1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/fs/open.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 */ 7 8 #include <linux/string.h> 9 #include <linux/mm.h> 10 #include <linux/file.h> 11 #include <linux/fdtable.h> 12 #include <linux/fsnotify.h> 13 #include <linux/module.h> 14 #include <linux/tty.h> 15 #include <linux/namei.h> 16 #include <linux/backing-dev.h> 17 #include <linux/capability.h> 18 #include <linux/securebits.h> 19 #include <linux/security.h> 20 #include <linux/mount.h> 21 #include <linux/fcntl.h> 22 #include <linux/slab.h> 23 #include <linux/uaccess.h> 24 #include <linux/fs.h> 25 #include <linux/personality.h> 26 #include <linux/pagemap.h> 27 #include <linux/syscalls.h> 28 #include <linux/rcupdate.h> 29 #include <linux/audit.h> 30 #include <linux/falloc.h> 31 #include <linux/fs_struct.h> 32 #include <linux/ima.h> 33 #include <linux/dnotify.h> 34 #include <linux/compat.h> 35 36 #include "internal.h" 37 38 int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry, 39 loff_t length, unsigned int time_attrs, struct file *filp) 40 { 41 int ret; 42 struct iattr newattrs; 43 44 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */ 45 if (length < 0) 46 return -EINVAL; 47 48 newattrs.ia_size = length; 49 newattrs.ia_valid = ATTR_SIZE | time_attrs; 50 if (filp) { 51 newattrs.ia_file = filp; 52 newattrs.ia_valid |= ATTR_FILE; 53 } 54 55 /* Remove suid, sgid, and file capabilities on truncate too */ 56 ret = dentry_needs_remove_privs(dentry); 57 if (ret < 0) 58 return ret; 59 if (ret) 60 newattrs.ia_valid |= ret | ATTR_FORCE; 61 62 inode_lock(dentry->d_inode); 63 /* Note any delegations or leases have already been broken: */ 64 ret = notify_change(mnt_userns, dentry, &newattrs, NULL); 65 inode_unlock(dentry->d_inode); 66 return ret; 67 } 68 69 long vfs_truncate(const struct path *path, loff_t length) 70 { 71 struct user_namespace *mnt_userns; 72 struct inode *inode; 73 long error; 74 75 inode = path->dentry->d_inode; 76 77 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */ 78 if (S_ISDIR(inode->i_mode)) 79 return -EISDIR; 80 if (!S_ISREG(inode->i_mode)) 81 return -EINVAL; 82 83 error = mnt_want_write(path->mnt); 84 if (error) 85 goto out; 86 87 mnt_userns = mnt_user_ns(path->mnt); 88 error = inode_permission(mnt_userns, inode, MAY_WRITE); 89 if (error) 90 goto mnt_drop_write_and_out; 91 92 error = -EPERM; 93 if (IS_APPEND(inode)) 94 goto mnt_drop_write_and_out; 95 96 error = get_write_access(inode); 97 if (error) 98 goto mnt_drop_write_and_out; 99 100 /* 101 * Make sure that there are no leases. get_write_access() protects 102 * against the truncate racing with a lease-granting setlease(). 103 */ 104 error = break_lease(inode, O_WRONLY); 105 if (error) 106 goto put_write_and_out; 107 108 error = security_path_truncate(path); 109 if (!error) 110 error = do_truncate(mnt_userns, path->dentry, length, 0, NULL); 111 112 put_write_and_out: 113 put_write_access(inode); 114 mnt_drop_write_and_out: 115 mnt_drop_write(path->mnt); 116 out: 117 return error; 118 } 119 EXPORT_SYMBOL_GPL(vfs_truncate); 120 121 long do_sys_truncate(const char __user *pathname, loff_t length) 122 { 123 unsigned int lookup_flags = LOOKUP_FOLLOW; 124 struct path path; 125 int error; 126 127 if (length < 0) /* sorry, but loff_t says... */ 128 return -EINVAL; 129 130 retry: 131 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path); 132 if (!error) { 133 error = vfs_truncate(&path, length); 134 path_put(&path); 135 } 136 if (retry_estale(error, lookup_flags)) { 137 lookup_flags |= LOOKUP_REVAL; 138 goto retry; 139 } 140 return error; 141 } 142 143 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length) 144 { 145 return do_sys_truncate(path, length); 146 } 147 148 #ifdef CONFIG_COMPAT 149 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length) 150 { 151 return do_sys_truncate(path, length); 152 } 153 #endif 154 155 long do_sys_ftruncate(unsigned int fd, loff_t length, int small) 156 { 157 struct inode *inode; 158 struct dentry *dentry; 159 struct fd f; 160 int error; 161 162 error = -EINVAL; 163 if (length < 0) 164 goto out; 165 error = -EBADF; 166 f = fdget(fd); 167 if (!f.file) 168 goto out; 169 170 /* explicitly opened as large or we are on 64-bit box */ 171 if (f.file->f_flags & O_LARGEFILE) 172 small = 0; 173 174 dentry = f.file->f_path.dentry; 175 inode = dentry->d_inode; 176 error = -EINVAL; 177 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE)) 178 goto out_putf; 179 180 error = -EINVAL; 181 /* Cannot ftruncate over 2^31 bytes without large file support */ 182 if (small && length > MAX_NON_LFS) 183 goto out_putf; 184 185 error = -EPERM; 186 /* Check IS_APPEND on real upper inode */ 187 if (IS_APPEND(file_inode(f.file))) 188 goto out_putf; 189 sb_start_write(inode->i_sb); 190 error = security_path_truncate(&f.file->f_path); 191 if (!error) 192 error = do_truncate(file_mnt_user_ns(f.file), dentry, length, 