cgroup: fix cgroup_path() vs rename() race
[linux-3.10.git] / block / ioctl.c
1 #include <linux/capability.h>
2 #include <linux/blkdev.h>
3 #include <linux/export.h>
4 #include <linux/gfp.h>
5 #include <linux/blkpg.h>
6 #include <linux/hdreg.h>
7 #include <linux/backing-dev.h>
8 #include <linux/fs.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, *lpart;
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 || partno > 65535)
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                 case BLKPG_RESIZE_PARTITION:
96                         start = p.start >> 9;
97                         /* new length of partition in bytes */
98                         length = p.length >> 9;
99                         /* check for fit in a hd_struct */
100                         if (sizeof(sector_t) == sizeof(long) &&
101                             sizeof(long long) > sizeof(long)) {
102                                 long pstart = start, plength = length;
103                                 if (pstart != start || plength != length
104                                     || pstart < 0 || plength < 0)
105                                         return -EINVAL;
106                         }
107                         part = disk_get_part(disk, partno);
108                         if (!part)
109                                 return -ENXIO;
110                         bdevp = bdget(part_devt(part));
111                         if (!bdevp) {
112                                 disk_put_part(part);
113                                 return -ENOMEM;
114                         }
115                         mutex_lock(&bdevp->bd_mutex);
116                         mutex_lock_nested(&bdev->bd_mutex, 1);
117                         if (start != part->start_sect) {
118                                 mutex_unlock(&bdevp->bd_mutex);
119                                 mutex_unlock(&bdev->bd_mutex);
120                                 bdput(bdevp);
121                                 disk_put_part(part);
122                                 return -EINVAL;
123                         }
124                         /* overlap? */
125                         disk_part_iter_init(&piter, disk,
126                                             DISK_PITER_INCL_EMPTY);
127                         while ((lpart = disk_part_iter_next(&piter))) {
128                                 if (lpart->partno != partno &&
129                                    !(start + length <= lpart->start_sect ||
130                                    start >= lpart->start_sect + lpart->nr_sects)
131                                    ) {
132                                         disk_part_iter_exit(&piter);
133                                         mutex_unlock(&bdevp->bd_mutex);
134                                         mutex_unlock(&bdev->bd_mutex);
135                                         bdput(bdevp);
136                                         disk_put_part(part);
137                                         return -EBUSY;
138                                 }
139                         }
140                         disk_part_iter_exit(&piter);
141                         part_nr_sects_write(part, (sector_t)length);
142                         i_size_write(bdevp->bd_inode, p.length);
143                         mutex_unlock(&bdevp->bd_mutex);
144                         mutex_unlock(&bdev->bd_mutex);
145                         bdput(bdevp);
146                         disk_put_part(part);
147                         return 0;
148                 default:
149                         return -EINVAL;
150         }
151 }
152
153 static int blkdev_reread_part(struct block_device *bdev)
154 {
155         struct gendisk *disk = bdev->bd_disk;
156         int res;
157
158         if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
159                 return -EINVAL;
160         if (!capable(CAP_SYS_ADMIN))
161                 return -EACCES;
162         if (!mutex_trylock(&bdev->bd_mutex))
163                 return -EBUSY;
164         res = rescan_partitions(disk, bdev);
165         mutex_unlock(&bdev->bd_mutex);
166         return res;
167 }
168
169 static int blk_ioctl_discard(struct block_device *bdev, uint64_t start,
170                              uint64_t len, int secure)
171 {
172         unsigned long flags = 0;
173
174         if (start & 511)
175                 return -EINVAL;
176         if (len & 511)
177                 return -EINVAL;
178         start >>= 9;
179         len >>= 9;
180
181         if (start + len > (i_size_read(bdev->bd_inode) >> 9))
182                 return -EINVAL;
183         if (secure)
184                 flags |= BLKDEV_DISCARD_SECURE;
185         return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
186 }
187
188 static int blk_ioctl_zeroout(struct block_device *bdev, uint64_t start,
189                              uint64_t len)
