1 /* 2 * linux/fs/open.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7 #include <linux/string.h> 8 #include <linux/mm.h> 9 #include <linux/file.h> 10 #include <linux/fdtable.h> 11 #include <linux/fsnotify.h> 12 #include <linux/module.h> 13 #include <linux/tty.h> 14 #include <linux/namei.h> 15 #include <linux/backing-dev.h> 16 #include <linux/capability.h> 17 #include <linux/securebits.h> 18 #include <linux/security.h> 19 #include <linux/mount.h> 20 #include <linux/fcntl.h> 21 #include <linux/slab.h> 22 #include <asm/uaccess.h> 23 #include <linux/fs.h> 24 #include <linux/personality.h> 25 #include <linux/pagemap.h> 26 #include <linux/syscalls.h> 27 #include <linux/rcupdate.h> 28 #include <linux/audit.h> 29 #include <linux/falloc.h> 30 #include <linux/fs_struct.h> 31 #include <linux/ima.h> 32 #include <linux/dnotify.h> 33 #include <linux/compat.h> 34 35 #include "internal.h" 36 37 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs, 38 struct file *filp) 39 { 40 int ret; 41 struct iattr newattrs; 42 43 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */ 44 if (length < 0) 45 return -EINVAL; 46 47 newattrs.ia_size = length; 48 newattrs.ia_valid = ATTR_SIZE | time_attrs; 49 if (filp) { 50 newattrs.ia_file = filp; 51 newattrs.ia_valid |= ATTR_FILE; 52 } 53 54 /* Remove suid/sgid on truncate too */ 55 ret = should_remove_suid(dentry); 56 if (ret) 57 newattrs.ia_valid |= ret | ATTR_FORCE; 58 59 mutex_lock(&dentry->d_inode->i_mutex); 60 /* Note any delegations or leases have already been broken: */ 61 ret = notify_change(dentry, &newattrs, NULL); 62 mutex_unlock(&dentry->d_inode->i_mutex); 63 return ret; 64 } 65 66 long vfs_truncate(struct path *path, loff_t length) 67 { 68 struct inode *inode; 69 long error; 70 71 inode = path->dentry->d_inode; 72 73 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */ 74 if (S_ISDIR(inode->i_mode)) 75 return -EISDIR; 76 if (!S_ISREG(inode->i_mode)) 77 return -EINVAL; 78 79 error = mnt_want_write(path->mnt); 80 if (error) 81 goto out; 82 83 error = inode_permission(inode, MAY_WRITE); 84 if (error) 85 goto mnt_drop_write_and_out; 86 87 error = -EPERM; 88 if (IS_APPEND(inode)) 89 goto mnt_drop_write_and_out; 90 91 error = get_write_access(inode); 92 if (error) 93 goto mnt_drop_write_and_out; 94 95 /* 96 * Make sure that there are no leases. get_write_access() protects 97 * against the truncate racing with a lease-granting setlease(). 98 */ 99 error = break_lease(inode, O_WRONLY); 100 if (error) 101 goto put_write_and_out; 102 103 error = locks_verify_truncate(inode, NULL, length); 104 if (!error) 105 error = security_path_truncate(path); 106 if (!error) 107 error = do_truncate(path->dentry, length, 0, NULL); 108 109 put_write_and_out: 110 put_write_access(inode); 111 mnt_drop_write_and_out: 112 mnt_drop_write(path->mnt); 113 out: 114 return error; 115 } 116 EXPORT_SYMBOL_GPL(vfs_truncate); 117 118 static long do_sys_truncate(const char __user *pathname, loff_t length) 119 { 120 unsigned int lookup_flags = LOOKUP_FOLLOW; 121 struct path path; 122 int error; 123 124 if (length < 0) /* sorry, but loff_t says... */ 125 return -EINVAL; 126 127 retry: 128 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path); 129 if (!error) { 130 error = vfs_truncate(&path, length); 131 path_put(&path); 132 } 133 if (retry_estale(error, lookup_flags)) { 134 lookup_flags |= LOOKUP_REVAL; 135 goto retry; 136 } 137 return error; 138 } 139 140 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length) 141 { 142 return do_sys_truncate(path, length); 143 } 144 145 #ifdef CONFIG_COMPAT 146 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length) 147 { 148 return