blob: 397960cf26afcd61fc7c2bba2365944c8485c767 [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>
Alexey Dobriyanf5009752008-10-04 23:53:21 +040013#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/seq_file.h>
15#include <linux/slab.h>
16#include <linux/kmod.h>
17#include <linux/kobj_map.h>
Christoph Hellwig2ef41632005-05-05 16:15:59 -070018#include <linux/buffer_head.h>
Jes Sorensen58383af2006-02-06 14:12:43 -080019#include <linux/mutex.h>
Tejun Heobcce3de2008-08-25 19:47:22 +090020#include <linux/idr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
Adrian Bunkff889722008-03-04 11:23:45 +010022#include "blk.h"
23
Kay Sieversedfaa7c2007-05-21 22:08:01 +020024static DEFINE_MUTEX(block_class_lock);
25#ifndef CONFIG_SYSFS_DEPRECATED
26struct kobject *block_depr;
27#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
Tejun Heobcce3de2008-08-25 19:47:22 +090029/* for extended dynamic devt allocation, currently only one major is used */
30#define MAX_EXT_DEVT (1 << MINORBITS)
31
32/* For extended devt allocation. ext_devt_mutex prevents look up
33 * results from going away underneath its user.
34 */
35static DEFINE_MUTEX(ext_devt_mutex);
36static DEFINE_IDR(ext_devt_idr);
37
Adrian Bunk1826ead2008-03-04 11:23:46 +010038static struct device_type disk_type;
39
Tejun Heoe71bf0d2008-09-03 09:03:02 +020040/**
41 * disk_get_part - get partition
42 * @disk: disk to look partition from
43 * @partno: partition number
44 *
45 * Look for partition @partno from @disk. If found, increment
46 * reference count and return it.
47 *
48 * CONTEXT:
49 * Don't care.
50 *
51 * RETURNS:
52 * Pointer to the found partition on success, NULL if not found.
53 */
54struct hd_struct *disk_get_part(struct gendisk *disk, int partno)
55{
Tejun Heo540eed52008-08-25 19:56:15 +090056 struct hd_struct *part = NULL;
57 struct disk_part_tbl *ptbl;
Tejun Heoe71bf0d2008-09-03 09:03:02 +020058
Tejun Heo540eed52008-08-25 19:56:15 +090059 if (unlikely(partno < 0))
Tejun Heoe71bf0d2008-09-03 09:03:02 +020060 return NULL;
Tejun Heo540eed52008-08-25 19:56:15 +090061
Tejun Heoe71bf0d2008-09-03 09:03:02 +020062 rcu_read_lock();
Tejun Heo540eed52008-08-25 19:56:15 +090063
64 ptbl = rcu_dereference(disk->part_tbl);
65 if (likely(partno < ptbl->len)) {
66 part = rcu_dereference(ptbl->part[partno]);
67 if (part)
68 get_device(part_to_dev(part));
69 }
70
Tejun Heoe71bf0d2008-09-03 09:03:02 +020071 rcu_read_unlock();
72
73 return part;
74}
75EXPORT_SYMBOL_GPL(disk_get_part);
76
77/**
78 * disk_part_iter_init - initialize partition iterator
79 * @piter: iterator to initialize
80 * @disk: disk to iterate over
81 * @flags: DISK_PITER_* flags
82 *
83 * Initialize @piter so that it iterates over partitions of @disk.
84 *
85 * CONTEXT:
86 * Don't care.
87 */
88void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk,
89 unsigned int flags)
90{
Tejun Heo540eed52008-08-25 19:56:15 +090091 struct disk_part_tbl *ptbl;
92
93 rcu_read_lock();
94 ptbl = rcu_dereference(disk->part_tbl);
95
Tejun Heoe71bf0d2008-09-03 09:03:02 +020096 piter->disk = disk;
97 piter->part = NULL;
98
99 if (flags & DISK_PITER_REVERSE)
Tejun Heo540eed52008-08-25 19:56:15 +0900100 piter->idx = ptbl->len - 1;
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200101 else if (flags & DISK_PITER_INCL_PART0)
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200102 piter->idx = 0;
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200103 else
104 piter->idx = 1;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200105
106 piter->flags = flags;
Tejun Heo540eed52008-08-25 19:56:15 +0900107
108 rcu_read_unlock();
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200109}
110EXPORT_SYMBOL_GPL(disk_part_iter_init);
111
112/**
113 * disk_part_iter_next - proceed iterator to the next partition and return it
114 * @piter: iterator of interest
115 *
116 * Proceed @piter to the next partition and return it.
117 *
118 * CONTEXT:
119 * Don't care.
120 */
121struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
122{
Tejun Heo540eed52008-08-25 19:56:15 +0900123 struct disk_part_tbl *ptbl;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200124 int inc, end;
125
126 /* put the last partition */
127 disk_put_part(piter->part);
128 piter->part = NULL;
129
Tejun Heo540eed52008-08-25 19:56:15 +0900130 /* get part_tbl */
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200131 rcu_read_lock();
Tejun Heo540eed52008-08-25 19:56:15 +0900132 ptbl = rcu_dereference(piter->disk->part_tbl);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200133
134 /* determine iteration parameters */
135 if (piter->flags & DISK_PITER_REVERSE) {
136 inc = -1;
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200137 if (piter->flags & DISK_PITER_INCL_PART0)
138 end = -1;
139 else
140 end = 0;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200141 } else {
142 inc = 1;
Tejun Heo540eed52008-08-25 19:56:15 +0900143 end = ptbl->len;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200144 }
145
146 /* iterate to the next partition */
147 for (; piter->idx != end; piter->idx += inc) {
148 struct hd_struct *part;
149
Tejun Heo540eed52008-08-25 19:56:15 +0900150 part = rcu_dereference(ptbl->part[piter->idx]);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200151 if (!part)
152 continue;
153 if (!(piter->flags & DISK_PITER_INCL_EMPTY) && !part->nr_sects)
154 continue;
155
Tejun Heoed9e1982008-08-25 19:56:05 +0900156 get_device(part_to_dev(part));
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200157 piter->part = part;
158 piter->idx += inc;
159 break;
160 }
161
162 rcu_read_unlock();
163
164 return piter->part;
165}
166EXPORT_SYMBOL_GPL(disk_part_iter_next);
167
168/**
169 * disk_part_iter_exit - finish up partition iteration
170 * @piter: iter of interest
171 *
172 * Called when iteration is over. Cleans up @piter.
