blob: 2f7feda61e35fc3cfd8dc74264300baeeaa56826 [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
184/**
185 * disk_map_sector_rcu - map sector to partition
186 * @disk: gendisk of interest
187 * @sector: sector to map
188 *
189 * Find out which partition @sector maps to on @disk. This is
190 * primarily used for stats accounting.
191 *
192 * CONTEXT:
193 * RCU read locked. The returned partition pointer is valid only
194 * while preemption is disabled.
195 *
196 * RETURNS:
Tejun Heo074a7ac2008-08-25 19:56:14 +0900197 * Found partition on success, part0 is returned if no partition matches
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200198 */
199struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
200{
Tejun Heo540eed52008-08-25 19:56:15 +0900201 struct disk_part_tbl *ptbl;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200202 int i;
203
Tejun Heo540eed52008-08-25 19:56:15 +0900204 ptbl = rcu_dereference(disk->part_tbl);
205
206 for (i = 1; i < ptbl->len; i++) {
207 struct hd_struct *part = rcu_dereference(ptbl->part[i]);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200208
209 if (part && part->start_sect <= sector &&
210 sector < part->start_sect + part->nr_sects)
211 return part;
212 }
Tejun Heo074a7ac2008-08-25 19:56:14 +0900213 return &disk->part0;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200214}
215EXPORT_SYMBOL_GPL(disk_map_sector_rcu);
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217/*
218 * Can be deleted altogether. Later.
219 *
220 */
221static struct blk_major_name {
222 struct blk_major_name *next;
223 int major;
224 char name[16];
Joe Korty68eef3b2006-03-31 02:30:32 -0800225} *major_names[BLKDEV_MAJOR_HASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
227/* index in the above - for now: assume no multimajor ranges */
228static inline int major_to_index(int major)
229{
Joe Korty68eef3b2006-03-31 02:30:32 -0800230 return major % BLKDEV_MAJOR_HASH_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
232
Joe Korty68eef3b2006-03-31 02:30:32 -0800233#ifdef CONFIG_PROC_FS
Tejun Heocf771cb2008-09-03 09:01:09 +0200234void blkdev_show(struct seq_file *seqf, off_t offset)
Neil Horman7170be52006-01-14 13:20:38 -0800235{
Joe Korty68eef3b2006-03-31 02:30:32 -0800236 struct blk_major_name *dp;
Neil Horman7170be52006-01-14 13:20:38 -0800237
Joe Korty68eef3b2006-03-31 02:30:32 -0800238 if (offset < BLKDEV_MAJOR_HASH_SIZE) {
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200239 mutex_lock(&block_class_lock);
Joe Korty68eef3b2006-03-31 02:30:32 -0800240 for (dp = major_names[offset]; dp; dp = dp->next)
Tejun Heocf771cb2008-09-03 09:01:09 +0200241 seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200242 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
Joe Korty68eef3b2006-03-31 02:30:32 -0800245#endif /* CONFIG_PROC_FS */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247int register_blkdev(unsigned int major, const char *name)
248{
249 struct blk_major_name **n, *p;
250 int index, ret = 0;
251
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200252 mutex_lock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
254 /* temporary */
255 if (major == 0) {
256 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
257 if (major_names[index] == NULL)
258 break;
259 }
260
261 if (index == 0) {
262 printk("register_blkdev: failed to get major for %s\n",
263 name);
264 ret = -EBUSY;
265 goto out;
266 }
267 major = index;
268 ret = major;
269 }
270
271 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
272 if (p == NULL) {
273 ret = -ENOMEM;
274 goto out;
275 }
276
277 p->major = major;
278 strlcpy(p->name, name, sizeof(p->name));
279 p->next = NULL;
280 index = major_to_index(major);
281
282 for (n = &major_names[index]; *n; n = &(*n)->next) {
283 if ((*n)->major == major)
284 break;
285 }
286 if (!*n)
287 *n = p;
288 else
289 ret = -EBUSY;
290
291 if (ret < 0) {
292 printk("register_blkdev: cannot get major %d for %s\n",
293 major, name);
294 kfree(p);
295 }
296out:
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200297 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 return ret;
299}
300
301EXPORT_SYMBOL(register_blkdev);
302
Akinobu Mitaf4480242007-07-17 04:03:47 -0700303void unregister_blkdev(unsigned int major, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304{
305 struct blk_major_name **n;
306 struct blk_major_name *p = NULL;
307 int index = major_to_index(major);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200309 mutex_lock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 for (n = &major_names[index]; *n; n = &(*n)->next)
311 if ((*n)->major == major)
312 break;
Akinobu Mita294462a2007-07-17 04:03:45 -0700313 if (!*n || strcmp((*n)->name, name)) {
314 WARN_ON(1);
Akinobu Mita294462a2007-07-17 04:03:45 -0700315 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 p = *n;
317 *n = p->next;
318 }
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200319 mutex_unlock(&block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 kfree(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
323EXPORT_SYMBOL(unregister_blkdev);
324
325static struct kobj_map *bdev_map;
326
Tejun Heobcce3de2008-08-25 19:47:22 +0900327/**
Tejun Heo870d6652008-08-25 19:47:25 +0900328 * blk_mangle_minor - scatter minor numbers apart
329 * @minor: minor number to mangle
330 *
331 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT
332 * is enabled. Mangling twice gives the original value.