do_sys_truncate(path, length); 149 } 150 #endif 151 152 static long do_sys_ftruncate(unsigned int fd, loff_t length, int small) 153 { 154 struct inode *inode; 155 struct dentry *dentry; 156 struct fd f; 157 int error; 158 159 error = -EINVAL; 160 if (length < 0) 161 goto out; 162 error = -EBADF; 163 f = fdget(fd); 164 if (!f.file) 165 goto out; 166 167 /* explicitly opened as large or we are on 64-bit box */ 168 if (f.file->f_flags & O_LARGEFILE) 169 small = 0; 170 171 dentry = f.file->f_path.dentry; 172 inode = dentry->d_inode; 173 error = -EINVAL; 174 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE)) 175 goto out_putf; 176 177 error = -EINVAL; 178 /* Cannot ftruncate over 2^31 bytes without large file support */ 179 if (small && length > MAX_NON_LFS) 180 goto out_putf; 181 182 error = -EPERM; 183 if (IS_APPEND(inode)) 184 goto out_putf; 185 186 sb_start_write(inode->i_sb); 187 error = locks_verify_truncate(inode, f.file, length); 188 if (!error) 189 error = security_path_truncate(&f.file->f_path); 190 if (!error) 191 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file); 192 sb_end_write(inode->i_sb); 193 out_putf: 194 fdput(f); 195 out: 196 return error; 197 } 198 199 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length) 200 { 201 return do_sys_ftruncate(fd, length, 1); 202 } 203 204 #ifdef CONFIG_COMPAT 205 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length) 206 { 207 return do_sys_ftruncate(fd, length, 1); 208 } 209 #endif 210 211 /* LFS versions of truncate are only needed on 32 bit machines */ 212 #if BITS_PER_LONG == 32 213 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length) 214 { 215 return do_sys_truncate(path, length); 216 } 217 218 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length) 219 { 220 return do_sys_ftruncate(fd, length, 0); 221 } 222 #endif /* BITS_PER_LONG == 32 */ 223 224 225 int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len) 226 { 227 struct inode *inode = file_inode(file); 228 long ret; 229 230 if (offset < 0 || len <= 0) 231 return -EINVAL; 232 233 /* Return error if mode is not supported */ 234 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | 235 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE)) 236 return -EOPNOTSUPP; 237 238 /* Punch hole and zero range are mutually exclusive */ 239 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) == 240 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) 241 return -EOPNOTSUPP; 242 243 /* Punch hole must have keep size set */ 244 if ((mode & FALLOC_FL_PUNCH_HOLE) && 245 !(mode & FALLOC_FL_KEEP_SIZE)) 246 return -EOPNOTSUPP; 247 248 /* Collapse range should only be used exclusively. */ 249 if ((mode & FALLOC_FL_COLLAPSE_RANGE) && 250 (mode & ~FALLOC_FL_COLLAPSE_RANGE)) 251 return -EINVAL; 252 253 if (!(file->f_mode & FMODE_WRITE)) 254 return -EBADF; 255 256 /* 257 * We can only allow pure fallocate on append only files 258 */ 259 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode)) 260 return -EPERM; 261 262 if (IS_IMMUTABLE(inode)) 263 return -EPERM; 264 265 /* 266 * We can not allow to do any fallocate operation on an active 267 * swapfile 268 */ 269 if (IS_SWAPFILE(inode)) 270 ret = -ETXTBSY; 271 272 /* 273 * Revalidate the write permissions, in case security policy has 274 * changed since the files were opened. 