block/ioctl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/capability.h>
   3 #include <linux/compat.h>
   4 #include <linux/blkdev.h>
   5 #include <linux/export.h>
   6 #include <linux/gfp.h>
   7 #include <linux/blkpg.h>
   8 #include <linux/hdreg.h>
   9 #include <linux/backing-dev.h>
  10 #include <linux/fs.h>
  11 #include <linux/blktrace_api.h>
  12 #include <linux/pr.h>
  13 #include <linux/uaccess.h>
  14 #include "blk.h"
  15
  16 static int blkpg_do_ioctl(struct block_device *bdev,
  17                           struct blkpg_partition __user *upart, int op)
  18 {
  19         struct blkpg_partition p;
  20         long long start, length;
  21
  22         if (!capable(CAP_SYS_ADMIN))
  23                 return -EACCES;
  24         if (copy_from_user(&p, upart, sizeof(struct blkpg_partition)))
  25                 return -EFAULT;
  26         if (bdev_is_partition(bdev))
  27                 return -EINVAL;
  28
  29         if (p.pno <= 0)
  30                 return -EINVAL;
  31
  32         if (op == BLKPG_DEL_PARTITION)
  33                 return bdev_del_partition(bdev, p.pno);
  34
  35         start = p.start >> SECTOR_SHIFT;
  36         length = p.length >> SECTOR_SHIFT;
  37
  38         switch (op) {
  39         case BLKPG_ADD_PARTITION:
  40                 /* check if partition is aligned to blocksize */
  41                 if (p.start & (bdev_logical_block_size(bdev) - 1))
  42                         return -EINVAL;
  43                 return bdev_add_partition(bdev, p.pno, start, length);
  44         case BLKPG_RESIZE_PARTITION:
  45                 return bdev_resize_partition(bdev, p.pno, start, length);
  46         default:
  47                 return -EINVAL;
  48         }
  49 }
  50
  51 static int blkpg_ioctl(struct block_device *bdev,
  52                        struct blkpg_ioctl_arg __user *arg)
  53 {
  54         struct blkpg_partition __user *udata;
  55         int op;
  56
  57         if (get_user(op, &arg->op) || get_user(udata, &arg->data))
  58                 return -EFAULT;
  59
  60         return blkpg_do_ioctl(bdev, udata, op);
  61 }
  62
  63 #ifdef CONFIG_COMPAT
  64 struct compat_blkpg_ioctl_arg {
  65         compat_int_t op;
  66         compat_int_t flags;
  67         compat_int_t datalen;
  68         compat_caddr_t data;
  69 };
  70
  71 static int compat_blkpg_ioctl(struct block_device *bdev,
  72                               struct compat_blkpg_ioctl_arg __user *arg)
  73 {
  74         compat_caddr_t udata;
  75         int op;
  76
  77         if (get_user(op, &arg->op) || get_user(udata, &arg->data))
  78                 return -EFAULT;
  79
  80         return blkpg_do_ioctl(bdev, compat_ptr(udata), op);
  81 }
  82 #endif
  83
  84 static int blkdev_reread_part(struct block_device *bdev)
  85 {
  86         int ret;
  87
  88         if (!disk_part_scan_enabled(bdev->bd_disk) || bdev_is_partition(bdev))
  89                 return -EINVAL;
  90         if (!capable(CAP_SYS_ADMIN))
  91                 return -EACCES;
  92
  93         mutex_lock(&bdev->bd_mutex);
  94         ret = bdev_disk_changed(bdev, false);
  95         mutex_unlock(&bdev->bd_mutex);
  96
  97         return ret;
  98 }
  99
 100 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
 101                 unsigned long arg, unsigned long flags)
 102 {
 103         uint64_t range[2];
 104         uint64_t start, len;
 105         struct request_queue *q = bdev_get_queue(bdev);
 106         int err;
 107
 108         if (!(mode & FMODE_WRITE))
 109                 return -EBADF;
 110
 111         if (!blk_queue_discard(q))
 112                 return -EOPNOTSUPP;
 113
 114         if (copy_from_user(range, (void __user *)arg, sizeof(range)))
 115                 return -EFAULT;
 116
 117         start = range[0];
 118         len = range[1];
 119
 120         if (start & 511)
 121                 return -EINVAL;
 122         if (len & 511)
