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 <linux/pr.h>
  11 #include <asm/uaccess.h>
  12
  13 static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
  14 {
  15         struct block_device *bdevp;
  16         struct gendisk *disk;
  17         struct hd_struct *part, *lpart;
  18         struct blkpg_ioctl_arg a;
  19         struct blkpg_partition p;
  20         struct disk_part_iter piter;
  21         long long start, length;
  22         int partno;
  23
  24         if (!capable(CAP_SYS_ADMIN))
  25                 return -EACCES;
  26         if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
  27                 return -EFAULT;
  28         if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
  29                 return -EFAULT;
  30         disk = bdev->bd_disk;
  31         if (bdev != bdev->bd_contains)
  32                 return -EINVAL;
  33         partno = p.pno;
  34         if (partno <= 0)
  35                 return -EINVAL;
  36         switch (a.op) {
  37                 case BLKPG_ADD_PARTITION:
  38                         start = p.start >> 9;
  39                         length = p.length >> 9;
  40                         /* check for fit in a hd_struct */
  41                         if (sizeof(sector_t) == sizeof(long) &&
  42                             sizeof(long long) > sizeof(long)) {
  43                                 long pstart = start, plength = length;
  44                                 if (pstart != start || plength != length
  45                                     || pstart < 0 || plength < 0 || partno > 65535)
  46                                         return -EINVAL;
  47                         }
  48
  49                         mutex_lock(&bdev->bd_mutex);
  50
  51                         /* overlap? */
  52                         disk_part_iter_init(&piter, disk,
  53                                             DISK_PITER_INCL_EMPTY);
  54                         while ((part = disk_part_iter_next(&piter))) {
  55                                 if (!(start + length <= part->start_sect ||
  56                                       start >= part->start_sect + part->nr_sects)) {
  57                                         disk_part_iter_exit(&piter);
  58                                         mutex_unlock(&bdev->bd_mutex);
  59                                         return -EBUSY;
  60                                 }
  61                         }
  62                         disk_part_iter_exit(&piter);
  63
  64                         /* all seems OK */
  65                         part = add_partition(disk, partno, start, length,
  66                                              ADDPART_FLAG_NONE, NULL);
  67                         mutex_unlock(&bdev->bd_mutex);
  68                         return PTR_ERR_OR_ZERO(part);
  69                 case BLKPG_DEL_PARTITION:
  70                         part = disk_get_part(disk, partno);
  71                         if (!part)
  72                                 return -ENXIO;
  73
  74                         bdevp = bdget(part_devt(part));
  75                         disk_put_part(part);
  76                         if (!bdevp)
  77                                 return -ENOMEM;
  78
  79                         mutex_lock(&bdevp->bd_mutex);
  80                         if (bdevp->bd_openers) {
  81                                 mutex_unlock(&bdevp->bd_mutex);
  82                                 bdput(bdevp);
  83                                 return -EBUSY;
  84                         }
  85                         /* all seems OK */
  86                         fsync_bdev(bdevp);
  87                         invalidate_bdev(bdevp);
  88
  89                         mutex_lock_nested(&bdev->bd_mutex, 1);
  90                         delete_partition(disk, partno);
  91                         mutex_unlock(&bdev->bd_mutex);
  92                         mutex_unlock(&bdevp->bd_mutex);
  93                         bdput(bdevp);
  94
  95                         return 0;
  96                 case BLKPG_RESIZE_PARTITION:
  97                         start = p.start >> 9;
  98                         /* new length of partition in bytes */
  99                         length = p.length >> 9;
 100                         /* check for fit in a hd_struct */
 101                         if (sizeof(sector_t) == sizeof(long) &&
 102                             sizeof(long long) > sizeof(long)) {
 103                                 long pstart = start, plength = length;
 104                                 if (pstart != start || plength != length
 105                                     || pstart < 0 || plength < 0)
 106                                         return -EINVAL;
 107                         }
 108                         part = disk_get_part(disk, partno);
 109                         if (!part)
 110                                 return -ENXIO;
 111                         bdevp = bdget(part_devt(part));
 112                         if (!bdevp) {
 113                                 disk_put_part(part);
 114                                 return -ENOMEM;
 115                         }
