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         printk("%s %d:%d %llu %llu", (flags & BLKDEV_DISCARD_SECURE)?"SECDIS":"DIS", MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev),
 222                 (unsigned long long)start, (unsigned long long)len);
 223         return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
 224 }
 225
 226 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
 227                 unsigned long arg)
 228 {
 229         uint64_t range[2];
 230         uint64_t start, len;
 231
 232         if (!(mode & FMODE_WRITE))
 233                 return -EBADF;
 234
 235         if (copy_from_user(range, (void __user *)arg, sizeof(range)))
 236                 return -EFAULT;
 237
 238         start = range[0];
 239         len = range[1];
 240
 241         if (start & 511)
 242                 return -EINVAL;
 243         if (len & 511)
 244                 return -EINVAL;
 245         start >>= 9;
 246         len >>= 9;
 247
 248         if (start + len > (i_size_read(bdev->bd_inode) >> 9))
 249                 return -EINVAL;
 250
 251         return blkdev_issue_zeroout(bdev, start, len, GFP_KERNEL, false);
 252 }
 253
 254 static int put_ushort(unsigned long arg, unsigned short val)
 255 {
 256         return put_user(val, (unsigned short __user *)arg);
 257 }
 258
 259 static int put_int(unsigned long arg, int val)
 260 {
 261         return put_user(val, (int __user *)arg);
 262 }
 263
 264 static int put_uint(unsigned long arg, unsigned int val)
 265 {
 266         return put_user(val, (unsigned int __user *)arg);
 267 }
 268
 269 static int put_long(unsigned long arg, long val)
 270 {
 271         return put_user(val, (long __user *)arg);
 272 }
 273
 274 static int put_ulong(unsigned long arg, unsigned long val)
 275 {
 276         return put_user(val, (unsigned long __user *)arg);
 277 }
 278
 279 static int put_u64(unsigned long arg, u64 val)
 280 {
 281         return put_user(val, (u64 __user *)arg);
 282 }
 283
 284 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
 285                         unsigned cmd, unsigned long arg)
 286 {
 287         struct gendisk *disk = bdev->bd_disk;
 288
 289         if (disk->fops->ioctl)
 290                 return disk->fops->ioctl(bdev, mode, cmd, arg);
 291
 292         return -ENOTTY;
 293 }
 294 /*
 295  * For the record: _GPL here is only because somebody decided to slap it
 296  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
 297  * at all and could be open-coded without any exports by anybody who cares.
 298  */
 299 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
 300
 301 static int blkdev_pr_register(struct block_device *bdev,
 302                 struct pr_registration __user *arg)
 303 {
 304         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 305         struct pr_registration reg;
 306
 307         if (!capable(CAP_SYS_ADMIN))
 308                 return -EPERM;
 309         if (!ops || !ops->pr_register)
 310                 return -EOPNOTSUPP;
 311         if (copy_from_user(&reg, arg, sizeof(reg)))
 312                 return -EFAULT;
 313
 314         if (reg.flags & ~PR_FL_IGNORE_KEY)
 315                 return -EOPNOTSUPP;
 316         return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
 317 }
 318
 319 static int blkdev_pr_reserve(struct block_device *bdev,
 320                 struct pr_reservation __user *arg)
 321 {
 322         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 323         struct pr_reservation rsv;
 324
 325         if (!capable(CAP_SYS_ADMIN))
 326                 return -EPERM;
 327         if (!ops || !ops->pr_reserve)
 328                 return -EOPNOTSUPP;
 329         if (copy_from_user(&rsv, arg, sizeof(rsv)))
 330                 return -EFAULT;
 331
 332         if (rsv.flags & ~PR_FL_IGNORE_KEY)
 333                 return -EOPNOTSUPP;
 334         return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
 335 }
 336
 337 static int blkdev_pr_release(struct block_device *bdev,
