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 #include <linux/mnt_idmapping.h> 36 37 #include "internal.h" 38 39 int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry, 40 loff_t length, unsigned int time_attrs, struct file *filp) 41 { 42 int ret; 43 struct iattr newattrs; 44 45 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */ 46 if (length < 0) 47 return -EINVAL; 48 49 newattrs.ia_size = length; 50 newattrs.ia_valid = ATTR_SIZE | time_attrs; 51 if (filp) { 52 newattrs.ia_file = filp; 53 newattrs.ia_valid |= ATTR_FILE; 54 } 55 56 /* Remove suid, sgid, and file capabilities on truncate too */ 57 ret = dentry_needs_remove_privs(dentry); 58 if (ret < 0) 59 return ret; 60 if (ret) 61 newattrs.ia_valid |= ret | ATTR_FORCE; 62 63 inode_lock(dentry->d_inode); 64 /* Note any delegations or leases have already been broken: */ 65 ret = notify_change(mnt_userns, dentry, &newattrs, NULL); 66 inode_unlock(dentry->d_inode); 67 return ret; 68 } 69 70 long vfs_truncate(const struct path *path, loff_t length) 71 { 72 struct user_namespace *mnt_userns; 73 struct inode *inode; 74 long error; 75 76 inode = path->dentry->d_inode; 77 78 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */ 79 if (S_ISDIR(inode->i_mode)) 80 return -EISDIR; 81 if (!S_ISREG(inode->i_mode)) 82 return -EINVAL; 83 84 error = mnt_want_write(path->mnt); 85 if (error) 86 goto out; 87 88 mnt_userns = mnt_user_ns(path->mnt); 89 error = inode_permission(mnt_userns, inode, MAY_WRITE); 90 if (error) 91 goto mnt_drop_write_and_out; 92 93 error = -EPERM; 94 if (IS_APPEND(inode)) 95 goto mnt_drop_write_and_out; 96 97 error = get_write_access(inode); 98 if (error) 99 goto mnt_drop_write_and_out; 100 101 /* 102 * Make sure that there are no leases. get_write_access() protects 103 * against the truncate racing with a lease-granting setlease(). 104 */ 105 error = break_lease(inode, O_WRONLY); 106 if (error) 107 goto put_write_and_out; 108 109 error = security_path_truncate(path); 110 if (!error) 111 error = do_truncate(mnt_userns, path->dentry, length, 0, NULL); 112 113 put_write_and_out: 114 put_write_access(inode); 115 mnt_drop_write_and_out: 116 mnt_drop_write(path->mnt); 117 out: 118 return error; 119 } 120 EXPORT_SYMBOL_GPL(vfs_truncate); 121 122 long do_sys_truncate(const char __user *pathname, loff_t length) 123 { 124 unsigned int lookup_flags = LOOKUP_FOLLOW; 125 struct path path; 126 int error; 127 128 if (length < 0) /* sorry, but loff_t says... */ 129 return -EINVAL; 130 131 retry: 132 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path); 133 if (!error) { 134 error = vfs_truncate(&path, length); 135 path_put(&path); 136 } 137 if (retry_estale(error, lookup_flags)) { 138 lookup_flags |= LOOKUP_REVAL; 139 goto retry; 140 } 141 return error; 142 } 143 144 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length) 145 { 146 return do_sys_truncate(path, length); 147 } 148 149 #ifdef CONFIG_COMPAT 150 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length) 151 { 152 return do_sys_truncate(path, length); 153 } 154 #endif 155 156 long do_sys_ftruncate(unsigned int fd, loff_t length, int small) 157 { 158 struct inode *inode; 159 struct dentry *dentry; 160 struct fd f; 161 int error; 162 163 error = -EINVAL; 164 if (length < 0) 165 goto out; 166 error = -EBADF; 167 f = fdget(fd); 168 if (!f.file) 169 goto out; 170 171 /* explicitly opened as large or we are on 64-bit box */ 172 if (f.file->f_flags & O_LARGEFILE) 173 small = 0; 174 175 dentry = f.file->f_path.dentry; 176 inode = dentry->d_inode; 177 error = -EINVAL; 178 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE)) 179 goto out_putf; 180 181 error = -EINVAL; 182 /* Cannot ftruncate over 2^31 bytes without large file support */ 183 if (small && length > MAX_NON_LFS) 184 goto out_putf; 185 186 error = -EPERM; 187 /* Check IS_APPEND on real upper inode */ 188 if (IS_APPEND(file_inode(f.file))) 189 goto out_putf; 190 sb_start_write(inode->i_sb); 191 error = security_path_truncate(&f.file->f_path); 192 if (!error) 193 error = do_truncate(file_mnt_user_ns(f.file), dentry, length, 194 ATTR_MTIME | ATTR_CTIME, f.file); 195 sb_end_write(inode->i_sb); 196 out_putf: 197 fdput(f); 198 out: 199 return error; 200 } 201 202 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length) 203 { 204 return do_sys_ftruncate(fd, length, 1); 205 } 206 207 #ifdef CONFIG_COMPAT 208 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length) 209 { 210 return do_sys_ftruncate(fd, length, 1); 211 } 212 #endif 213 214 /* LFS versions of truncate are only needed on 32 bit machines */ 215 #if BITS_PER_LONG == 32 216 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length) 217 { 218 return do_sys_truncate(path, length); 219 } 220 221 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length) 222 { 223 return do_sys_ftruncate(fd, length, 0); 224 } 225 #endif /* BITS_PER_LONG == 32 */ 226 227 228 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 229 { 230 struct inode *inode = file_inode(file); 231 long ret; 232 233 if (offset < 0 || len <= 0) 234 return -EINVAL; 235 236 /* Return error if mode is not supported */ 237 if (mode & ~FALLOC_FL_SUPPORTED_MASK) 238 return -EOPNOTSUPP; 239 240 /* Punch hole and zero range are mutually exclusive */ 241 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) == 242 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) 243 return -EOPNOTSUPP; 244 245 /* Punch hole must have keep size set */ 246 if ((mode & FALLOC_FL_PUNCH_HOLE) && 247 !