193 ATTR_MTIME | ATTR_CTIME, f.file); 194 sb_end_write(inode->i_sb); 195 out_putf: 196 fdput(f); 197 out: 198 return error; 199 } 200 201 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length) 202 { 203 return do_sys_ftruncate(fd, length, 1); 204 } 205 206 #ifdef CONFIG_COMPAT 207 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length) 208 { 209 return do_sys_ftruncate(fd, length, 1); 210 } 211 #endif 212 213 /* LFS versions of truncate are only needed on 32 bit machines */ 214 #if BITS_PER_LONG == 32 215 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length) 216 { 217 return do_sys_truncate(path, length); 218 } 219 220 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length) 221 { 222 return do_sys_ftruncate(fd, length, 0); 223 } 224 #endif /* BITS_PER_LONG == 32 */ 225 226 227 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 228 { 229 struct inode *inode = file_inode(file); 230 long ret; 231 232 if (offset < 0 || len <= 0) 233 return -EINVAL; 234 235 /* Return error if mode is not supported */ 236 if (mode & ~FALLOC_FL_SUPPORTED_MASK) 237 return -EOPNOTSUPP; 238 239 /* Punch hole and zero range are mutually exclusive */ 240 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) == 241 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) 242 return -EOPNOTSUPP; 243 244 /* Punch hole must have keep size set */ 245 if ((mode & FALLOC_FL_PUNCH_HOLE) && 246 !(mode & FALLOC_FL_KEEP_SIZE)) 247 return -EOPNOTSUPP; 248 249 /* Collapse range should only be used exclusively. */ 250 if ((mode & FALLOC_FL_COLLAPSE_RANGE) && 251 (mode & ~FALLOC_FL_COLLAPSE_RANGE)) 252 return -EINVAL; 253 254 /* Insert range should only be used exclusively. */ 255 if ((mode & FALLOC_FL_INSERT_RANGE) && 256 (mode & ~FALLOC_FL_INSERT_RANGE)) 257 return -EINVAL; 258 259 /* Unshare range should only be used with allocate mode. */ 260 if ((mode & FALLOC_FL_UNSHARE_RANGE) && 261 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE))) 262 return -EINVAL; 263 264 if (!(file->f_mode & FMODE_WRITE)) 265 return -EBADF; 266 267 /* 268 * We can only allow pure fallocate on append only files 269 */ 270 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode)) 271 return -EPERM; 272 273 if (IS_IMMUTABLE(inode)) 274 return -EPERM; 275 276 /* 277 * We cannot allow any fallocate operation on an active swapfile 278 */ 279 if (IS_SWAPFILE(inode)) 280 return -ETXTBSY; 281 282 /* 283 * Revalidate the write permissions, in case security policy has 284 * changed since the files were opened. 285 */ 286 ret = security_file_permission(file, MAY_WRITE); 287 if (ret) 288 return ret; 289 290 if (S_ISFIFO(inode->i_mode)) 291 return -ESPIPE; 292 293 if (S_ISDIR(inode->i_mode)) 294 return -EISDIR; 295 296 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 297 return -ENODEV; 298 299 /* Check for wrap through zero too */ 300 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0)) 301 return -EFBIG; 302 303 if (!file->f_op->fallocate) 304 return -EOPNOTSUPP; 305 306 file_start_write(file); 307 ret = file->f_op->fallocate(file, mode, offset, len); 308 309 /* 310 * Create inotify and fanotify events. 311 * 312 * To keep the logic simple always create events if fallocate succeeds. 313 * This implies that events are even created if the file size remains 314 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE. 315 */ 316 if (ret == 0) 317 fsnotify_modify(file); 318 319 file_end_write(file); 320 return ret; 321 } 322 EXPORT_SYMBOL_GPL(vfs_fallocate); 323 324 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len) 325 { 326 struct fd f = fdget(fd); 327 int error = -EBADF; 328 329 if (f.file) { 330 error = vfs_fallocate(f.file, mode, offset, len); 331 fdput(f); 332 } 333 return error; 334 } 335 336 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len) 337 { 338 return ksys_fallocate(fd, mode, offset, len); 339 } 340 341 /* 342 * access() needs to use the real uid/gid, not the effective uid/gid. 343 * We do this by temporarily clearing all FS-related capabilities and 344 * switching the fsuid/fsgid around to the real ones. 345 */ 346 static const struct cred *access_override_creds(void) 347 { 348 const struct cred *old_cred; 349 struct cred *override_cred; 350 351 override_cred = prepare_creds(); 352 if (!override_cred) 353 return NULL; 354 355 override_cred->fsuid = override_cred->uid; 356 override_cred->fsgid = override_cred->gid; 357 358 if (!issecure(SECURE_NO_SETUID_FIXUP)) { 359 /* Clear the capabilities if we switch to a non-root user */ 360 kuid_t root_uid = make_kuid(override_cred->user_ns, 0); 361 if (!uid_eq(override_cred->uid, root_uid)) 362 cap_clear(override_cred->cap_effective); 363 else 364 override_cred->cap_effective = 365 override_cred->cap_permitted; 366 } 367 368 /* 369 * The new set of credentials can *only* be used in 370 * task-synchronous circumstances, and does not need 371 * RCU