block: make blkdev_get/put() handle exclusive access
[linux-2.6.git] / block / ioctl.c
1 #include <linux/capability.h>
2 #include <linux/blkdev.h>
3 #include <linux/gfp.h>
4 #include <linux/blkpg.h>
5 #include <linux/hdreg.h>
6 #include <linux/backing-dev.h>
7 #include <linux/buffer_head.h>
8 #include <linux/smp_lock.h>
9 #include <linux/blktrace_api.h>
10 #include <asm/uaccess.h>
11
12 static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
13 {
14         struct block_device *bdevp;
15         struct gendisk *disk;
16         struct hd_struct *part;
17         struct blkpg_ioctl_arg a;
18         struct blkpg_partition p;
19         struct disk_part_iter piter;
20         long long start, length;
21         int partno;
22
23         if (!capable(CAP_SYS_ADMIN))
24                 return -EACCES;
25         if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
26                 return -EFAULT;
27         if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
28                 return -EFAULT;
29         disk = bdev->bd_disk;
30         if (bdev != bdev->bd_contains)
31                 return -EINVAL;
32         partno = p.pno;
33         if (partno <= 0)
34                 return -EINVAL;
35         switch (a.op) {
36                 case BLKPG_ADD_PARTITION:
37                         start = p.start >> 9;
38                         length = p.length >> 9;
39                         /* check for fit in a hd_struct */ 
40                         if (sizeof(sector_t) == sizeof(long) && 
41                             sizeof(long long) > sizeof(long)) {
42                                 long pstart = start, plength = length;
43                                 if (pstart != start || plength != length
44                                     || pstart < 0 || plength < 0)
45                                         return -EINVAL;
46                         }
47
48                         mutex_lock(&bdev->bd_mutex);
49
50                         /* overlap? */
51                         disk_part_iter_init(&piter, disk,
52                                             DISK_PITER_INCL_EMPTY);
53                         while ((part = disk_part_iter_next(&piter))) {
54                                 if (!(start + length <= part->start_sect ||
55                                       start >= part->start_sect + part->nr_sects)) {
56                                         disk_part_iter_exit(&piter);
57                                         mutex_unlock(&bdev->bd_mutex);
58                                         return -EBUSY;
59                                 }
60                         }
61                         disk_part_iter_exit(&piter);
62
63                         /* all seems OK */
64                         part = add_partition(disk, partno, start, length,
65                                              ADDPART_FLAG_NONE, NULL);
66                         mutex_unlock(&bdev->bd_mutex);
67                         return IS_ERR(part) ? PTR_ERR(part) : 0;
68                 case BLKPG_DEL_PARTITION:
69                         part = disk_get_part(disk, partno);
70                         if (!part)
71                                 return -ENXIO;
72
73                         bdevp = bdget(part_devt(part));
74                         disk_put_part(part);
75                         if (!bdevp)
76                                 return -ENOMEM;
77
78                         mutex_lock(&bdevp->bd_mutex);
79                         if (bdevp->bd_openers) {
80                                 mutex_unlock(&bdevp->bd_mutex);
81                                 bdput(bdevp);
82                                 return -EBUSY;
83                         }
84                         /* all seems OK */
85                         fsync_bdev(bdevp);
86                         invalidate_bdev(bdevp);
87
88                         mutex_lock_nested(&bdev->bd_mutex, 1);
89                         delete_partition(disk, partno);
90                         mutex_unlock(&bdev->bd_mutex);
91                         mutex_unlock(&bdevp->bd_mutex);
92                         bdput(bdevp);
93
94                         return 0;
95                 default:
96                         return -EINVAL;
97         }
98 }
99
100 static int blkdev_reread_part(struct block_device *bdev)
101 {
102         struct gendisk *disk = bdev->bd_disk;
103         int res;
104
105         if (!disk_partitionable(disk) || bdev != bdev->bd_contains)
106                 return -EINVAL;
107         if (!capable(CAP_SYS_ADMIN))
108                 return -EACCES;
109         if (!mutex_trylock(&bdev->bd_mutex))
110                 return -EBUSY;
111         res = rescan_partitions(disk, bdev);
112         mutex_unlock(&bdev->bd_mutex);
113         return res;
114 }
115
116 static int blk_ioctl_discard(struct block_device *bdev, uint64_t start,
117                              uint64_t len, int secure)
118 {
119         unsigned long flags = 0;
120
121         if (start & 511)
122                 return -EINVAL;
123         if (len & 511)
124                 return -EINVAL;
125         start >>= 9;
126         len >>= 9;
127
128         if (start + len > (bdev->bd_inode->i_size >> 9))
129                 return -EINVAL;
130         if (secure)
131                 flags |= BLKDEV_DISCARD_SECURE;
132         return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
133 }
134
135 static int put_ushort(unsigned long arg, unsigned short val)
136 {
137         return put_user(val, (unsigned short __user *)arg);
138 }
139
140 static int put_int(unsigned long arg, int val)
141 {
142         return put_user(val, (int __user *)arg);
143 }
144
145 static int put_uint(unsigned long arg, unsigned int val)
146 {
147         return put_user(val, (unsigned int __user *)arg);
148 }
149
150 static int put_long(unsigned long arg, long val)
151 {
152         return put_user(val, (long __user *)arg);
153 }
154
155 static int put_ulong(unsigned long arg, unsigned long val)
156 {
157         return put_user(val, (unsigned long __user *)arg);
158 }
159
160 static int put_u64(unsigned long arg, u64 val)
161 {
162         return put_user(val, (u64 __user *)arg);
163 }
164
165 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
166                         unsigned cmd, unsigned long arg)
167 {
168         struct gendisk *disk = bdev->bd_disk;
169
170         if (disk->fops->ioctl)
171                 return disk->fops->ioctl(bdev, mode, cmd, arg);
172
173         return -ENOTTY;
174 }
175 /*
176  * For the record: _GPL here is only because somebody decided to slap it
177  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
178  * at all and could be open-coded without any exports by anybody who cares.