(mode & FALLOC_FL_KEEP_SIZE)) 248 return -EOPNOTSUPP; 249 250 /* Collapse range should only be used exclusively. */ 251 if ((mode & FALLOC_FL_COLLAPSE_RANGE) && 252 (mode & ~FALLOC_FL_COLLAPSE_RANGE)) 253 return -EINVAL; 254 255 /* Insert range should only be used exclusively. */ 256 if ((mode & FALLOC_FL_INSERT_RANGE) && 257 (mode & ~FALLOC_FL_INSERT_RANGE)) 258 return -EINVAL; 259 260 /* Unshare range should only be used with allocate mode. */ 261 if ((mode & FALLOC_FL_UNSHARE_RANGE) && 262 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE))) 263 return -EINVAL; 264 265 if (!(file->f_mode & FMODE_WRITE)) 266 return -EBADF; 267 268 /* 269 * We can only allow pure fallocate on append only files 270 */ 271 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode)) 272 return -EPERM; 273 274 if (IS_IMMUTABLE(inode)) 275 return -EPERM; 276 277 /* 278 * We cannot allow any fallocate operation on an active swapfile 279 */ 280 if (IS_SWAPFILE(inode)) 281 return -ETXTBSY; 282 283 /* 284 * Revalidate the write permissions, in case security policy has 285 * changed since the files were opened. 286 */ 287 ret = security_file_permission(file, MAY_WRITE); 288 if (ret) 289 return ret; 290 291 if (S_ISFIFO(inode->i_mode)) 292 return -ESPIPE; 293 294 if (S_ISDIR(inode->i_mode)) 295 return -EISDIR; 296 297 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 298 return -ENODEV; 299 300 /* Check for wrap through zero too */ 301 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0)) 302 return -EFBIG; 303 304 if (!file->f_op->fallocate) 305 return -EOPNOTSUPP; 306 307 file_start_write(file); 308 ret = file->f_op->fallocate(file, mode, offset, len); 309 310 /* 311 * Create inotify and fanotify events. 312 * 313 * To keep the logic simple always create events if fallocate succeeds. 314 * This implies that events are even created if the file size remains 315 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE. 316 */ 317 if (ret == 0) 318 fsnotify_modify(file); 319 320 file_end_write(file); 321 return ret; 322 } 323 EXPORT_SYMBOL_GPL(vfs_fallocate); 324 325 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len) 326 { 327 struct fd f = fdget(fd); 328 int error = -EBADF; 329 330 if (f.file) { 331 error = vfs_fallocate(f.file, mode, offset, len); 332 fdput(f); 333 } 334 return error; 335 } 336 337 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len) 338 { 339 return ksys_fallocate(fd, mode, offset, len); 340 } 341 342 /* 343 * access() needs to use the real uid/gid, not the effective uid/gid. 344 * We do this by temporarily clearing all FS-related capabilities and 345 * switching the fsuid/fsgid around to the real ones. 346 */ 347 static const struct cred *access_override_creds(void) 348 { 349 const struct cred *old_cred; 350 struct cred *override_cred; 351 352 override_cred = prepare_creds(); 353 if (!override_cred) 354 return NULL; 355 356 override_cred->fsuid = override_cred->uid; 357 override_cred->fsgid = override_cred->gid; 358 359 if (!issecure(SECURE_NO_SETUID_FIXUP)) { 360 /* Clear the capabilities if we switch to a non-root user */ 361 kuid_t root_uid = make_kuid(override_cred->user_ns, 0); 362 if (!uid_eq(override_cred->uid, root_uid)) 363 cap_clear(override_cred->cap_effective); 364 else 365 override_cred->cap_effective = 366 override_cred->cap_permitted; 367 } 368 369 /* 370 * The new set of credentials can *only* be used in 371 * task-synchronous circumstances, and does not need 372 * RCU freeing, unless somebody then takes a separate 373 * reference to it. 374 * 375 * NOTE! This is _only_ true because this credential 376 * is used purely for override_creds() that installs 377 * it as the subjective cred. Other threads will be 378 * accessing ->real_cred, not the subjective cred. 379 * 380 * If somebody _does_ make a copy of this (using the 381 * 'get_current_cred()' function), that will clear the 382 * non_rcu field, because now that other user may be 383 * expecting RCU freeing. But normal thread-synchronous 384 * cred accesses will keep things non-RCY. 385 */ 386 override_cred->non_rcu = 1; 387 388 old_cred = override_creds(override_cred); 389 390 /* override_cred() gets its own ref */ 391 put_cred(override_cred); 392 393 return old_cred; 394 } 395 396 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags) 397 { 398 struct path path; 399 struct inode *inode; 400 int res; 401 unsigned int lookup_flags = LOOKUP_FOLLOW; 402 const struct cred *old_cred = NULL; 403 404 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */ 405 return -EINVAL; 406 407 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) 408 return -EINVAL; 409 410 if (flags & AT_SYMLINK_NOFOLLOW) 411 lookup_flags &= ~LOOKUP_FOLLOW; 412 if (flags & AT_EMPTY_PATH) 413 lookup_flags |= LOOKUP_EMPTY; 414 415 if (!