173 *
174 * CONTEXT:
175 * Don't care.
176 */
177void disk_part_iter_exit(struct disk_part_iter *piter)
178{
179 disk_put_part(piter->part);
180 piter->part = NULL;
181}
182EXPORT_SYMBOL_GPL(disk_part_iter_exit);
183
Jens Axboea6f23652008-10-24 12:52:42 +0200184static inline int sector_in_part(struct hd_struct *part, sector_t sector)
185{
186 return part->start_sect <= sector &&
187 sector < part->start_sect + part->nr_sects;
188}
189
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200190/**
191 * disk_map_sector_rcu - map sector to partition
192 * @disk: gendisk of interest
193 * @sector: sector to map
194 *
195 * Find out which partition @sector maps to on @disk. This is
196 * primarily used for stats accounting.
197 *
198 * CONTEXT:
199 * RCU read locked. The returned partition pointer is valid only
200 * while preemption is disabled.
201 *
202 * RETURNS:
Tejun Heo074a7ac2008-08-25 19:56:14 +0900203 * Found partition on success, part0 is returned if no partition matches
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200204 */
205struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
206{
Tejun Heo540eed52008-08-25 19:56:15 +0900207 struct disk_part_tbl *ptbl;
Jens Axboea6f23652008-10-24 12:52:42 +0200208 struct hd_struct *part;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200209 int i;
210
Tejun Heo540eed52008-08-25 19:56:15 +0900211 ptbl = rcu_dereference(disk->part_tbl);
212
Jens Axboea6f23652008-10-24 12:52:42 +0200213 part = rcu_dereference(ptbl->last_lookup);
214 if (part && sector_in_part(part, sector))
215 return part;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200216
Jens Axboea6f23652008-10-24 12:52:42 +0200217 for (i = 1; i < ptbl->len; i++) {
218 part = rcu_dereference(ptbl->part[i]);
219
220 if (part && sector_in_part(part, sector)) {
221 rcu_assign_pointer(ptbl->last_lookup, part);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200222 return part;
Jens Axboea6f23652008-10-24 12:52:42 +0200223 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200224 }
Tejun Heo074a7ac2008-08-25 19:56:14 +0900225 return &disk->part0;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200226}
227EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
228
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229/*
230 * Can be deleted altogether. Later.
231 *
232 */
233static struct blk_major_name {
234 struct blk_major_name *next;
235 int major;
236 char name[16];
Joe Korty68eef3b2006-03-31 02:30:32 -0800237} *major_names[BLKDEV_MAJOR_HASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239/* index in the above - for now: assume no multimajor ranges */
240static inline int major_to_index(int major)
241{
Joe Korty68eef3b2006-03-31 02:30:32 -0800242 return major % BLKDEV_MAJOR_HASH_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
Joe Korty68eef3b2006-03-31 02:30:32 -0800245#ifdef CONFIG_PROC_FS
Tejun Heocf771cb2008-09-03 09:01:09 +0200246void blkdev_show(struct seq_file *seqf, off_t offset)
Neil Horman7170be52006-01-14 13:20:38 -0800247{
Joe Korty68eef3b2006-03-31 02:30:32 -0800248 struct blk_major_name *dp;
Neil Horman7170be52006-01-14 13:20:38 -0800249
Joe Korty68eef3b2006-03-31 02:30:32 -0800250 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200251 mutex_lock(&block_class_lock);
Joe Korty68eef3b2006-03-31 02:30:32 -0800252 for (dp = major_names[offset]; dp; dp = dp->next)
Tejun Heocf771cb2008-09-03 09:01:09 +0200253 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200254 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
Joe Korty68eef3b2006-03-31 02:30:32 -0800257#endif /* CONFIG_PROC_FS */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259int register_blkdev(unsigned int major, const char *name)
260{
261 struct blk_major_name **n, *p;
262 int index, ret = 0;
263
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200264 mutex_lock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
266 /* temporary */
267 if (major == 0) {
268 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
269 if (major_names[index] == NULL)
270 break;
271 }
272
273 if (index == 0) {
274 printk("register_blkdev: failed to get major for %s\n",
275 name);
276 ret = -EBUSY;
277 goto out;
278 }
279 major = index;
280 ret = major;
281 }
282
283 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
284 if (p == NULL) {
285 ret = -ENOMEM;
286 goto out;
287 }
288
289 p->major = major;
290 strlcpy(p->name, name, sizeof(p->name));
291 p->next = NULL;
292 index = major_to_index(major);
293
294 for (n = &major_names[index]; *n; n = &(*n)->next) {
295 if ((*n)->major == major)
296 break;
297 }
298 if (!*n)
299 *n = p;
300 else
301 ret = -EBUSY;
302
303 if (ret < 0) {
304 printk("register_blkdev: cannot get major %d for %s\n",
305 major, name);
306 kfree(p);
307 }
308out:
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200309 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 return ret;
311}
312
313EXPORT_SYMBOL(register_blkdev);
314
Akinobu Mitaf4480242007-07-17 04:03:47 -0700315void unregister_blkdev(unsigned int major, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316{
317 struct blk_major_name **n;
318 struct blk_major_name *p = NULL;
319 int index = major_to_index(major);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200321 mutex_lock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 for (n = &major_names[index]; *n; n = &(*n)->next)
323 if ((*n)->major == major)
324 break;
Akinobu Mita294462a2007-07-17 04:03:45 -0700325 if (!*n || strcmp((*n)->name, name)) {
326 WARN_ON(1);
Akinobu Mita294462a2007-07-17 04:03:45 -0700327 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 p = *n;
329 *n = p->next;
330 }
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200331 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334
335EXPORT_SYMBOL(unregister_blkdev);
336
337static struct kobj_map *bdev_map;
338
Tejun Heobcce3de2008-08-25 19:47:22 +0900339/**
Tejun Heo870d6652008-08-25 19:47:25 +0900340 * blk_mangle_minor - scatter minor numbers apart
341 * @minor: minor number to mangle
342 *
343 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
344 * is enabled. Mangling twice gives the original value.