275 */ 276 ret = security_file_permission(file, MAY_WRITE); 277 if (ret) 278 return ret; 279 280 if (S_ISFIFO(inode->i_mode)) 281 return -ESPIPE; 282 283 /* 284 * Let individual file system decide if it supports preallocation 285 * for directories or not. 286 */ 287 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode)) 288 return -ENODEV; 289 290 /* Check for wrap through zero too */ 291 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0)) 292 return -EFBIG; 293 294 if (!file->f_op->fallocate) 295 return -EOPNOTSUPP; 296 297 sb_start_write(inode->i_sb); 298 ret = file->f_op->fallocate(file, mode, offset, len); 299 sb_end_write(inode->i_sb); 300 return ret; 301 } 302 303 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len) 304 { 305 struct fd f = fdget(fd); 306 int error = -EBADF; 307 308 if (f.file) { 309 error = do_fallocate(f.file, mode, offset, len); 310 fdput(f); 311 } 312 return error; 313 } 314 315 /* 316 * access() needs to use the real uid/gid, not the effective uid/gid. 317 * We do this by temporarily clearing all FS-related capabilities and 318 * switching the fsuid/fsgid around to the real ones. 319 */ 320 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode) 321 { 322 const struct cred *old_cred; 323 struct cred *override_cred; 324 struct path path; 325 struct inode *inode; 326 int res; 327 unsigned int lookup_flags = LOOKUP_FOLLOW; 328 329 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */ 330 return -EINVAL; 331 332 override_cred = prepare_creds(); 333 if (!override_cred) 334 return -ENOMEM; 335 336 override_cred->fsuid = override_cred->uid; 337 override_cred->fsgid = override_cred->gid; 338 339 if (!issecure(SECURE_NO_SETUID_FIXUP)) { 340 /* Clear the capabilities if we switch to a non-root user */ 341 kuid_t root_uid = make_kuid(override_cred->user_ns, 0); 342 if (!uid_eq(override_cred->uid, root_uid)) 343 cap_clear(override_cred->cap_effective); 344 else 345 override_cred->cap_effective = 346 override_cred->cap_permitted; 347 } 348 349 old_cred = override_creds(override_cred); 350 retry: 351 res = user_path_at(dfd, filename, lookup_flags, &path); 352 if (res) 353 goto out; 354 355 inode = path.dentry->d_inode; 356 357 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) { 358 /* 359 * MAY_EXEC on regular files is denied if the fs is mounted 360 * with the "noexec" flag. 361 */ 362 res = -EACCES; 363 if (path.mnt->mnt_flags & MNT_NOEXEC) 364 goto out_path_release; 365 } 366 367 res = inode_permission(inode, mode | MAY_ACCESS); 368 /* SuS v2 requires we report a read only fs too */ 369 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode)) 370 goto out_path_release; 371 /* 372 * This is a rare case where using __mnt_is_readonly() 373 * is OK without a mnt_want/drop_write() pair. Since 374 * no actual write to the fs is performed here, we do 375 * not need to telegraph to that to anyone. 376 * 377 * By doing this, we accept that this access is 378 * inherently racy and know that the fs may change 379 * state before we even see this result. 380 */ 381 if (__mnt_is_readonly(path.mnt)) 382 res = -EROFS; 383 384 out_path_release: 385 path_put(&path); 386 if (retry_estale(res, lookup_flags)) { 387 lookup_flags |= LOOKUP_REVAL; 388 goto retry; 389 } 390 out: 391 revert_creds(old_cred); 392 put_cred(override_cred); 393 return res; 394 } 395 396 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode) 397 { 398 return sys_faccessat(AT_FDCWD, filename, mode); 399 } 400 401 SYSCALL_DEFINE1(chdir, const char __user *, filename) 402 { 403 struct path path; 404 int error; 405 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 406 retry: 407 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); 408 if (error) 409 goto out; 410 411 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR); 412 if (error) 413 goto dput_and_out; 414 415 set_fs_pwd(current->fs, &path); 416 417 dput_and_out: 418 path_put(&path); 419 if (retry_estale(error, lookup_flags)) { 420 lookup_flags |= LOOKUP_REVAL; 