(flags & AT_EACCESS)) { 416 old_cred = access_override_creds(); 417 if (!old_cred) 418 return -ENOMEM; 419 } 420 421 retry: 422 res = user_path_at(dfd, filename, lookup_flags, &path); 423 if (res) 424 goto out; 425 426 inode = d_backing_inode(path.dentry); 427 428 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) { 429 /* 430 * MAY_EXEC on regular files is denied if the fs is mounted 431 * with the "noexec" flag. 432 */ 433 res = -EACCES; 434 if (path_noexec(&path)) 435 goto out_path_release; 436 } 437 438 res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS); 439 /* SuS v2 requires we report a read only fs too */ 440 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode)) 441 goto out_path_release; 442 /* 443 * This is a rare case where using __mnt_is_readonly() 444 * is OK without a mnt_want/drop_write() pair. Since 445 * no actual write to the fs is performed here, we do 446 * not need to telegraph to that to anyone. 447 * 448 * By doing this, we accept that this access is 449 * inherently racy and know that the fs may change 450 * state before we even see this result. 451 */ 452 if (__mnt_is_readonly(path.mnt)) 453 res = -EROFS; 454 455 out_path_release: 456 path_put(&path); 457 if (retry_estale(res, lookup_flags)) { 458 lookup_flags |= LOOKUP_REVAL; 459 goto retry; 460 } 461 out: 462 if (old_cred) 463 revert_creds(old_cred); 464 465 return res; 466 } 467 468 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode) 469 { 470 return do_faccessat(dfd, filename, mode, 0); 471 } 472 473 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode, 474 int, flags) 475 { 476 return do_faccessat(dfd, filename, mode, flags); 477 } 478 479 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode) 480 { 481 return do_faccessat(AT_FDCWD, filename, mode, 0); 482 } 483 484 SYSCALL_DEFINE1(chdir, const char __user *, filename) 485 { 486 struct path path; 487 int error; 488 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 489 retry: 490 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); 491 if (error) 492 goto out; 493 494 error = path_permission(&path, MAY_EXEC | MAY_CHDIR); 495 if (error) 496 goto dput_and_out; 497 498 set_fs_pwd(current->fs, &path); 499 500 dput_and_out: 501 path_put(&path); 502 if (retry_estale(error, lookup_flags)) { 503 lookup_flags |= LOOKUP_REVAL; 504 goto retry; 505 } 506 out: 507 return error; 508 } 509 510 SYSCALL_DEFINE1(fchdir, unsigned int, fd) 511 { 512 struct fd f = fdget_raw(fd); 513 int error; 514 515 error = -EBADF; 516 if (!f.file) 517 goto out; 518 519 error = -ENOTDIR; 520 if (!d_can_lookup(f.file->f_path.dentry)) 521 goto out_putf; 522 523 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR); 524 if (!error) 525 set_fs_pwd(current->fs, &f.file->f_path); 526 out_putf: 527 fdput(f); 528 out: 529 return error; 530 } 531 532 SYSCALL_DEFINE1(chroot, const char __user *, filename) 533 { 534 struct path path; 535 int error; 536 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 537 retry: 538 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); 539 if (error) 540 goto out; 541 542 error = path_permission(&path, MAY_EXEC | MAY_CHDIR); 543 if (error) 544 goto dput_and_out; 545 546 error = -EPERM; 547 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT)) 548 goto dput_and_out; 549 error = security_path_chroot(&path); 550 if (error) 551 goto dput_and_out; 552 553 set_fs_root(current->fs, &path); 554 error = 0; 555 dput_and_out: 556 path_put(&path); 557 if (retry_estale(error, lookup_flags)) { 558 lookup_flags |= LOOKUP_REVAL; 559 goto retry; 560 } 561 out: 562 return error; 563 } 564 565 int chmod_common(const struct path *path, umode_t mode) 566 { 567 struct inode *inode = path->dentry->d_inode; 568 struct inode *delegated_inode = NULL; 569 struct iattr newattrs; 570 int error; 571 572 error = mnt_want_write(path->mnt); 573 if (error) 574 return error; 575 retry_deleg: 576 inode_lock(inode); 577 error = security_path_chmod(path, mode); 578 if (error) 579 goto out_unlock; 580 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO); 581 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME; 