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/badblocks.h> 8 #include <linux/backing-dev.h> 9 #include <linux/fs.h> 10 #include <linux/blktrace_api.h> 11 #include <linux/pr.h> 12 #include <asm/uaccess.h> 13 14 static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg) 15 { 16 struct block_device *bdevp; 17 struct gendisk *disk; 18 struct hd_struct *part, *lpart; 19 struct blkpg_ioctl_arg a; 20 struct blkpg_partition p; 21 struct disk_part_iter piter; 22 long long start, length; 23 int partno; 24 25 if (!capable(CAP_SYS_ADMIN)) 26 return -EACCES; 27 if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg))) 28 return -EFAULT; 29 if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition))) 30 return -EFAULT; 31 disk = bdev->bd_disk; 32 if (bdev != bdev->bd_contains) 33 return -EINVAL; 34 partno = p.pno; 35 if (partno <= 0) 36 return -EINVAL; 37 switch (a.op) { 38 case BLKPG_ADD_PARTITION: 39 start = p.start >> 9; 40 length = p.length >> 9; 41 /* check for fit in a hd_struct */ 42 if (sizeof(sector_t) == sizeof(long) && 43 sizeof(long long) > sizeof(long)) { 44 long pstart = start, plength = length; 45 if (pstart != start || plength != length 46 || pstart < 0 || plength < 0 || partno > 65535) 47 return -EINVAL; 48 } 49 50 mutex_lock(&bdev->bd_mutex); 51 52 /* overlap? */ 53 disk_part_iter_init(&piter, disk, 54 DISK_PITER_INCL_EMPTY); 55 while ((part = disk_part_iter_next(&piter))) { 56 if (!(start + length <= part->start_sect || 57 start >= part->start_sect + part->nr_sects)) { 58 disk_part_iter_exit(&piter); 59 mutex_unlock(&bdev->bd_mutex); 60 return -EBUSY; 61 } 62 } 63 disk_part_iter_exit(&piter); 64 65 /* all seems OK */ 66 part = add_partition(disk, partno, start, length, 67 ADDPART_FLAG_NONE, NULL); 68 mutex_unlock(&bdev->bd_mutex); 69 return PTR_ERR_OR_ZERO(part); 70 case BLKPG_DEL_PARTITION: 71 part = disk_get_part(disk, partno); 72 if (!part) 73 return -ENXIO; 74 75 bdevp = bdget(part_devt(part)); 76 disk_put_part(part); 77 if (!bdevp) 78 return -ENOMEM; 79 80 mutex_lock(&bdevp->bd_mutex); 81 if (bdevp->bd_openers) { 82 mutex_unlock(&bdevp->bd_mutex); 83 bdput(bdevp); 84 return -EBUSY; 85 } 86 /* all seems OK */ 87 fsync_bdev(bdevp); 88 invalidate_bdev(bdevp); 89 90 mutex_lock_nested(&bdev->bd_mutex, 1); 91 delete_partition(disk, partno); 92 mutex_unlock(&bdev->bd_mutex); 93 mutex_unlock(&bdevp->bd_mutex); 94 bdput(bdevp); 95 96 return 0; 97 case BLKPG_RESIZE_PARTITION: 98 start = p.start >> 9; 99 /* new length of partition in bytes */ 100 length = p.length >> 9; 101 /* check for fit in a hd_struct */ 102 if (sizeof(sector_t) == sizeof(long) && 103 sizeof(long long) > sizeof(long)) { 104 long pstart = start, plength = length; 105 if (pstart != start || plength != length 106 || pstart < 0 || plength < 0) 107 return -EINVAL; 108 } 109 part = disk_get_part(disk, partno); 110 if (!part) 111 return -ENXIO; 112 bdevp = bdget(part_devt(part)); 113 if (!bdevp) { 114 disk_put_part(part); 115 return -ENOMEM; 116 } 117 mutex_lock(&bdevp->bd_mutex); 118 mutex_lock_nested(&bdev->bd_mutex, 1); 119 if (start != part->start_sect) { 120 mutex_unlock(&bdevp->bd_mutex); 121 mutex_unlock(&bdev->bd_mutex); 122 bdput(bdevp); 123 disk_put_part(part); 124 return -EINVAL; 125 } 126 /* overlap? */ 127 disk_part_iter_init(&piter, disk, 128 DISK_PITER_INCL_EMPTY); 129 while ((lpart = disk_part_iter_next(&piter))) { 130 if (lpart->partno != partno && 131 !