 116                         mutex_lock(&bdevp->bd_mutex);
 117                         mutex_lock_nested(&bdev->bd_mutex, 1);
 118                         if (start != part->start_sect) {
 119                                 mutex_unlock(&bdevp->bd_mutex);
 120                                 mutex_unlock(&bdev->bd_mutex);
 121                                 bdput(bdevp);
 122                                 disk_put_part(part);
 123                                 return -EINVAL;
 124                         }
 125                         /* overlap? */
 126                         disk_part_iter_init(&piter, disk,
 127                                             DISK_PITER_INCL_EMPTY);
 128                         while ((lpart = disk_part_iter_next(&piter))) {
 129                                 if (lpart->partno != partno &&
 130                                    !(start + length <= lpart->start_sect ||
 131                                    start >= lpart->start_sect + lpart->nr_sects)
 132                                    ) {
 133                                         disk_part_iter_exit(&piter);
 134                                         mutex_unlock(&bdevp->bd_mutex);
 135                                         mutex_unlock(&bdev->bd_mutex);
 136                                         bdput(bdevp);
 137                                         disk_put_part(part);
 138                                         return -EBUSY;
 139                                 }
 140                         }
 141                         disk_part_iter_exit(&piter);
 142                         part_nr_sects_write(part, (sector_t)length);
 143                         i_size_write(bdevp->bd_inode, p.length);
 144                         mutex_unlock(&bdevp->bd_mutex);
 145                         mutex_unlock(&bdev->bd_mutex);
 146                         bdput(bdevp);
 147                         disk_put_part(part);
 148                         return 0;
 149                 default:
 150                         return -EINVAL;
 151         }
 152 }
 153
 154 /*
 155  * This is an exported API for the block driver, and will not
 156  * acquire bd_mutex. This API should be used in case that
 157  * caller has held bd_mutex already.
 158  */
 159 int __blkdev_reread_part(struct block_device *bdev)
 160 {
 161         struct gendisk *disk = bdev->bd_disk;
 162
 163         if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
 164                 return -EINVAL;
 165         if (!capable(CAP_SYS_ADMIN))
 166                 return -EACCES;
 167
 168         lockdep_assert_held(&bdev->bd_mutex);
 169
 170         return rescan_partitions(disk, bdev);
 171 }
 172 EXPORT_SYMBOL(__blkdev_reread_part);
 173
 174 /*
 175  * This is an exported API for the block driver, and will
 176  * try to acquire bd_mutex. If bd_mutex has been held already
 177  * in current context, please call __blkdev_reread_part().
 178  *
 179  * Make sure the held locks in current context aren't required
 180  * in open()/close() handler and I/O path for avoiding ABBA deadlock:
 181  * - bd_mutex is held before calling block driver's open/close
 182  *   handler
 183  * - reading partition table may submit I/O to the block device
 184  */
 185 int blkdev_reread_part(struct block_device *bdev)
 186 {
 187         int res;
 188
 189         mutex_lock(&bdev->bd_mutex);
 190         res = __blkdev_reread_part(bdev);
 191         mutex_unlock(&bdev->bd_mutex);
 192
 193         return res;
 194 }
 195 EXPORT_SYMBOL(blkdev_reread_part);
 196
 197 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
 198                 unsigned long arg, unsigned long flags)
 199 {
 200         uint64_t range[2];
 201         uint64_t start, len;
 202
 203         if (!(mode & FMODE_WRITE))
 204                 return -EBADF;
 205
 206         if (copy_from_user(range, (void __user *)arg, sizeof(range)))
 207                 return -EFAULT;
 208
 209         start = range[0];
 210         len = range[1];
 211
 212         if (start & 511)
 213                 return -EINVAL;
 214         if (len & 511)
 215                 return -EINVAL;
 216         start >>= 9;
 217         len >>= 9;
 218
 219         if (start + len > (i_size_read(bdev->bd_inode) >> 9))
 220                 return -EINVAL;
 221         return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
 222 }
 223
 224 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
 225                 unsigned long arg)
 226 {
 227         uint64_t range[2];
 228         struct address_space *mapping;
 229         uint64_t start, end, len;
 230
 231         if (!(mode & FMODE_WRITE))
 232                 return -EBADF;
 233
 234         if (copy_from_user(range, (void __user *)arg, sizeof(range)))
 235                 return -EFAULT;
 236
 237         start = range[0];
 238         len = range[1];
 239         end = start + len - 1;
 240
 241         if (start & 511)
 242                 return -EINVAL;
 243         if (len & 511)
 244                 return -EINVAL;
 245         if (end >= (uint64_t)i_size_read(bdev->bd_inode))
 246                 return -EINVAL;
 247         if (end < start)
 248                 return -EINVAL;