582 error = notify_change(mnt_user_ns(path->mnt), path->dentry, 583 &newattrs, &delegated_inode); 584 out_unlock: 585 inode_unlock(inode); 586 if (delegated_inode) { 587 error = break_deleg_wait(&delegated_inode); 588 if (!error) 589 goto retry_deleg; 590 } 591 mnt_drop_write(path->mnt); 592 return error; 593 } 594 595 int vfs_fchmod(struct file *file, umode_t mode) 596 { 597 audit_file(file); 598 return chmod_common(&file->f_path, mode); 599 } 600 601 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode) 602 { 603 struct fd f = fdget(fd); 604 int err = -EBADF; 605 606 if (f.file) { 607 err = vfs_fchmod(f.file, mode); 608 fdput(f); 609 } 610 return err; 611 } 612 613 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode) 614 { 615 struct path path; 616 int error; 617 unsigned int lookup_flags = LOOKUP_FOLLOW; 618 retry: 619 error = user_path_at(dfd, filename, lookup_flags, &path); 620 if (!error) { 621 error = chmod_common(&path, mode); 622 path_put(&path); 623 if (retry_estale(error, lookup_flags)) { 624 lookup_flags |= LOOKUP_REVAL; 625 goto retry; 626 } 627 } 628 return error; 629 } 630 631 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, 632 umode_t, mode) 633 { 634 return do_fchmodat(dfd, filename, mode); 635 } 636 637 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode) 638 { 639 return do_fchmodat(AT_FDCWD, filename, mode); 640 } 641 642 int chown_common(const struct path *path, uid_t user, gid_t group) 643 { 644 struct user_namespace *mnt_userns, *fs_userns; 645 struct inode *inode = path->dentry->d_inode; 646 struct inode *delegated_inode = NULL; 647 int error; 648 struct iattr newattrs; 649 kuid_t uid; 650 kgid_t gid; 651 652 uid = make_kuid(current_user_ns(), user); 653 gid = make_kgid(current_user_ns(), group); 654 655 mnt_userns = mnt_user_ns(path->mnt); 656 fs_userns = i_user_ns(inode); 657 uid = mapped_kuid_user(mnt_userns, fs_userns, uid); 658 gid = mapped_kgid_user(mnt_userns, fs_userns, gid); 659 660 retry_deleg: 661 newattrs.ia_valid = ATTR_CTIME; 662 if (user != (uid_t) -1) { 663 if (!uid_valid(uid)) 664 return -EINVAL; 665 newattrs.ia_valid |= ATTR_UID; 666 newattrs.ia_uid = uid; 667 } 668 if (group != (gid_t) -1) { 669 if (!gid_valid(gid)) 670 return -EINVAL; 671 newattrs.ia_valid |= ATTR_GID; 672 newattrs.ia_gid = gid; 673 } 674 if (!S_ISDIR(inode->i_mode)) 675 newattrs.ia_valid |= 676 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 677 inode_lock(inode); 678 error = security_path_chown(path, uid, gid); 679 if (!error) 680 error = notify_change(mnt_userns, path->dentry, &newattrs, 681 &delegated_inode); 682 inode_unlock(inode); 683 if (delegated_inode) { 684 error = break_deleg_wait(&delegated_inode); 685 if (!error) 686 goto retry_deleg; 687 } 688 return error; 689 } 690 691 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group, 692 int flag) 693 { 694 struct path path; 695 int error = -EINVAL; 696 int lookup_flags; 697 698 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0) 699 goto out; 700 701 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW; 702 if (flag & AT_EMPTY_PATH) 703 lookup_flags |= LOOKUP_EMPTY; 704 retry: 705 error = user_path_at(dfd, filename, lookup_flags, &path); 706 if (error) 707 goto out; 708 error = mnt_want_write(path.mnt); 709 if (error) 710 goto out_release; 711 error = chown_common(&path, user, group); 712 mnt_drop_write(path.mnt); 713 out_release: 714 path_put(&path); 715 if (retry_estale(error, lookup_flags)) { 716 lookup_flags |= LOOKUP_REVAL; 717 goto retry; 718 } 719 out: 720 return error; 721 } 722 723 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user, 724 gid_t, group, int, flag) 725 { 726 return do_fchownat(dfd, filename, user, group, flag); 727 } 728 729 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group) 730 { 731 return do_fchownat(AT_FDCWD, filename, user, group, 0); 732 } 733 734 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group) 735 { 736 return do_fchownat(AT_FDCWD, filename, user, group, 737 AT_SYMLINK_NOFOLLOW); 738 } 739 740 int vfs_fchown(struct file *file, uid_t user, gid_t group) 741 { 742 int error; 743 744 error = mnt_want_write_file(file); 745 if (error) 746 return error; 747 audit_file(file); 748 error = chown_common(&file->f_path, user, group); 749 mnt_drop_write_file(file); 750 return error; 751 } 752 753 int ksys_fchown(unsigned int fd, uid_t user, gid_t group) 754 { 755 struct fd f = fdget(fd); 756 int error = -EBADF; 757 758 if (f.file) { 759 error = vfs_fchown(f.file, user, group); 760 fdput(f); 761 } 762 return error; 763 } 764 765 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group) 766 { 767 return ksys_fchown(fd, user, group); 768 } 769 770 static int do_dentry_open(struct file *f, 771 struct inode *inode, 772 int (*open)(struct inode *, struct file *)) 773 { 774 static const struct file_operations empty_fops = {}; 775 int error; 776 777 path_get(&f->f_path); 778 f->f_inode = inode; 779 f->f_mapping = inode->i_mapping; 780 f->f_wb_err = filemap_sample_wb_err(f->f_mapping); 781 f->f_sb_err = file_sample_sb_err(f); 782 783 if (unlikely(f->f_flags & O_PATH)) { 784 f->f_mode = FMODE_PATH | FMODE_OPENED; 785 f->f_op = &empty_fops; 786 return 0; 787 } 788 789 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) { 790 error = get_write_access(inode); 791 if (unlikely(error)) 792 goto cleanup_file; 793 error = __mnt_want_write(f->f_path.mnt); 794 if (unlikely(error)) { 795 put_write_access(inode); 796 goto cleanup_file; 797 } 798 f->f_mode |= FMODE_WRITER; 799 } 800 801 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */ 802 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)) 803 f->f_mode |= FMODE_ATOMIC_POS; 804 805 f->f_op = fops_get(inode->i_fop); 806 if (WARN_ON(!f->f_op)) { 807 error = -ENODEV; 808 goto cleanup_all; 809 } 810 811 error = security_file_open(f); 812 if (error) 813 goto cleanup_all; 814 815 error = break_lease(locks_inode(f), f->f_flags); 816 if (error) 817 goto cleanup_all; 818 819 /* normally all 3 are set; ->open() can clear them if needed */ 820 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE; 821 if (!open) 822 open = f->f_op->open; 823 if (open) { 824 error = open(inode, f); 825 if (error) 826 goto cleanup_all; 827 } 828 f->f_mode |= FMODE_OPENED; 829 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) 830 i_readcount_inc(inode); 831 if ((f->f_mode & FMODE_READ) && 832 likely(f->f_op->read || f->f_op->read_iter)) 833 f->f_mode |= FMODE_CAN_READ; 834 if ((f->f_mode & FMODE_WRITE) && 835 likely(f->f_op->write || f->f_op->write_iter)) 836 f->f_mode |= FMODE_CAN_WRITE; 837 838 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC); 839 840 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping); 841 842 /* NB: we're sure to have correct a_ops only after f_op->open */ 843 if (f->f_flags & O_DIRECT) { 844 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO) 845 return -EINVAL; 846 } 847 848 /* 849 * XXX: Huge page cache doesn't support writing yet. Drop all page 850 * cache for this file before processing writes. 851 */ 852 if (f->f_mode & FMODE_WRITE) { 853 /* 854 * Paired with smp_mb() in collapse_file() to ensure nr_thps 855 * is up to date and the update to i_writecount by 856 * get_write_access() is visible. Ensures subsequent insertion 857 * of THPs into the page cache will fail. 858 */ 859 smp_mb(); 860 if (filemap_nr_thps(inode->i_mapping)) { 861 struct address_space *mapping = inode->i_mapping; 862 863 filemap_invalidate_lock(inode->i_mapping); 864 /* 865 * unmap_mapping_range just need to be called once 866 * here, because the private pages is not need to be 867 * unmapped mapping (e.g. data segment of dynamic 868 * shared libraries here). 869 */ 870 unmap_mapping_range(mapping, 0, 0, 0); 871 truncate_inode_pages(mapping, 0); 872 filemap_invalidate_unlock(inode->i_mapping); 873 } 874 } 875 876 return 0; 877 878 cleanup_all: 879 if (WARN_ON_ONCE(error > 0)) 880 error = -EINVAL; 881 fops_put(f->f_op); 882 if (f->f_mode & FMODE_WRITER) { 883 put_write_access(inode); 884 __mnt_drop_write(f->f_path.mnt); 885 } 886 cleanup_file: 887 path_put(&f->f_path); 888 f->f_path.mnt = NULL; 889 f->f_path.dentry = NULL; 890 f->f_inode = NULL; 891 return error; 892 } 893 894 /** 895 * finish_open - finish opening a file 896 * @file: file pointer 897 * @dentry: pointer to dentry 898 * @open: open callback 899 * @opened: state of open 900 * 901 * This can be used to finish opening a file passed to i_op->atomic_open(). 902 * 903 * If the open callback is set to NULL, then the standard f_op->open() 904 * filesystem callback is substituted. 905 * 906 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is 907 * the return value of d_splice_alias(), then the caller needs to perform dput() 908 * on it after finish_open(). 909 * 910 * Returns zero on success or -errno if the open failed. 911 */ 912 int finish_open(struct file *file, struct dentry *dentry, 913 int (*open)(struct inode *, struct file *)) 914 { 915 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 916 917 file->f_path.dentry = dentry; 918 return do_dentry_open(file, d_backing_inode(dentry), open); 919 } 920 EXPORT_SYMBOL(finish_open); 921 922 /** 923 * finish_no_open - finish ->atomic_open() without opening the file 924 * 925 * @file: file pointer 926 * @dentry: dentry or NULL (as returned from ->lookup()) 927 * 928 * This can be used to set the result of a successful lookup in ->atomic_open(). 929 * 930 * NB: unlike finish_open() this function does consume the dentry reference and 931 * the caller need not dput() it. 932 * 933 * Returns "0" which must be the return value of ->atomic_open() after having 934 * called this function. 