(start + length <= lpart->start_sect || 132 start >= lpart->start_sect + lpart->nr_sects) 133 ) { 134 disk_part_iter_exit(&piter); 135 mutex_unlock(&bdevp->bd_mutex); 136 mutex_unlock(&bdev->bd_mutex); 137 bdput(bdevp); 138 disk_put_part(part); 139 return -EBUSY; 140 } 141 } 142 disk_part_iter_exit(&piter); 143 part_nr_sects_write(part, (sector_t)length); 144 i_size_write(bdevp->bd_inode, p.length); 145 mutex_unlock(&bdevp->bd_mutex); 146 mutex_unlock(&bdev->bd_mutex); 147 bdput(bdevp); 148 disk_put_part(part); 149 return 0; 150 default: 151 return -EINVAL; 152 } 153 } 154 155 /* 156 * This is an exported API for the block driver, and will not 157 * acquire bd_mutex. This API should be used in case that 158 * caller has held bd_mutex already. 159 */ 160 int __blkdev_reread_part(struct block_device *bdev) 161 { 162 struct gendisk *disk = bdev->bd_disk; 163 164 if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains) 165 return -EINVAL; 166 if (!capable(CAP_SYS_ADMIN)) 167 return -EACCES; 168 169 lockdep_assert_held(&bdev->bd_mutex); 170 171 return rescan_partitions(disk, bdev); 172 } 173 EXPORT_SYMBOL(__blkdev_reread_part); 174 175 /* 176 * This is an exported API for the block driver, and will 177 * try to acquire bd_mutex. If bd_mutex has been held already 178 * in current context, please call __blkdev_reread_part(). 179 * 180 * Make sure the held locks in current context aren't required 181 * in open()/close() handler and I/O path for avoiding ABBA deadlock: 182 * - bd_mutex is held before calling block driver's open/close 183 * handler 184 * - reading partition table may submit I/O to the block device 185 */ 186 int blkdev_reread_part(struct block_device *bdev) 187 { 188 int res; 189 190 mutex_lock(&bdev->bd_mutex); 191 res = __blkdev_reread_part(bdev); 192 mutex_unlock(&bdev->bd_mutex); 193 194 return res; 195 } 196 EXPORT_SYMBOL(blkdev_reread_part); 197 198 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode, 199 unsigned long arg, unsigned long flags) 200 { 201 uint64_t range[2]; 202 uint64_t start, len; 203 204 if (!(mode & FMODE_WRITE)) 205 return -EBADF; 206 207 if (copy_from_user(range, (void __user *)arg, sizeof(range))) 208 return -EFAULT; 209 210 start = range[0]; 211 len = range[1]; 212 213 if (start & 511) 214 return -EINVAL; 215 if (len & 511) 216 return -EINVAL; 217 start >>= 9; 218 len >>= 9; 219 220 if (start + len > (i_size_read(bdev->bd_inode) >> 9)) 221 return -EINVAL; 222 return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags); 223 } 224 225 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode, 226 unsigned long arg) 227 { 228 uint64_t range[2]; 229 uint64_t start, 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]; 239 240 if (start & 511) 241 return -EINVAL; 242 if (len & 511) 243 return -EINVAL; 244 start >>= 9; 245 len >>= 9; 246 247 if (start + len > (i_size_read(bdev->bd_inode) >> 9)) 248 return -EINVAL; 249 250 return blkdev_issue_zeroout(bdev, start, len, GFP_KERNEL, false); 251 } 252 253 static int put_ushort(unsigned long arg, unsigned short val) 254 { 255 return put_user(val, (unsigned short __user *)arg); 256 } 257 258 static int put_int(unsigned long arg, int val) 259 { 260 return put_user(val, (int __user *)arg); 261 } 262 263 static int put_uint(unsigned long arg, unsigned int val) 264 { 265 return put_user(val, (unsigned int __user *)arg); 266 } 267 268 static int put_long(unsigned long arg, long val) 269 { 270 return put_user(val, (long __user *)arg); 271 } 272 273 static int put_ulong(unsigned long arg, unsigned long val) 274 { 275 return put_user(val, (unsigned long __user *)arg); 276 } 277 278 static int put_u64(unsigned long arg, u64 val) 279 { 280 return put_user(val, (u64 __user *)arg); 281 } 282 283 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode, 284 unsigned cmd, unsigned long arg) 285 { 286 struct gendisk *disk = bdev->bd_disk; 287 288 if (disk->fops->ioctl) 289 return disk->fops->ioctl(bdev, mode, cmd, arg); 290 291 return -ENOTTY; 292 } 293 /* 294 * For the record: _GPL here is only because somebody decided to slap it 295 * on the previous export. Sheer idiocy, since it wasn't copyrightable 296 * at all and could be open-coded without any exports by anybody who cares. 297 */ 298 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl); 299 300 static int blkdev_pr_register(struct block_device *bdev, 301 struct pr_registration __user *arg) 302 { 303 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 304 struct pr_registration reg; 305 306 if (!capable(CAP_SYS_ADMIN)) 307 return -EPERM; 308 if (!ops || !ops->pr_register) 309 return -EOPNOTSUPP; 310 if (copy_from_user(®, arg, sizeof(reg))) 311 return -EFAULT; 312 313 if (reg.flags & ~PR_FL_IGNORE_KEY) 314 return -EOPNOTSUPP; 315 return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags); 316 } 317 318 static int blkdev_pr_reserve(struct block_device *bdev, 319 struct pr_reservation __user *arg) 320 { 321 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 322 struct pr_reservation rsv; 323 324 if (!capable(CAP_SYS_ADMIN)) 325 return -EPERM; 326 if (!ops || !ops->pr_reserve) 327 return -EOPNOTSUPP; 328 if (copy_from_user(&rsv, arg, sizeof(rsv))) 329 return -EFAULT; 330 331 if (rsv.flags & ~PR_FL_IGNORE_KEY) 332 return -EOPNOTSUPP; 333 return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags); 334 } 335 336 static int blkdev_pr_release(struct block_device *bdev, 337 struct pr_reservation __user *arg) 338 { 339 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 340 struct pr_reservation rsv; 341 342 if (!capable(CAP_SYS_ADMIN)) 343 return -EPERM; 344 if (!ops || !ops->pr_release) 345 return -EOPNOTSUPP; 346 if (copy_from_user(&rsv, arg, sizeof(rsv))) 347 return -EFAULT; 348 349 if (rsv.flags) 350 return -EOPNOTSUPP; 351 return ops->pr_release(bdev, rsv.key, rsv.type); 352 } 353 354 static int blkdev_pr_preempt(struct block_device *bdev, 355 struct pr_preempt __user *arg, bool abort) 356 { 357 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 358 struct pr_preempt p; 359 360 if (!capable(CAP_SYS_ADMIN)) 361 return -EPERM; 362 if (!ops || !ops->pr_preempt) 363 return -EOPNOTSUPP; 364 if (copy_from_user(&p, arg, sizeof(p))) 365 return -EFAULT; 366 367 if (p.flags) 368 return -EOPNOTSUPP; 369 return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort); 370 } 371 372 static int blkdev_pr_clear(struct block_device *bdev, 373 struct pr_clear __user *arg) 374 { 375 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 376 struct pr_clear c; 377 378 if (!capable(CAP_SYS_ADMIN)) 379 return -EPERM; 380 if (!ops || !ops->pr_clear) 381 return -EOPNOTSUPP; 382 if (copy_from_user(&c, arg, sizeof(c))) 383 return -EFAULT; 384 385 if (c.flags) 386 return -EOPNOTSUPP; 387 return ops->pr_clear(bdev, c.key); 388 } 389 390 /* 391 * Is it an unrecognized ioctl? The correct returns are either 392 * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a 393 * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl 394 * code before returning. 395 * 396 * Confused drivers sometimes return EINVAL, which is wrong. It 397 * means "I understood the ioctl command, but the parameters to 398 * it were wrong". 