 249
 250         /* Invalidate the page cache, including dirty pages */
 251         mapping = bdev->bd_inode->i_mapping;
 252         truncate_inode_pages_range(mapping, start, end);
 253
 254         return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
 255                                     false);
 256 }
 257
 258 static int put_ushort(unsigned long arg, unsigned short val)
 259 {
 260         return put_user(val, (unsigned short __user *)arg);
 261 }
 262
 263 static int put_int(unsigned long arg, int val)
 264 {
 265         return put_user(val, (int __user *)arg);
 266 }
 267
 268 static int put_uint(unsigned long arg, unsigned int val)
 269 {
 270         return put_user(val, (unsigned int __user *)arg);
 271 }
 272
 273 static int put_long(unsigned long arg, long val)
 274 {
 275         return put_user(val, (long __user *)arg);
 276 }
 277
 278 static int put_ulong(unsigned long arg, unsigned long val)
 279 {
 280         return put_user(val, (unsigned long __user *)arg);
 281 }
 282
 283 static int put_u64(unsigned long arg, u64 val)
 284 {
 285         return put_user(val, (u64 __user *)arg);
 286 }
 287
 288 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
 289                         unsigned cmd, unsigned long arg)
 290 {
 291         struct gendisk *disk = bdev->bd_disk;
 292
 293         if (disk->fops->ioctl)
 294                 return disk->fops->ioctl(bdev, mode, cmd, arg);
 295
 296         return -ENOTTY;
 297 }
 298 /*
 299  * For the record: _GPL here is only because somebody decided to slap it
 300  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
 301  * at all and could be open-coded without any exports by anybody who cares.
 302  */
 303 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
 304
 305 static int blkdev_pr_register(struct block_device *bdev,
 306                 struct pr_registration __user *arg)
 307 {
 308         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 309         struct pr_registration reg;
 310
 311         if (!capable(CAP_SYS_ADMIN))
 312                 return -EPERM;
 313         if (!ops || !ops->pr_register)
 314                 return -EOPNOTSUPP;
 315         if (copy_from_user(&reg, arg, sizeof(reg)))
 316                 return -EFAULT;
 317
 318         if (reg.flags & ~PR_FL_IGNORE_KEY)
 319                 return -EOPNOTSUPP;
 320         return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
 321 }
 322
 323 static int blkdev_pr_reserve(struct block_device *bdev,
 324                 struct pr_reservation __user *arg)
 325 {
 326         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 327         struct pr_reservation rsv;
 328
 329         if (!capable(CAP_SYS_ADMIN))
 330                 return -EPERM;
 331         if (!ops || !ops->pr_reserve)
 332                 return -EOPNOTSUPP;
 333         if (copy_from_user(&rsv, arg, sizeof(rsv)))
 334                 return -EFAULT;
 335
 336         if (rsv.flags & ~PR_FL_IGNORE_KEY)
 337                 return -EOPNOTSUPP;
 338         return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
 339 }
 340
 341 static int blkdev_pr_release(struct block_device *bdev,
 342                 struct pr_reservation __user *arg)
 343 {
 344         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 345         struct pr_reservation rsv;
 346
 347         if (!capable(CAP_SYS_ADMIN))
 348                 return -EPERM;
 349         if (!ops || !ops->pr_release)
 350                 return -EOPNOTSUPP;
 351         if (copy_from_user(&rsv, arg, sizeof(rsv)))
 352                 return -EFAULT;
 353
 354         if (rsv.flags)
 355                 return -EOPNOTSUPP;
 356         return ops->pr_release(bdev, rsv.key, rsv.type);
 357 }
 358
 359 static int blkdev_pr_preempt(struct block_device *bdev,
 360                 struct pr_preempt __user *arg, bool abort)
 361 {
 362         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 363         struct pr_preempt p;
 364
 365         if (!capable(CAP_SYS_ADMIN))
 366                 return -EPERM;
 367         if (!ops || !ops->pr_preempt)
 368                 return -EOPNOTSUPP;
 369         if (copy_from_user(&p, arg, sizeof(p)))
 370                 return -EFAULT;
 371
 372         if (p.flags)
 373                 return -EOPNOTSUPP;
 374         return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
 375 }
 376
 377 static int blkdev_pr_clear(struct block_device *bdev,
 378                 struct pr_clear __user *arg)
 379 {
 380         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 381         struct pr_clear c;
 382
 383         if (!capable(CAP_SYS_ADMIN))
 384                 return -EPERM;
 385         if (!ops || !ops->pr_clear)
 386                 return -EOPNOTSUPP;
 387         if (copy_from_user(&c, arg, sizeof(c)))
 388                 return -EFAULT;
 389
 390         if (c.flags)
 391                 return -EOPNOTSUPP;
 392         return ops->pr_clear(bdev, c.key);
 393 }
 394
 395 /*
 396  * Is it an unrecognized ioctl? The correct returns are either
 397  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
 398  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
 399  * code before returning.