 123                 return -EINVAL;
 124
 125         if (start + len > i_size_read(bdev->bd_inode))
 126                 return -EINVAL;
 127
 128         err = truncate_bdev_range(bdev, mode, start, start + len - 1);
 129         if (err)
 130                 return err;
 131
 132         return blkdev_issue_discard(bdev, start >> 9, len >> 9,
 133                                     GFP_KERNEL, flags);
 134 }
 135
 136 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
 137                 unsigned long arg)
 138 {
 139         uint64_t range[2];
 140         uint64_t start, end, len;
 141         int err;
 142
 143         if (!(mode & FMODE_WRITE))
 144                 return -EBADF;
 145
 146         if (copy_from_user(range, (void __user *)arg, sizeof(range)))
 147                 return -EFAULT;
 148
 149         start = range[0];
 150         len = range[1];
 151         end = start + len - 1;
 152
 153         if (start & 511)
 154                 return -EINVAL;
 155         if (len & 511)
 156                 return -EINVAL;
 157         if (end >= (uint64_t)i_size_read(bdev->bd_inode))
 158                 return -EINVAL;
 159         if (end < start)
 160                 return -EINVAL;
 161
 162         /* Invalidate the page cache, including dirty pages */
 163         err = truncate_bdev_range(bdev, mode, start, end);
 164         if (err)
 165                 return err;
 166
 167         return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
 168                         BLKDEV_ZERO_NOUNMAP);
 169 }
 170
 171 static int put_ushort(unsigned short __user *argp, unsigned short val)
 172 {
 173         return put_user(val, argp);
 174 }
 175
 176 static int put_int(int __user *argp, int val)
 177 {
 178         return put_user(val, argp);
 179 }
 180
 181 static int put_uint(unsigned int __user *argp, unsigned int val)
 182 {
 183         return put_user(val, argp);
 184 }
 185
 186 static int put_long(long __user *argp, long val)
 187 {
 188         return put_user(val, argp);
 189 }
 190
 191 static int put_ulong(unsigned long __user *argp, unsigned long val)
 192 {
 193         return put_user(val, argp);
 194 }
 195
 196 static int put_u64(u64 __user *argp, u64 val)
 197 {
 198         return put_user(val, argp);
 199 }
 200
 201 #ifdef CONFIG_COMPAT
 202 static int compat_put_long(compat_long_t __user *argp, long val)
 203 {
 204         return put_user(val, argp);
 205 }
 206
 207 static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val)
 208 {
 209         return put_user(val, argp);
 210 }
 211 #endif
 212
 213 #ifdef CONFIG_COMPAT
 214 /*
 215  * This is the equivalent of compat_ptr_ioctl(), to be used by block
 216  * drivers that implement only commands that are completely compatible
 217  * between 32-bit and 64-bit user space
 218  */
 219 int blkdev_compat_ptr_ioctl(struct block_device *bdev, fmode_t mode,
 220                         unsigned cmd, unsigned long arg)
 221 {
 222         struct gendisk *disk = bdev->bd_disk;
 223
 224         if (disk->fops->ioctl)
 225                 return disk->fops->ioctl(bdev, mode, cmd,
 226                                          (unsigned long)compat_ptr(arg));
 227
 228         return -ENOIOCTLCMD;
 229 }
 230 EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
 231 #endif
 232
 233 static int blkdev_pr_register(struct block_device *bdev,
 234                 struct pr_registration __user *arg)
 235 {
 236         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 237         struct pr_registration reg;
 238
 239         if (!capable(CAP_SYS_ADMIN))
 240                 return -EPERM;
 241         if (!ops || !ops->pr_register)
 242                 return -EOPNOTSUPP;
 243         if (copy_from_user(&reg, arg, sizeof(reg)))
 244                 return -EFAULT;
 245
 246         if (reg.flags & ~PR_FL_IGNORE_KEY)
 247                 return -EOPNOTSUPP;
 248         return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