333 *
334 * RETURNS:
335 * Mangled value.
336 *
337 * CONTEXT:
338 * Don't care.
339 */
340static int blk_mangle_minor(int minor)
341{
342#ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
343 int i;
344
345 for (i = 0; i < MINORBITS / 2; i++) {
346 int low = minor & (1 << i);
347 int high = minor & (1 << (MINORBITS - 1 - i));
348 int distance = MINORBITS - 1 - 2 * i;
349
350 minor ^= low | high; /* clear both bits */
351 low <<= distance; /* swap the positions */
352 high >>= distance;
353 minor |= low | high; /* and set */
354 }
355#endif
356 return minor;
357}
358
359/**
Tejun Heobcce3de2008-08-25 19:47:22 +0900360 * blk_alloc_devt - allocate a dev_t for a partition
361 * @part: partition to allocate dev_t for
Tejun Heobcce3de2008-08-25 19:47:22 +0900362 * @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
Randy Dunlap496aa8a2008-10-16 07:46:23 +0200538 * @partno: 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
Alexey Dobriyanf5009752008-10-04 23:53:21 +0400731static const 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};
Alexey Dobriyanf5009752008-10-04 23:53:21 +0400737
738static int partitions_open(struct inode *inode, struct file *file)
739{
740 return seq_open(file, &partitions_op);
741}
742
743static const struct file_operations proc_partitions_operations = {
744 .open = partitions_open,
745 .read = seq_read,
746 .llseek = seq_lseek,
747 .release = seq_release,
748};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749#endif
750
751
Tejun Heocf771cb2008-09-03 09:01:09 +0200752static struct kobject *base_probe(dev_t devt, int *partno, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200754 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 /* Make old-style 2.4 aliases work */
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200756 request_module("block-major-%d", MAJOR(devt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return NULL;
758}
759
760static int __init genhd_device_init(void)
761{
Dan Williamse105b8b2008-04-21 10:51:07 -0700762 int error;
763
764 block_class.dev_kobj = sysfs_dev_block_kobj;
765 error = class_register(&block_class);
Roland McGrathee27a552008-03-11 17:13:15 -0700766 if (unlikely(error))
767 return error;
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200768 bdev_map = kobj_map_init(base_probe, &block_class_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 blk_dev_init();
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200770
Zhang, Yanmin561ec682008-11-14 08:26:30 +0100771 register_blkdev(BLOCK_EXT_MAJOR, "blkext");
772
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200773#ifndef CONFIG_SYSFS_DEPRECATED
774 /* create top-level block dir */
775 block_depr = kobject_create_and_add("block", NULL);
776#endif
Greg Kroah-Hartman830d3cf2007-11-06 10:36:58 -0800777 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
780subsys_initcall(genhd_device_init);
781
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200782static ssize_t disk_range_show(struct device *dev,
783 struct device_attribute *attr, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200785 struct gendisk *disk = dev_to_disk(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200787 return sprintf(buf, "%d\n", disk->minors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
789
Tejun Heo1f014292008-08-25 19:47:23 +0900790static ssize_t disk_ext_range_show(struct device *dev,
791 struct device_attribute *attr, char *buf)
792{
793 struct gendisk *disk = dev_to_disk(dev);
794
Tejun Heob5d0b9d2008-09-03 09:06:42 +0200795 return sprintf(buf, "%d\n", disk_max_parts(disk));
Tejun Heo1f014292008-08-25 19:47:23 +0900796}
797
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200798static ssize_t disk_removable_show(struct device *dev,
799 struct device_attribute *attr, char *buf)
Kay Sieversa7fd6702005-10-01 14:49:43 +0200800{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200801 struct gendisk *disk = dev_to_disk(dev);
Kay Sieversa7fd6702005-10-01 14:49:43 +0200802
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200803 return sprintf(buf, "%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200805}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Kay Sievers1c9ce522008-06-13 09:41:00 +0200807static ssize_t disk_ro_show(struct device *dev,