345 *
346 * RETURNS:
347 * Mangled value.
348 *
349 * CONTEXT:
350 * Don't care.
351 */
352static int blk_mangle_minor(int minor)
353{
354#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
355 int i;
356
357 for (i = 0; i < MINORBITS / 2; i++) {
358 int low = minor & (1 << i);
359 int high = minor & (1 << (MINORBITS - 1 - i));
360 int distance = MINORBITS - 1 - 2 * i;
361
362 minor ^= low | high; /* clear both bits */
363 low <<= distance; /* swap the positions */
364 high >>= distance;
365 minor |= low | high; /* and set */
366 }
367#endif
368 return minor;
369}
370
371/**
Tejun Heobcce3de2008-08-25 19:47:22 +0900372 * blk_alloc_devt - allocate a dev_t for a partition
373 * @part: partition to allocate dev_t for
Tejun Heobcce3de2008-08-25 19:47:22 +0900374 * @devt: out parameter for resulting dev_t
375 *
376 * Allocate a dev_t for block device.
377 *
378 * RETURNS:
379 * 0 on success, allocated dev_t is returned in *@devt. -errno on
380 * failure.
381 *
382 * CONTEXT:
383 * Might sleep.
384 */
385int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
386{
387 struct gendisk *disk = part_to_disk(part);
388 int idx, rc;
389
390 /* in consecutive minor range? */
391 if (part->partno < disk->minors) {
392 *devt = MKDEV(disk->major, disk->first_minor + part->partno);
393 return 0;
394 }
395
396 /* allocate ext devt */
397 do {
398 if (!idr_pre_get(&ext_devt_idr, GFP_KERNEL))
399 return -ENOMEM;
400 rc = idr_get_new(&ext_devt_idr, part, &idx);
401 } while (rc == -EAGAIN);
402
403 if (rc)
404 return rc;
405
406 if (idx > MAX_EXT_DEVT) {
407 idr_remove(&ext_devt_idr, idx);
408 return -EBUSY;
409 }
410
Tejun Heo870d6652008-08-25 19:47:25 +0900411 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx));
Tejun Heobcce3de2008-08-25 19:47:22 +0900412 return 0;
413}
414
415/**
416 * blk_free_devt - free a dev_t
417 * @devt: dev_t to free
418 *
419 * Free @devt which was allocated using blk_alloc_devt().
420 *
421 * CONTEXT:
422 * Might sleep.
423 */
424void blk_free_devt(dev_t devt)
425{
426 might_sleep();
427
428 if (devt == MKDEV(0, 0))
429 return;
430
431 if (MAJOR(devt) == BLOCK_EXT_MAJOR) {
432 mutex_lock(&ext_devt_mutex);
Tejun Heo870d6652008-08-25 19:47:25 +0900433 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
Tejun Heobcce3de2008-08-25 19:47:22 +0900434 mutex_unlock(&ext_devt_mutex);
435 }
436}
437
Tejun Heo1f014292008-08-25 19:47:23 +0900438static char *bdevt_str(dev_t devt, char *buf)
439{
440 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
441 char tbuf[BDEVT_SIZE];
442 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
443 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
444 } else
445 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
446
447 return buf;
448}
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450/*
451 * Register device numbers dev..(dev+range-1)
452 * range must be nonzero
453 * The hash chain is sorted on range, so that subranges can override.
454 */
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200455void blk_register_region(dev_t devt, unsigned long range, struct module *module,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 struct kobject *(*probe)(dev_t, int *, void *),
457 int (*lock)(dev_t, void *), void *data)
458{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200459 kobj_map(bdev_map, devt, range, module, probe, lock, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460}
461
462EXPORT_SYMBOL(blk_register_region);
463
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200464void blk_unregister_region(dev_t devt, unsigned long range)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200466 kobj_unmap(bdev_map, devt, range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469EXPORT_SYMBOL(blk_unregister_region);
470
Tejun Heocf771cb2008-09-03 09:01:09 +0200471static struct kobject *exact_match(dev_t devt, int *partno, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472{
473 struct gendisk *p = data;
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200474
Tejun Heoed9e1982008-08-25 19:56:05 +0900475 return &disk_to_dev(p)->kobj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476}
477
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200478static int exact_lock(dev_t devt, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
480 struct gendisk *p = data;
481
482 if (!get_disk(p))
483 return -1;
484 return 0;
485}
486
487/**
488 * add_disk - add partitioning information to kernel list
489 * @disk: per-device partitioning information
490 *
491 * This function registers the partitioning information in @disk
492 * with the kernel.
Tejun Heo3e1a7ff2008-08-25 19:56:17 +0900493 *
494 * FIXME: error handling
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 */
496void add_disk(struct gendisk *disk)
497{
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700498 struct backing_dev_info *bdi;
Tejun Heo3e1a7ff2008-08-25 19:56:17 +0900499 dev_t devt;
Greg Kroah-Hartman6ffeea72008-05-22 17:21:08 -0400500 int retval;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700501
Tejun Heo3e1a7ff2008-08-25 19:56:17 +0900502 /* minors == 0 indicates to use ext devt from part0 and should
503 * be accompanied with EXT_DEVT flag. Make sure all
504 * parameters make sense.
505 */
506 WARN_ON(disk->minors && !(disk->major || disk->first_minor));
507 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT));
508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 disk->flags |= GENHD_FL_UP;
Tejun Heo3e1a7ff2008-08-25 19:56:17 +0900510
511 retval = blk_alloc_devt(&disk->part0, &devt);
512 if (retval) {
513 WARN_ON(1);
514 return;
515 }
516 disk_to_dev(disk)->devt = devt;
517
518 /* ->major and ->first_minor aren't supposed to be
519 * dereferenced from here on, but set them just in case.
