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 != bdev->bd_contains) 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 /* check for fit in a hd_struct */ 39 if (sizeof(sector_t) < sizeof(long long)) { 40 long pstart = start, plength = length; 41 42 if (pstart != start || plength != length || pstart < 0 || 43 plength < 0 || p.pno > 65535) 44 return -EINVAL; 45 } 46 47 switch (op) { 48 case BLKPG_ADD_PARTITION: 49 /* check if partition is aligned to blocksize */ 50 if (p.start & (bdev_logical_block_size(bdev) - 1)) 51 return -EINVAL; 52 return bdev_add_partition(bdev, p.pno, start, length); 53 case BLKPG_RESIZE_PARTITION: 54 return bdev_resize_partition(bdev, p.pno, start, length); 55 default: 56 return -EINVAL; 57 } 58 } 59 60 static int blkpg_ioctl(struct block_device *bdev, 61 struct blkpg_ioctl_arg __user *arg) 62 { 63 struct blkpg_partition __user *udata; 64 int op; 65 66 if (get_user(op, &arg->op) || get_user(udata, &arg->data)) 67 return -EFAULT; 68 69 return blkpg_do_ioctl(bdev, udata, op); 70 } 71 72 #ifdef CONFIG_COMPAT 73 struct compat_blkpg_ioctl_arg { 74 compat_int_t op; 75 compat_int_t flags; 76 compat_int_t datalen; 77 compat_caddr_t data; 78 }; 79 80 static int compat_blkpg_ioctl(struct block_device *bdev, 81 struct compat_blkpg_ioctl_arg __user *arg) 82 { 83 compat_caddr_t udata; 84 int op; 85 86 if (get_user(op, &arg->op) || get_user(udata, &arg->data)) 87 return -EFAULT; 88 89 return blkpg_do_ioctl(bdev, compat_ptr(udata), op); 90 } 91 #endif 92 93 static int blkdev_reread_part(struct block_device *bdev) 94 { 95 int ret; 96 97 if (!disk_part_scan_enabled(bdev->bd_disk) || bdev != bdev->bd_contains) 98 return -EINVAL; 99 if (!capable(CAP_SYS_ADMIN)) 100 return -EACCES; 101 102 mutex_lock(&bdev->bd_mutex); 103 ret = bdev_disk_changed(bdev, false); 104 mutex_unlock(&bdev->bd_mutex); 105 106 return ret; 107 } 108 109 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode, 110 unsigned long arg, unsigned long flags) 111 { 112 uint64_t range[2]; 113 uint64_t start, len; 114 struct request_queue *q = bdev_get_queue(bdev); 115 struct address_space *mapping = bdev->bd_inode->i_mapping; 116 117 118 if (!(mode & FMODE_WRITE)) 119 return -EBADF; 120 121 if (!blk_queue_discard(q)) 122 return -EOPNOTSUPP; 123 124 if (copy_from_user(range, (void __user *)arg, sizeof(range))) 125 return -EFAULT; 126 127 start = range[0]; 128 len = range[1]; 129 130 if (start & 511) 131 return -EINVAL; 132 if (len & 511) 133 return -EINVAL; 134 135 if (start + len > i_size_read(bdev->bd_inode)) 136 return -EINVAL; 137 truncate_inode_pages_range(mapping, start, start + len - 1); 138 return blkdev_issue_discard(bdev, start >> 9, len >> 9, 139 GFP_KERNEL, flags); 140 } 141 142 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode, 143 unsigned long arg) 144 { 145 uint64_t range[2]; 146 struct address_space *mapping; 147 uint64_t start, end, len; 148 149 if (!(mode & FMODE_WRITE)) 150 return -EBADF; 151 152 if (copy_from_user(range, (void __user *)arg, sizeof(range))) 153 return -EFAULT; 154 155 start = range[0]; 156 len = range[1]; 157 end = start + len - 1; 158 159 if (start & 511) 160 return -EINVAL; 161 if (len & 511) 162 return -EINVAL; 163 if (end >= (uint64_t)i_size_read(bdev->bd_inode)) 164 return -EINVAL; 165 if (end < start) 166 return -EINVAL; 167 168 /* Invalidate the page cache, including dirty pages */ 169 mapping = bdev->bd_inode->i_mapping; 170 truncate_inode_pages_range(mapping, start, end); 171 172 return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL, 173 BLKDEV_ZERO_NOUNMAP); 174 } 175 176 static int put_ushort(unsigned short __user *argp, unsigned short