399 * 400 * We should aim to just fix the broken drivers, the EINVAL case 401 * should go away. 402 */ 403 static inline int is_unrecognized_ioctl(int ret) 404 { 405 return ret == -EINVAL || 406 ret == -ENOTTY || 407 ret == -ENOIOCTLCMD; 408 } 409 410 #ifdef CONFIG_FS_DAX 411 bool blkdev_dax_capable(struct block_device *bdev) 412 { 413 struct gendisk *disk = bdev->bd_disk; 414 415 if (!disk->fops->direct_access) 416 return false; 417 418 /* 419 * If the partition is not aligned on a page boundary, we can't 420 * do dax I/O to it. 421 */ 422 if ((bdev->bd_part->start_sect % (PAGE_SIZE / 512)) 423 || (bdev->bd_part->nr_sects % (PAGE_SIZE / 512))) 424 return false; 425 426 /* 427 * If the device has known bad blocks, force all I/O through the 428 * driver / page cache. 429 * 430 * TODO: support finer grained dax error handling 431 */ 432 if (disk->bb && disk->bb->count) 433 return false; 434 435 return true; 436 } 437 438 static int blkdev_daxset(struct block_device *bdev, unsigned long argp) 439 { 440 unsigned long arg; 441 int rc = 0; 442 443 if (!capable(CAP_SYS_ADMIN)) 444 return -EACCES; 445 446 if (get_user(arg, (int __user *)(argp))) 447 return -EFAULT; 448 arg = !!arg; 449 if (arg == !!(bdev->bd_inode->i_flags & S_DAX)) 450 return 0; 451 452 if (arg) 453 arg = S_DAX; 454 455 if (arg && !blkdev_dax_capable(bdev)) 456 return -ENOTTY; 457 458 inode_lock(bdev->bd_inode); 459 if (bdev->bd_map_count == 0) 460 inode_set_flags(bdev->bd_inode, arg, S_DAX); 461 else 462 rc = -EBUSY; 463 inode_unlock(bdev->bd_inode); 464 return rc; 465 } 466 #else 467 static int blkdev_daxset(struct block_device *bdev, int arg) 468 { 469 if (arg) 470 return -ENOTTY; 471 return 0; 472 } 473 #endif 474 475 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode, 476 unsigned cmd, unsigned long arg) 477 { 478 int ret; 479 480 if (!capable(CAP_SYS_ADMIN)) 481 return -EACCES; 482 483 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg); 484 if (!is_unrecognized_ioctl(ret)) 485 return ret; 486 487 fsync_bdev(bdev); 488 invalidate_bdev(bdev); 489 return 0; 490 } 491 492 static int blkdev_roset(struct block_device *bdev, fmode_t mode, 493 unsigned cmd, unsigned long arg) 494 { 495 int ret, n; 496 497 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg); 498 if (!is_unrecognized_ioctl(ret)) 499 return ret; 500 if (!capable(CAP_SYS_ADMIN)) 501 return -EACCES; 502 if (get_user(n, (int __user *)arg)) 503 return -EFAULT; 504 set_device_ro(bdev, n); 505 return 0; 506 } 507 508 static int blkdev_getgeo(struct block_device *bdev, 509 struct hd_geometry __user *argp) 510 { 511 struct gendisk *disk = bdev->bd_disk; 512 struct hd_geometry geo; 513 int ret; 514 515 if (!argp) 516 return -EINVAL; 517 if (!disk->fops->getgeo) 518 return -ENOTTY; 519 520 /* 521 * We need to set the startsect first, the driver may 522 * want to override it. 523 */ 524 memset(&geo, 0, sizeof(geo)); 525 geo.start = get_start_sect(bdev); 526 ret = disk->fops->getgeo(bdev, &geo); 527 if (ret) 528 return ret; 529 if (copy_to_user(argp, &geo, sizeof(geo))) 530 return -EFAULT; 531 return 0; 532 } 533 534 /* set the logical block size */ 535 static int blkdev_bszset(struct block_device *bdev, fmode_t mode, 536 int __user *argp) 537 { 538 int ret, n; 539 540 if (!capable(CAP_SYS_ADMIN)) 541 return -EACCES; 542 if (!argp) 543 return -EINVAL; 544 if (get_user(n, argp)) 545 return -EFAULT; 546 547 if (!