520 */
521 disk->major = MAJOR(devt);
522 disk->first_minor = MINOR(devt);
523
Tejun Heof331c022008-09-03 09:01:48 +0200524 blk_register_region(disk_devt(disk), disk->minors, NULL,
525 exact_match, exact_lock, disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 register_disk(disk);
527 blk_register_queue(disk);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700528
529 bdi = &disk->queue->backing_dev_info;
Tejun Heof331c022008-09-03 09:01:48 +0200530 bdi_register_dev(bdi, disk_devt(disk));
Tejun Heoed9e1982008-08-25 19:56:05 +0900531 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
532 "bdi");
Greg Kroah-Hartman6ffeea72008-05-22 17:21:08 -0400533 WARN_ON(retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534}
535
536EXPORT_SYMBOL(add_disk);
537EXPORT_SYMBOL(del_gendisk); /* in partitions/check.c */
538
539void unlink_gendisk(struct gendisk *disk)
540{
Tejun Heoed9e1982008-08-25 19:56:05 +0900541 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700542 bdi_unregister(&disk->queue->backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 blk_unregister_queue(disk);
Tejun Heof331c022008-09-03 09:01:48 +0200544 blk_unregister_region(disk_devt(disk), disk->minors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545}
546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547/**
548 * get_gendisk - get partitioning information for a given device
Randy Dunlap710027a2008-08-19 20:13:11 +0200549 * @devt: device to get partitioning information for
Randy Dunlap496aa8a2008-10-16 07:46:23 +0200550 * @partno: returned partition index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 *
552 * This function gets the structure containing partitioning
Randy Dunlap710027a2008-08-19 20:13:11 +0200553 * information for the given device @devt.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 */
Tejun Heocf771cb2008-09-03 09:01:09 +0200555struct gendisk *get_gendisk(dev_t devt, int *partno)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Tejun Heobcce3de2008-08-25 19:47:22 +0900557 struct gendisk *disk = NULL;
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200558
Tejun Heobcce3de2008-08-25 19:47:22 +0900559 if (MAJOR(devt) != BLOCK_EXT_MAJOR) {
560 struct kobject *kobj;
561
562 kobj = kobj_lookup(bdev_map, devt, partno);
563 if (kobj)
564 disk = dev_to_disk(kobj_to_dev(kobj));
565 } else {
566 struct hd_struct *part;
567
568 mutex_lock(&ext_devt_mutex);
Tejun Heo870d6652008-08-25 19:47:25 +0900569 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt)));
Tejun Heobcce3de2008-08-25 19:47:22 +0900570 if (part && get_disk(part_to_disk(part))) {
571 *partno = part->partno;
572 disk = part_to_disk(part);
573 }
574 mutex_unlock(&ext_devt_mutex);
575 }
576
577 return disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
Tejun Heof331c022008-09-03 09:01:48 +0200580/**
581 * bdget_disk - do bdget() by gendisk and partition number
582 * @disk: gendisk of interest
583 * @partno: partition number
584 *
585 * Find partition @partno from @disk, do bdget() on it.
586 *
587 * CONTEXT:
588 * Don't care.
589 *
590 * RETURNS:
591 * Resulting block_device on success, NULL on failure.
592 */
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200593struct block_device *bdget_disk(struct gendisk *disk, int partno)
Tejun Heof331c022008-09-03 09:01:48 +0200594{
Tejun Heo548b10e2008-08-29 09:01:47 +0200595 struct hd_struct *part;
596 struct block_device *bdev = NULL;
Tejun Heof331c022008-09-03 09:01:48 +0200597
Tejun Heo548b10e2008-08-29 09:01:47 +0200598 part = disk_get_part(disk, partno);
Tejun Heo2bbedcb2008-08-29 11:41:51 +0200599 if (part)
Tejun Heo548b10e2008-08-29 09:01:47 +0200600 bdev = bdget(part_devt(part));
601 disk_put_part(part);
Tejun Heof331c022008-09-03 09:01:48 +0200602
Tejun Heo548b10e2008-08-29 09:01:47 +0200603 return bdev;
Tejun Heof331c022008-09-03 09:01:48 +0200604}
605EXPORT_SYMBOL(bdget_disk);
606
Dave Gilbertdd2a3452007-05-09 02:33:24 -0700607/*
608 * print a full list of all partitions - intended for places where the root
609 * filesystem can't be mounted and thus to give the victim some idea of what
610 * went wrong
611 */
612void __init printk_all_partitions(void)
613{
Tejun Heodef4e382008-09-03 08:57:12 +0200614 struct class_dev_iter iter;
615 struct device *dev;
616
617 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
618 while ((dev = class_dev_iter_next(&iter))) {
619 struct gendisk *disk = dev_to_disk(dev);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200620 struct disk_part_iter piter;
621 struct hd_struct *part;
Tejun Heo1f014292008-08-25 19:47:23 +0900622 char name_buf[BDEVNAME_SIZE];
623 char devt_buf[BDEVT_SIZE];
Tejun Heodef4e382008-09-03 08:57:12 +0200624
625 /*
626 * Don't show empty devices or things that have been
627 * surpressed
628 */
629 if (get_capacity(disk) == 0 ||
630 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
631 continue;
632
633 /*
634 * Note, unlike /proc/partitions, I am showing the
635 * numbers in hex - the same format as the root=
636 * option takes.