935 */ 936 int finish_no_open(struct file *file, struct dentry *dentry) 937 { 938 file->f_path.dentry = dentry; 939 return 0; 940 } 941 EXPORT_SYMBOL(finish_no_open); 942 943 char *file_path(struct file *filp, char *buf, int buflen) 944 { 945 return d_path(&filp->f_path, buf, buflen); 946 } 947 EXPORT_SYMBOL(file_path); 948 949 /** 950 * vfs_open - open the file at the given path 951 * @path: path to open 952 * @file: newly allocated file with f_flag initialized 953 * @cred: credentials to use 954 */ 955 int vfs_open(const struct path *path, struct file *file) 956 { 957 file->f_path = *path; 958 return do_dentry_open(file, d_backing_inode(path->dentry), NULL); 959 } 960 961 struct file *dentry_open(const struct path *path, int flags, 962 const struct cred *cred) 963 { 964 int error; 965 struct file *f; 966 967 validate_creds(cred); 968 969 /* We must always pass in a valid mount pointer. */ 970 BUG_ON(!path->mnt); 971 972 f = alloc_empty_file(flags, cred); 973 if (!IS_ERR(f)) { 974 error = vfs_open(path, f); 975 if (error) { 976 fput(f); 977 f = ERR_PTR(error); 978 } 979 } 980 return f; 981 } 982 EXPORT_SYMBOL(dentry_open); 983 984 struct file *open_with_fake_path(const struct path *path, int flags, 985 struct inode *inode, const struct cred *cred) 986 { 987 struct file *f = alloc_empty_file_noaccount(flags, cred); 988 if (!IS_ERR(f)) { 989 int error; 990 991 f->f_path = *path; 992 error = do_dentry_open(f, inode, NULL); 993 if (error) { 994 fput(f); 995 f = ERR_PTR(error); 996 } 997 } 998 return f; 999 } 1000 EXPORT_SYMBOL(open_with_fake_path); 1001 1002 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE)) 1003 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC) 1004 1005 inline struct open_how build_open_how(int flags, umode_t mode) 1006 { 1007 struct open_how how = { 1008 .flags = flags & VALID_OPEN_FLAGS, 1009 .mode = mode & S_IALLUGO, 1010 }; 1011 1012 /* O_PATH beats everything else. */ 1013 if (how.flags & O_PATH) 1014 how.flags &= O_PATH_FLAGS; 1015 /* Modes should only be set for create-like flags. */ 1016 if (!WILL_CREATE(how.flags)) 1017 how.mode = 0; 1018 return how; 1019 } 1020 1021 inline int build_open_flags(const struct open_how *how, struct open_flags *op) 1022 { 1023 u64 flags = how->flags; 1024 u64 strip = FMODE_NONOTIFY | O_CLOEXEC; 1025 int lookup_flags = 0; 1026 int acc_mode = ACC_MODE(flags); 1027 1028 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS), 1029 "struct open_flags doesn't yet handle flags > 32 bits"); 1030 1031 /* 1032 * Strip flags that either shouldn't be set by userspace like 1033 * FMODE_NONOTIFY or that aren't relevant in determining struct 1034 * open_flags like O_CLOEXEC. 1035 */ 1036 flags &= ~strip; 1037 1038 /* 1039 * Older syscalls implicitly clear all of the invalid flags or argument 1040 * values before calling build_open_flags(), but openat2(2) checks all 1041 * of its arguments. 1042 */ 1043 if (flags & ~VALID_OPEN_FLAGS) 1044 return -EINVAL; 1045 if (how->resolve & ~VALID_RESOLVE_FLAGS) 1046 return -EINVAL; 1047 1048 /* Scoping flags are mutually exclusive. */ 1049 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT)) 1050 return -EINVAL; 1051 1052 /* Deal with the mode. */ 1053 if (WILL_CREATE(flags)) { 1054 if (how->mode & ~S_IALLUGO) 1055 return -EINVAL; 1056 op->mode = how->mode | S_IFREG; 1057 } else { 1058 if (how->mode != 0) 1059 return -EINVAL; 1060 op->mode = 0; 1061 } 1062 1063 /* 1064 * In order to ensure programs get explicit errors when trying to use 1065 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it 1066 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we 1067 * have to require userspace to explicitly set it. 1068 */ 1069 if (flags & __O_TMPFILE) { 1070 if ((flags & O_TMPFILE_MASK) != O_TMPFILE) 1071 return -EINVAL; 1072 if (!(acc_mode & MAY_WRITE)) 1073 return -EINVAL; 1074 } 1075 if (flags & O_PATH) { 1076 /* O_PATH only permits certain other flags to be set. */ 1077 if (flags & ~O_PATH_FLAGS) 1078 return -EINVAL; 1079 acc_mode = 0; 1080 } 1081 1082 /* 1083 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only 1084 * check for O_DSYNC if the need any syncing at all we enforce it's 1085 * always set instead of having to deal with possibly weird behaviour 1086 * for malicious applications setting only __O_SYNC. 1087 */ 1088 if (flags & __O_SYNC) 1089 flags |= O_DSYNC; 1090 1091 op->open_flag = flags; 1092 1093 /* O_TRUNC implies we need access checks for write permissions */ 1094 if (flags & O_TRUNC) 1095 acc_mode |= MAY_WRITE; 1096 1097 /* Allow the LSM permission hook to distinguish append 1098 access from general write access. */ 1099 if (flags & O_APPEND) 1100 acc_mode |= MAY_APPEND; 1101 1102 op->acc_mode = acc_mode; 1103 1104 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN; 1105 1106 if (flags & O_CREAT) { 1107 op->intent |= LOOKUP_CREATE; 1108 if (flags & O_EXCL) { 1109 op->intent |= LOOKUP_EXCL; 1110 flags |= O_NOFOLLOW; 1111 } 1112 } 1113 1114 if (flags & O_DIRECTORY) 1115 lookup_flags |= LOOKUP_DIRECTORY; 1116 if (!