val) 177 { 178 return put_user(val, argp); 179 } 180 181 static int put_int(int __user *argp, int val) 182 { 183 return put_user(val, argp); 184 } 185 186 static int put_uint(unsigned int __user *argp, unsigned int val) 187 { 188 return put_user(val, argp); 189 } 190 191 static int put_long(long __user *argp, long val) 192 { 193 return put_user(val, argp); 194 } 195 196 static int put_ulong(unsigned long __user *argp, unsigned long val) 197 { 198 return put_user(val, argp); 199 } 200 201 static int put_u64(u64 __user *argp, u64 val) 202 { 203 return put_user(val, argp); 204 } 205 206 #ifdef CONFIG_COMPAT 207 static int compat_put_long(compat_long_t __user *argp, long val) 208 { 209 return put_user(val, argp); 210 } 211 212 static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val) 213 { 214 return put_user(val, argp); 215 } 216 #endif 217 218 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode, 219 unsigned cmd, unsigned long arg) 220 { 221 struct gendisk *disk = bdev->bd_disk; 222 223 if (disk->fops->ioctl) 224 return disk->fops->ioctl(bdev, mode, cmd, arg); 225 226 return -ENOTTY; 227 } 228 /* 229 * For the record: _GPL here is only because somebody decided to slap it 230 * on the previous export. Sheer idiocy, since it wasn't copyrightable 231 * at all and could be open-coded without any exports by anybody who cares. 232 */ 233 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl); 234 235 #ifdef CONFIG_COMPAT 236 /* 237 * This is the equivalent of compat_ptr_ioctl(), to be used by block 238 * drivers that implement only commands that are completely compatible 239 * between 32-bit and 64-bit user space 240 */ 241 int blkdev_compat_ptr_ioctl(struct block_device *bdev, fmode_t mode, 242 unsigned cmd, unsigned long arg) 243 { 244 struct gendisk *disk = bdev->bd_disk; 245 246 if (disk->fops->ioctl) 247 return disk->fops->ioctl(bdev, mode, cmd, 248 (unsigned long)compat_ptr(arg)); 249 250 return -ENOIOCTLCMD; 251 } 252 EXPORT_SYMBOL(blkdev_compat_ptr_ioctl); 253 #endif 254 255 static int blkdev_pr_register(struct block_device *bdev, 256 struct pr_registration __user *arg) 257 { 258 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 259 struct pr_registration reg; 260 261 if (!capable(CAP_SYS_ADMIN)) 262 return -EPERM; 263 if (!ops || !ops->pr_register) 264 return -EOPNOTSUPP; 265 if (copy_from_user(®, arg, sizeof(reg))) 266 return -EFAULT; 267 268 if (reg.flags & ~PR_FL_IGNORE_KEY) 269 return -EOPNOTSUPP; 270 return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags); 271 } 272 273 static int blkdev_pr_reserve(struct block_device *bdev, 274 struct pr_reservation __user *arg) 275 { 276 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 277 struct pr_reservation rsv; 278 279 if (!capable(CAP_SYS_ADMIN)) 280 return -EPERM; 281 if (!ops || !ops->pr_reserve) 282 return -EOPNOTSUPP; 283 if (copy_from_user(&rsv, arg, sizeof(rsv))) 284 return -EFAULT; 285 286 if (rsv.flags & ~PR_FL_IGNORE_KEY) 287 return -EOPNOTSUPP; 288 return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags); 289 } 290 291 static int blkdev_pr_release(struct block_device *bdev, 292 struct pr_reservation __user *arg) 293 { 294 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 295 struct pr_reservation rsv; 296 297 if (!capable(CAP_SYS_ADMIN)) 298 return -EPERM; 299 if (!ops || !ops->pr_release) 300 return -EOPNOTSUPP; 301 if (copy_from_user(&rsv, arg, sizeof(rsv))) 302 return -EFAULT; 303 304 if (rsv.flags) 305 return -EOPNOTSUPP; 306 return ops->pr_release(bdev, rsv.key, rsv.type); 307 } 308 309 static int blkdev_pr_preempt(struct block_device *bdev, 310 struct pr_preempt __user *arg, bool abort) 311 { 312 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 313 struct pr_preempt p; 314 315 if (!capable(CAP_SYS_ADMIN)) 316 return -EPERM; 317 if (!ops || !ops->pr_preempt) 318 return -EOPNOTSUPP; 319 if (copy_from_user(&p, arg, sizeof(p))) 320 return -EFAULT; 321 322 if (p.flags) 323 return -EOPNOTSUPP; 324 return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort); 325 } 326 327 static int blkdev_pr_clear(struct block_device *bdev, 328 struct pr_clear __user *arg) 329 { 330 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 331 struct pr_clear c; 332 333 if (!capable(CAP_SYS_ADMIN)) 334 return -EPERM; 335 if (!ops || !ops->pr_clear) 336 return -EOPNOTSUPP; 337 if (copy_from_user(&c, arg, sizeof(c))) 338 return -EFAULT; 339 340 if (c.flags) 341 return -EOPNOTSUPP; 342 return ops->pr_clear(bdev, c.key); 343 } 344 345 /* 346 * Is it an unrecognized ioctl? The correct returns are either 347 * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a 348 * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl 349 * code before returning. 350 * 351 * Confused drivers sometimes return EINVAL, which is wrong. It 352 * means "I understood the ioctl command, but the parameters to 353 * it were wrong". 354 * 355 * We should aim to just fix the broken drivers, the EINVAL case 356 * should go away. 357 */ 358 static inline int is_unrecognized_ioctl(int ret) 359 { 360 return ret == -EINVAL || 361 ret == -ENOTTY || 362 ret == -ENOIOCTLCMD; 363 } 364 365 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode, 366 unsigned cmd, unsigned long arg) 367 { 368 int ret; 369 370 if (!capable(CAP_SYS_ADMIN)) 371 return -EACCES; 372 373 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg); 374 if (!is_unrecognized_ioctl(ret)) 375 return ret; 376 377 fsync_bdev(bdev); 378 invalidate_bdev(bdev); 379 return 0; 380 } 381 382 static int blkdev_roset(struct block_device *bdev, fmode_t mode, 383 unsigned cmd, unsigned long arg) 384 { 385 int ret, n; 386 387 if (!capable(CAP_SYS_ADMIN)) 388 return -EACCES; 389 390 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg); 391 if (!is_unrecognized_ioctl(ret)) 392 return ret; 393 if (get_user(n, (int __user *)arg)) 394 return -EFAULT; 395 set_device_ro(bdev, n); 396 return 0; 397 } 398 399 static int blkdev_getgeo(struct block_device *bdev, 400 struct hd_geometry __user *argp) 401 { 402 struct gendisk *disk = bdev->bd_disk; 403 struct hd_geometry geo; 404 int ret; 405 406 if (!argp) 407 return -EINVAL; 408 if (!disk->fops->getgeo) 409 return -ENOTTY; 410 411 /* 412 * We need to set the startsect first, the driver may 413 * want to override it. 414 */ 415 memset(&geo, 0, sizeof(geo)); 416 geo.start = get_start_sect(bdev); 417 ret = disk->fops->getgeo(bdev, &geo); 418 if (ret) 419 return ret; 420 if (copy_to_user(argp, &geo, sizeof(geo))) 421 return -EFAULT; 422 return 0; 423 } 424 425 #ifdef CONFIG_COMPAT 426 struct compat_hd_geometry { 427 unsigned char heads; 428 unsigned char sectors; 429 unsigned short cylinders; 430 u32 start; 431 }; 432 433 static int compat_hdio_getgeo(struct block_device *bdev, 434 struct compat_hd_geometry __user *ugeo) 435 { 436 struct gendisk *disk = bdev->bd_disk; 437 struct hd_geometry geo; 438 int ret; 439 440 if (!ugeo) 441 return -EINVAL; 442 if (!disk->fops->getgeo) 443 return -ENOTTY; 444 445 memset(&geo, 0, sizeof(geo)); 446 /* 447 * We need to set the startsect first, the driver may 448 * want to override it. 449 */ 450 geo.start = get_start_sect(bdev); 451 ret = disk->fops->getgeo(bdev, &geo); 452 if (ret) 453 return ret; 454 455 ret = copy_to_user(ugeo, &geo, 4); 456 ret |= put_user(geo.start, &ugeo->start); 457 if (ret) 458 ret = -EFAULT; 459 460 return ret; 461 } 462 #endif 463 464 /* set the logical block size */ 465 static int blkdev_bszset(struct block_device *bdev, fmode_t mode, 466 int __user *argp) 467 { 468 int ret, n; 469 470 if (!capable(CAP_SYS_ADMIN)) 471 return -EACCES; 472 if (!argp) 473 return -EINVAL; 474 if (get_user(n, argp)) 475 return -EFAULT; 476 477 if (!