freeing, unless somebody then takes a separate 372 * reference to it. 373 * 374 * NOTE! This is _only_ true because this credential 375 * is used purely for override_creds() that installs 376 * it as the subjective cred. Other threads will be 377 * accessing ->real_cred, not the subjective cred. 378 * 379 * If somebody _does_ make a copy of this (using the 380 * 'get_current_cred()' function), that will clear the 381 * non_rcu field, because now that other user may be 382 * expecting RCU freeing. But normal thread-synchronous 383 * cred accesses will keep things non-RCY. 384 */ 385 override_cred->non_rcu = 1; 386 387 old_cred = override_creds(override_cred); 388 389 /* override_cred() gets its own ref */ 390 put_cred(override_cred); 391 392 return old_cred; 393 } 394 395 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags) 396 { 397 struct path path; 398 struct inode *inode; 399 int res; 400 unsigned int lookup_flags = LOOKUP_FOLLOW; 401 const struct cred *old_cred = NULL; 402 403 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */ 404 return -EINVAL; 405 406 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) 407 return -EINVAL; 408 409 if (flags & AT_SYMLINK_NOFOLLOW) 410 lookup_flags &= ~LOOKUP_FOLLOW; 411 if (flags & AT_EMPTY_PATH) 412 lookup_flags |= LOOKUP_EMPTY; 413 414 if (!(flags & AT_EACCESS)) { 415 old_cred = access_override_creds(); 416 if (!old_cred) 417 return -ENOMEM; 418 } 419 420 retry: 421 res = user_path_at(dfd, filename, lookup_flags, &path); 422 if (res) 423 goto out; 424 425 inode = d_backing_inode(path.dentry); 426 427 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) { 428 /* 429 * MAY_EXEC on regular files is denied if the fs is mounted 430 * with the "noexec" flag. 431 */ 432 res = -EACCES; 433 if (path_noexec(&path)) 434 goto out_path_release; 435 } 436 437 res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS); 438 /* SuS v2 requires we report a read only fs too */ 439 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode)) 440 goto out_path_release; 441 /* 442 * This is a rare case where using __mnt_is_readonly() 443 * is OK without a mnt_want/drop_write() pair. Since 444 * no actual write to the fs is performed here, we do 445 * not need to telegraph to that to anyone. 446 * 447 * By doing this, we accept that this access is 448 * inherently racy and know that the fs may change 449 * state before we even see this result. 450 */ 451 if (__mnt_is_readonly(path.mnt)) 452 res = -EROFS; 453 454 out_path_release: 455 path_put(&path); 456 if (retry_estale(res, lookup_flags)) { 457 lookup_flags |= LOOKUP_REVAL; 458 goto retry; 459 } 460 out: 461 if (old_cred) 462 revert_creds(old_cred); 463 464 return res; 465 } 466 467 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode) 468 { 469 return do_faccessat(dfd, filename, mode, 0); 470 } 471 472 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode, 473 int, flags) 474 { 475 return do_faccessat(dfd, filename, mode, flags); 476 } 477 478 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode) 479 { 480 return do_faccessat(AT_FDCWD, filename, mode, 0); 481 } 482 483 SYSCALL_DEFINE1(chdir, const char __user *, filename) 484 { 485 struct path path; 486 int error; 487 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 488 retry: 489 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); 490 if (error) 491 goto out; 492 493 error = path_permission(&path, MAY_EXEC | MAY_CHDIR); 494 if (error) 495 goto dput_and_out; 496 497 set_fs_pwd(current->fs, &path); 498 499 dput_and_out: 500 path_put(&path); 501 if (retry_estale(error, lookup_flags)) { 502 lookup_flags |= LOOKUP_REVAL; 503 goto retry; 504 } 505 out: 506 return error; 507 } 508 509 SYSCALL_DEFINE1(fchdir, unsigned int, fd) 510 { 511 struct fd f = fdget_raw(fd); 512 int error; 513 514 error = -EBADF; 515 if (!f.file) 516 goto out; 517 518 error = -ENOTDIR; 519 if (!d_can_lookup(f.file->f_path.dentry)) 520 goto out_putf; 521 522 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR); 523 if (!error) 524 set_fs_pwd(current->fs, &f.file->f_path); 525 out_putf: 526 fdput(f); 527 out: 528 return error; 529 } 530 531 SYSCALL_DEFINE1(chroot, const char __user *, filename) 532 { 533 struct path path; 534 int error; 535 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 536 retry: 537 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); 538 if (error) 539 goto out; 540 541 error = path_permission(&path, MAY_EXEC | MAY_CHDIR); 542 if (error) 543 goto dput_and_out; 544 545 error = -EPERM; 546 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT)) 547 goto dput_and_out; 548 error = security_path_chroot(&path); 549 if (error) 550 goto dput_and_out; 551 552 set_fs_root(current->fs, &path); 553 error = 0; 554 dput_and_out: 555 path_put(&path); 556 if (retry_estale(error, lookup_flags)) { 557 lookup_flags |= LOOKUP_REVAL; 558 goto retry; 559 } 560 out: 561 