808 struct device_attribute *attr, char *buf)
809{
810 struct gendisk *disk = dev_to_disk(dev);
811
Tejun Heob7db9952008-08-25 19:56:10 +0900812 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
Kay Sievers1c9ce522008-06-13 09:41:00 +0200813}
814
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200815static ssize_t disk_capability_show(struct device *dev,
816 struct device_attribute *attr, char *buf)
Kristen Carlson Accardi86ce18d2007-05-23 13:57:38 -0700817{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200818 struct gendisk *disk = dev_to_disk(dev);
819
820 return sprintf(buf, "%x\n", disk->flags);
Kristen Carlson Accardi86ce18d2007-05-23 13:57:38 -0700821}
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200822
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200823static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL);
Tejun Heo1f014292008-08-25 19:47:23 +0900824static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200825static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL);
Kay Sievers1c9ce522008-06-13 09:41:00 +0200826static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL);
Tejun Heoe5610522008-08-25 19:56:09 +0900827static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL);
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200828static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL);
Tejun Heo074a7ac2008-08-25 19:56:14 +0900829static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800830#ifdef CONFIG_FAIL_MAKE_REQUEST
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200831static struct device_attribute dev_attr_fail =
Tejun Heoeddb2e22008-08-25 19:56:13 +0900832 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800833#endif
Jens Axboe581d4e22008-09-14 05:56:33 -0700834#ifdef CONFIG_FAIL_IO_TIMEOUT
835static struct device_attribute dev_attr_fail_timeout =
836 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show,
837 part_timeout_store);
838#endif
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200839
840static struct attribute *disk_attrs[] = {
841 &dev_attr_range.attr,
Tejun Heo1f014292008-08-25 19:47:23 +0900842 &dev_attr_ext_range.attr,
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200843 &dev_attr_removable.attr,
Kay Sievers1c9ce522008-06-13 09:41:00 +0200844 &dev_attr_ro.attr,
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200845 &dev_attr_size.attr,
846 &dev_attr_capability.attr,
847 &dev_attr_stat.attr,
848#ifdef CONFIG_FAIL_MAKE_REQUEST
849 &dev_attr_fail.attr,
850#endif
Jens Axboe581d4e22008-09-14 05:56:33 -0700851#ifdef CONFIG_FAIL_IO_TIMEOUT
852 &dev_attr_fail_timeout.attr,
853#endif
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200854 NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855};
856
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200857static struct attribute_group disk_attr_group = {
858 .attrs = disk_attrs,
859};
860
861static struct attribute_group *disk_attr_groups[] = {
862 &disk_attr_group,
863 NULL
864};
865
Tejun Heo540eed52008-08-25 19:56:15 +0900866static void disk_free_ptbl_rcu_cb(struct rcu_head *head)
867{
868 struct disk_part_tbl *ptbl =
869 container_of(head, struct disk_part_tbl, rcu_head);
870
871 kfree(ptbl);
872}
873
874/**
875 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way
876 * @disk: disk to replace part_tbl for
877 * @new_ptbl: new part_tbl to install
878 *
879 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The
880 * original ptbl is freed using RCU callback.
881 *
882 * LOCKING:
883 * Matching bd_mutx locked.
884 */
885static void disk_replace_part_tbl(struct gendisk *disk,
886 struct disk_part_tbl *new_ptbl)
887{
888 struct disk_part_tbl *old_ptbl = disk->part_tbl;
889
890 rcu_assign_pointer(disk->part_tbl, new_ptbl);
891 if (old_ptbl)
892 call_rcu(&old_ptbl->rcu_head, disk_free_ptbl_rcu_cb);
893}
894
895/**
896 * disk_expand_part_tbl - expand disk->part_tbl
897 * @disk: disk to expand part_tbl for
898 * @partno: expand such that this partno can fit in
899 *
900 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl
901 * uses RCU to allow unlocked dereferencing for stats and other stuff.
902 *
903 * LOCKING:
904 * Matching bd_mutex locked, might sleep.
905 *
906 * RETURNS:
907 * 0 on success, -errno on failure.
