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 PTR_ERR_OR_ZERO(part);
  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 /*
 154  * This is an exported API for the block driver, and will not
 155  * acquire bd_mutex. This API should be used in case that
 156  * caller has held bd_mutex already.
 157  */
 158 int __blkdev_reread_part(struct block_device *bdev)
 159 {
 160         struct gendisk *disk = bdev->bd_disk;
 161
 162         if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
 163                 return -EINVAL;
 164         if (!capable(CAP_SYS_ADMIN))
 165                 return -EACCES;
 166
 167         lockdep_assert_held(&bdev->bd_mutex);
 168
 169         return rescan_partitions(disk, bdev);
 170 }
 171 EXPORT_SYMBOL(__blkdev_reread_part);
 172
 173 /*
 174  * This is an exported API for the block driver, and will
 175  * try to acquire bd_mutex. If bd_mutex has been held already
 176  * in current context, please call __blkdev_reread_part().
 177  *
 178  * Make sure the held locks in current context aren't required
 179  * in open()/close() handler and I/O path for avoiding ABBA deadlock:
 180  * - bd_mutex is held before calling block driver's open/close
 181  *   handler
 182  * - reading partition table may submit I/O to the block device
 183  */
 184 int blkdev_reread_part(struct block_device *bdev)
 185 {
 186         int res;
 187
 188         mutex_lock(&bdev->bd_mutex);
 189         res = __blkdev_reread_part(bdev);
 190         mutex_unlock(&bdev->bd_mutex);
 191
 192         return res;
 193 }
 194 EXPORT_SYMBOL(blkdev_reread_part);
 195
 196 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
 197                 unsigned long arg, unsigned long flags)
 198 {
 199         uint64_t range[2];
 200         uint64_t start, len;
 201
 202         if (!(mode & FMODE_WRITE))
 203                 return -EBADF;
 204
 205         if (copy_from_user(range, (void __user *)arg, sizeof(range)))
 206                 return -EFAULT;
 207
 208         start = range[0];
 209         len = range[1];
 210
 211         if (start & 511)
 212                 return -EINVAL;
 213         if (len & 511)
 214                 return -EINVAL;
 215         start >>= 9;
 216         len >>= 9;
 217
 218         if (start + len > (i_size_read(bdev->bd_inode) >> 9))
 219                 return -EINVAL;
 220         return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
 221 }
 222
 223 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
 224                 unsigned long arg)
 225 {
 226         uint64_t range[2];
 227         uint64_t start, len;
 228
 229         if (!(mode & FMODE_WRITE))
 230                 return -EBADF;
 231
 232         if (copy_from_user(range, (void __user *)arg, sizeof(range)))
 233                 return -EFAULT;
 234
 235         start = range[0];
 236         len = range[1];
 237
 238         if (start & 511)
 239                 return -EINVAL;
 240         if (len & 511)
 241                 return -EINVAL;
 242         start >>= 9;
 243         len >>= 9;
 244
 245         if (start + len > (i_size_read(bdev->bd_inode) >> 9))
 246                 return -EINVAL;
 247
 248         return blkdev_issue_zeroout(bdev, start, len, GFP_KERNEL, false);
 249 }
 250
 251 static int put_ushort(unsigned long arg, unsigned short val)
 252 {
 253         return put_user(val, (unsigned short __user *)arg);
 254 }
 255
 256 static int put_int(unsigned long arg, int val)
 257 {
 258         return put_user(val, (int __user *)arg);
 259 }
 260
 261 static int put_uint(unsigned long arg, unsigned int val)
 262 {
 263         return put_user(val, (unsigned int __user *)arg);
 264 }
 265
 266 static int put_long(unsigned long arg, long val)
 267 {
 268         return put_user(val, (long __user *)arg);
 269 }
 270
 271 static int put_ulong(unsigned long arg, unsigned long val)
 272 {
 273         return put_user(val, (unsigned long __user *)arg);
 274 }
 275
 276 static int put_u64(unsigned long arg, u64 val)
 277 {
 278         return put_user(val, (u64 __user *)arg);
 279 }
 280
 281 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
 282                         unsigned cmd, unsigned long arg)
 283 {
 284         struct gendisk *disk = bdev->bd_disk;
 285
 286         if (disk->fops->ioctl)
 287                 return disk->fops->ioctl(bdev, mode, cmd, arg);
 288
 289         return -ENOTTY;
 290 }
 291 /*
 292  * For the record: _GPL here is only because somebody decided to slap it
 293  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
 294  * at all and could be open-coded without any exports by anybody who cares.
 295  */
 296 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
 297
 298 /*
 299  * Is it an unrecognized ioctl? The correct returns are either
 300  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
 301  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
 302  * code before returning.
 303  *
 304  * Confused drivers sometimes return EINVAL, which is wrong. It
 305  * means "I understood the ioctl command, but the parameters to
 306  * it were wrong".
 307  *
 308  * We should aim to just fix the broken drivers, the EINVAL case
 309  * should go away.