637 */
Tejun Heo074a7ac2008-08-25 19:56:14 +0900638 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
639 while ((part = disk_part_iter_next(&piter))) {
640 bool is_part0 = part == &disk->part0;
Tejun Heodef4e382008-09-03 08:57:12 +0200641
Tejun Heo074a7ac2008-08-25 19:56:14 +0900642 printk("%s%s %10llu %s", is_part0 ? "" : " ",
Tejun Heo1f014292008-08-25 19:47:23 +0900643 bdevt_str(part_devt(part), devt_buf),
Tejun Heof331c022008-09-03 09:01:48 +0200644 (unsigned long long)part->nr_sects >> 1,
Tejun Heo1f014292008-08-25 19:47:23 +0900645 disk_name(disk, part->partno, name_buf));
Tejun Heo074a7ac2008-08-25 19:56:14 +0900646 if (is_part0) {
647 if (disk->driverfs_dev != NULL &&
648 disk->driverfs_dev->driver != NULL)
649 printk(" driver: %s\n",
650 disk->driverfs_dev->driver->name);
651 else
652 printk(" (driver?)\n");
653 } else
654 printk("\n");
655 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200656 disk_part_iter_exit(&piter);
Tejun Heodef4e382008-09-03 08:57:12 +0200657 }
658 class_dev_iter_exit(&iter);
Dave Gilbertdd2a3452007-05-09 02:33:24 -0700659}
660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661#ifdef CONFIG_PROC_FS
662/* iterator */
Tejun Heodef4e382008-09-03 08:57:12 +0200663static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
Greg Kroah-Hartman68c4d4a2008-05-22 17:21:08 -0400664{
Tejun Heodef4e382008-09-03 08:57:12 +0200665 loff_t skip = *pos;
666 struct class_dev_iter *iter;
667 struct device *dev;
Greg Kroah-Hartman68c4d4a2008-05-22 17:21:08 -0400668
Harvey Harrisonaeb3d3a2008-08-28 09:27:42 +0200669 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
Tejun Heodef4e382008-09-03 08:57:12 +0200670 if (!iter)
671 return ERR_PTR(-ENOMEM);
672
673 seqf->private = iter;
674 class_dev_iter_init(iter, &block_class, NULL, &disk_type);
675 do {
676 dev = class_dev_iter_next(iter);
677 if (!dev)
678 return NULL;
679 } while (skip--);
680
681 return dev_to_disk(dev);
Greg Kroah-Hartman68c4d4a2008-05-22 17:21:08 -0400682}
683
Tejun Heodef4e382008-09-03 08:57:12 +0200684static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200686 struct device *dev;
Greg Kroah-Hartman66c64af2008-05-22 17:21:08 -0400687
Tejun Heodef4e382008-09-03 08:57:12 +0200688 (*pos)++;
689 dev = class_dev_iter_next(seqf->private);
Tejun Heo2ac3cee2008-09-03 08:53:37 +0200690 if (dev)
Greg Kroah-Hartman68c4d4a2008-05-22 17:21:08 -0400691 return dev_to_disk(dev);
Tejun Heo2ac3cee2008-09-03 08:53:37 +0200692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 return NULL;
694}
695
Tejun Heodef4e382008-09-03 08:57:12 +0200696static void disk_seqf_stop(struct seq_file *seqf, void *v)
Greg Kroah-Hartman27f30252008-05-22 17:21:08 -0400697{
Tejun Heodef4e382008-09-03 08:57:12 +0200698 struct class_dev_iter *iter = seqf->private;
Greg Kroah-Hartman27f30252008-05-22 17:21:08 -0400699
Tejun Heodef4e382008-09-03 08:57:12 +0200700 /* stop is called even after start failed :-( */
701 if (iter) {
702 class_dev_iter_exit(iter);
703 kfree(iter);
Kay Sievers5c0ef6d2008-08-16 14:30:30 +0200704 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705}
706
Tejun Heodef4e382008-09-03 08:57:12 +0200707static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708{
Tejun Heodef4e382008-09-03 08:57:12 +0200709 static void *p;
710
711 p = disk_seqf_start(seqf, pos);
Tejun Heo243294d2008-09-04 09:17:31 +0200712 if (!IS_ERR(p) && p && !*pos)
Tejun Heodef4e382008-09-03 08:57:12 +0200713 seq_puts(seqf, "major minor #blocks name\n\n");
714 return p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715}
716
Tejun Heocf771cb2008-09-03 09:01:09 +0200717static int show_partition(struct seq_file *seqf, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
719 struct gendisk *sgp = v;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200720 struct disk_part_iter piter;
721 struct hd_struct *part;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 char buf[BDEVNAME_SIZE];
723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 /* Don't show non-partitionable removeable devices or empty devices */
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200725 if (!get_capacity(sgp) || (!disk_partitionable(sgp) &&
Tejun Heof331c022008-09-03 09:01:48 +0200726 (sgp->flags & GENHD_FL_REMOVABLE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 return 0;
728 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)
729 return 0;
730
731 /* show the full disk and all non-0 size partitions of it */
Tejun Heo074a7ac2008-08-25 19:56:14 +0900732 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200733 while ((part = disk_part_iter_next(&piter)))
Tejun Heo1f014292008-08-25 19:47:23 +0900734 seq_printf(seqf, "%4d %7d %10llu %s\n",
Tejun Heof331c022008-09-03 09:01:48 +0200735 MAJOR(part_devt(part)), MINOR(part_devt(part)),
736 (unsigned long long)part->nr_sects >> 1,
737 disk_name(sgp, part->partno, buf));
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200738 disk_part_iter_exit(&piter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
740 return 0;
741}
742
Alexey Dobriyanf5009752008-10-04 23:53:21 +0400743static const struct seq_operations partitions_op = {
Tejun Heodef4e382008-09-03 08:57:12 +0200744 .start = show_partition_start,
745 .next = disk_seqf_next,
746 .stop = disk_seqf_stop,
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200747 .show = show_partition
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748};
Alexey Dobriyanf5009752008-10-04 23:53:21 +0400749
750static int partitions_open(struct inode *inode, struct file *file)
751{
752 return seq_open(file, &partitions_op);
753}
754
755static const struct file_operations proc_partitions_operations = {
756 .open = partitions_open,
757 .read = seq_read,
758 .llseek = seq_lseek,
759 .release = seq_release,
760};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761#endif
762
763
Tejun Heocf771cb2008-09-03 09:01:09 +0200764static struct kobject *base_probe(dev_t devt, int *partno, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200766 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 /* Make old-style 2.4 aliases work */
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200768 request_module("block-major-%d", MAJOR(devt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 return NULL;
770}
771
772static int __init genhd_device_init(void)
773{
Dan Williamse105b8b2008-04-21 10:51:07 -0700774 int error;
775
776 block_class.dev_kobj = sysfs_dev_block_kobj;
777 error = class_register(&block_class);
Roland McGrathee27a552008-03-11 17:13:15 -0700778 if (unlikely(error))