 338                 struct pr_reservation __user *arg)
 339 {
 340         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 341         struct pr_reservation rsv;
 342
 343         if (!capable(CAP_SYS_ADMIN))
 344                 return -EPERM;
 345         if (!ops || !ops->pr_release)
 346                 return -EOPNOTSUPP;
 347         if (copy_from_user(&rsv, arg, sizeof(rsv)))
 348                 return -EFAULT;
 349
 350         if (rsv.flags)
 351                 return -EOPNOTSUPP;
 352         return ops->pr_release(bdev, rsv.key, rsv.type);
 353 }
 354
 355 static int blkdev_pr_preempt(struct block_device *bdev,
 356                 struct pr_preempt __user *arg, bool abort)
 357 {
 358         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 359         struct pr_preempt p;
 360
 361         if (!capable(CAP_SYS_ADMIN))
 362                 return -EPERM;
 363         if (!ops || !ops->pr_preempt)
 364                 return -EOPNOTSUPP;
 365         if (copy_from_user(&p, arg, sizeof(p)))
 366                 return -EFAULT;
 367
 368         if (p.flags)
 369                 return -EOPNOTSUPP;
 370         return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
 371 }
 372
 373 static int blkdev_pr_clear(struct block_device *bdev,
 374                 struct pr_clear __user *arg)
 375 {
 376         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 377         struct pr_clear c;
 378
 379         if (!capable(CAP_SYS_ADMIN))
 380                 return -EPERM;
 381         if (!ops || !ops->pr_clear)
 382                 return -EOPNOTSUPP;
 383         if (copy_from_user(&c, arg, sizeof(c)))
 384                 return -EFAULT;
 385
 386         if (c.flags)
 387                 return -EOPNOTSUPP;
 388         return ops->pr_clear(bdev, c.key);
 389 }
 390
 391 /*
 392  * Is it an unrecognized ioctl? The correct returns are either
 393  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
 394  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
 395  * code before returning.
 396  *
 397  * Confused drivers sometimes return EINVAL, which is wrong. It
 398  * means "I understood the ioctl command, but the parameters to
 399  * it were wrong".
 400  *
 401  * We should aim to just fix the broken drivers, the EINVAL case
 402  * should go away.
 403  */
 404 static inline int is_unrecognized_ioctl(int ret)
 405 {
 406         return  ret == -EINVAL ||
 407                 ret == -ENOTTY ||
 408                 ret == -ENOIOCTLCMD;
 409 }
 410
 411 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
 412                 unsigned cmd, unsigned long arg)
 413 {
 414         int ret;
 415
 416         if (!capable(CAP_SYS_ADMIN))
 417                 return -EACCES;
 418
 419         ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 420         if (!is_unrecognized_ioctl(ret))
 421                 return ret;
 422
 423         fsync_bdev(bdev);
 424         invalidate_bdev(bdev);
 425         return 0;
 426 }
 427
 428 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
 429                 unsigned cmd, unsigned long arg)
 430 {
 431         int ret, n;
 432
 433         ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 434         if (!is_unrecognized_ioctl(ret))
 435                 return ret;
 436         if (!capable(CAP_SYS_ADMIN))
 437                 return -EACCES;
 438         if (get_user(n, (int __user *)arg))
 439                 return -EFAULT;
 440         set_device_ro(bdev, n);
 441         return 0;
 442 }
 443
 444 static int blkdev_getgeo(struct block_device *bdev,
 445                 struct hd_geometry __user *argp)
 446 {
 447         struct gendisk *disk = bdev->bd_disk;
 448         struct hd_geometry geo;
 449         int ret;
 450
 451         if (!argp)
 452                 return -EINVAL;
 453         if (!disk->fops->getgeo)
 454                 return -ENOTTY;
 455
 456         /*
 457          * We need to set the startsect first, the driver may
 458          * want to override it.
