph Lameter94f60302007-07-17 04:03:29 -07001064 disk = kmalloc_node(sizeof(struct gendisk),
1065 GFP_KERNEL | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (disk) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001067 if (!init_part_stats(&disk->part0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 kfree(disk);
1069 return NULL;
1070 }
Tejun Heo540eed52008-08-25 19:56:15 +09001071 if (disk_expand_part_tbl(disk, 0)) {
1072 free_part_stats(&disk->part0);
Tejun Heob5d0b9d2008-09-03 09:06:42 +02001073 kfree(disk);
1074 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 }
Tejun Heo540eed52008-08-25 19:56:15 +09001076 disk->part_tbl->part[0] = &disk->part0;
Tejun Heob5d0b9d2008-09-03 09:06:42 +02001077
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 disk->minors = minors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 rand_initialize_disk(disk);
Tejun Heoed9e1982008-08-25 19:56:05 +09001080 disk_to_dev(disk)->class = &block_class;
1081 disk_to_dev(disk)->type = &disk_type;
1082 device_initialize(disk_to_dev(disk));
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001083 INIT_WORK(&disk->async_notify,
1084 media_change_notify_thread);
Tejun Heo540eed52008-08-25 19:56:15 +09001085 disk->node_id = node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 }
1087 return disk;
1088}
Christoph Lameter19460892005-06-23 00:08:19 -07001089EXPORT_SYMBOL(alloc_disk_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
1091struct kobject *get_disk(struct gendisk *disk)
1092{
1093 struct module *owner;
1094 struct kobject *kobj;
1095
1096 if (!disk->fops)
1097 return NULL;
1098 owner = disk->fops->owner;
1099 if (owner && !try_module_get(owner))
1100 return NULL;
Tejun Heoed9e1982008-08-25 19:56:05 +09001101 kobj = kobject_get(&disk_to_dev(disk)->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 if (kobj == NULL) {
1103 module_put(owner);
1104 return NULL;
1105 }
1106 return kobj;
1107
1108}
1109
1110EXPORT_SYMBOL(get_disk);
1111
1112void put_disk(struct gendisk *disk)
1113{
1114 if (disk)
Tejun Heoed9e1982008-08-25 19:56:05 +09001115 kobject_put(&disk_to_dev(disk)->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116}
1117
1118EXPORT_SYMBOL(put_disk);
1119
1120void set_device_ro(struct block_device *bdev, int flag)
1121{
Tejun Heob7db9952008-08-25 19:56:10 +09001122 bdev->bd_part->policy = flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123}
1124
1125EXPORT_SYMBOL(set_device_ro);
1126
1127void set_disk_ro(struct gendisk *disk, int flag)
1128{
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001129 struct disk_part_iter piter;
1130 struct hd_struct *part;
1131
Tejun Heob7db9952008-08-25 19:56:10 +09001132 disk_part_iter_init(&piter, disk,
1133 DISK_PITER_INCL_EMPTY | DISK_PITER_INCL_PART0);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001134 while ((part = disk_part_iter_next(&piter)))
1135 part->policy = flag;
1136 disk_part_iter_exit(&piter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
1139EXPORT_SYMBOL(set_disk_ro);
1140
1141int bdev_read_only(struct block_device *bdev)
1142{
1143 if (!bdev)
1144 return 0;
Tejun Heob7db9952008-08-25 19:56:10 +09001145 return bdev->bd_part->policy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
1147
1148EXPORT_SYMBOL(bdev_read_only);
1149
Tejun Heocf771cb2008-09-03 09:01:09 +02001150int invalidate_partition(struct gendisk *disk, int partno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151{
1152 int res = 0;
Tejun Heocf771cb2008-09-03 09:01:09 +02001153 struct block_device *bdev = bdget_disk(disk, partno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (bdev) {
Christoph Hellwig2ef41632005-05-05 16:15:59 -07001155 fsync_bdev(bdev);
1156 res = __invalidate_device(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 bdput(bdev);
1158 }
1159 return res;
1160}
1161
1162EXPORT_SYMBOL(invalidate_partition);