421 goto retry; 422 } 423 out: 424 return error; 425 } 426 427 SYSCALL_DEFINE1(fchdir, unsigned int, fd) 428 { 429 struct fd f = fdget_raw(fd); 430 struct inode *inode; 431 int error = -EBADF; 432 433 error = -EBADF; 434 if (!f.file) 435 goto out; 436 437 inode = file_inode(f.file); 438 439 error = -ENOTDIR; 440 if (!S_ISDIR(inode->i_mode)) 441 goto out_putf; 442 443 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR); 444 if (!error) 445 set_fs_pwd(current->fs, &f.file->f_path); 446 out_putf: 447 fdput(f); 448 out: 449 return error; 450 } 451 452 SYSCALL_DEFINE1(chroot, const char __user *, filename) 453 { 454 struct path path; 455 int error; 456 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 457 retry: 458 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path); 459 if (error) 460 goto out; 461 462 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR); 463 if (error) 464 goto dput_and_out; 465 466 error = -EPERM; 467 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT)) 468 goto dput_and_out; 469 error = security_path_chroot(&path); 470 if (error) 471 goto dput_and_out; 472 473 set_fs_root(current->fs, &path); 474 error = 0; 475 dput_and_out: 476 path_put(&path); 477 if (retry_estale(error, lookup_flags)) { 478 lookup_flags |= LOOKUP_REVAL; 479 goto retry; 480 } 481 out: 482 return error; 483 } 484 485 static int chmod_common(struct path *path, umode_t mode) 486 { 487 struct inode *inode = path->dentry->d_inode; 488 struct inode *delegated_inode = NULL; 489 struct iattr newattrs; 490 int error; 491 492 error = mnt_want_write(path->mnt); 493 if (error) 494 return error; 495 retry_deleg: 496 mutex_lock(&inode->i_mutex); 497 error = security_path_chmod(path, mode); 498 if (error) 499 goto out_unlock; 500 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO); 501 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME; 502 error = notify_change(path->dentry, &newattrs, &delegated_inode); 503 out_unlock: 504 mutex_unlock(&inode->i_mutex); 505 if (delegated_inode) { 506 error = break_deleg_wait(&delegated_inode); 507 if (!error) 508 goto retry_deleg; 509 } 510 mnt_drop_write(path->mnt); 511 return error; 512 } 513 514 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode) 515 { 516 struct fd f = fdget(fd); 517 int err = -EBADF; 518 519 if (f.file) { 520 audit_inode(NULL, f.file->f_path.dentry, 0); 521 err = chmod_common(&f.file->f_path, mode); 522 fdput(f); 523 } 524 return err; 525 } 526 527 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode) 528 { 529 struct path path; 530 int error; 531 unsigned int lookup_flags = LOOKUP_FOLLOW; 532 retry: 533 error = user_path_at(dfd, filename, lookup_flags, &path); 534 if (!error) { 535 error = chmod_common(&path, mode); 536 path_put(&path); 537 if (retry_estale(error, lookup_flags)) { 538 lookup_flags |= LOOKUP_REVAL; 539 goto retry; 540 } 541 } 542 return error; 543 } 544 545 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode) 546 { 547 return sys_fchmodat(AT_FDCWD, filename, mode); 548 } 549 550 static int chown_common(struct path *path, uid_t user, gid_t group) 551 { 552 struct inode *inode = path->dentry->d_inode; 553 struct inode *delegated_inode = NULL; 554 int error; 555 struct iattr newattrs; 556 kuid_t uid; 557 kgid_t gid; 558 559 uid = make_kuid(current_user_ns(), user); 560 gid = make_kgid(current_user_ns(), group); 561 562 newattrs.ia_valid = ATTR_CTIME; 563 if (user != (uid_t) -1) { 564 if (!uid_valid(uid)) 565 return -EINVAL; 566 newattrs.ia_valid |= ATTR_UID; 567 newattrs.ia_uid = uid; 568 } 569 if (group != (gid_t) -1) { 570 if (!gid_valid(gid)) 571 return -EINVAL; 572 newattrs.ia_valid |= ATTR_GID; 573 