return error; 562 } 563 564 int chmod_common(const struct path *path, umode_t mode) 565 { 566 struct inode *inode = path->dentry->d_inode; 567 struct inode *delegated_inode = NULL; 568 struct iattr newattrs; 569 int error; 570 571 error = mnt_want_write(path->mnt); 572 if (error) 573 return error; 574 retry_deleg: 575 inode_lock(inode); 576 error = security_path_chmod(path, mode); 577 if (error) 578 goto out_unlock; 579 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO); 580 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME; 581 error = notify_change(mnt_user_ns(path->mnt), path->dentry, 582 &newattrs, &delegated_inode); 583 out_unlock: 584 inode_unlock(inode); 585 if (delegated_inode) { 586 error = break_deleg_wait(&delegated_inode); 587 if (!error) 588 goto retry_deleg; 589 } 590 mnt_drop_write(path->mnt); 591 return error; 592 } 593 594 int vfs_fchmod(struct file *file, umode_t mode) 595 { 596 audit_file(file); 597 return chmod_common(&file->f_path, mode); 598 } 599 600 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode) 601 { 602 struct fd f = fdget(fd); 603 int err = -EBADF; 604 605 if (f.file) { 606 err = vfs_fchmod(f.file, mode); 607 fdput(f); 608 } 609 return err; 610 } 611 612 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode) 613 { 614 struct path path; 615 int error; 616 unsigned int lookup_flags = LOOKUP_FOLLOW; 617 retry: 618 error = user_path_at(dfd, filename, lookup_flags, &path); 619 if (!error) { 620 error = chmod_common(&path, mode); 621 path_put(&path); 622 if (retry_estale(error, lookup_flags)) { 623 lookup_flags |= LOOKUP_REVAL; 624 goto retry; 625 } 626 } 627 return error; 628 } 629 630 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, 631 umode_t, mode) 632 { 633 return do_fchmodat(dfd, filename, mode); 634 } 635 636 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode) 637 { 638 return do_fchmodat(AT_FDCWD, filename, mode); 639 } 640 641 int chown_common(const struct path *path, uid_t user, gid_t group) 642 { 643 struct user_namespace *mnt_userns; 644 struct inode *inode = path->dentry->d_inode; 645 struct inode *delegated_inode = NULL; 646 int error; 647 struct iattr newattrs; 648 kuid_t uid; 649 kgid_t gid; 650 651 uid = make_kuid(current_user_ns(), user); 652 gid = make_kgid(current_user_ns(), group); 653 654 mnt_userns = mnt_user_ns(path->mnt); 655 uid = kuid_from_mnt(mnt_userns, uid); 656 gid = kgid_from_mnt(mnt_userns, gid); 657 658 retry_deleg: 659 newattrs.ia_valid = ATTR_CTIME; 660 if (user != (uid_t) -1) { 661 if (!uid_valid(uid)) 662 return -EINVAL; 663 newattrs.ia_valid |= ATTR_UID; 664 newattrs.ia_uid = uid; 665 } 666 if (group != (gid_t) -1) { 667 if (!gid_valid(gid)) 668 return -EINVAL; 669 newattrs.ia_valid |= ATTR_GID; 670 newattrs.ia_gid = gid; 671 } 672 if (!S_ISDIR(inode->i_mode)) 673 newattrs.ia_valid |= 674 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 675 inode_lock(inode); 676 error = security_path_chown(path, uid, gid); 677 if (!error) 678 error = notify_change(mnt_userns, path->dentry, &newattrs, 679 &delegated_inode); 680 inode_unlock(inode); 681 if (delegated_inode) { 682 error = break_deleg_wait(&delegated_inode); 683 if (!error) 684 goto retry_deleg; 685 } 686 return error; 687 } 688 689 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group, 690 int flag) 691 { 692 struct path path; 693 int error = -EINVAL; 694 int lookup_flags; 695 696 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0) 697 goto out; 698 699 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW; 700 if (flag & AT_EMPTY_PATH) 701 lookup_flags |= LOOKUP_EMPTY; 702 retry: 703 error = user_path_at(dfd, filename, lookup_flags, &path); 704 if (error) 705 goto out; 706 error = mnt_want_write(path.mnt); 707 if (error) 708 goto out_release; 709 error = chown_common(&path, user, group); 710 mnt_drop_write(path.mnt); 711 out_release: 712 path_put(&path); 713 if (retry_estale(error, lookup_flags)) { 714 lookup_flags |= LOOKUP_REVAL; 715 goto retry; 716 } 717 out: 718 return error; 719 } 720 721 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user, 722 gid_t, group, int, flag) 723 { 724 return do_fchownat(dfd, filename, user, group, flag); 725 } 726 727 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group) 728 { 729 return do_fchownat(AT_FDCWD, filename, user, group, 0); 730 } 731 732 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group) 733 { 734 return do_fchownat(AT_FDCWD, filename, user, group, 735 AT_SYMLINK_NOFOLLOW); 736 } 737 738 int vfs_fchown(struct file *file, uid_t user, gid_t group) 739 { 740 int error; 741 742 error = mnt_want_write_file(file); 743 if (error) 744 return error; 745 audit_file(file); 746 error = chown_common(&file->f_path, user, group); 747 mnt_drop_write_file(file); 748 return error; 749 } 750 751 int ksys_fchown(unsigned int fd, uid_t user, gid_t group) 752 { 753 struct fd f = fdget(fd); 754 int