(mode & FMODE_EXCL)) { 548 bdgrab(bdev); 549 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0) 550 return -EBUSY; 551 } 552 553 ret = set_blocksize(bdev, n); 554 if (!(mode & FMODE_EXCL)) 555 blkdev_put(bdev, mode | FMODE_EXCL); 556 return ret; 557 } 558 559 /* 560 * always keep this in sync with compat_blkdev_ioctl() 561 */ 562 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd, 563 unsigned long arg) 564 { 565 struct backing_dev_info *bdi; 566 void __user *argp = (void __user *)arg; 567 loff_t size; 568 unsigned int max_sectors; 569 570 switch (cmd) { 571 case BLKFLSBUF: 572 return blkdev_flushbuf(bdev, mode, cmd, arg); 573 case BLKROSET: 574 return blkdev_roset(bdev, mode, cmd, arg); 575 case BLKDISCARD: 576 return blk_ioctl_discard(bdev, mode, arg, 0); 577 case BLKSECDISCARD: 578 return blk_ioctl_discard(bdev, mode, arg, 579 BLKDEV_DISCARD_SECURE); 580 case BLKZEROOUT: 581 return blk_ioctl_zeroout(bdev, mode, arg); 582 case HDIO_GETGEO: 583 return blkdev_getgeo(bdev, argp); 584 case BLKRAGET: 585 case BLKFRAGET: 586 if (!arg) 587 return -EINVAL; 588 bdi = blk_get_backing_dev_info(bdev); 589 return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512); 590 case BLKROGET: 591 return put_int(arg, bdev_read_only(bdev) != 0); 592 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */ 593 return put_int(arg, block_size(bdev)); 594 case BLKSSZGET: /* get block device logical block size */ 595 return put_int(arg, bdev_logical_block_size(bdev)); 596 case BLKPBSZGET: /* get block device physical block size */ 597 return put_uint(arg, bdev_physical_block_size(bdev)); 598 case BLKIOMIN: 599 return put_uint(arg, bdev_io_min(bdev)); 600 case BLKIOOPT: 601 return put_uint(arg, bdev_io_opt(bdev)); 602 case BLKALIGNOFF: 603 return put_int(arg, bdev_alignment_offset(bdev)); 604 case BLKDISCARDZEROES: 605 return put_uint(arg, bdev_discard_zeroes_data(bdev)); 606 case BLKSECTGET: 607 max_sectors = min_t(unsigned int, USHRT_MAX, 608 queue_max_sectors(bdev_get_queue(bdev))); 609 return put_ushort(arg, max_sectors); 610 case BLKROTATIONAL: 611 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev))); 612 case BLKRASET: 613 case BLKFRASET: 614 if(!capable(CAP_SYS_ADMIN)) 615 return -EACCES; 616 bdi = blk_get_backing_dev_info(bdev); 617 bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE; 618 return 0; 619 case BLKBSZSET: 620 return blkdev_bszset(bdev, mode, argp); 621 case BLKPG: 622 return blkpg_ioctl(bdev, argp); 623 case BLKRRPART: 624 return blkdev_reread_part(bdev); 625 case BLKGETSIZE: 626 size = i_size_read(bdev->bd_inode); 627 if ((size >> 9) > ~0UL) 628 return -EFBIG; 629 return put_ulong(arg, size >> 9); 630 case BLKGETSIZE64: 631 return put_u64(arg, i_size_read(bdev->bd_inode)); 632 case BLKTRACESTART: 633 case BLKTRACESTOP: 634 case BLKTRACESETUP: 635 case BLKTRACETEARDOWN: 636 return blk_trace_ioctl(bdev, cmd, argp); 637 case BLKDAXSET: 638 return blkdev_daxset(bdev, arg); 639 case BLKDAXGET: 640 return put_int(arg, !!(bdev->bd_inode->i_flags & S_DAX)); 641 break; 642 case IOC_PR_REGISTER: 643 return blkdev_pr_register(bdev, argp); 644 case IOC_PR_RESERVE: 645 return blkdev_pr_reserve(bdev, argp); 646 case IOC_PR_RELEASE: 647 return blkdev_pr_release(bdev, argp); 648 case IOC_PR_PREEMPT: 649 return blkdev_pr_preempt(bdev, argp, false); 650 case IOC_PR_PREEMPT_ABORT: 651 return blkdev_pr_preempt(bdev, argp, true); 652 case IOC_PR_CLEAR: 653 return blkdev_pr_clear(bdev, argp); 654 default: 655 return __blkdev_driver_ioctl(bdev, mode, cmd, arg); 656 } 657 } 658 EXPORT_SYMBOL_GPL(blkdev_ioctl); 659