(mode & FMODE_EXCL)) { 478 bdgrab(bdev); 479 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0) 480 return -EBUSY; 481 } 482 483 ret = set_blocksize(bdev, n); 484 if (!(mode & FMODE_EXCL)) 485 blkdev_put(bdev, mode | FMODE_EXCL); 486 return ret; 487 } 488 489 /* 490 * Common commands that are handled the same way on native and compat 491 * user space. Note the separate arg/argp parameters that are needed 492 * to deal with the compat_ptr() conversion. 493 */ 494 static int blkdev_common_ioctl(struct block_device *bdev, fmode_t mode, 495 unsigned cmd, unsigned long arg, void __user *argp) 496 { 497 unsigned int max_sectors; 498 499 switch (cmd) { 500 case BLKFLSBUF: 501 return blkdev_flushbuf(bdev, mode, cmd, arg); 502 case BLKROSET: 503 return blkdev_roset(bdev, mode, cmd, arg); 504 case BLKDISCARD: 505 return blk_ioctl_discard(bdev, mode, arg, 0); 506 case BLKSECDISCARD: 507 return blk_ioctl_discard(bdev, mode, arg, 508 BLKDEV_DISCARD_SECURE); 509 case BLKZEROOUT: 510 return blk_ioctl_zeroout(bdev, mode, arg); 511 case BLKREPORTZONE: 512 return blkdev_report_zones_ioctl(bdev, mode, cmd, arg); 513 case BLKRESETZONE: 514 case BLKOPENZONE: 515 case BLKCLOSEZONE: 516 case BLKFINISHZONE: 517 return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg); 518 case BLKGETZONESZ: 519 return put_uint(argp, bdev_zone_sectors(bdev)); 520 case BLKGETNRZONES: 521 return put_uint(argp, blkdev_nr_zones(bdev->bd_disk)); 522 case BLKROGET: 523 return put_int(argp, bdev_read_only(bdev) != 0); 524 case BLKSSZGET: /* get block device logical block size */ 525 return put_int(argp, bdev_logical_block_size(bdev)); 526 case BLKPBSZGET: /* get block device physical block size */ 527 return put_uint(argp, bdev_physical_block_size(bdev)); 528 case BLKIOMIN: 529 return put_uint(argp, bdev_io_min(bdev)); 530 case BLKIOOPT: 531 return put_uint(argp, bdev_io_opt(bdev)); 532 case BLKALIGNOFF: 533 return put_int(argp, bdev_alignment_offset(bdev)); 534 case BLKDISCARDZEROES: 535 return put_uint(argp, 0); 536 case BLKSECTGET: 537 max_sectors = min_t(unsigned int, USHRT_MAX, 538 queue_max_sectors(bdev_get_queue(bdev))); 539 return put_ushort(argp, max_sectors); 540 case BLKROTATIONAL: 541 return put_ushort(argp, !blk_queue_nonrot(bdev_get_queue(bdev))); 542 case BLKRASET: 543 case BLKFRASET: 544 if(!capable(CAP_SYS_ADMIN)) 545 return -EACCES; 546 bdev->bd_bdi->ra_pages = (arg * 512) / PAGE_SIZE; 547 return 0; 548 case BLKRRPART: 549 return blkdev_reread_part(bdev); 550 case BLKTRACESTART: 551 case BLKTRACESTOP: 552 case BLKTRACETEARDOWN: 553 return blk_trace_ioctl(bdev, cmd, argp); 554 case IOC_PR_REGISTER: 555 return blkdev_pr_register(bdev, argp); 556 case IOC_PR_RESERVE: 557 return blkdev_pr_reserve(bdev, argp); 558 case IOC_PR_RELEASE: 559 return blkdev_pr_release(bdev, argp); 560 case IOC_PR_PREEMPT: 561 return blkdev_pr_preempt(bdev, argp, false); 562 case IOC_PR_PREEMPT_ABORT: 563 return blkdev_pr_preempt(bdev, argp, true); 564 case IOC_PR_CLEAR: 565 return blkdev_pr_clear(bdev, argp); 566 default: 567 return -ENOIOCTLCMD; 568 } 569 } 570 571 /* 572 * Always keep this in sync with compat_blkdev_ioctl() 573 * to handle all incompatible commands in both functions. 