779 return error;
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200780 bdev_map = kobj_map_init(base_probe, &block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 blk_dev_init();
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200782
Zhang, Yanmin561ec682008-11-14 08:26:30 +0100783 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
784
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200785#ifndef CONFIG_SYSFS_DEPRECATED
786 /* create top-level block dir */
787 block_depr = kobject_create_and_add("block", NULL);
788#endif
Greg Kroah-Hartman830d3cf2007-11-06 10:36:58 -0800789 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
792subsys_initcall(genhd_device_init);
793
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200794static ssize_t disk_range_show(struct device *dev,
795 struct device_attribute *attr, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200797 struct gendisk *disk = dev_to_disk(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200799 return sprintf(buf, "%d\n", disk->minors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800}
801
Tejun Heo1f014292008-08-25 19:47:23 +0900802static ssize_t disk_ext_range_show(struct device *dev,
803 struct device_attribute *attr, char *buf)
804{
805 struct gendisk *disk = dev_to_disk(dev);
806
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200807 return sprintf(buf, "%d\n", disk_max_parts(disk));
Tejun Heo1f014292008-08-25 19:47:23 +0900808}
809
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200810static ssize_t disk_removable_show(struct device *dev,
811 struct device_attribute *attr, char *buf)
Kay Sieversa7fd6702005-10-01 14:49:43 +0200812{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200813 struct gendisk *disk = dev_to_disk(dev);
Kay Sieversa7fd6702005-10-01 14:49:43 +0200814
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200815 return sprintf(buf, "%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200817}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Kay Sievers1c9ce522008-06-13 09:41:00 +0200819static ssize_t disk_ro_show(struct device *dev,
820 struct device_attribute *attr, char *buf)
821{
822 struct gendisk *disk = dev_to_disk(dev);
823
Tejun Heob7db9952008-08-25 19:56:10 +0900824 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
Kay Sievers1c9ce522008-06-13 09:41:00 +0200825}
826
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200827static ssize_t disk_capability_show(struct device *dev,
828 struct device_attribute *attr, char *buf)
Kristen Carlson Accardi86ce18d2007-05-23 13:57:38 -0700829{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200830 struct gendisk *disk = dev_to_disk(dev);
831
832 return sprintf(buf, "%x\n", disk->flags);
Kristen Carlson Accardi86ce18d2007-05-23 13:57:38 -0700833}
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200834
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200835static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
Tejun Heo1f014292008-08-25 19:47:23 +0900836static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200837static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
Kay Sievers1c9ce522008-06-13 09:41:00 +0200838static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
Tejun Heoe5610522008-08-25 19:56:09 +0900839static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200840static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
Tejun Heo074a7ac2008-08-25 19:56:14 +0900841static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800842#ifdef CONFIG_FAIL_MAKE_REQUEST
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200843static struct device_attribute dev_attr_fail =
Tejun Heoeddb2e22008-08-25 19:56:13 +0900844 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800845#endif
Jens Axboe581d4e22008-09-14 05:56:33 -0700846#ifdef CONFIG_FAIL_IO_TIMEOUT
847static struct device_attribute dev_attr_fail_timeout =
848 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
849 part_timeout_store);
850#endif
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200851
852static struct attribute *disk_attrs[] = {
853 &dev_attr_range.attr,
Tejun Heo1f014292008-08-25 19:47:23 +0900854 &dev_attr_ext_range.attr,
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200855 &dev_attr_removable.attr,
Kay Sievers1c9ce522008-06-13 09:41:00 +0200856 &dev_attr_ro.attr,
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200857 &dev_attr_size.attr,
858 &dev_attr_capability.attr,
859 &dev_attr_stat.attr,
860#ifdef CONFIG_FAIL_MAKE_REQUEST
861 &dev_attr_fail.attr,
862#endif
Jens Axboe581d4e22008-09-14 05:56:33 -0700863#ifdef CONFIG_FAIL_IO_TIMEOUT
864 &dev_attr_fail_timeout.attr,
865#endif
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200866 NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867};
868
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200869static struct attribute_group disk_attr_group = {
870 .attrs = disk_attrs,
871};
872
873static struct attribute_group *disk_attr_groups[] = {
874 &disk_attr_group,
875 NULL
876};
877
Tejun Heo540eed52008-08-25 19:56:15 +0900878static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
879{
880 struct disk_part_tbl *ptbl =
881 container_of(head, struct disk_part_tbl, rcu_head);
882
883 kfree(ptbl);
884}
885
886/**
887 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
888 * @disk: disk to replace part_tbl for
889 * @new_ptbl: new part_tbl to install
890 *
891 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
892 * original ptbl is freed using RCU callback.
893 *
894 * LOCKING:
895 * Matching bd_mutx locked.
896 */
897static void disk_replace_part_tbl(struct gendisk *disk,
898 struct disk_part_tbl *new_ptbl)
899{
900 struct disk_part_tbl *old_ptbl = disk->part_tbl;
901
902 rcu_assign_pointer(disk->part_tbl, new_ptbl);
Jens Axboea6f23652008-10-24 12:52:42 +0200903
904 if (old_ptbl) {
905 rcu_assign_pointer(old_ptbl->last_lookup, NULL);
Tejun Heo540eed52008-08-25 19:56:15 +0900906 call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
Jens Axboea6f23652008-10-24 12:52:42 +0200907 }
Tejun Heo540eed52008-08-25 19:56:15 +0900908}
909
910/**
911 * disk_expand_part_tbl - expand disk->part_tbl
912 * @disk: disk to expand part_tbl for
913 * @partno: expand such that this partno can fit in
914 *
915 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
916 * uses RCU to allow unlocked dereferencing for stats and other stuff.
917 *
918 * LOCKING:
919 * Matching bd_mutex locked, might sleep.
920 *
921 * RETURNS:
922 * 0 on success, -errno on failure.