179  */
180 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
181
182 /*
183  * always keep this in sync with compat_blkdev_ioctl()
184  */
185 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
186                         unsigned long arg)
187 {
188         struct gendisk *disk = bdev->bd_disk;
189         struct backing_dev_info *bdi;
190         loff_t size;
191         int ret, n;
192
193         switch(cmd) {
194         case BLKFLSBUF:
195                 if (!capable(CAP_SYS_ADMIN))
196                         return -EACCES;
197
198                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
199                 /* -EINVAL to handle old uncorrected drivers */
200                 if (ret != -EINVAL && ret != -ENOTTY)
201                         return ret;
202
203                 fsync_bdev(bdev);
204                 invalidate_bdev(bdev);
205                 return 0;
206
207         case BLKROSET:
208                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
209                 /* -EINVAL to handle old uncorrected drivers */
210                 if (ret != -EINVAL && ret != -ENOTTY)
211                         return ret;
212                 if (!capable(CAP_SYS_ADMIN))
213                         return -EACCES;
214                 if (get_user(n, (int __user *)(arg)))
215                         return -EFAULT;
216                 set_device_ro(bdev, n);
217                 return 0;
218
219         case BLKDISCARD:
220         case BLKSECDISCARD: {
221                 uint64_t range[2];
222
223                 if (!(mode & FMODE_WRITE))
224                         return -EBADF;
225
226                 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
227                         return -EFAULT;
228
229                 return blk_ioctl_discard(bdev, range[0], range[1],
230                                          cmd == BLKSECDISCARD);
231         }
232
233         case HDIO_GETGEO: {
234                 struct hd_geometry geo;
235
236                 if (!arg)
237                         return -EINVAL;
238                 if (!disk->fops->getgeo)
239                         return -ENOTTY;
240
241                 /*
242                  * We need to set the startsect first, the driver may
243                  * want to override it.
244                  */
245                 geo.start = get_start_sect(bdev);
246                 ret = disk->fops->getgeo(bdev, &geo);
247                 if (ret)
248                         return ret;
249                 if (copy_to_user((struct hd_geometry __user *)arg, &geo,
250                                         sizeof(geo)))
251                         return -EFAULT;
252                 return 0;
253         }
254         case BLKRAGET:
255         case BLKFRAGET:
256                 if (!arg)
257                         return -EINVAL;
258                 bdi = blk_get_backing_dev_info(bdev);
259                 if (bdi == NULL)
260                         return -ENOTTY;
261                 return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512);
262         case BLKROGET:
263                 return put_int(arg, bdev_read_only(bdev) != 0);
264         case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
265                 return put_int(arg, block_size(bdev));
266         case BLKSSZGET: /* get block device logical block size */
267                 return put_int(arg, bdev_logical_block_size(bdev));
268         case BLKPBSZGET: /* get block device physical block size */
269                 return put_uint(arg, bdev_physical_block_size(bdev));
270         case BLKIOMIN:
271                 return put_uint(arg, bdev_io_min(bdev));
272         case BLKIOOPT:
273                 return put_uint(arg, bdev_io_opt(bdev));
274         case BLKALIGNOFF:
275                 return put_int(arg, bdev_alignment_offset(bdev));
276         case BLKDISCARDZEROES:
277                 return put_uint(arg, bdev_discard_zeroes_data(bdev));
278         case BLKSECTGET:
279                 return put_ushort(arg, queue_max_sectors(bdev_get_queue(bdev)));
280         case BLKRASET:
281         case BLKFRASET:
282                 if(!capable(CAP_SYS_ADMIN))
283                         return -EACCES;
284                 bdi = blk_get_backing_dev_info(bdev);
285                 if (bdi == NULL)
286                         return -ENOTTY;
287                 bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
288                 return 0;
289         case BLKBSZSET:
290                 /* set the logical block size */
291                 if (!capable(CAP_SYS_ADMIN))
292                         return -EACCES;
293                 if (!arg)
294                         return -EINVAL;
295                 if (get_user(n, (int __user *) arg))
296                         return -EFAULT;
297                 if (!(mode & FMODE_EXCL) &&
298                     blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
299                         return -EBUSY;
300                 ret = set_blocksize(bdev, n);
301                 if (!(mode & FMODE_EXCL))
302                         blkdev_put(bdev, mode | FMODE_EXCL);
303                 return ret;
304         case BLKPG:
305                 ret = blkpg_ioctl(bdev, (struct blkpg_ioctl_arg __user *) arg);
306                 break;
307         case BLKRRPART:
308                 ret = blkdev_reread_part(bdev);
309                 break;
310         case BLKGETSIZE:
311                 size = bdev->bd_inode->i_size;
312                 if ((size >> 9) > ~0UL)
313                         return -EFBIG;
314                 return put_ulong(arg, size >> 9);
315         case BLKGETSIZE64:
316                 return put_u64(arg, bdev->bd_inode->i_size);
317         case BLKTRACESTART:
318         case BLKTRACESTOP:
319         case BLKTRACESETUP:
320         case BLKTRACETEARDOWN:
321                 ret = blk_trace_ioctl(bdev, cmd, (char __user *) arg);
322                 break;
323         default:
324                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
325         }
326         return ret;
327 }
328 EXPORT_SYMBOL_GPL(blkdev_ioctl);