 459          */
 460         memset(&geo, 0, sizeof(geo));
 461         geo.start = get_start_sect(bdev);
 462         ret = disk->fops->getgeo(bdev, &geo);
 463         if (ret)
 464                 return ret;
 465         if (copy_to_user(argp, &geo, sizeof(geo)))
 466                 return -EFAULT;
 467         return 0;
 468 }
 469
 470 /* set the logical block size */
 471 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
 472                 int __user *argp)
 473 {
 474         int ret, n;
 475
 476         if (!capable(CAP_SYS_ADMIN))
 477                 return -EACCES;
 478         if (!argp)
 479                 return -EINVAL;
 480         if (get_user(n, argp))
 481                 return -EFAULT;
 482
 483         if (!(mode & FMODE_EXCL)) {
 484                 bdgrab(bdev);
 485                 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
 486                         return -EBUSY;
 487         }
 488
 489         ret = set_blocksize(bdev, n);
 490         if (!(mode & FMODE_EXCL))
 491                 blkdev_put(bdev, mode | FMODE_EXCL);
 492         return ret;
 493 }
 494
 495 /*
 496  * always keep this in sync with compat_blkdev_ioctl()
 497  */
 498 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
 499                         unsigned long arg)
 500 {
 501         struct backing_dev_info *bdi;
 502         void __user *argp = (void __user *)arg;
 503         loff_t size;
 504         unsigned int max_sectors;
 505
 506         switch (cmd) {
 507         case BLKFLSBUF:
 508                 return blkdev_flushbuf(bdev, mode, cmd, arg);
 509         case BLKROSET:
 510                 return blkdev_roset(bdev, mode, cmd, arg);
 511         case BLKDISCARD:
 512                 return blk_ioctl_discard(bdev, mode, arg, 0);
 513         case BLKSECDISCARD:
 514                 return blk_ioctl_discard(bdev, mode, arg,
 515                                 BLKDEV_DISCARD_SECURE);
 516         case BLKZEROOUT:
 517                 return blk_ioctl_zeroout(bdev, mode, arg);
 518         case HDIO_GETGEO:
 519                 return blkdev_getgeo(bdev, argp);
 520         case BLKRAGET:
 521         case BLKFRAGET:
 522                 if (!arg)
 523                         return -EINVAL;
 524                 bdi = blk_get_backing_dev_info(bdev);
 525                 return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512);
 526         case BLKROGET:
 527                 return put_int(arg, bdev_read_only(bdev) != 0);
 528         case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
 529                 return put_int(arg, block_size(bdev));
 530         case BLKSSZGET: /* get block device logical block size */
 531                 return put_int(arg, bdev_logical_block_size(bdev));
 532         case BLKPBSZGET: /* get block device physical block size */
 533                 return put_uint(arg, bdev_physical_block_size(bdev));
 534         case BLKIOMIN:
 535                 return put_uint(arg, bdev_io_min(bdev));
 536         case BLKIOOPT:
 537                 return put_uint(arg, bdev_io_opt(bdev));
 538         case BLKALIGNOFF:
 539                 return put_int(arg, bdev_alignment_offset(bdev));
 540         case BLKDISCARDZEROES:
 541                 return put_uint(arg, bdev_discard_zeroes_data(bdev));
 542         case BLKSECTGET:
 543                 max_sectors = min_t(unsigned int, USHRT_MAX,
 544                                     queue_max_sectors(bdev_get_queue(bdev)));
 545                 return put_ushort(arg, max_sectors);
 546         case BLKROTATIONAL:
 547                 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
 548         case BLKRASET:
 549         case BLKFRASET:
 550                 if(!capable(CAP_SYS_ADMIN))
 551                         return -EACCES;
 552                 bdi = blk_get_backing_dev_info(bdev);
 553                 bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
 554                 return 0;
 555         case BLKBSZSET:
 556                 return blkdev_bszset(bdev, mode, argp);
 557         case BLKPG:
 558                 return blkpg_ioctl(bdev, argp);
 559         case BLKRRPART:
 560                 return blkdev_reread_part(bdev);
 561         case BLKGETSIZE:
 562                 size = i_size_read(bdev->bd_inode);
 563                 if ((size >> 9) > ~0UL)
 564                         return -EFBIG;
 565                 return put_ulong(arg, size >> 9);
 566         case BLKGETSIZE64:
 567                 return put_u64(arg, i_size_read(bdev->bd_inode));
 568         case BLKTRACESTART:
 569         case BLKTRACESTOP:
 570         case BLKTRACESETUP:
 571         case BLKTRACETEARDOWN:
 572                 return blk_trace_ioctl(bdev, cmd, argp);
 573         case IOC_PR_REGISTER:
 574                 return blkdev_pr_register(bdev, argp);
 575         case IOC_PR_RESERVE:
 576                 return blkdev_pr_reserve(bdev, argp);
 577         case IOC_PR_RELEASE:
 578                 return blkdev_pr_release(bdev, argp);
 579         case IOC_PR_PREEMPT:
 580                 return blkdev_pr_preempt(bdev, argp, false);
 581         case IOC_PR_PREEMPT_ABORT:
 582                 return blkdev_pr_preempt(bdev, argp, true);
 583         case IOC_PR_CLEAR:
 584                 return blkdev_pr_clear(bdev, argp);
 585         default:
 586                 return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 587         }
 588 }
 589 EXPORT_SYMBOL_GPL(blkdev_ioctl);