923 */
924int disk_expand_part_tbl(struct gendisk *disk, int partno)
925{
926 struct disk_part_tbl *old_ptbl = disk->part_tbl;
927 struct disk_part_tbl *new_ptbl;
928 int len = old_ptbl ? old_ptbl->len : 0;
929 int target = partno + 1;
930 size_t size;
931 int i;
932
933 /* disk_max_parts() is zero during initialization, ignore if so */
934 if (disk_max_parts(disk) && target > disk_max_parts(disk))
935 return -EINVAL;
936
937 if (target <= len)
938 return 0;
939
940 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
941 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
942 if (!new_ptbl)
943 return -ENOMEM;
944
945 INIT_RCU_HEAD(&new_ptbl->rcu_head);
946 new_ptbl->len = target;
947
948 for (i = 0; i < len; i++)
949 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
950
951 disk_replace_part_tbl(disk, new_ptbl);
952 return 0;
953}
954
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200955static void disk_release(struct device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200957 struct gendisk *disk = dev_to_disk(dev);
958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 kfree(disk->random);
Tejun Heo540eed52008-08-25 19:56:15 +0900960 disk_replace_part_tbl(disk, NULL);
Tejun Heo074a7ac2008-08-25 19:56:14 +0900961 free_part_stats(&disk->part0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 kfree(disk);
963}
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200964struct class block_class = {
965 .name = "block",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966};
967
Adrian Bunk1826ead2008-03-04 11:23:46 +0100968static struct device_type disk_type = {
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200969 .name = "disk",
970 .groups = disk_attr_groups,
971 .release = disk_release,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972};
973
Randy Dunlapa6e2ba82008-05-23 09:44:11 -0700974#ifdef CONFIG_PROC_FS
Tejun Heocf771cb2008-09-03 09:01:09 +0200975/*
976 * aggregate disk stat collector. Uses the same stats that the sysfs
977 * entries do, above, but makes them available through one seq_file.
978 *
979 * The output looks suspiciously like /proc/partitions with a bunch of
980 * extra fields.
981 */
982static int diskstats_show(struct seq_file *seqf, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983{
984 struct gendisk *gp = v;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200985 struct disk_part_iter piter;
986 struct hd_struct *hd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 char buf[BDEVNAME_SIZE];
Tejun Heoc9959052008-08-25 19:47:21 +0900988 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
990 /*
Tejun Heoed9e1982008-08-25 19:56:05 +0900991 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
Tejun Heocf771cb2008-09-03 09:01:09 +0200992 seq_puts(seqf, "major minor name"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 " rio rmerge rsect ruse wio wmerge "
994 "wsect wuse running use aveq"
995 "\n\n");
996 */
997
Tejun Heo074a7ac2008-08-25 19:56:14 +0900998 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_PART0);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200999 while ((hd = disk_part_iter_next(&piter))) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001000 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +09001001 part_round_stats(cpu, hd);
Tejun Heo074a7ac2008-08-25 19:56:14 +09001002 part_stat_unlock();
Tejun Heo1f014292008-08-25 19:47:23 +09001003 seq_printf(seqf, "%4d %7d %s %lu %lu %llu "
Jerome Marchand28f39d52008-02-08 11:04:56 +01001004 "%u %lu %lu %llu %u %u %u %u\n",
Tejun Heof331c022008-09-03 09:01:48 +02001005 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
1006 disk_name(gp, hd->partno, buf),
Jerome Marchand28f39d52008-02-08 11:04:56 +01001007 part_stat_read(hd, ios[0]),
1008 part_stat_read(hd, merges[0]),
1009 (unsigned long long)part_stat_read(hd, sectors[0]),
1010 jiffies_to_msecs(part_stat_read(hd, ticks[0])),
1011 part_stat_read(hd, ios[1]),
1012 part_stat_read(hd, merges[1]),
1013 (unsigned long long)part_stat_read(hd, sectors[1]),
1014 jiffies_to_msecs(part_stat_read(hd, ticks[1])),
1015 hd->in_flight,
1016 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1017 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1018 );
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001020 disk_part_iter_exit(&piter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
1022 return 0;
1023}
1024
Alexey Dobriyan31d85ab2008-10-06 12:55:38 +04001025static const struct seq_operations diskstats_op = {
Tejun Heodef4e382008-09-03 08:57:12 +02001026 .start = disk_seqf_start,
1027 .next = disk_seqf_next,
1028 .stop = disk_seqf_stop,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 .show = diskstats_show
1030};
Alexey Dobriyanf5009752008-10-04 23:53:21 +04001031
Alexey Dobriyan31d85ab2008-10-06 12:55:38 +04001032static int diskstats_open(struct inode *inode, struct file *file)
1033{
1034 return seq_open(file, &diskstats_op);
1035}
1036
1037static const struct file_operations proc_diskstats_operations = {
1038 .open = diskstats_open,
1039 .read = seq_read,
1040 .llseek = seq_lseek,
1041 .release = seq_release,
1042};
1043
Alexey Dobriyanf5009752008-10-04 23:53:21 +04001044static int __init proc_genhd_init(void)
1045{
Alexey Dobriyan31d85ab2008-10-06 12:55:38 +04001046 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
Alexey Dobriyanf5009752008-10-04 23:53:21 +04001047 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1048 return 0;
1049}
1050module_init(proc_genhd_init);
Randy Dunlapa6e2ba82008-05-23 09:44:11 -07001051#endif /* CONFIG_PROC_FS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052
Kristen Carlson Accardi8ce7ad7b2007-05-23 13:57:38 -07001053static void media_change_notify_thread(struct work_struct *work)
1054{
1055 struct gendisk *gd = container_of(work, struct gendisk, async_notify);
1056 char event[] = "MEDIA_CHANGE=1";
1057 char *envp[] = { event, NULL };
1058
1059 /*
1060 * set enviroment vars to indicate which event this is for
1061 * so that user space will know to go check the media status.