error = -EBADF; 755 756 if (f.file) { 757 error = vfs_fchown(f.file, user, group); 758 fdput(f); 759 } 760 return error; 761 } 762 763 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group) 764 { 765 return ksys_fchown(fd, user, group); 766 } 767 768 static int do_dentry_open(struct file *f, 769 struct inode *inode, 770 int (*open)(struct inode *, struct file *)) 771 { 772 static const struct file_operations empty_fops = {}; 773 int error; 774 775 path_get(&f->f_path); 776 f->f_inode = inode; 777 f->f_mapping = inode->i_mapping; 778 f->f_wb_err = filemap_sample_wb_err(f->f_mapping); 779 f->f_sb_err = file_sample_sb_err(f); 780 781 if (unlikely(f->f_flags & O_PATH)) { 782 f->f_mode = FMODE_PATH | FMODE_OPENED; 783 f->f_op = &empty_fops; 784 return 0; 785 } 786 787 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) { 788 error = get_write_access(inode); 789 if (unlikely(error)) 790 goto cleanup_file; 791 error = __mnt_want_write(f->f_path.mnt); 792 if (unlikely(error)) { 793 put_write_access(inode); 794 goto cleanup_file; 795 } 796 f->f_mode |= FMODE_WRITER; 797 } 798 799 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */ 800 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)) 801 f->f_mode |= FMODE_ATOMIC_POS; 802 803 f->f_op = fops_get(inode->i_fop); 804 if (WARN_ON(!f->f_op)) { 805 error = -ENODEV; 806 goto cleanup_all; 807 } 808 809 error = security_file_open(f); 810 if (error) 811 goto cleanup_all; 812 813 error = break_lease(locks_inode(f), f->f_flags); 814 if (error) 815 goto cleanup_all; 816 817 /* normally all 3 are set; ->open() can clear them if needed */ 818 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; 819 if (!open) 820 open = f->f_op->open; 821 if (open) { 822 error = open(inode, f); 823 if (error) 824 goto cleanup_all; 825 } 826 f->f_mode |= FMODE_OPENED; 827 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) 828 i_readcount_inc(inode); 829 if ((f->f_mode & FMODE_READ) && 830 likely(f->f_op->read || f->f_op->read_iter)) 831 f->f_mode |= FMODE_CAN_READ; 832 if ((f->f_mode & FMODE_WRITE) && 833 likely(f->f_op->write || f->f_op->write_iter)) 834 f->f_mode |= FMODE_CAN_WRITE; 835 836 f->f_write_hint = WRITE_LIFE_NOT_SET; 837 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC); 838 839 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping); 840 841 /* NB: we're sure to have correct a_ops only after f_op->open */ 842 if (f->f_flags & O_DIRECT) { 843 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO) 844 return -EINVAL; 845 } 846 847 /* 848 * XXX: Huge page cache doesn't support writing yet. Drop all page 849 * cache for this file before processing writes. 850 */ 851 if (f->f_mode & FMODE_WRITE) { 852 /* 853 * Paired with smp_mb() in collapse_file() to ensure nr_thps 854 * is up to date and the update to i_writecount by 855 * get_write_access() is visible. Ensures subsequent insertion 856 * of THPs into the page cache will fail. 857 */ 858 smp_mb(); 859 if (filemap_nr_thps(inode->i_mapping)) { 860 struct address_space *mapping = inode->i_mapping; 861 862 filemap_invalidate_lock(inode->i_mapping); 863 /* 864 * unmap_mapping_range just need to be called once 865 * here, because the private pages is not need to be 866 * unmapped mapping (e.g. data segment of dynamic 867 * shared libraries here). 868 */ 869 unmap_mapping_range(mapping, 0, 0, 0); 870 truncate_inode_pages(mapping, 0); 871 filemap_invalidate_unlock(inode->i_mapping); 872 } 873 } 874 875 return 0; 876 877 cleanup_all: 878 if (WARN_ON_ONCE(error > 0)) 879 error = -EINVAL; 880 fops_put(f->f_op); 881 if (f->f_mode & FMODE_WRITER) { 882 put_write_access(inode); 883 __mnt_drop_write(f->f_path.mnt); 884 } 885 cleanup_file: 886 path_put(&f->f_path); 887 f->f_path.mnt = NULL; 888 f->f_path.dentry = NULL; 889 f->f_inode = NULL; 890 return error; 891 } 892 893 /** 894 * finish_open - finish opening a file 895 * @file: file pointer 896 * @dentry: pointer to dentry 897 * @open: open callback 898 * @opened: state of open 899 * 900 * This can be used to finish opening a file passed to i_op->atomic_open(). 901 * 902 * If the open callback is set to NULL, then the standard f_op->open() 903 * filesystem callback is substituted. 904 * 905 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is 906 * the return value of d_splice_alias(), then the caller needs to perform dput() 907 * on it after finish_open(). 908 * 909 * Returns zero on success or -errno if the open failed. 910 */ 911 int finish_open(struct file *file, struct dentry *dentry, 912 int (*open)(struct inode *, struct file *)) 913 { 914 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 915 916 file->f_path.dentry = dentry; 917 return do_dentry_open(file, d_backing_inode(dentry), open); 918 } 919 EXPORT_SYMBOL(finish_open); 920 921 /** 922 * finish_no_open - finish ->atomic_open() without opening the file 923 * 924 * @file: file pointer 925 * @dentry: dentry or NULL (as returned from ->lookup()) 926 * 927 * This can be used to set the result of a successful lookup in ->atomic_open(). 