908 */
909int disk_expand_part_tbl(struct gendisk *disk, int partno)
910{
911 struct disk_part_tbl *old_ptbl = disk->part_tbl;
912 struct disk_part_tbl *new_ptbl;
913 int len = old_ptbl ? old_ptbl->len : 0;
914 int target = partno + 1;
915 size_t size;
916 int i;
917
918 /* disk_max_parts() is zero during initialization, ignore if so */
919 if (disk_max_parts(disk) && target > disk_max_parts(disk))
920 return -EINVAL;
921
922 if (target <= len)
923 return 0;
924
925 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]);
926 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id);
927 if (!new_ptbl)
928 return -ENOMEM;
929
930 INIT_RCU_HEAD(&new_ptbl->rcu_head);
931 new_ptbl->len = target;
932
933 for (i = 0; i < len; i++)
934 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]);
935
936 disk_replace_part_tbl(disk, new_ptbl);
937 return 0;
938}
939
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200940static void disk_release(struct device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941{
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200942 struct gendisk *disk = dev_to_disk(dev);
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 kfree(disk->random);
Tejun Heo540eed52008-08-25 19:56:15 +0900945 disk_replace_part_tbl(disk, NULL);
Tejun Heo074a7ac2008-08-25 19:56:14 +0900946 free_part_stats(&disk->part0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 kfree(disk);
948}
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200949struct class block_class = {
950 .name = "block",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951};
952
Adrian Bunk1826ead2008-03-04 11:23:46 +0100953static struct device_type disk_type = {
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200954 .name = "disk",
955 .groups = disk_attr_groups,
956 .release = disk_release,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957};
958
Randy Dunlapa6e2ba82008-05-23 09:44:11 -0700959#ifdef CONFIG_PROC_FS
Tejun Heocf771cb2008-09-03 09:01:09 +0200960/*
961 * aggregate disk stat collector. Uses the same stats that the sysfs
962 * entries do, above, but makes them available through one seq_file.
963 *
964 * The output looks suspiciously like /proc/partitions with a bunch of
965 * extra fields.
966 */
967static int diskstats_show(struct seq_file *seqf, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968{
969 struct gendisk *gp = v;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200970 struct disk_part_iter piter;
971 struct hd_struct *hd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 char buf[BDEVNAME_SIZE];
Tejun Heoc9959052008-08-25 19:47:21 +0900973 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
975 /*
Tejun Heoed9e1982008-08-25 19:56:05 +0900976 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
Tejun Heocf771cb2008-09-03 09:01:09 +0200977 seq_puts(seqf, "major minor name"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 " rio rmerge rsect ruse wio wmerge "
979 "wsect wuse running use aveq"
980 "\n\n");
981 */
982
Tejun Heo074a7ac2008-08-25 19:56:14 +0900983 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_PART0);
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200984 while ((hd = disk_part_iter_next(&piter))) {
Tejun Heo074a7ac2008-08-25 19:56:14 +0900985 cpu = part_stat_lock();
Tejun Heoc9959052008-08-25 19:47:21 +0900986 part_round_stats(cpu, hd);
Tejun Heo074a7ac2008-08-25 19:56:14 +0900987 part_stat_unlock();
Tejun Heo1f014292008-08-25 19:47:23 +0900988 seq_printf(seqf, "%4d %7d %s %lu %lu %llu "
Jerome Marchand28f39d52008-02-08 11:04:56 +0100989 "%u %lu %lu %llu %u %u %u %u\n",
Tejun Heof331c022008-09-03 09:01:48 +0200990 MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
991 disk_name(gp, hd->partno, buf),
Jerome Marchand28f39d52008-02-08 11:04:56 +0100992 part_stat_read(hd, ios[0]),
993 part_stat_read(hd, merges[0]),
994 (unsigned long long)part_stat_read(hd, sectors[0]),
995 jiffies_to_msecs(part_stat_read(hd, ticks[0])),
996 part_stat_read(hd, ios[1]),
997 part_stat_read(hd, merges[1]),
998 (unsigned long long)part_stat_read(hd, sectors[1]),
999 jiffies_to_msecs(part_stat_read(hd, ticks[1])),
1000 hd->in_flight,
1001 jiffies_to_msecs(part_stat_read(hd, io_ticks)),