1062 */
Tejun Heoed9e1982008-08-25 19:56:05 +09001063 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
Kristen Carlson Accardi8ce7ad7b2007-05-23 13:57:38 -07001064 put_device(gd->driverfs_dev);
1065}
1066
Adrian Bunk1826ead2008-03-04 11:23:46 +01001067#if 0
Kristen Carlson Accardi8ce7ad7b2007-05-23 13:57:38 -07001068void genhd_media_change_notify(struct gendisk *disk)
1069{
1070 get_device(disk->driverfs_dev);
1071 schedule_work(&disk->async_notify);
1072}
1073EXPORT_SYMBOL_GPL(genhd_media_change_notify);
Adrian Bunk1826ead2008-03-04 11:23:46 +01001074#endif /* 0 */
Kristen Carlson Accardi8ce7ad7b2007-05-23 13:57:38 -07001075
Tejun Heocf771cb2008-09-03 09:01:09 +02001076dev_t blk_lookup_devt(const char *name, int partno)
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001077{
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001078 dev_t devt = MKDEV(0, 0);
Tejun Heodef4e382008-09-03 08:57:12 +02001079 struct class_dev_iter iter;
1080 struct device *dev;
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001081
Tejun Heodef4e382008-09-03 08:57:12 +02001082 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1083 while ((dev = class_dev_iter_next(&iter))) {
1084 struct gendisk *disk = dev_to_disk(dev);
Tejun Heo548b10e2008-08-29 09:01:47 +02001085 struct hd_struct *part;
Tejun Heodef4e382008-09-03 08:57:12 +02001086
Kay Sievers3ada8b72009-01-06 10:44:43 -08001087 if (strcmp(dev_name(dev), name))
Tejun Heof331c022008-09-03 09:01:48 +02001088 continue;
Tejun Heof331c022008-09-03 09:01:48 +02001089
Tejun Heo548b10e2008-08-29 09:01:47 +02001090 part = disk_get_part(disk, partno);
Tejun Heo2bbedcb2008-08-29 11:41:51 +02001091 if (part) {
Tejun Heof331c022008-09-03 09:01:48 +02001092 devt = part_devt(part);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001093 disk_put_part(part);
Tejun Heo548b10e2008-08-29 09:01:47 +02001094 break;
Tejun Heodef4e382008-09-03 08:57:12 +02001095 }
Tejun Heo548b10e2008-08-29 09:01:47 +02001096 disk_put_part(part);
Kay Sievers5c0ef6d2008-08-16 14:30:30 +02001097 }
Tejun Heodef4e382008-09-03 08:57:12 +02001098 class_dev_iter_exit(&iter);
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001099 return devt;
1100}
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001101EXPORT_SYMBOL(blk_lookup_devt);
1102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103struct gendisk *alloc_disk(int minors)
1104{
Christoph Lameter19460892005-06-23 00:08:19 -07001105 return alloc_disk_node(minors, -1);
1106}
Tejun Heo689d6fa2008-08-25 19:56:16 +09001107EXPORT_SYMBOL(alloc_disk);
Christoph Lameter19460892005-06-23 00:08:19 -07001108
1109struct gendisk *alloc_disk_node(int minors, int node_id)
1110{
1111 struct gendisk *disk;
1112
Christoph Lameter94f60302007-07-17 04:03:29 -07001113 disk = kmalloc_node(sizeof(struct gendisk),
1114 GFP_KERNEL | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 if (disk) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001116 if (!init_part_stats(&disk->part0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 kfree(disk);
1118 return NULL;
1119 }
Cheng Renquanbf91db12008-11-20 08:37:37 +01001120 disk->node_id = node_id;
Tejun Heo540eed52008-08-25 19:56:15 +09001121 if (disk_expand_part_tbl(disk, 0)) {
1122 free_part_stats(&disk->part0);
Tejun Heob5d0b9d2008-09-03 09:06:42 +02001123 kfree(disk);
1124 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
Tejun Heo540eed52008-08-25 19:56:15 +09001126 disk->part_tbl->part[0] = &disk->part0;
Tejun Heob5d0b9d2008-09-03 09:06:42 +02001127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 disk->minors = minors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 rand_initialize_disk(disk);
Tejun Heoed9e1982008-08-25 19:56:05 +09001130 disk_to_dev(disk)->class = &block_class;
1131 disk_to_dev(disk)->type = &disk_type;
1132 device_initialize(disk_to_dev(disk));
Kristen Carlson Accardi8ce7ad7b2007-05-23 13:57:38 -07001133 INIT_WORK(&disk->async_notify,
1134 media_change_notify_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 }
1136 return disk;
1137}
Christoph Lameter19460892005-06-23 00:08:19 -07001138EXPORT_SYMBOL(alloc_disk_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140struct kobject *get_disk(struct gendisk *disk)
1141{
1142 struct module *owner;
1143 struct kobject *kobj;
1144
1145 if (!disk->fops)
1146 return NULL;
1147 owner = disk->fops->owner;
1148 if (owner && !try_module_get(owner))
1149 return NULL;
Tejun Heoed9e1982008-08-25 19:56:05 +09001150 kobj = kobject_get(&disk_to_dev(disk)->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 if (kobj == NULL) {
1152 module_put(owner);
1153 return NULL;
1154 }
1155 return kobj;
1156
1157}
1158
1159EXPORT_SYMBOL(get_disk);
1160
1161void put_disk(struct gendisk *disk)
1162{
1163 if (disk)
Tejun Heoed9e1982008-08-25 19:56:05 +09001164 kobject_put(&disk_to_dev(disk)->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165}
1166
1167EXPORT_SYMBOL(put_disk);
1168
1169void set_device_ro(struct block_device *bdev, int flag)
1170{
Tejun Heob7db9952008-08-25 19:56:10 +09001171 bdev->bd_part->policy = flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172}
1173
1174EXPORT_SYMBOL(set_device_ro);
1175
1176void set_disk_ro(struct gendisk *disk, int flag)
1177{
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001178 struct disk_part_iter piter;
1179 struct hd_struct *part;
1180
Tejun Heob7db9952008-08-25 19:56:10 +09001181 disk_part_iter_init(&piter, disk,
1182 DISK_PITER_INCL_EMPTY | DISK_PITER_INCL_PART0);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001183 while ((part = disk_part_iter_next(&piter)))
1184 part->policy = flag;
1185 disk_part_iter_exit(&piter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186}
1187
1188EXPORT_SYMBOL(set_disk_ro);
1189
1190int bdev_read_only(struct block_device *bdev)
1191{
1192 if (!bdev)
1193 return 0;
Tejun Heob7db9952008-08-25 19:56:10 +09001194 return bdev->bd_part->policy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195}
1196
1197EXPORT_SYMBOL(bdev_read_only);
1198
Tejun Heocf771cb2008-09-03 09:01:09 +02001199int invalidate_partition(struct gendisk *disk, int partno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
1201 int res = 0;
Tejun Heocf771cb2008-09-03 09:01:09 +02001202 struct block_device *bdev = bdget_disk(disk, partno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 if (bdev) {
Christoph Hellwig2ef41632005-05-05 16:15:59 -07001204 fsync_bdev(bdev);
1205 res = __invalidate_device(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 bdput(bdev);
1207 }
1208 return res;
1209}
1210
1211EXPORT_SYMBOL(invalidate_partition);