 249 }
 250
 251 static int blkdev_pr_reserve(struct block_device *bdev,
 252                 struct pr_reservation __user *arg)
 253 {
 254         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 255         struct pr_reservation rsv;
 256
 257         if (!capable(CAP_SYS_ADMIN))
 258                 return -EPERM;
 259         if (!ops || !ops->pr_reserve)
 260                 return -EOPNOTSUPP;
 261         if (copy_from_user(&rsv, arg, sizeof(rsv)))
 262                 return -EFAULT;
 263
 264         if (rsv.flags & ~PR_FL_IGNORE_KEY)
 265                 return -EOPNOTSUPP;
 266         return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
 267 }
 268
 269 static int blkdev_pr_release(struct block_device *bdev,
 270                 struct pr_reservation __user *arg)
 271 {
 272         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 273         struct pr_reservation rsv;
 274
 275         if (!capable(CAP_SYS_ADMIN))
 276                 return -EPERM;
 277         if (!ops || !ops->pr_release)
 278                 return -EOPNOTSUPP;
 279         if (copy_from_user(&rsv, arg, sizeof(rsv)))
 280                 return -EFAULT;
 281
 282         if (rsv.flags)
 283                 return -EOPNOTSUPP;
 284         return ops->pr_release(bdev, rsv.key, rsv.type);
 285 }
 286
 287 static int blkdev_pr_preempt(struct block_device *bdev,
 288                 struct pr_preempt __user *arg, bool abort)
 289 {
 290         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 291         struct pr_preempt p;
 292
 293         if (!capable(CAP_SYS_ADMIN))
 294                 return -EPERM;
 295         if (!ops || !ops->pr_preempt)
 296                 return -EOPNOTSUPP;
 297         if (copy_from_user(&p, arg, sizeof(p)))
 298                 return -EFAULT;
 299
 300         if (p.flags)
 301                 return -EOPNOTSUPP;
 302         return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
 303 }
 304
 305 static int blkdev_pr_clear(struct block_device *bdev,
 306                 struct pr_clear __user *arg)
 307 {
 308         const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
 309         struct pr_clear c;
 310
 311         if (!capable(CAP_SYS_ADMIN))
 312                 return -EPERM;
 313         if (!ops || !ops->pr_clear)
 314                 return -EOPNOTSUPP;
 315         if (copy_from_user(&c, arg, sizeof(c)))
 316                 return -EFAULT;
 317
 318         if (c.flags)
 319                 return -EOPNOTSUPP;
 320         return ops->pr_clear(bdev, c.key);
 321 }
 322
 323 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
 324                 unsigned cmd, unsigned long arg)
 325 {
 326         if (!capable(CAP_SYS_ADMIN))
 327                 return -EACCES;
 328         fsync_bdev(bdev);
 329         invalidate_bdev(bdev);
 330         return 0;
 331 }
 332
 333 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
 334                 unsigned cmd, unsigned long arg)
 335 {
 336         int ret, n;
 337
 338         if (!capable(CAP_SYS_ADMIN))
 339                 return -EACCES;
 340
 341         if (get_user(n, (int __user *)arg))
 342                 return -EFAULT;
 343         if (bdev->bd_disk->fops->set_read_only) {
 344                 ret = bdev->bd_disk->fops->set_read_only(bdev, n);
 345                 if (ret)
 346                         return ret;
 347         }
 348         bdev->bd_read_only = n;
 349         return 0;
 350 }
 351
 352 static int blkdev_getgeo(struct block_device *bdev,
 353                 struct hd_geometry __user *argp)
 354 {
 355         struct gendisk *disk = bdev->bd_disk;
 356         struct hd_geometry geo;
 357         int ret;
 358
 359         if (!argp)
 360                 return -EINVAL;
 361         if (!disk->fops->getgeo)
 362                 return -ENOTTY;
 363
 364         /*
 365          * We need to set the startsect first, the driver may
 366          * want to override it.