190 {
191         if (start & 511)
192                 return -EINVAL;
193         if (len & 511)
194                 return -EINVAL;
195         start >>= 9;
196         len >>= 9;
197
198         if (start + len > (i_size_read(bdev->bd_inode) >> 9))
199                 return -EINVAL;
200
201         return blkdev_issue_zeroout(bdev, start, len, GFP_KERNEL);
202 }
203
204 static int put_ushort(unsigned long arg, unsigned short val)
205 {
206         return put_user(val, (unsigned short __user *)arg);
207 }
208
209 static int put_int(unsigned long arg, int val)
210 {
211         return put_user(val, (int __user *)arg);
212 }
213
214 static int put_uint(unsigned long arg, unsigned int val)
215 {
216         return put_user(val, (unsigned int __user *)arg);
217 }
218
219 static int put_long(unsigned long arg, long val)
220 {
221         return put_user(val, (long __user *)arg);
222 }
223
224 static int put_ulong(unsigned long arg, unsigned long val)
225 {
226         return put_user(val, (unsigned long __user *)arg);
227 }
228
229 static int put_u64(unsigned long arg, u64 val)
230 {
231         return put_user(val, (u64 __user *)arg);
232 }
233
234 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
235                         unsigned cmd, unsigned long arg)
236 {
237         struct gendisk *disk = bdev->bd_disk;
238
239         if (disk->fops->ioctl)
240                 return disk->fops->ioctl(bdev, mode, cmd, arg);
241
242         return -ENOTTY;
243 }
244 /*
245  * For the record: _GPL here is only because somebody decided to slap it
246  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
247  * at all and could be open-coded without any exports by anybody who cares.
248  */
249 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
250
251 /*
252  * Is it an unrecognized ioctl? The correct returns are either
253  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
254  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
255  * code before returning.
256  *
257  * Confused drivers sometimes return EINVAL, which is wrong. It
258  * means "I understood the ioctl command, but the parameters to
259  * it were wrong".
260  *
261  * We should aim to just fix the broken drivers, the EINVAL case
262  * should go away.
263  */
264 static inline int is_unrecognized_ioctl(int ret)
265 {
266         return  ret == -EINVAL ||
267                 ret == -ENOTTY ||
268                 ret == -ENOIOCTLCMD;
269 }
270
271 /*
272  * always keep this in sync with compat_blkdev_ioctl()
273  */
274 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
275                         unsigned long arg)
276 {
277         struct gendisk *disk = bdev->bd_disk;
278         struct backing_dev_info *bdi;
279         loff_t size;
280         int ret, n;
281
282         switch(cmd) {
283         case BLKFLSBUF:
284                 if (!capable(CAP_SYS_ADMIN))
285                         return -EACCES;
286
287                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
288                 if (!is_unrecognized_ioctl(ret))
289                         return ret;
290
291                 fsync_bdev(bdev);
292                 invalidate_bdev(bdev);
293                 return 0;
294
295         case BLKROSET:
296                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
297                 if (!is_unrecognized_ioctl(ret))
298                         return ret;
299                 if (!capable(CAP_SYS_ADMIN))
300                         return -EACCES;
301                 if (get_user(n, (int __user *)(arg)))
302                         return -EFAULT;
303                 set_device_ro(bdev, n);
304                 return 0;
305
306         case BLKDISCARD:
307         case BLKSECDISCARD: {
308                 uint64_t range[2];
309
310                 if (!(mode & FMODE_WRITE))
311                         return -EBADF;
312
313                 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
314                         return -EFAULT;
315
316                 return blk_ioctl_discard(bdev, range[0], range[1],
317                                          cmd == BLKSECDISCARD);
318         }
319         case BLKZEROOUT: {
320                 uint64_t range[2];
321
322                 if (!(mode & FMODE_WRITE))
323                         return -EBADF;
324
325                 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
326                         return -EFAULT;
327
328                 return blk_ioctl_zeroout(bdev, range[0], range[1]);
329         }
330
331         case HDIO_GETGEO: {
332                 struct hd_geometry geo;
333
334                 if (!arg)
335                         return -EINVAL;
336                 if (!disk->fops->getgeo)
337                         return -ENOTTY;
338
339                 /*
340                  * We need to set the startsect first, the driver may
341                  * want to override it.