newattrs.ia_gid = gid; 574 } 575 if (!S_ISDIR(inode->i_mode)) 576 newattrs.ia_valid |= 577 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV; 578 retry_deleg: 579 mutex_lock(&inode->i_mutex); 580 error = security_path_chown(path, uid, gid); 581 if (!error) 582 error = notify_change(path->dentry, &newattrs, &delegated_inode); 583 mutex_unlock(&inode->i_mutex); 584 if (delegated_inode) { 585 error = break_deleg_wait(&delegated_inode); 586 if (!error) 587 goto retry_deleg; 588 } 589 return error; 590 } 591 592 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user, 593 gid_t, group, int, flag) 594 { 595 struct path path; 596 int error = -EINVAL; 597 int lookup_flags; 598 599 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0) 600 goto out; 601 602 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW; 603 if (flag & AT_EMPTY_PATH) 604 lookup_flags |= LOOKUP_EMPTY; 605 retry: 606 error = user_path_at(dfd, filename, lookup_flags, &path); 607 if (error) 608 goto out; 609 error = mnt_want_write(path.mnt); 610 if (error) 611 goto out_release; 612 error = chown_common(&path, user, group); 613 mnt_drop_write(path.mnt); 614 out_release: 615 path_put(&path); 616 if (retry_estale(error, lookup_flags)) { 617 lookup_flags |= LOOKUP_REVAL; 618 goto retry; 619 } 620 out: 621 return error; 622 } 623 624 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group) 625 { 626 return sys_fchownat(AT_FDCWD, filename, user, group, 0); 627 } 628 629 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group) 630 { 631 return sys_fchownat(AT_FDCWD, filename, user, group, 632 AT_SYMLINK_NOFOLLOW); 633 } 634 635 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group) 636 { 637 struct fd f = fdget(fd); 638 int error = -EBADF; 639 640 if (!f.file) 641 goto out; 642 643 error = mnt_want_write_file(f.file); 644 if (error) 645 goto out_fput; 646 audit_inode(NULL, f.file->f_path.dentry, 0); 647 error = chown_common(&f.file->f_path, user, group); 648 mnt_drop_write_file(f.file); 649 out_fput: 650 fdput(f); 651 out: 652 return error; 653 } 654 655 int open_check_o_direct(struct file *f) 656 { 657 /* NB: we're sure to have correct a_ops only after f_op->open */ 658 if (f->f_flags & O_DIRECT) { 659 if (!f->f_mapping->a_ops || 660 ((!f->f_mapping->a_ops->direct_IO) && 661 (!f->f_mapping->a_ops->get_xip_mem))) { 662 return -EINVAL; 663 } 664 } 665 return 0; 666 } 667 668 static int do_dentry_open(struct file *f, 669 int (*open)(struct inode *, struct file *), 670 const struct cred *cred) 671 { 672 static const struct file_operations empty_fops = {}; 673 struct inode *inode; 674 int error; 675 676 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK | 677 FMODE_PREAD | FMODE_PWRITE; 678 679 path_get(&f->f_path); 680 inode = f->f_inode = f->f_path.dentry->d_inode; 681 f->f_mapping = inode->i_mapping; 682 683 if (unlikely(f->f_flags & O_PATH)) { 684 f->f_mode = FMODE_PATH; 685 f->f_op = &empty_fops; 686 return 0; 687 } 688 689 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) { 690 error = get_write_access(inode); 691 if (unlikely(error)) 692 goto cleanup_file; 693 error = __mnt_want_write(f->f_path.mnt); 694 if (unlikely(error)) { 695 put_write_access(inode); 696 goto cleanup_file; 697 } 698 f->f_mode |= FMODE_WRITER; 699 } 700 701 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */ 702 if (S_ISREG(inode->i_mode)) 703 f->f_mode |= FMODE_ATOMIC_POS; 704 705 f->f_op = fops_get(inode->i_fop); 706 if (unlikely(WARN_ON(!f->f_op))) { 707 error = -ENODEV; 708 goto cleanup_all; 709 } 710 711 error = security_file_open(f, cred); 712 if (error) 713 goto cleanup_all; 714 715 error = break_lease(inode, f->f_flags); 