 367          */
 368         memset(&geo, 0, sizeof(geo));
 369         geo.start = get_start_sect(bdev);
 370         ret = disk->fops->getgeo(bdev, &geo);
 371         if (ret)
 372                 return ret;
 373         if (copy_to_user(argp, &geo, sizeof(geo)))
 374                 return -EFAULT;
 375         return 0;
 376 }
 377
 378 #ifdef CONFIG_COMPAT
 379 struct compat_hd_geometry {
 380         unsigned char heads;
 381         unsigned char sectors;
 382         unsigned short cylinders;
 383         u32 start;
 384 };
 385
 386 static int compat_hdio_getgeo(struct block_device *bdev,
 387                               struct compat_hd_geometry __user *ugeo)
 388 {
 389         struct gendisk *disk = bdev->bd_disk;
 390         struct hd_geometry geo;
 391         int ret;
 392
 393         if (!ugeo)
 394                 return -EINVAL;
 395         if (!disk->fops->getgeo)
 396                 return -ENOTTY;
 397
 398         memset(&geo, 0, sizeof(geo));
 399         /*
 400          * We need to set the startsect first, the driver may
 401          * want to override it.
 402          */
 403         geo.start = get_start_sect(bdev);
 404         ret = disk->fops->getgeo(bdev, &geo);
 405         if (ret)
 406                 return ret;
 407
 408         ret = copy_to_user(ugeo, &geo, 4);
 409         ret |= put_user(geo.start, &ugeo->start);
 410         if (ret)
 411                 ret = -EFAULT;
 412
 413         return ret;
 414 }
 415 #endif
 416
 417 /* set the logical block size */
 418 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
 419                 int __user *argp)
 420 {
 421         int ret, n;
 422
 423         if (!capable(CAP_SYS_ADMIN))
 424                 return -EACCES;
 425         if (!argp)
 426                 return -EINVAL;
 427         if (get_user(n, argp))
 428                 return -EFAULT;
 429
 430         if (mode & FMODE_EXCL)
 431                 return set_blocksize(bdev, n);
 432
 433         if (IS_ERR(blkdev_get_by_dev(bdev->bd_dev, mode | FMODE_EXCL, &bdev)))
 434                 return -EBUSY;
 435         ret = set_blocksize(bdev, n);
 436         blkdev_put(bdev, mode | FMODE_EXCL);
 437
 438         return ret;
 439 }
 440
 441 /*
 442  * Common commands that are handled the same way on native and compat
 443  * user space. Note the separate arg/argp parameters that are needed
 444  * to deal with the compat_ptr() conversion.
