1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/string.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>
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>
38 int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
42 struct iattr newattrs;
44 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
48 newattrs.ia_size = length;
49 newattrs.ia_valid = ATTR_SIZE | time_attrs;
51 newattrs.ia_file = filp;
52 newattrs.ia_valid |= ATTR_FILE;
55 /* Remove suid, sgid, and file capabilities on truncate too */
56 ret = dentry_needs_remove_privs(dentry);
60 newattrs.ia_valid |= ret | ATTR_FORCE;
62 inode_lock(dentry->d_inode);
63 /* Note any delegations or leases have already been broken: */
64 ret = notify_change(dentry, &newattrs, NULL);
65 inode_unlock(dentry->d_inode);
69 long vfs_truncate(const struct path *path, loff_t length)
74 inode = path->dentry->d_inode;
76 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
77 if (S_ISDIR(inode->i_mode))
79 if (!S_ISREG(inode->i_mode))
82 error = mnt_want_write(path->mnt);
86 error = inode_permission(inode, MAY_WRITE);
88 goto mnt_drop_write_and_out;
92 goto mnt_drop_write_and_out;
94 error = get_write_access(inode);
96 goto mnt_drop_write_and_out;
99 * Make sure that there are no leases. get_write_access() protects
100 * against the truncate racing with a lease-granting setlease().
102 error = break_lease(inode, O_WRONLY);
104 goto put_write_and_out;
106 error = locks_verify_truncate(inode, NULL, length);
108 error = security_path_truncate(path);
110 error = do_truncate(path->dentry, length, 0, NULL);
113 put_write_access(inode);
114 mnt_drop_write_and_out:
115 mnt_drop_write(path->mnt);
119 EXPORT_SYMBOL_GPL(vfs_truncate);
121 long do_sys_truncate(const char __user *pathname, loff_t length)
123 unsigned int lookup_flags = LOOKUP_FOLLOW;
127 if (length < 0) /* sorry, but loff_t says... */
131 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
133 error = vfs_truncate(&path, length);
136 if (retry_estale(error, lookup_flags)) {
137 lookup_flags |= LOOKUP_REVAL;
143 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
145 return do_sys_truncate(path, length);
149 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
151 return do_sys_truncate(path, length);
155 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
158 struct dentry *dentry;
170 /* explicitly opened as large or we are on 64-bit box */
171 if (f.file->f_flags & O_LARGEFILE)
174 dentry = f.file->f_path.dentry;
175 inode = dentry->d_inode;
177 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
181 /* Cannot ftruncate over 2^31 bytes without large file support */
182 if (small && length > MAX_NON_LFS)
186 /* Check IS_APPEND on real upper inode */
187 if (IS_APPEND(file_inode(f.file)))
190 sb_start_write(inode->i_sb);
191 error = locks_verify_truncate(inode, f.file, length);
193 error = security_path_truncate(&f.file->f_path);
195 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
196 sb_end_write(inode->i_sb);
203 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
205 return do_sys_ftruncate(fd, length, 1);
209 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
211 return do_sys_ftruncate(fd, length, 1);
215 /* LFS versions of truncate are only needed on 32 bit machines */
216 #if BITS_PER_LONG == 32
217 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
219 return do_sys_truncate(path, length);
222 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
224 return do_sys_ftruncate(fd, length, 0);
226 #endif /* BITS_PER_LONG == 32 */
229 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
231 struct inode *inode = file_inode(file);
234 if (offset < 0 || len <= 0)
237 /* Return error if mode is not supported */
238 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
241 /* Punch hole and zero range are mutually exclusive */
242 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
243 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
246 /* Punch hole must have keep size set */
247 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
248 !(mode & FALLOC_FL_KEEP_SIZE))
251 /* Collapse range should only be used exclusively. */
252 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
253 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
256 /* Insert range should only be used exclusively. */
257 if ((mode & FALLOC_FL_INSERT_RANGE) &&
258 (mode & ~FALLOC_FL_INSERT_RANGE))
261 /* Unshare range should only be used with allocate mode. */
262 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
263 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
266 if (!(file->f_mode & FMODE_WRITE))
270 * We can only allow pure fallocate on append only files
272 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
275 if (IS_IMMUTABLE(inode))
279 * We cannot allow any fallocate operation on an active swapfile
281 if (IS_SWAPFILE(inode))
285 * Revalidate the write permissions, in case security policy has
286 * changed since the files were opened.
288 ret = security_file_permission(file, MAY_WRITE);
292 if (S_ISFIFO(inode->i_mode))
295 if (S_ISDIR(inode->i_mode))
298 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
301 /* Check for wrap through zero too */
302 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
305 if (!file->f_op->fallocate)
308 file_start_write(file);
309 ret = file->f_op->fallocate(file, mode, offset, len);
312 * Create inotify and fanotify events.
314 * To keep the logic simple always create events if fallocate succeeds.
315 * This implies that events are even created if the file size remains
316 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
319 fsnotify_modify(file);
321 file_end_write(file);
324 EXPORT_SYMBOL_GPL(vfs_fallocate);
326 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
328 struct fd f = fdget(fd);
332 error = vfs_fallocate(f.file, mode, offset, len);
338 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
340 return ksys_fallocate(fd, mode, offset, len);
344 * access() needs to use the real uid/gid, not the effective uid/gid.
345 * We do this by temporarily clearing all FS-related capabilities and
346 * switching the fsuid/fsgid around to the real ones.