(flags & O_NOFOLLOW)) 1117 lookup_flags |= LOOKUP_FOLLOW; 1118 1119 if (how->resolve & RESOLVE_NO_XDEV) 1120 lookup_flags |= LOOKUP_NO_XDEV; 1121 if (how->resolve & RESOLVE_NO_MAGICLINKS) 1122 lookup_flags |= LOOKUP_NO_MAGICLINKS; 1123 if (how->resolve & RESOLVE_NO_SYMLINKS) 1124 lookup_flags |= LOOKUP_NO_SYMLINKS; 1125 if (how->resolve & RESOLVE_BENEATH) 1126 lookup_flags |= LOOKUP_BENEATH; 1127 if (how->resolve & RESOLVE_IN_ROOT) 1128 lookup_flags |= LOOKUP_IN_ROOT; 1129 if (how->resolve & RESOLVE_CACHED) { 1130 /* Don't bother even trying for create/truncate/tmpfile open */ 1131 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE)) 1132 return -EAGAIN; 1133 lookup_flags |= LOOKUP_CACHED; 1134 } 1135 1136 op->lookup_flags = lookup_flags; 1137 return 0; 1138 } 1139 1140 /** 1141 * file_open_name - open file and return file pointer 1142 * 1143 * @name: struct filename containing path to open 1144 * @flags: open flags as per the open(2) second argument 1145 * @mode: mode for the new file if O_CREAT is set, else ignored 1146 * 1147 * This is the helper to open a file from kernelspace if you really 1148 * have to. But in generally you should not do this, so please move 1149 * along, nothing to see here.. 1150 */ 1151 struct file *file_open_name(struct filename *name, int flags, umode_t mode) 1152 { 1153 struct open_flags op; 1154 struct open_how how = build_open_how(flags, mode); 1155 int err = build_open_flags(&how, &op); 1156 if (err) 1157 return ERR_PTR(err); 1158 return do_filp_open(AT_FDCWD, name, &op); 1159 } 1160 1161 /** 1162 * filp_open - open file and return file pointer 1163 * 1164 * @filename: path to open 1165 * @flags: open flags as per the open(2) second argument 1166 * @mode: mode for the new file if O_CREAT is set, else ignored 1167 * 1168 * This is the helper to open a file from kernelspace if you really 1169 * have to. But in generally you should not do this, so please move 1170 * along, nothing to see here.. 1171 */ 1172 struct file *filp_open(const char *filename, int flags, umode_t mode) 1173 { 1174 struct filename *name = getname_kernel(filename); 1175 struct file *file = ERR_CAST(name); 1176 1177 if (!IS_ERR(name)) { 1178 file = file_open_name(name, flags, mode); 1179 putname(name); 1180 } 1181 return file; 1182 } 1183 EXPORT_SYMBOL(filp_open); 1184 1185 struct file *file_open_root(const struct path *root, 1186 const char *filename, int flags, umode_t mode) 1187 { 1188 struct open_flags op; 1189 struct open_how how = build_open_how(flags, mode); 1190 int err = build_open_flags(&how, &op); 1191 if (err) 1192 return ERR_PTR(err); 1193 return do_file_open_root(root, filename, &op); 1194 } 1195 EXPORT_SYMBOL(file_open_root); 1196 1197 static long do_sys_openat2(int dfd, const char __user *filename, 1198 struct open_how *how) 1199 { 1200 struct open_flags op; 1201 int fd = build_open_flags(how, &op); 1202 struct filename *tmp; 1203 1204 if (fd) 1205 return fd; 1206 1207 tmp = getname(filename); 1208 if (IS_ERR(tmp)) 1209 return PTR_ERR(tmp); 1210 1211 fd = get_unused_fd_flags(how->flags); 1212 if (fd >= 0) { 1213 struct file *f = do_filp_open(dfd, tmp, &op); 1214 if (IS_ERR(f)) { 1215 put_unused_fd(fd); 1216 fd = PTR_ERR(f); 1217 } else { 1218 fsnotify_open(f); 1219 fd_install(fd, f); 1220 } 1221 } 1222 putname(tmp); 1223 return fd; 1224 } 1225 1226 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode) 1227 { 1228 struct open_how how = build_open_how(flags, mode); 1229 return do_sys_openat2(dfd, filename, &how); 1230 } 1231 1232 1233 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode) 1234 { 1235 if (force_o_largefile()) 1236 flags |= O_LARGEFILE; 1237 return do_sys_open(AT_FDCWD, filename, flags, mode); 1238 } 1239 1240 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, 1241 umode_t, mode) 1242 { 1243 if (force_o_largefile()) 1244 flags |= O_LARGEFILE; 1245 return do_sys_open(dfd, filename, flags, mode); 1246 } 1247 1248 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename, 1249 struct open_how __user *, how, size_t, usize) 1250 { 1251 int err; 1252 struct open_how tmp; 1253 1254 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0); 1255 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST); 1256 1257 if (unlikely(usize < OPEN_HOW_SIZE_VER0)) 1258 return -EINVAL; 1259 1260 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize); 1261 if (err) 1262 return err; 1263 1264 audit_openat2_how(&tmp); 1265 1266 /* O_LARGEFILE is only allowed for non-O_PATH. */ 1267 if (!