928 * 929 * NB: unlike finish_open() this function does consume the dentry reference and 930 * the caller need not dput() it. 931 * 932 * Returns "0" which must be the return value of ->atomic_open() after having 933 * called this function. 934 */ 935 int finish_no_open(struct file *file, struct dentry *dentry) 936 { 937 file->f_path.dentry = dentry; 938 return 0; 939 } 940 EXPORT_SYMBOL(finish_no_open); 941 942 char *file_path(struct file *filp, char *buf, int buflen) 943 { 944 return d_path(&filp->f_path, buf, buflen); 945 } 946 EXPORT_SYMBOL(file_path); 947 948 /** 949 * vfs_open - open the file at the given path 950 * @path: path to open 951 * @file: newly allocated file with f_flag initialized 952 * @cred: credentials to use 953 */ 954 int vfs_open(const struct path *path, struct file *file) 955 { 956 file->f_path = *path; 957 return do_dentry_open(file, d_backing_inode(path->dentry), NULL); 958 } 959 960 struct file *dentry_open(const struct path *path, int flags, 961 const struct cred *cred) 962 { 963 int error; 964 struct file *f; 965 966 validate_creds(cred); 967 968 /* We must always pass in a valid mount pointer. */ 969 BUG_ON(!path->mnt); 970 971 f = alloc_empty_file(flags, cred); 972 if (!IS_ERR(f)) { 973 error = vfs_open(path, f); 974 if (error) { 975 fput(f); 976 f = ERR_PTR(error); 977 } 978 } 979 return f; 980 } 981 EXPORT_SYMBOL(dentry_open); 982 983 struct file *open_with_fake_path(const struct path *path, int flags, 984 struct inode *inode, const struct cred *cred) 985 { 986 struct file *f = alloc_empty_file_noaccount(flags, cred); 987 if (!IS_ERR(f)) { 988 int error; 989 990 f->f_path = *path; 991 error = do_dentry_open(f, inode, NULL); 992 if (error) { 993 fput(f); 994 f = ERR_PTR(error); 995 } 996 } 997 return f; 998 } 999 EXPORT_SYMBOL(open_with_fake_path); 1000 1001 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE)) 1002 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC) 1003 1004 inline struct open_how build_open_how(int flags, umode_t mode) 1005 { 1006 struct open_how how = { 1007 .flags = flags & VALID_OPEN_FLAGS, 1008 .mode = mode & S_IALLUGO, 1009 }; 1010 1011 /* O_PATH beats everything else. */ 1012 if (how.flags & O_PATH) 1013 how.flags &= O_PATH_FLAGS; 1014 /* Modes should only be set for create-like flags. */ 1015 if (!WILL_CREATE(how.flags)) 1016 how.mode = 0; 1017 return how; 1018 } 1019 1020 inline int build_open_flags(const struct open_how *how, struct open_flags *op) 1021 { 1022 u64 flags = how->flags; 1023 u64 strip = FMODE_NONOTIFY | O_CLOEXEC; 1024 int lookup_flags = 0; 1025 int acc_mode = ACC_MODE(flags); 1026 1027 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS), 1028 "struct open_flags doesn't yet handle flags > 32 bits"); 1029 1030 /* 1031 * Strip flags that either shouldn't be set by userspace like 1032 * FMODE_NONOTIFY or that aren't relevant in determining struct 1033 * open_flags like O_CLOEXEC. 1034 */ 1035 flags &= ~strip; 1036 1037 /* 1038 * Older syscalls implicitly clear all of the invalid flags or argument 1039 * values before calling build_open_flags(), but openat2(2) checks all 1040 * of its arguments. 1041 */ 1042 if (flags & ~VALID_OPEN_FLAGS) 1043 return -EINVAL; 1044 if (how->resolve & ~VALID_RESOLVE_FLAGS) 1045 return -EINVAL; 1046 1047 /* Scoping flags are mutually exclusive. */ 1048 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT)) 1049 return -EINVAL; 1050 1051 /* Deal with the mode. */ 1052 if (WILL_CREATE(flags)) { 1053 if (how->mode & ~S_IALLUGO) 1054 return -EINVAL; 1055 op->mode = how->mode | S_IFREG; 1056 } else { 1057 if (how->mode != 0) 1058 return -EINVAL; 1059 op->mode = 0; 1060 } 1061 1062 /* 1063 * In order to ensure programs get explicit errors when trying to use 1064 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it 1065 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we 1066 * have to require userspace to explicitly set it. 1067 */ 1068 if (flags & __O_TMPFILE) { 1069 if ((flags & O_TMPFILE_MASK) != O_TMPFILE) 1070 return -EINVAL; 1071 if (!(acc_mode & MAY_WRITE)) 1072 return -EINVAL; 1073 } 1074 if (flags & O_PATH) { 1075 /* O_PATH only permits certain other flags to be set. */ 1076 if (flags & ~O_PATH_FLAGS) 1077 return -EINVAL; 1078 acc_mode = 0; 1079 } 1080 1081 /* 1082 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only 1083 * check for O_DSYNC if the need any syncing at all we enforce it's 1084 * always set instead of having to deal with possibly weird behaviour 1085 * for malicious applications setting only __O_SYNC. 1086 */ 1087 if (flags & __O_SYNC) 1088 flags |= O_DSYNC; 1089 1090 op->open_flag = flags; 1091 1092 /* O_TRUNC implies we need access checks for write permissions */ 1093 if (flags & O_TRUNC) 1094 acc_mode |= MAY_WRITE; 1095 1096 /* Allow the LSM permission hook to distinguish append 1097 access from general write access. */ 1098 if (flags & O_APPEND) 1099 acc_mode |= MAY_APPEND; 1100 1101 op->acc_mode = acc_mode; 1102 1103 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN; 1104 1105 if (flags & O_CREAT) { 1106 op->intent |= LOOKUP_CREATE; 1107 if (flags & O_EXCL) { 1108 op->intent |= LOOKUP_EXCL; 1109 flags |= O_NOFOLLOW; 1110 } 1111 } 1112 1113 if (flags & O_DIRECTORY) 1114 lookup_flags |= LOOKUP_DIRECTORY; 1115 if (!