 310  */
 311 static inline int is_unrecognized_ioctl(int ret)
 312 {
 313         return  ret == -EINVAL ||
 314                 ret == -ENOTTY ||
 315                 ret == -ENOIOCTLCMD;
 316 }
 317
 318 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
 319                 unsigned cmd, unsigned long arg)
 320 {
 321         int ret;
 322
 323         if (!capable(CAP_SYS_ADMIN))
 324                 return -EACCES;
 325
 326         ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 327         if (!is_unrecognized_ioctl(ret))
 328                 return ret;
 329
 330         fsync_bdev(bdev);
 331         invalidate_bdev(bdev);
 332         return 0;
 333 }
 334
 335 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
 336                 unsigned cmd, unsigned long arg)
 337 {
 338         int ret, n;
 339
 340         ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 341         if (!is_unrecognized_ioctl(ret))
 342                 return ret;
 343         if (!capable(CAP_SYS_ADMIN))
 344                 return -EACCES;
 345         if (get_user(n, (int __user *)arg))
 346                 return -EFAULT;
 347         set_device_ro(bdev, n);
 348         return 0;
 349 }
 350
 351 static int blkdev_getgeo(struct block_device *bdev,
 352                 struct hd_geometry __user *argp)
 353 {
 354         struct gendisk *disk = bdev->bd_disk;
 355         struct hd_geometry geo;
 356         int ret;
 357
 358         if (!argp)
 359                 return -EINVAL;
 360         if (!disk->fops->getgeo)
 361                 return -ENOTTY;
 362
 363         /*
 364          * We need to set the startsect first, the driver may
 365          * want to override it.
 366          */
 367         memset(&geo, 0, sizeof(geo));
 368         geo.start = get_start_sect(bdev);
 369         ret = disk->fops->getgeo(bdev, &geo);
 370         if (ret)
 371                 return ret;
 372         if (copy_to_user(argp, &geo, sizeof(geo)))
 373                 return -EFAULT;
 374         return 0;
 375 }
 376
 377 /* set the logical block size */
 378 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
 379                 int __user *argp)
 380 {
 381         int ret, n;
 382
 383         if (!capable(CAP_SYS_ADMIN))
 384                 return -EACCES;
 385         if (!argp)
 386                 return -EINVAL;
 387         if (get_user(n, argp))
 388                 return -EFAULT;
 389
 390         if (!(mode & FMODE_EXCL)) {
 391                 bdgrab(bdev);
 392                 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
 393                         return -EBUSY;
 394         }
 395
 396         ret = set_blocksize(bdev, n);
 397         if (!(mode & FMODE_EXCL))
 398                 blkdev_put(bdev, mode | FMODE_EXCL);
 399         return ret;
 400 }
 401
 402 /*
 403  * always keep this in sync with compat_blkdev_ioctl()
 404  */
 405 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
 406                         unsigned long arg)
 407 {
 408         struct backing_dev_info *bdi;
 409         void __user *argp = (void __user *)arg;
 410         loff_t size;
 411         unsigned int max_sectors;
 412
 413         switch (cmd) {
 414         case BLKFLSBUF:
 415                 return blkdev_flushbuf(bdev, mode, cmd, arg);
 416         case BLKROSET:
 417                 return blkdev_roset(bdev, mode, cmd, arg);
 418         case BLKDISCARD:
 419                 return blk_ioctl_discard(bdev, mode, arg, 0);
 420         case BLKSECDISCARD:
 421                 return blk_ioctl_discard(bdev, mode, arg,
 422                                 BLKDEV_DISCARD_SECURE);
 423         case BLKZEROOUT:
 424                 return blk_ioctl_zeroout(bdev, mode, arg);
 425         case HDIO_GETGEO:
 426                 return blkdev_getgeo(bdev, argp);
 427         case BLKRAGET:
 428         case BLKFRAGET:
 429                 if (!arg)
 430                         return -EINVAL;
 431                 bdi = blk_get_backing_dev_info(bdev);
 432                 return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512);
 433         case BLKROGET:
 434                 return put_int(arg, bdev_read_only(bdev) != 0);
 435         case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
 436                 return put_int(arg, block_size(bdev));
 437         case BLKSSZGET: /* get block device logical block size */
 438                 return put_int(arg, bdev_logical_block_size(bdev));
 439         case BLKPBSZGET: /* get block device physical block size */
 440                 return put_uint(arg, bdev_physical_block_size(bdev));
 441         case BLKIOMIN:
 442                 return put_uint(arg, bdev_io_min(bdev));
 443         case BLKIOOPT:
 444                 return put_uint(arg, bdev_io_opt(bdev));
 445         case BLKALIGNOFF:
 446                 return put_int(arg, bdev_alignment_offset(bdev));
 447         case BLKDISCARDZEROES:
 448                 return put_uint(arg, bdev_discard_zeroes_data(bdev));
 449         case BLKSECTGET:
 450                 max_sectors = min_t(unsigned int, USHRT_MAX,
 451                                     queue_max_sectors(bdev_get_queue(bdev)));
 452                 return put_ushort(arg, max_sectors);
 453         case BLKROTATIONAL:
 454                 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
 455         case BLKRASET:
 456         case BLKFRASET:
 457                 if(!capable(CAP_SYS_ADMIN))
 458                         return -EACCES;
 459                 bdi = blk_get_backing_dev_info(bdev);
 460                 bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
 461                 return 0;
 462         case BLKBSZSET:
 463                 return blkdev_bszset(bdev, mode, argp);
 464         case BLKPG:
 465                 return blkpg_ioctl(bdev, argp);
 466         case BLKRRPART:
 467                 return blkdev_reread_part(bdev);
 468         case BLKGETSIZE:
 469                 size = i_size_read(bdev->bd_inode);
 470                 if ((size >> 9) > ~0UL)
 471                         return -EFBIG;
 472                 return put_ulong(arg, size >> 9);
 473         case BLKGETSIZE64:
 474                 return put_u64(arg, i_size_read(bdev->bd_inode));
 475         case BLKTRACESTART:
 476         case BLKTRACESTOP:
 477         case BLKTRACESETUP:
 478         case BLKTRACETEARDOWN:
 479                 return blk_trace_ioctl(bdev, cmd, argp);
 480         default:
 481                 return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 482         }
 483 }
 484 EXPORT_SYMBOL_GPL(blkdev_ioctl);