342                  */
343                 memset(&geo, 0, sizeof(geo));
344                 geo.start = get_start_sect(bdev);
345                 ret = disk->fops->getgeo(bdev, &geo);
346                 if (ret)
347                         return ret;
348                 if (copy_to_user((struct hd_geometry __user *)arg, &geo,
349                                         sizeof(geo)))
350                         return -EFAULT;
351                 return 0;
352         }
353         case BLKRAGET:
354         case BLKFRAGET:
355                 if (!arg)
356                         return -EINVAL;
357                 bdi = blk_get_backing_dev_info(bdev);
358                 if (bdi == NULL)
359                         return -ENOTTY;
360                 return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512);
361         case BLKROGET:
362                 return put_int(arg, bdev_read_only(bdev) != 0);
363         case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
364                 return put_int(arg, block_size(bdev));
365         case BLKSSZGET: /* get block device logical block size */
366                 return put_int(arg, bdev_logical_block_size(bdev));
367         case BLKPBSZGET: /* get block device physical block size */
368                 return put_uint(arg, bdev_physical_block_size(bdev));
369         case BLKIOMIN:
370                 return put_uint(arg, bdev_io_min(bdev));
371         case BLKIOOPT:
372                 return put_uint(arg, bdev_io_opt(bdev));
373         case BLKALIGNOFF:
374                 return put_int(arg, bdev_alignment_offset(bdev));
375         case BLKDISCARDZEROES:
376                 return put_uint(arg, bdev_discard_zeroes_data(bdev));
377         case BLKSECTGET:
378                 return put_ushort(arg, queue_max_sectors(bdev_get_queue(bdev)));
379         case BLKROTATIONAL:
380                 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
381         case BLKRASET:
382         case BLKFRASET:
383                 if(!capable(CAP_SYS_ADMIN))
384                         return -EACCES;
385                 bdi = blk_get_backing_dev_info(bdev);
386                 if (bdi == NULL)
387                         return -ENOTTY;
388                 bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
389                 return 0;
390         case BLKBSZSET:
391                 /* set the logical block size */
392                 if (!capable(CAP_SYS_ADMIN))
393                         return -EACCES;
394                 if (!arg)
395                         return -EINVAL;
396                 if (get_user(n, (int __user *) arg))
397                         return -EFAULT;
398                 if (!(mode & FMODE_EXCL)) {
399                         bdgrab(bdev);
400                         if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
401                                 return -EBUSY;
402                 }
403                 ret = set_blocksize(bdev, n);
404                 if (!(mode & FMODE_EXCL))
405                         blkdev_put(bdev, mode | FMODE_EXCL);
406                 return ret;
407         case BLKPG:
408                 ret = blkpg_ioctl(bdev, (struct blkpg_ioctl_arg __user *) arg);
409                 break;
410         case BLKRRPART:
411                 ret = blkdev_reread_part(bdev);
412                 break;
413         case BLKGETSIZE:
414                 size = i_size_read(bdev->bd_inode);
415                 if ((size >> 9) > ~0UL)
416                         return -EFBIG;
417                 return put_ulong(arg, size >> 9);
418         case BLKGETSIZE64:
419                 return put_u64(arg, i_size_read(bdev->bd_inode));
420         case BLKTRACESTART:
421         case BLKTRACESTOP:
422         case BLKTRACESETUP:
423         case BLKTRACETEARDOWN:
424                 ret = blk_trace_ioctl(bdev, cmd, (char __user *) arg);
425                 break;
426         default:
427                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
428         }
429         return ret;
430 }
431 EXPORT_SYMBOL_GPL(blkdev_ioctl);