(flags & O_NOFOLLOW)) 1116 lookup_flags |= LOOKUP_FOLLOW; 1117 1118 if (how->resolve & RESOLVE_NO_XDEV) 1119 lookup_flags |= LOOKUP_NO_XDEV; 1120 if (how->resolve & RESOLVE_NO_MAGICLINKS) 1121 lookup_flags |= LOOKUP_NO_MAGICLINKS; 1122 if (how->resolve & RESOLVE_NO_SYMLINKS) 1123 lookup_flags |= LOOKUP_NO_SYMLINKS; 1124 if (how->resolve & RESOLVE_BENEATH) 1125 lookup_flags |= LOOKUP_BENEATH; 1126 if (how->resolve & RESOLVE_IN_ROOT) 1127 lookup_flags |= LOOKUP_IN_ROOT; 1128 if (how->resolve & RESOLVE_CACHED) { 1129 /* Don't bother even trying for create/truncate/tmpfile open */ 1130 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE)) 1131 return -EAGAIN; 1132 lookup_flags |= LOOKUP_CACHED; 1133 } 1134 1135 op->lookup_flags = lookup_flags; 1136 return 0; 1137 } 1138 1139 /** 1140 * file_open_name - open file and return file pointer 1141 * 1142 * @name: struct filename containing path to open 1143 * @flags: open flags as per the open(2) second argument 1144 * @mode: mode for the new file if O_CREAT is set, else ignored 1145 * 1146 * This is the helper to open a file from kernelspace if you really 1147 * have to. But in generally you should not do this, so please move 1148 * along, nothing to see here.. 1149 */ 1150 struct file *file_open_name(struct filename *name, int flags, umode_t mode) 1151 { 1152 struct open_flags op; 1153 struct open_how how = build_open_how(flags, mode); 1154 int err = build_open_flags(&how, &op); 1155 if (err) 1156 return ERR_PTR(err); 1157 return do_filp_open(AT_FDCWD, name, &op); 1158 } 1159 1160 /** 1161 * filp_open - open file and return file pointer 1162 * 1163 * @filename: path to open 1164 * @flags: open flags as per the open(2) second argument 1165 * @mode: mode for the new file if O_CREAT is set, else ignored 1166 * 1167 * This is the helper to open a file from kernelspace if you really 1168 * have to. But in generally you should not do this, so please move 1169 * along, nothing to see here.. 1170 */ 1171 struct file *filp_open(const char *filename, int flags, umode_t mode) 1172 { 1173 struct filename *name = getname_kernel(filename); 1174 struct file *file = ERR_CAST(name); 1175 1176 if (!IS_ERR(name)) { 1177 file = file_open_name(name, flags, mode); 1178 putname(name); 1179 } 1180 return file; 1181 } 1182 EXPORT_SYMBOL(filp_open); 1183 1184 struct file *file_open_root(const struct path *root, 1185 const char *filename, int flags, umode_t mode) 1186 { 1187 struct open_flags op; 1188 struct open_how how = build_open_how(flags, mode); 1189 int err = build_open_flags(&how, &op); 1190 if (err) 1191 return ERR_PTR(err); 1192 return do_file_open_root(root, filename, &op); 1193 } 1194 EXPORT_SYMBOL(file_open_root); 1195 1196 static long do_sys_openat2(int dfd, const char __user *filename, 1197 struct open_how *how) 1198 { 1199 struct open_flags op; 1200 int fd = build_open_flags(how, &op); 1201 struct filename *tmp; 1202 1203 if (fd) 1204 return fd; 1205 1206 tmp = getname(filename); 1207 if (IS_ERR(tmp)) 1208 return PTR_ERR(tmp); 1209 1210 fd = get_unused_fd_flags(how->flags); 1211 if (fd >= 0) { 1212 struct file *f = do_filp_open(dfd, tmp, &op); 1213 if (IS_ERR(f)) { 1214 put_unused_fd(fd); 1215 fd = PTR_ERR(f); 1216 } else { 1217 fsnotify_open(f); 1218 fd_install(fd, f); 1219 } 1220 } 1221 putname(tmp); 1222 return fd; 1223 } 1224 1225 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode) 1226 { 1227 struct open_how how = build_open_how(flags, mode); 1228 return do_sys_openat2(dfd, filename, &how); 1229 } 1230 1231 1232 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode) 1233 { 1234 if (force_o_largefile()) 1235 flags |= O_LARGEFILE; 1236 return do_sys_open(AT_FDCWD, filename, flags, mode); 1237 } 1238 1239 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, 1240 umode_t, mode) 1241 { 1242 if (force_o_largefile()) 1243 flags |= O_LARGEFILE; 1244 return do_sys_open(dfd, filename, flags, mode); 1245 } 1246 1247 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename, 1248 struct open_how __user *, how, size_t, usize) 1249 { 1250 int err; 1251 struct open_how tmp; 1252 1253 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0); 1254 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST); 1255 1256 if (unlikely(usize < OPEN_HOW_SIZE_VER0)) 1257 return -EINVAL; 1258 1259 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize); 1260 if (err) 1261 return err; 1262 1263 audit_openat2_how(&tmp); 1264 1265 /* O_LARGEFILE is only allowed for non-O_PATH. */ 1266 if (!