716 if (error) 717 goto cleanup_all; 718 719 if (!open) 720 open = f->f_op->open; 721 if (open) { 722 error = open(inode, f); 723 if (error) 724 goto cleanup_all; 725 } 726 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) 727 i_readcount_inc(inode); 728 if ((f->f_mode & FMODE_READ) && 729 likely(f->f_op->read || f->f_op->aio_read || f->f_op->read_iter)) 730 f->f_mode |= FMODE_CAN_READ; 731 if ((f->f_mode & FMODE_WRITE) && 732 likely(f->f_op->write || f->f_op->aio_write || f->f_op->write_iter)) 733 f->f_mode |= FMODE_CAN_WRITE; 734 735 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC); 736 737 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping); 738 739 return 0; 740 741 cleanup_all: 742 fops_put(f->f_op); 743 if (f->f_mode & FMODE_WRITER) { 744 put_write_access(inode); 745 __mnt_drop_write(f->f_path.mnt); 746 } 747 cleanup_file: 748 path_put(&f->f_path); 749 f->f_path.mnt = NULL; 750 f->f_path.dentry = NULL; 751 f->f_inode = NULL; 752 return error; 753 } 754 755 /** 756 * finish_open - finish opening a file 757 * @file: file pointer 758 * @dentry: pointer to dentry 759 * @open: open callback 760 * @opened: state of open 761 * 762 * This can be used to finish opening a file passed to i_op->atomic_open(). 763 * 764 * If the open callback is set to NULL, then the standard f_op->open() 765 * filesystem callback is substituted. 766 * 767 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is 768 * the return value of d_splice_alias(), then the caller needs to perform dput() 769 * on it after finish_open(). 770 * 771 * On successful return @file is a fully instantiated open file. After this, if 772 * an error occurs in ->atomic_open(), it needs to clean up with fput(). 773 * 774 * Returns zero on success or -errno if the open failed. 775 */ 776 int finish_open(struct file *file, struct dentry *dentry, 777 int (*open)(struct inode *, struct file *), 778 int *opened) 779 { 780 int error; 781 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 782 783 file->f_path.dentry = dentry; 784 error = do_dentry_open(file, open, current_cred()); 785 if (!error) 786 *opened |= FILE_OPENED; 787 788 return error; 789 } 790 EXPORT_SYMBOL(finish_open); 791 792 /** 793 * finish_no_open - finish ->atomic_open() without opening the file 794 * 795 * @file: file pointer 796 * @dentry: dentry or NULL (as returned from ->lookup()) 797 * 798 * This can be used to set the result of a successful lookup in ->atomic_open(). 799 * 800 * NB: unlike finish_open() this function does consume the dentry reference and 801 * the caller need not dput() it. 802 * 803 * Returns "1" which must be the return value of ->atomic_open() after having 804 * called this function. 805 */ 806 int finish_no_open(struct file *file, struct dentry *dentry) 807 { 808 file->f_path.dentry = dentry; 809 return 1; 810 } 811 EXPORT_SYMBOL(finish_no_open); 812 813 struct file *dentry_open(const struct path *path, int flags, 814 const struct cred *cred) 815 { 816 int error; 817 struct file *f; 818 819 validate_creds(cred); 820 821 /* We must always pass in a valid mount pointer. */ 822 BUG_ON(!path->mnt); 823 824 f = get_empty_filp(); 825 if (!IS_ERR(f)) { 826 f->f_flags = flags; 827 f->f_path = *path; 828 error = do_dentry_open(f, NULL, cred); 829 if (!error) { 830 /* from now on we need fput() to dispose of f */ 831 error = open_check_o_direct(f); 832 if (error) { 833 fput(f); 834 f = ERR_PTR(error); 835 } 836 } else { 837 put_filp(f); 838 f = ERR_PTR(error); 839 } 840 } 841 return f; 842 } 843 EXPORT_SYMBOL(dentry_open); 844 845 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op) 846 { 847 int lookup_flags = 0; 848 int acc_mode; 