 445  */
 446 static int blkdev_common_ioctl(struct block_device *bdev, fmode_t mode,
 447                                 unsigned cmd, unsigned long arg, void __user *argp)
 448 {
 449         unsigned int max_sectors;
 450
 451         switch (cmd) {
 452         case BLKFLSBUF:
 453                 return blkdev_flushbuf(bdev, mode, cmd, arg);
 454         case BLKROSET:
 455                 return blkdev_roset(bdev, mode, cmd, arg);
 456         case BLKDISCARD:
 457                 return blk_ioctl_discard(bdev, mode, arg, 0);
 458         case BLKSECDISCARD:
 459                 return blk_ioctl_discard(bdev, mode, arg,
 460                                 BLKDEV_DISCARD_SECURE);
 461         case BLKZEROOUT:
 462                 return blk_ioctl_zeroout(bdev, mode, arg);
 463         case BLKREPORTZONE:
 464                 return blkdev_report_zones_ioctl(bdev, mode, cmd, arg);
 465         case BLKRESETZONE:
 466         case BLKOPENZONE:
 467         case BLKCLOSEZONE:
 468         case BLKFINISHZONE:
 469                 return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
 470         case BLKGETZONESZ:
 471                 return put_uint(argp, bdev_zone_sectors(bdev));
 472         case BLKGETNRZONES:
 473                 return put_uint(argp, blkdev_nr_zones(bdev->bd_disk));
 474         case BLKROGET:
 475                 return put_int(argp, bdev_read_only(bdev) != 0);
 476         case BLKSSZGET: /* get block device logical block size */
 477                 return put_int(argp, bdev_logical_block_size(bdev));
 478         case BLKPBSZGET: /* get block device physical block size */
 479                 return put_uint(argp, bdev_physical_block_size(bdev));
 480         case BLKIOMIN:
 481                 return put_uint(argp, bdev_io_min(bdev));
 482         case BLKIOOPT:
 483                 return put_uint(argp, bdev_io_opt(bdev));
 484         case BLKALIGNOFF:
 485                 return put_int(argp, bdev_alignment_offset(bdev));
 486         case BLKDISCARDZEROES:
 487                 return put_uint(argp, 0);
 488         case BLKSECTGET:
 489                 max_sectors = min_t(unsigned int, USHRT_MAX,
 490                                     queue_max_sectors(bdev_get_queue(bdev)));
 491                 return put_ushort(argp, max_sectors);
 492         case BLKROTATIONAL:
 493                 return put_ushort(argp, !blk_queue_nonrot(bdev_get_queue(bdev)));
 494         case BLKRASET:
 495         case BLKFRASET:
 496                 if(!capable(CAP_SYS_ADMIN))
 497                         return -EACCES;
 498                 bdev->bd_bdi->ra_pages = (arg * 512) / PAGE_SIZE;
 499                 return 0;
 500         case BLKRRPART:
 501                 return blkdev_reread_part(bdev);
 502         case BLKTRACESTART:
 503         case BLKTRACESTOP:
 504         case BLKTRACETEARDOWN:
 505                 return blk_trace_ioctl(bdev, cmd, argp);
 506         case IOC_PR_REGISTER:
 507                 return blkdev_pr_register(bdev, argp);
 508         case IOC_PR_RESERVE:
 509                 return blkdev_pr_reserve(bdev, argp);
 510         case IOC_PR_RELEASE:
 511                 return blkdev_pr_release(bdev, argp);
 512         case IOC_PR_PREEMPT:
 513                 return blkdev_pr_preempt(bdev, argp, false);
 514         case IOC_PR_PREEMPT_ABORT:
 515                 return blkdev_pr_preempt(bdev, argp, true);
 516         case IOC_PR_CLEAR:
 517                 return blkdev_pr_clear(bdev, argp);
 518         default:
 519                 return -ENOIOCTLCMD;
 520         }
 521 }
 522
 523 /*
 524  * Always keep this in sync with compat_blkdev_ioctl()
 525  * to handle all incompatible commands in both functions.
