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