(tmp.flags & O_PATH) && force_o_largefile()) 1268 tmp.flags |= O_LARGEFILE; 1269 1270 return do_sys_openat2(dfd, filename, &tmp); 1271 } 1272 1273 #ifdef CONFIG_COMPAT 1274 /* 1275 * Exactly like sys_open(), except that it doesn't set the 1276 * O_LARGEFILE flag. 1277 */ 1278 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode) 1279 { 1280 return do_sys_open(AT_FDCWD, filename, flags, mode); 1281 } 1282 1283 /* 1284 * Exactly like sys_openat(), except that it doesn't set the 1285 * O_LARGEFILE flag. 1286 */ 1287 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode) 1288 { 1289 return do_sys_open(dfd, filename, flags, mode); 1290 } 1291 #endif 1292 1293 #ifndef __alpha__ 1294 1295 /* 1296 * For backward compatibility? Maybe this should be moved 1297 * into arch/i386 instead? 1298 */ 1299 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode) 1300 { 1301 int flags = O_CREAT | O_WRONLY | O_TRUNC; 1302 1303 if (force_o_largefile()) 1304 flags |= O_LARGEFILE; 1305 return do_sys_open(AT_FDCWD, pathname, flags, mode); 1306 } 1307 #endif 1308 1309 /* 1310 * "id" is the POSIX thread ID. We use the 1311 * files pointer for this.. 1312 */ 1313 int filp_close(struct file *filp, fl_owner_t id) 1314 { 1315 int retval = 0; 1316 1317 if (!file_count(filp)) { 1318 printk(KERN_ERR "VFS: Close: file count is 0\n"); 1319 return 0; 1320 } 1321 1322 if (filp->f_op->flush) 1323 retval = filp->f_op->flush(filp, id); 1324 1325 if (likely(!(filp->f_mode & FMODE_PATH))) { 1326 dnotify_flush(filp, id); 1327 locks_remove_posix(filp, id); 1328 } 1329 fput(filp); 1330 return retval; 1331 } 1332 1333 EXPORT_SYMBOL(filp_close); 1334 1335 /* 1336 * Careful here! We test whether the file pointer is NULL before 1337 * releasing the fd. This ensures that one clone task can't release 1338 * an fd while another clone is opening it. 1339 */ 1340 SYSCALL_DEFINE1(close, unsigned int, fd) 1341 { 1342 int retval = close_fd(fd); 1343 1344 /* can't restart close syscall because file table entry was cleared */ 1345 if (unlikely(retval == -ERESTARTSYS || 1346 retval == -ERESTARTNOINTR || 1347 retval == -ERESTARTNOHAND || 1348 retval == -ERESTART_RESTARTBLOCK)) 1349 retval = -EINTR; 1350 1351 return retval; 1352 } 1353 1354 /** 1355 * close_range() - Close all file descriptors in a given range. 1356 * 1357 * @fd: starting file descriptor to close 1358 * @max_fd: last file descriptor to close 1359 * @flags: reserved for future extensions 1360 * 1361 * This closes a range of file descriptors. All file descriptors 1362 * from @fd up to and including @max_fd are closed. 1363 * Currently, errors to close a given file descriptor are ignored. 1364 */ 1365 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd, 1366 unsigned int, flags) 1367 { 1368 return __close_range(fd, max_fd, flags); 1369 } 1370 1371 /* 1372 * This routine simulates a hangup on the tty, to arrange that users 1373 * are given clean terminals at login time. 1374 */ 1375 SYSCALL_DEFINE0(vhangup) 1376 { 1377 if (capable(CAP_SYS_TTY_CONFIG)) { 1378 tty_vhangup_self(); 1379 return 0; 1380 } 1381 return -EPERM; 1382 } 1383 1384 /* 1385 * Called when an inode is about to be open. 1386 * We use this to disallow opening large files on 32bit systems if 1387 * the caller didn't specify O_LARGEFILE. On 64bit systems we force 1388 * on this flag in sys_open. 1389 */ 1390 int generic_file_open(struct inode * inode, struct file * filp) 1391 { 1392 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) 1393 return -EOVERFLOW; 1394 return 0; 1395 } 1396 1397 EXPORT_SYMBOL(generic_file_open); 1398 1399 /* 1400 * This is used by subsystems that don't want seekable 1401 * file descriptors. The function is not supposed to ever fail, the only 1402 * reason it returns an 'int' and not 'void' is so that it can be plugged 1403 * directly into file_operations structure. 1404 */ 1405 int nonseekable_open(struct inode *inode, struct file *filp) 1406 { 1407 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE); 1408 return 0; 1409 } 1410 1411 EXPORT_SYMBOL(nonseekable_open); 1412 1413 /* 1414 * stream_open is used by subsystems that want stream-like file descriptors. 1415 * Such file descriptors are not seekable and don't have notion of position 1416 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL). 1417 * Contrary to file descriptors of other regular files, .read() and .write() 1418 * can run simultaneously. 1419 * 1420 * stream_open never fails and is marked to return int so that it could be 1421 * directly used as file_operations.open . 1422 */ 1423 int stream_open(struct inode *inode, struct file *filp) 1424 { 1425 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS); 1426 filp->f_mode |= FMODE_STREAM; 1427 return 0; 1428 } 1429 1430 EXPORT_SYMBOL(stream_open); 1431