 526  *
 527  * New commands must be compatible and go into blkdev_common_ioctl
 528  */
 529 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
 530                         unsigned long arg)
 531 {
 532         int ret;
 533         loff_t size;
 534         void __user *argp = (void __user *)arg;
 535
 536         switch (cmd) {
 537         /* These need separate implementations for the data structure */
 538         case HDIO_GETGEO:
 539                 return blkdev_getgeo(bdev, argp);
 540         case BLKPG:
 541                 return blkpg_ioctl(bdev, argp);
 542
 543         /* Compat mode returns 32-bit data instead of 'long' */
 544         case BLKRAGET:
 545         case BLKFRAGET:
 546                 if (!argp)
 547                         return -EINVAL;
 548                 return put_long(argp, (bdev->bd_bdi->ra_pages*PAGE_SIZE) / 512);
 549         case BLKGETSIZE:
 550                 size = i_size_read(bdev->bd_inode);
 551                 if ((size >> 9) > ~0UL)
 552                         return -EFBIG;
 553                 return put_ulong(argp, size >> 9);
 554
 555         /* The data is compatible, but the command number is different */
 556         case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
 557                 return put_int(argp, block_size(bdev));
 558         case BLKBSZSET:
 559                 return blkdev_bszset(bdev, mode, argp);
 560         case BLKGETSIZE64:
 561                 return put_u64(argp, i_size_read(bdev->bd_inode));
 562
 563         /* Incompatible alignment on i386 */
 564         case BLKTRACESETUP:
 565                 return blk_trace_ioctl(bdev, cmd, argp);
 566         default:
 567                 break;
 568         }
 569
 570         ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
 571         if (ret != -ENOIOCTLCMD)
 572                 return ret;
 573
 574         if (!bdev->bd_disk->fops->ioctl)
 575                 return -ENOTTY;
 576         return bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
 577 }
 578 EXPORT_SYMBOL_GPL(blkdev_ioctl); /* for /dev/raw */
 579
 580 #ifdef CONFIG_COMPAT
 581
 582 #define BLKBSZGET_32            _IOR(0x12, 112, int)
 583 #define BLKBSZSET_32            _IOW(0x12, 113, int)
 584 #define BLKGETSIZE64_32         _IOR(0x12, 114, int)
 585
 586 /* Most of the generic ioctls are handled in the normal fallback path.
 587    This assumes the blkdev's low level compat_ioctl always returns
 588    ENOIOCTLCMD for unknown ioctls. */
 589 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 590 {
 591         int ret;
 592         void __user *argp = compat_ptr(arg);
 593         struct block_device *bdev = I_BDEV(file->f_mapping->host);
 594         struct gendisk *disk = bdev->bd_disk;
 595         fmode_t mode = file->f_mode;
 596         loff_t size;
 597
 598         /*
 599          * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
 600          * to updated it before every ioctl.
 601          */
 602         if (file->f_flags & O_NDELAY)
 603                 mode |= FMODE_NDELAY;
 604         else
 605                 mode &= ~FMODE_NDELAY;
 606
 607         switch (cmd) {
 608         /* These need separate implementations for the data structure */
 609         case HDIO_GETGEO:
 610                 return compat_hdio_getgeo(bdev, argp);
 611         case BLKPG:
 612                 return compat_blkpg_ioctl(bdev, argp);
 613
 614         /* Compat mode returns 32-bit data instead of 'long' */
 615         case BLKRAGET:
 616         case BLKFRAGET:
 617                 if (!argp)
 618                         return -EINVAL;
 619                 return compat_put_long(argp,
 620                                (bdev->bd_bdi->ra_pages * PAGE_SIZE) / 512);
 621         case BLKGETSIZE:
 622                 size = i_size_read(bdev->bd_inode);
 623                 if ((size >> 9) > ~0UL)
 624                         return -EFBIG;
 625                 return compat_put_ulong(argp, size >> 9);
 626
 627         /* The data is compatible, but the command number is different */
 628         case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
 629                 return put_int(argp, bdev_logical_block_size(bdev));
 630         case BLKBSZSET_32:
 631                 return blkdev_bszset(bdev, mode, argp);
 632         case BLKGETSIZE64_32:
 633                 return put_u64(argp, i_size_read(bdev->bd_inode));
 634
 635         /* Incompatible alignment on i386 */
 636         case BLKTRACESETUP32:
 637                 return blk_trace_ioctl(bdev, cmd, argp);
 638         default:
 639                 break;
 640         }
 641
 642         ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
 643         if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
 644                 ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
 645
 646         return ret;
 647 }
 648 #endif