1002 jiffies_to_msecs(part_stat_read(hd, time_in_queue))
1003 );
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001005 disk_part_iter_exit(&piter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
1007 return 0;
1008}
1009
Alexey Dobriyan31d85ab2008-10-06 12:55:38 +04001010static const struct seq_operations diskstats_op = {
Tejun Heodef4e382008-09-03 08:57:12 +02001011 .start = disk_seqf_start,
1012 .next = disk_seqf_next,
1013 .stop = disk_seqf_stop,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 .show = diskstats_show
1015};
Alexey Dobriyanf5009752008-10-04 23:53:21 +04001016
Alexey Dobriyan31d85ab2008-10-06 12:55:38 +04001017static int diskstats_open(struct inode *inode, struct file *file)
1018{
1019 return seq_open(file, &diskstats_op);
1020}
1021
1022static const struct file_operations proc_diskstats_operations = {
1023 .open = diskstats_open,
1024 .read = seq_read,
1025 .llseek = seq_lseek,
1026 .release = seq_release,
1027};
1028
Alexey Dobriyanf5009752008-10-04 23:53:21 +04001029static int __init proc_genhd_init(void)
1030{
Alexey Dobriyan31d85ab2008-10-06 12:55:38 +04001031 proc_create("diskstats", 0, NULL, &proc_diskstats_operations);
Alexey Dobriyanf5009752008-10-04 23:53:21 +04001032 proc_create("partitions", 0, NULL, &proc_partitions_operations);
1033 return 0;
1034}
1035module_init(proc_genhd_init);
Randy Dunlapa6e2ba82008-05-23 09:44:11 -07001036#endif /* CONFIG_PROC_FS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001038static void media_change_notify_thread(struct work_struct *work)
1039{
1040 struct gendisk *gd = container_of(work, struct gendisk, async_notify);
1041 char event[] = "MEDIA_CHANGE=1";
1042 char *envp[] = { event, NULL };
1043
1044 /*
1045 * set enviroment vars to indicate which event this is for
1046 * so that user space will know to go check the media status.
1047 */
Tejun Heoed9e1982008-08-25 19:56:05 +09001048 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001049 put_device(gd->driverfs_dev);
1050}
1051
Adrian Bunk1826ead2008-03-04 11:23:46 +01001052#if 0
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001053void genhd_media_change_notify(struct gendisk *disk)
1054{
1055 get_device(disk->driverfs_dev);
1056 schedule_work(&disk->async_notify);
1057}
1058EXPORT_SYMBOL_GPL(genhd_media_change_notify);
Adrian Bunk1826ead2008-03-04 11:23:46 +01001059#endif /* 0 */
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001060
Tejun Heocf771cb2008-09-03 09:01:09 +02001061dev_t blk_lookup_devt(const char *name, int partno)
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001062{
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001063 dev_t devt = MKDEV(0, 0);
Tejun Heodef4e382008-09-03 08:57:12 +02001064 struct class_dev_iter iter;
1065 struct device *dev;
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001066
Tejun Heodef4e382008-09-03 08:57:12 +02001067 class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1068 while ((dev = class_dev_iter_next(&iter))) {
1069 struct gendisk *disk = dev_to_disk(dev);
Tejun Heo548b10e2008-08-29 09:01:47 +02001070 struct hd_struct *part;
Tejun Heodef4e382008-09-03 08:57:12 +02001071
Tejun Heof331c022008-09-03 09:01:48 +02001072 if (strcmp(dev->bus_id, name))
1073 continue;
Tejun Heof331c022008-09-03 09:01:48 +02001074
Tejun Heo548b10e2008-08-29 09:01:47 +02001075 part = disk_get_part(disk, partno);
Tejun Heo2bbedcb2008-08-29 11:41:51 +02001076 if (part) {
Tejun Heof331c022008-09-03 09:01:48 +02001077 devt = part_devt(part);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001078 disk_put_part(part);
Tejun Heo548b10e2008-08-29 09:01:47 +02001079 break;
Tejun Heodef4e382008-09-03 08:57:12 +02001080 }
Tejun Heo548b10e2008-08-29 09:01:47 +02001081 disk_put_part(part);
Kay Sievers5c0ef6d2008-08-16 14:30:30 +02001082 }
Tejun Heodef4e382008-09-03 08:57:12 +02001083 class_dev_iter_exit(&iter);
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001084 return devt;
1085}
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001086EXPORT_SYMBOL(blk_lookup_devt);
1087
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088struct gendisk *alloc_disk(int minors)
1089{