(tmp.flags & O_PATH) && force_o_largefile()) 1267 tmp.flags |= O_LARGEFILE; 1268 1269 return do_sys_openat2(dfd, filename, &tmp); 1270 } 1271 1272 #ifdef CONFIG_COMPAT 1273 /* 1274 * Exactly like sys_open(), except that it doesn't set the 1275 * O_LARGEFILE flag. 1276 */ 1277 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode) 1278 { 1279 return do_sys_open(AT_FDCWD, filename, flags, mode); 1280 } 1281 1282 /* 1283 * Exactly like sys_openat(), except that it doesn't set the 1284 * O_LARGEFILE flag. 1285 */ 1286 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode) 1287 { 1288 return do_sys_open(dfd, filename, flags, mode); 1289 } 1290 #endif 1291 1292 #ifndef __alpha__ 1293 1294 /* 1295 * For backward compatibility? Maybe this should be moved 1296 * into arch/i386 instead? 1297 */ 1298 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode) 1299 { 1300 int flags = O_CREAT | O_WRONLY | O_TRUNC; 1301 1302 if (force_o_largefile()) 1303 flags |= O_LARGEFILE; 1304 return do_sys_open(AT_FDCWD, pathname, flags, mode); 1305 } 1306 #endif 1307 1308 /* 1309 * "id" is the POSIX thread ID. We use the 1310 * files pointer for this.. 1311 */ 1312 int filp_close(struct file *filp, fl_owner_t id) 1313 { 1314 int retval = 0; 1315 1316 if (!file_count(filp)) { 1317 printk(KERN_ERR "VFS: Close: file count is 0\n"); 1318 return 0; 1319 } 1320 1321 if (filp->f_op->flush) 1322 retval = filp->f_op->flush(filp, id); 1323 1324 if (likely(!(filp->f_mode & FMODE_PATH))) { 1325 dnotify_flush(filp, id); 1326 locks_remove_posix(filp, id); 1327 } 1328 fput(filp); 1329 return retval; 1330 } 1331 1332 EXPORT_SYMBOL(filp_close); 1333 1334 /* 1335 * Careful here! We test whether the file pointer is NULL before 1336 * releasing the fd. This ensures that one clone task can't release 1337 * an fd while another clone is opening it. 1338 */ 1339 SYSCALL_DEFINE1(close, unsigned int, fd) 1340 { 1341 int retval = close_fd(fd); 1342 1343 /* can't restart close syscall because file table entry was cleared */ 1344 if (unlikely(retval == -ERESTARTSYS || 1345 retval == -ERESTARTNOINTR || 1346 retval == -ERESTARTNOHAND || 1347 retval == -ERESTART_RESTARTBLOCK)) 1348 retval = -EINTR; 1349 1350 return retval; 1351 } 1352 1353 /** 1354 * close_range() - Close all file descriptors in a given range. 1355 * 1356 * @fd: starting file descriptor to close 1357 * @max_fd: last file descriptor to close 1358 * @flags: reserved for future extensions 1359 * 1360 * This closes a range of file descriptors. All file descriptors 1361 * from @fd up to and including @max_fd are closed. 1362 * Currently, errors to close a given file descriptor are ignored. 1363 */ 1364 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd, 1365 unsigned int, flags) 1366 { 1367 return __close_range(fd, max_fd, flags); 1368 } 1369 1370 /* 1371 * This routine simulates a hangup on the tty, to arrange that users 1372 * are given clean terminals at login time. 1373 */ 1374 SYSCALL_DEFINE0(vhangup) 1375 { 1376 if (capable(CAP_SYS_TTY_CONFIG)) { 1377 tty_vhangup_self(); 1378 return 0; 1379 } 1380 return -EPERM; 1381 } 1382 1383 /* 1384 * Called when an inode is about to be open. 1385 * We use this to disallow opening large files on 32bit systems if 1386 * the caller didn't specify O_LARGEFILE. On 64bit systems we force 1387 * on this flag in sys_open. 1388 */ 1389 int generic_file_open(struct inode * inode, struct file * filp) 1390 { 1391 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) 1392 return -EOVERFLOW; 1393 return 0; 1394 } 1395 1396 EXPORT_SYMBOL(generic_file_open); 1397 1398 /* 1399 * This is used by subsystems that don't want seekable 1400 * file descriptors. The function is not supposed to ever fail, the only 1401 * reason it returns an 'int' and not 'void' is so that it can be plugged 1402 * directly into file_operations structure. 1403 */ 1404 int nonseekable_open(struct inode *inode, struct file *filp) 1405 { 1406 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE); 1407 return 0; 1408 } 1409 1410 EXPORT_SYMBOL(nonseekable_open); 1411 1412 /* 1413 * stream_open is used by subsystems that want stream-like file descriptors. 1414 * Such file descriptors are not seekable and don't have notion of position 1415 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL). 1416 * Contrary to file descriptors of other regular files, .read() and .write() 1417 * can run simultaneously. 1418 * 1419 * stream_open never fails and is marked to return int so that it could be 1420 * directly used as file_operations.open . 1421 */ 1422 int stream_open(struct inode *inode, struct file *filp) 1423 { 1424 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS); 1425 filp->f_mode |= FMODE_STREAM; 1426 return 0; 1427 } 1428 1429 EXPORT_SYMBOL(stream_open); 1430