849 850 if (flags & (O_CREAT | __O_TMPFILE)) 851 op->mode = (mode & S_IALLUGO) | S_IFREG; 852 else 853 op->mode = 0; 854 855 /* Must never be set by userspace */ 856 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC; 857 858 /* 859 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only 860 * check for O_DSYNC if the need any syncing at all we enforce it's 861 * always set instead of having to deal with possibly weird behaviour 862 * for malicious applications setting only __O_SYNC. 863 */ 864 if (flags & __O_SYNC) 865 flags |= O_DSYNC; 866 867 if (flags & __O_TMPFILE) { 868 if ((flags & O_TMPFILE_MASK) != O_TMPFILE) 869 return -EINVAL; 870 acc_mode = MAY_OPEN | ACC_MODE(flags); 871 if (!(acc_mode & MAY_WRITE)) 872 return -EINVAL; 873 } else if (flags & O_PATH) { 874 /* 875 * If we have O_PATH in the open flag. Then we 876 * cannot have anything other than the below set of flags 877 */ 878 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH; 879 acc_mode = 0; 880 } else { 881 acc_mode = MAY_OPEN | ACC_MODE(flags); 882 } 883 884 op->open_flag = flags; 885 886 /* O_TRUNC implies we need access checks for write permissions */ 887 if (flags & O_TRUNC) 888 acc_mode |= MAY_WRITE; 889 890 /* Allow the LSM permission hook to distinguish append 891 access from general write access. */ 892 if (flags & O_APPEND) 893 acc_mode |= MAY_APPEND; 894 895 op->acc_mode = acc_mode; 896 897 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN; 898 899 if (flags & O_CREAT) { 900 op->intent |= LOOKUP_CREATE; 901 if (flags & O_EXCL) 902 op->intent |= LOOKUP_EXCL; 903 } 904 905 if (flags & O_DIRECTORY) 906 lookup_flags |= LOOKUP_DIRECTORY; 907 if (!(flags & O_NOFOLLOW)) 908 lookup_flags |= LOOKUP_FOLLOW; 909 op->lookup_flags = lookup_flags; 910 return 0; 911 } 912 913 /** 914 * file_open_name - open file and return file pointer 915 * 916 * @name: struct filename containing path to open 917 * @flags: open flags as per the open(2) second argument 918 * @mode: mode for the new file if O_CREAT is set, else ignored 919 * 920 * This is the helper to open a file from kernelspace if you really 921 * have to. But in generally you should not do this, so please move 922 * along, nothing to see here.. 923 */ 924 struct file *file_open_name(struct filename *name, int flags, umode_t mode) 925 { 926 struct open_flags op; 927 int err = build_open_flags(flags, mode, &op); 928 return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op); 929 } 930 931 /** 932 * filp_open - open file and return file pointer 933 * 934 * @filename: path to open 935 * @flags: open flags as per the open(2) second argument 936 * @mode: mode for the new file if O_CREAT is set, else ignored 937 * 938 * This is the helper to open a file from kernelspace if you really 939 * have to. But in generally you should not do this, so please move 940 * along, nothing to see here.. 941 */ 942 struct file *filp_open(const char *filename, int flags, umode_t mode) 943 { 944 struct filename name = {.name = filename}; 945 return file_open_name(&name, flags, mode); 946 } 947 EXPORT_SYMBOL(filp_open); 948 949 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt, 950 const char *filename, int flags) 951 { 952 struct open_flags op; 953 int err = build_open_flags(flags, 0, &op); 954 if (err) 955 return ERR_PTR(err); 956 if (flags & O_CREAT) 957 return ERR_PTR(-EINVAL); 958 if (!filename && (flags & O_DIRECTORY)) 959 if (!dentry->d_inode->i_op->lookup) 960 return ERR_PTR(-ENOTDIR); 961 return do_file_open_root(dentry, mnt, filename, &op); 962 } 963 EXPORT_SYMBOL(file_open_root); 964 965 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode) 966 { 967 struct