 400  *
 401  * Confused drivers sometimes return EINVAL, which is wrong. It
 402  * means "I understood the ioctl command, but the parameters to
 403  * it were wrong".
 404  *
 405  * We should aim to just fix the broken drivers, the EINVAL case
 406  * should go away.
 407  */
 408 static inline int is_unrecognized_ioctl(int ret)
 409 {
 410         return  ret == -EINVAL ||
 411                 ret == -ENOTTY ||
 412                 ret == -ENOIOCTLCMD;
 413 }
 414
 415 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
 416                 unsigned cmd, unsigned long arg)
 417 {
 418         int ret;
 419
 420         if (!capable(CAP_SYS_ADMIN))
 421                 return -EACCES;
 422
 423         ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 424         if (!is_unrecognized_ioctl(ret))
 425                 return ret;
 426
 427         fsync_bdev(bdev);
 428         invalidate_bdev(bdev);
 429         return 0;
 430 }
 431
 432 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
 433                 unsigned cmd, unsigned long arg)
 434 {
 435         int ret, n;
 436
 437         ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 438         if (!is_unrecognized_ioctl(ret))
 439                 return ret;
 440         if (!capable(CAP_SYS_ADMIN))
 441                 return -EACCES;
 442         if (get_user(n, (int __user *)arg))
 443                 return -EFAULT;
 444         set_device_ro(bdev, n);
 445         return 0;
 446 }
 447
 448 static int blkdev_getgeo(struct block_device *bdev,
 449                 struct hd_geometry __user *argp)
 450 {
 451         struct gendisk *disk = bdev->bd_disk;
 452         struct hd_geometry geo;
 453         int ret;
 454
 455         if (!argp)
 456                 return -EINVAL;
 457         if (!disk->fops->getgeo)
 458                 return -ENOTTY;
 459
 460         /*
 461          * We need to set the startsect first, the driver may
 462          * want to override it.
 463          */
 464         memset(&geo, 0, sizeof(geo));
 465         geo.start = get_start_sect(bdev);
 466         ret = disk->fops->getgeo(bdev, &geo);
 467         if (ret)
 468                 return ret;
 469         if (copy_to_user(argp, &geo, sizeof(geo)))
 470                 return -EFAULT;
 471         return 0;
 472 }
 473
 474 /* set the logical block size */
 475 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
 476                 int __user *argp)
 477 {
 478         int ret, n;
 479
 480         if (!capable(CAP_SYS_ADMIN))
 481                 return -EACCES;
 482         if (!argp)
 483                 return -EINVAL;
 484         if (get_user(n, argp))
 485                 return -EFAULT;
 486
 487         if (!(mode & FMODE_EXCL)) {
 488                 bdgrab(bdev);
 489                 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
 490                         return -EBUSY;
 491         }
 492
 493         ret = set_blocksize(bdev, n);
 494         if (!(mode & FMODE_EXCL))
 495                 blkdev_put(bdev, mode | FMODE_EXCL);
 496         return ret;
 497 }
 498
 499 /*
 500  * always keep this in sync with compat_blkdev_ioctl()
 501  */
 502 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
 503                         unsigned long arg)
 504 {
 505         struct backing_dev_info *bdi;
 506         void __user *argp = (void __user *)arg;
 507         loff_t size;
 508         unsigned int max_sectors;
 509
 510         switch (cmd) {
 511         case BLKFLSBUF:
 512                 return blkdev_flushbuf(bdev, mode, cmd, arg);
 513         case BLKROSET:
 514                 return blkdev_roset(bdev, mode, cmd, arg);
 515         case BLKDISCARD:
 516                 return blk_ioctl_discard(bdev, mode, arg, 0);
 517         case BLKSECDISCARD:
 518                 return blk_ioctl_discard(bdev, mode, arg,
 519                                 BLKDEV_DISCARD_SECURE);
 520         case BLKZEROOUT:
 521                 return blk_ioctl_zeroout(bdev, mode, arg);
 522         case HDIO_GETGEO:
 523                 return blkdev_getgeo(bdev, argp);
 524         case BLKRAGET:
 525         case BLKFRAGET:
 526                 if (!arg)
 527                         return -EINVAL;
 528                 bdi = blk_get_backing_dev_info(bdev);
 529                 return put_long(arg, (bdi->ra_pages * PAGE_SIZE) / 512);
 530         case BLKROGET:
 531                 return put_int(arg, bdev_read_only(bdev) != 0);
 532         case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
 533                 return put_int(arg, block_size(bdev));
 534         case BLKSSZGET: /* get block device logical block size */
 535                 return put_int(arg, bdev_logical_block_size(bdev));
 536         case BLKPBSZGET: /* get block device physical block size */
 537                 return put_uint(arg, bdev_physical_block_size(bdev));
 538         case BLKIOMIN:
 539                 return put_uint(arg, bdev_io_min(bdev));
 540         case BLKIOOPT:
 541                 return put_uint(arg, bdev_io_opt(bdev));
 542         case BLKALIGNOFF:
 543                 return put_int(arg, bdev_alignment_offset(bdev));
 544         case BLKDISCARDZEROES:
 545                 return put_uint(arg, bdev_discard_zeroes_data(bdev));
 546         case BLKSECTGET:
 547                 max_sectors = min_t(unsigned int, USHRT_MAX,
 548                                     queue_max_sectors(bdev_get_queue(bdev)));
 549                 return put_ushort(arg, max_sectors);
 550         case BLKROTATIONAL:
 551                 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
 552         case BLKRASET:
 553         case BLKFRASET:
 554                 if(!capable(CAP_SYS_ADMIN))
 555                         return -EACCES;
 556                 bdi = blk_get_backing_dev_info(bdev);
 557                 bdi->ra_pages = (arg * 512) / PAGE_SIZE;
 558                 return 0;
 559         case BLKBSZSET:
 560                 return blkdev_bszset(bdev, mode, argp);
 561         case BLKPG:
 562                 return blkpg_ioctl(bdev, argp);
 563         case BLKRRPART:
 564                 return blkdev_reread_part(bdev);
 565         case BLKGETSIZE:
 566                 size = i_size_read(bdev->bd_inode);
 567                 if ((size >> 9) > ~0UL)
 568                         return -EFBIG;
 569                 return put_ulong(arg, size >> 9);
 570         case BLKGETSIZE64:
 571                 return put_u64(arg, i_size_read(bdev->bd_inode));
 572         case BLKTRACESTART:
 573         case BLKTRACESTOP:
 574         case BLKTRACESETUP:
 575         case BLKTRACETEARDOWN:
 576                 return blk_trace_ioctl(bdev, cmd, argp);
 577         case IOC_PR_REGISTER:
 578                 return blkdev_pr_register(bdev, argp);
 579         case IOC_PR_RESERVE:
 580                 return blkdev_pr_reserve(bdev, argp);
 581         case IOC_PR_RELEASE:
 582                 return blkdev_pr_release(bdev, argp);
 583         case IOC_PR_PREEMPT:
 584                 return blkdev_pr_preempt(bdev, argp, false);
 585         case IOC_PR_PREEMPT_ABORT:
 586                 return blkdev_pr_preempt(bdev, argp, true);
 587         case IOC_PR_CLEAR:
 588                 return blkdev_pr_clear(bdev, argp);
 589         default:
 590                 return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 591         }
 592 }
 593 EXPORT_SYMBOL_GPL(blkdev_ioctl);