Christoph Lameter19460892005-06-23 00:08:19 -07001090 return alloc_disk_node(minors, -1);
1091}
Tejun Heo689d6fa2008-08-25 19:56:16 +09001092EXPORT_SYMBOL(alloc_disk);
Christoph Lameter19460892005-06-23 00:08:19 -07001093
1094struct gendisk *alloc_disk_node(int minors, int node_id)
1095{
1096 struct gendisk *disk;
1097
Christoph Lameter94f60302007-07-17 04:03:29 -07001098 disk = kmalloc_node(sizeof(struct gendisk),
1099 GFP_KERNEL | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 if (disk) {
Tejun Heo074a7ac2008-08-25 19:56:14 +09001101 if (!init_part_stats(&disk->part0)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 kfree(disk);
1103 return NULL;
1104 }
Cheng Renquanbf91db12008-11-20 08:37:37 +01001105 disk->node_id = node_id;
Tejun Heo540eed52008-08-25 19:56:15 +09001106 if (disk_expand_part_tbl(disk, 0)) {
1107 free_part_stats(&disk->part0);
Tejun Heob5d0b9d2008-09-03 09:06:42 +02001108 kfree(disk);
1109 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
Tejun Heo540eed52008-08-25 19:56:15 +09001111 disk->part_tbl->part[0] = &disk->part0;
Tejun Heob5d0b9d2008-09-03 09:06:42 +02001112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 disk->minors = minors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 rand_initialize_disk(disk);
Tejun Heoed9e1982008-08-25 19:56:05 +09001115 disk_to_dev(disk)->class = &block_class;
1116 disk_to_dev(disk)->type = &disk_type;
1117 device_initialize(disk_to_dev(disk));
Kristen Carlson Accardi8ce7ad72007-05-23 13:57:38 -07001118 INIT_WORK(&disk->async_notify,
1119 media_change_notify_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 }
1121 return disk;
1122}
Christoph Lameter19460892005-06-23 00:08:19 -07001123EXPORT_SYMBOL(alloc_disk_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125struct kobject *get_disk(struct gendisk *disk)
1126{
1127 struct module *owner;
1128 struct kobject *kobj;
1129
1130 if (!disk->fops)
1131 return NULL;
1132 owner = disk->fops->owner;
1133 if (owner && !try_module_get(owner))
1134 return NULL;
Tejun Heoed9e1982008-08-25 19:56:05 +09001135 kobj = kobject_get(&disk_to_dev(disk)->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 if (kobj == NULL) {
1137 module_put(owner);
1138 return NULL;
1139 }
1140 return kobj;
1141
1142}
1143
1144EXPORT_SYMBOL(get_disk);
1145
1146void put_disk(struct gendisk *disk)
1147{
1148 if (disk)
Tejun Heoed9e1982008-08-25 19:56:05 +09001149 kobject_put(&disk_to_dev(disk)->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150}
1151
1152EXPORT_SYMBOL(put_disk);
1153
1154void set_device_ro(struct block_device *bdev, int flag)
1155{
Tejun Heob7db9952008-08-25 19:56:10 +09001156 bdev->bd_part->policy = flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157}
1158
1159EXPORT_SYMBOL(set_device_ro);
1160
1161void set_disk_ro(struct gendisk *disk, int flag)
1162{
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001163 struct disk_part_iter piter;
1164 struct hd_struct *part;
1165
Tejun Heob7db9952008-08-25 19:56:10 +09001166 disk_part_iter_init(&piter, disk,
1167 DISK_PITER_INCL_EMPTY | DISK_PITER_INCL_PART0);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001168 while ((part = disk_part_iter_next(&piter)))
1169 part->policy = flag;
1170 disk_part_iter_exit(&piter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171}
1172
1173EXPORT_SYMBOL(set_disk_ro);
1174
1175int bdev_read_only(struct block_device *bdev)
1176{
1177 if (!bdev)
1178 return 0;
Tejun Heob7db9952008-08-25 19:56:10 +09001179 return bdev->bd_part->policy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
1181
1182EXPORT_SYMBOL(bdev_read_only);
1183
Tejun Heocf771cb2008-09-03 09:01:09 +02001184int invalidate_partition(struct gendisk *disk, int partno)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185{
1186 int res = 0;
Tejun Heocf771cb2008-09-03 09:01:09 +02001187 struct block_device *bdev = bdget_disk(disk, partno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 if (bdev) {
Christoph Hellwig2ef41632005-05-05 16:15:59 -07001189 fsync_bdev(bdev);
1190 res = __invalidate_device(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 bdput(bdev);
1192 }
1193 return res;
1194}
1195
1196EXPORT_SYMBOL(invalidate_partition);