open_flags op; 968 int fd = build_open_flags(flags, mode, &op); 969 struct filename *tmp; 970 971 if (fd) 972 return fd; 973 974 tmp = getname(filename); 975 if (IS_ERR(tmp)) 976 return PTR_ERR(tmp); 977 978 fd = get_unused_fd_flags(flags); 979 if (fd >= 0) { 980 struct file *f = do_filp_open(dfd, tmp, &op); 981 if (IS_ERR(f)) { 982 put_unused_fd(fd); 983 fd = PTR_ERR(f); 984 } else { 985 fsnotify_open(f); 986 fd_install(fd, f); 987 } 988 } 989 putname(tmp); 990 return fd; 991 } 992 993 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode) 994 { 995 if (force_o_largefile()) 996 flags |= O_LARGEFILE; 997 998 return do_sys_open(AT_FDCWD, filename, flags, mode); 999 } 1000 1001 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, 1002 umode_t, mode) 1003 { 1004 if (force_o_largefile()) 1005 flags |= O_LARGEFILE; 1006 1007 return do_sys_open(dfd, filename, flags, mode); 1008 } 1009 1010 #ifndef __alpha__ 1011 1012 /* 1013 * For backward compatibility? Maybe this should be moved 1014 * into arch/i386 instead? 1015 */ 1016 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode) 1017 { 1018 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode); 1019 } 1020 1021 #endif 1022 1023 /* 1024 * "id" is the POSIX thread ID. We use the 1025 * files pointer for this.. 1026 */ 1027 int filp_close(struct file *filp, fl_owner_t id) 1028 { 1029 int retval = 0; 1030 1031 if (!file_count(filp)) { 1032 printk(KERN_ERR "VFS: Close: file count is 0\n"); 1033 return 0; 1034 } 1035 1036 if (filp->f_op->flush) 1037 retval = filp->f_op->flush(filp, id); 1038 1039 if (likely(!(filp->f_mode & FMODE_PATH))) { 1040 dnotify_flush(filp, id); 1041 locks_remove_posix(filp, id); 1042 } 1043 fput(filp); 1044 return retval; 1045 } 1046 1047 EXPORT_SYMBOL(filp_close); 1048 1049 /* 1050 * Careful here! We test whether the file pointer is NULL before 1051 * releasing the fd. This ensures that one clone task can't release 1052 * an fd while another clone is opening it. 1053 */ 1054 SYSCALL_DEFINE1(close, unsigned int, fd) 1055 { 1056 int retval = __close_fd(current->files, fd); 1057 1058 /* can't restart close syscall because file table entry was cleared */ 1059 if (unlikely(retval == -ERESTARTSYS || 1060 retval == -ERESTARTNOINTR || 1061 retval == -ERESTARTNOHAND || 1062 retval == -ERESTART_RESTARTBLOCK)) 1063 retval = -EINTR; 1064 1065 return retval; 1066 } 1067 EXPORT_SYMBOL(sys_close); 1068 1069 /* 1070 * This routine simulates a hangup on the tty, to arrange that users 1071 * are given clean terminals at login time. 1072 */ 1073 SYSCALL_DEFINE0(vhangup) 1074 { 1075 if (capable(CAP_SYS_TTY_CONFIG)) { 1076 tty_vhangup_self(); 1077 return 0; 1078 } 1079 return -EPERM; 1080 } 1081 1082 /* 1083 * Called when an inode is about to be open. 1084 * We use this to disallow opening large files on 32bit systems if 1085 * the caller didn't specify O_LARGEFILE. On 64bit systems we force 1086 * on this flag in sys_open. 1087 */ 1088 int generic_file_open(struct inode * inode, struct file * filp) 1089 { 1090 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS) 1091 return -EOVERFLOW; 1092 return 0; 1093 } 1094 1095 EXPORT_SYMBOL(generic_file_open); 1096 1097 /* 1098 * This is used by subsystems that don't want seekable 1099 * file descriptors. The function is not supposed to ever fail, the only 1100 * reason it returns an 'int' and not 'void' is so that it can be plugged 1101 * directly into file_operations structure. 1102 */ 1103 int nonseekable_open(struct inode *inode, struct file *filp) 1104 { 1105 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE); 1106 return 0; 1107 } 1108 1109 EXPORT_SYMBOL(nonseekable_open); 1110