574 * 575 * New commands must be compatible and go into blkdev_common_ioctl 576 */ 577 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd, 578 unsigned long arg) 579 { 580 int ret; 581 loff_t size; 582 void __user *argp = (void __user *)arg; 583 584 switch (cmd) { 585 /* These need separate implementations for the data structure */ 586 case HDIO_GETGEO: 587 return blkdev_getgeo(bdev, argp); 588 case BLKPG: 589 return blkpg_ioctl(bdev, argp); 590 591 /* Compat mode returns 32-bit data instead of 'long' */ 592 case BLKRAGET: 593 case BLKFRAGET: 594 if (!argp) 595 return -EINVAL; 596 return put_long(argp, (bdev->bd_bdi->ra_pages*PAGE_SIZE) / 512); 597 case BLKGETSIZE: 598 size = i_size_read(bdev->bd_inode); 599 if ((size >> 9) > ~0UL) 600 return -EFBIG; 601 return put_ulong(argp, size >> 9); 602 603 /* The data is compatible, but the command number is different */ 604 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */ 605 return put_int(argp, block_size(bdev)); 606 case BLKBSZSET: 607 return blkdev_bszset(bdev, mode, argp); 608 case BLKGETSIZE64: 609 return put_u64(argp, i_size_read(bdev->bd_inode)); 610 611 /* Incompatible alignment on i386 */ 612 case BLKTRACESETUP: 613 return blk_trace_ioctl(bdev, cmd, argp); 614 default: 615 break; 616 } 617 618 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp); 619 if (ret == -ENOIOCTLCMD) 620 return __blkdev_driver_ioctl(bdev, mode, cmd, arg); 621 622 return ret; 623 } 624 EXPORT_SYMBOL_GPL(blkdev_ioctl); /* for /dev/raw */ 625 626 #ifdef CONFIG_COMPAT 627 628 #define BLKBSZGET_32 _IOR(0x12, 112, int) 629 #define BLKBSZSET_32 _IOW(0x12, 113, int) 630 #define BLKGETSIZE64_32 _IOR(0x12, 114, int) 631 632 /* Most of the generic ioctls are handled in the normal fallback path. 633 This assumes the blkdev's low level compat_ioctl always returns 634 ENOIOCTLCMD for unknown ioctls. */ 635 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg) 636 { 637 int ret; 638 void __user *argp = compat_ptr(arg); 639 struct inode *inode = file->f_mapping->host; 640 struct block_device *bdev = inode->i_bdev; 641 struct gendisk *disk = bdev->bd_disk; 642 fmode_t mode = file->f_mode; 643 loff_t size; 644 645 /* 646 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have 647 * to updated it before every ioctl. 648 */ 649 if (file->f_flags & O_NDELAY) 650 mode |= FMODE_NDELAY; 651 else 652 mode &= ~FMODE_NDELAY; 653 654 switch (cmd) { 655 /* These need separate implementations for the data structure */ 656 case HDIO_GETGEO: 657 return compat_hdio_getgeo(bdev, argp); 658 case BLKPG: 659 return compat_blkpg_ioctl(bdev, argp); 660 661 /* Compat mode returns 32-bit data instead of 'long' */ 662 case BLKRAGET: 663 case BLKFRAGET: 664 if (!argp) 665 return -EINVAL; 666 return compat_put_long(argp, 667 (bdev->bd_bdi->ra_pages * PAGE_SIZE) / 512); 668 case BLKGETSIZE: 669 size = i_size_read(bdev->bd_inode); 670 if ((size >> 9) > ~0UL) 671 return -EFBIG; 672 return compat_put_ulong(argp, size >> 9); 673 674 /* The data is compatible, but the command number is different */ 675 case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */ 676 return put_int(argp, bdev_logical_block_size(bdev)); 677 case BLKBSZSET_32: 678 return blkdev_bszset(bdev, mode, argp); 679 case BLKGETSIZE64_32: 680 return put_u64(argp, i_size_read(bdev->bd_inode)); 681 682 /* Incompatible alignment on i386 */ 683 case BLKTRACESETUP32: 684 return blk_trace_ioctl(bdev, cmd, argp); 685 default: 686 break; 687 } 688 689 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp); 690 if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl) 691 ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg); 692 693 return ret; 694 } 695 #endif 696