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;
  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_part_scan_enabled(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 > (i_size_read(bdev->bd_inode) >> 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  * Is it an unrecognized ioctl? The correct returns are either
 184  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
 185  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
 186  * code before returning.
 187  *
 188  * Confused drivers sometimes return EINVAL, which is wrong. It
 189  * means "I understood the ioctl command, but the parameters to
 190  * it were wrong".
 191  *
 192  * We should aim to just fix the broken drivers, the EINVAL case
 193  * should go away.
 194  */
 195 static inline int is_unrecognized_ioctl(int ret)
 196 {
 197         return  ret == -EINVAL ||
 198                 ret == -ENOTTY ||
 199                 ret == -ENOIOCTLCMD;
 200 }
 201
 202 /*
 203  * always keep this in sync with compat_blkdev_ioctl()
 204  */
 205 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
 206                         unsigned long arg)
 207 {
 208         struct gendisk *disk = bdev->bd_disk;
 209         struct backing_dev_info *bdi;
 210         loff_t size;
 211         int ret, n;
 212
 213         switch(cmd) {
 214         case BLKFLSBUF:
 215                 if (!capable(CAP_SYS_ADMIN))
 216                         return -EACCES;
 217
 218                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 219                 if (!is_unrecognized_ioctl(ret))
 220                         return ret;
 221
 222                 fsync_bdev(bdev);
 223                 invalidate_bdev(bdev);
 224                 return 0;
 225
 226         case BLKROSET:
 227                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 228                 if (!is_unrecognized_ioctl(ret))
 229                         return ret;
 230                 if (!capable(CAP_SYS_ADMIN))
 231                         return -EACCES;
 232                 if (get_user(n, (int __user *)(arg)))
 233                         return -EFAULT;
 234                 set_device_ro(bdev, n);
 235                 return 0;
 236
 237         case BLKDISCARD:
 238         case BLKSECDISCARD: {
 239                 uint64_t range[2];
 240
 241                 if (!(mode & FMODE_WRITE))
 242                         return -EBADF;
 243
 244                 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
 245                         return -EFAULT;
 246
 247                 return blk_ioctl_discard(bdev, range[0], range[1],
 248                                          cmd == BLKSECDISCARD);
 249         }
 250
 251         case HDIO_GETGEO: {
 252                 struct hd_geometry geo;
 253
 254                 if (!arg)
 255                         return -EINVAL;
 256                 if (!disk->fops->getgeo)
 257                         return -ENOTTY;
 258
 259                 /*
 260                  * We need to set the startsect first, the driver may
 261                  * want to override it.
 262                  */
 263                 memset(&geo, 0, sizeof(geo));
 264                 geo.start = get_start_sect(bdev);
 265                 ret = disk->fops->getgeo(bdev, &geo);
 266                 if (ret)
 267                         return ret;
 268                 if (copy_to_user((struct hd_geometry __user *)arg, &geo,
 269                                         sizeof(geo)))
 270                         return -EFAULT;
 271                 return 0;
 272         }
 273         case BLKRAGET:
 274         case BLKFRAGET:
 275                 if (!arg)
 276                         return -EINVAL;
 277                 bdi = blk_get_backing_dev_info(bdev);
 278                 if (bdi == NULL)
 279                         return -ENOTTY;
 280                 return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512);
 281         case BLKROGET:
 282                 return put_int(arg, bdev_read_only(bdev) != 0);
 283         case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
 284                 return put_int(arg, block_size(bdev));
 285         case BLKSSZGET: /* get block device logical block size */
 286                 return put_int(arg, bdev_logical_block_size(bdev));
 287         case BLKPBSZGET: /* get block device physical block size */
 288                 return put_uint(arg, bdev_physical_block_size(bdev));
 289         case BLKIOMIN:
 290                 return put_uint(arg, bdev_io_min(bdev));
 291         case BLKIOOPT:
 292                 return put_uint(arg, bdev_io_opt(bdev));
 293         case BLKALIGNOFF:
 294                 return put_int(arg, bdev_alignment_offset(bdev));
 295         case BLKDISCARDZEROES:
 296                 return put_uint(arg, bdev_discard_zeroes_data(bdev));
 297         case BLKSECTGET:
 298                 return put_ushort(arg, queue_max_sectors(bdev_get_queue(bdev)));
 299         case BLKRASET:
 300         case BLKFRASET:
 301                 if(!capable(CAP_SYS_ADMIN))
 302                         return -EACCES;
 303                 bdi = blk_get_backing_dev_info(bdev);
 304                 if (bdi == NULL)
 305                         return -ENOTTY;
 306                 bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
 307                 return 0;
 308         case BLKBSZSET:
 309                 /* set the logical block size */
 310                 if (!capable(CAP_SYS_ADMIN))
 311                         return -EACCES;
 312                 if (!arg)
 313                         return -EINVAL;
 314                 if (get_user(n, (int __user *) arg))
 315                         return -EFAULT;
 316                 if (!(mode & FMODE_EXCL)) {
 317                         bdgrab(bdev);
 318                         if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
 319                                 return -EBUSY;
 320                 }
 321                 ret = set_blocksize(bdev, n);
 322                 if (!(mode & FMODE_EXCL))
 323                         blkdev_put(bdev, mode | FMODE_EXCL);
 324                 return ret;
 325         case BLKPG:
 326                 ret = blkpg_ioctl(bdev, (struct blkpg_ioctl_arg __user *) arg);
 327                 break;
 328         case BLKRRPART:
 329                 ret = blkdev_reread_part(bdev);
 330                 break;
 331         case BLKGETSIZE:
 332                 size = i_size_read(bdev->bd_inode);
 333                 if ((size >> 9) > ~0UL)
 334                         return -EFBIG;
 335                 return put_ulong(arg, size >> 9);
 336         case BLKGETSIZE64:
 337                 return put_u64(arg, i_size_read(bdev->bd_inode));
 338         case BLKTRACESTART:
 339         case BLKTRACESTOP:
 340         case BLKTRACESETUP:
 341         case BLKTRACETEARDOWN:
 342                 ret = blk_trace_ioctl(bdev, cmd, (char __user *) arg);
 343                 break;
 344         default:
 345                 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 346         }
 347         return ret;
 348 }
 349 EXPORT_SYMBOL_GPL(blkdev_ioctl);