348 static const struct cred *access_override_creds(void)
350 const struct cred *old_cred;
351 struct cred *override_cred;
353 override_cred = prepare_creds();
357 override_cred->fsuid = override_cred->uid;
358 override_cred->fsgid = override_cred->gid;
360 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
361 /* Clear the capabilities if we switch to a non-root user */
362 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
363 if (!uid_eq(override_cred->uid, root_uid))
364 cap_clear(override_cred->cap_effective);
366 override_cred->cap_effective =
367 override_cred->cap_permitted;
371 * The new set of credentials can *only* be used in
372 * task-synchronous circumstances, and does not need
373 * RCU freeing, unless somebody then takes a separate
376 * NOTE! This is _only_ true because this credential
377 * is used purely for override_creds() that installs
378 * it as the subjective cred. Other threads will be
379 * accessing ->real_cred, not the subjective cred.
381 * If somebody _does_ make a copy of this (using the
382 * 'get_current_cred()' function), that will clear the
383 * non_rcu field, because now that other user may be
384 * expecting RCU freeing. But normal thread-synchronous
385 * cred accesses will keep things non-RCY.
387 override_cred->non_rcu = 1;
389 old_cred = override_creds(override_cred);
391 /* override_cred() gets its own ref */
392 put_cred(override_cred);
397 long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
402 unsigned int lookup_flags = LOOKUP_FOLLOW;
403 const struct cred *old_cred = NULL;
405 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
408 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
411 if (flags & AT_SYMLINK_NOFOLLOW)
412 lookup_flags &= ~LOOKUP_FOLLOW;
413 if (flags & AT_EMPTY_PATH)
414 lookup_flags |= LOOKUP_EMPTY;
416 if (!(flags & AT_EACCESS)) {
417 old_cred = access_override_creds();
423 res = user_path_at(dfd, filename, lookup_flags, &path);
427 inode = d_backing_inode(path.dentry);
429 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
431 * MAY_EXEC on regular files is denied if the fs is mounted
432 * with the "noexec" flag.
435 if (path_noexec(&path))
436 goto out_path_release;
439 res = inode_permission(inode, mode | MAY_ACCESS);
440 /* SuS v2 requires we report a read only fs too */
441 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
442 goto out_path_release;
444 * This is a rare case where using __mnt_is_readonly()
445 * is OK without a mnt_want/drop_write() pair. Since
446 * no actual write to the fs is performed here, we do
447 * not need to telegraph to that to anyone.
449 * By doing this, we accept that this access is
450 * inherently racy and know that the fs may change
451 * state before we even see this result.
453 if (__mnt_is_readonly(path.mnt))
458 if (retry_estale(res, lookup_flags)) {
459 lookup_flags |= LOOKUP_REVAL;
464 revert_creds(old_cred);
469 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
471 return do_faccessat(dfd, filename, mode, 0);
474 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
477 return do_faccessat(dfd, filename, mode, flags);
480 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
482 return do_faccessat(AT_FDCWD, filename, mode, 0);
485 SYSCALL_DEFINE1(chdir, const char __user *, filename)
489 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
491 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
495 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
499 set_fs_pwd(current->fs, &path);
503 if (retry_estale(error, lookup_flags)) {
504 lookup_flags |= LOOKUP_REVAL;
511 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
513 struct fd f = fdget_raw(fd);
521 if (!d_can_lookup(f.file->f_path.dentry))
524 error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
526 set_fs_pwd(current->fs, &f.file->f_path);
533 SYSCALL_DEFINE1(chroot, const char __user *, filename)
537 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
539 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
543 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
548 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
550 error = security_path_chroot(&path);
554 set_fs_root(current->fs, &path);
558 if (retry_estale(error, lookup_flags)) {
559 lookup_flags |= LOOKUP_REVAL;
566 static int chmod_common(const struct path *path, umode_t mode)
568 struct inode *inode = path->dentry->d_inode;
569 struct inode *delegated_inode = NULL;
570 struct iattr newattrs;
573 error = mnt_want_write(path->mnt);
578 error = security_path_chmod(path, mode);
581 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
582 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
583 error = notify_change(path->dentry, &newattrs, &delegated_inode);
586 if (delegated_inode) {
587 error = break_deleg_wait(&delegated_inode);
591 mnt_drop_write(path->mnt);
595 int vfs_fchmod(struct file *file, umode_t mode)
598 return chmod_common(&file->f_path, mode);
601 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
603 struct fd f = fdget(fd);
607 err = vfs_fchmod(f.file, mode);
613 int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
617 unsigned int lookup_flags = LOOKUP_FOLLOW;
619 error = user_path_at(dfd, filename, lookup_flags, &path);
621 error = chmod_common(&path, mode);
623 if (retry_estale(error, lookup_flags)) {
624 lookup_flags |= LOOKUP_REVAL;
631 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
634 return do_fchmodat(dfd, filename, mode);
637 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
639 return do_fchmodat(AT_FDCWD, filename, mode);
642 int chown_common(const struct path *path, uid_t user, gid_t group)
644 struct inode *inode = path->dentry->d_inode;
645 struct inode *delegated_inode = NULL;
647 struct iattr newattrs;
651 uid = make_kuid(current_user_ns(), user);
652 gid = make_kgid(current_user_ns(), group);
655 newattrs.ia_valid = ATTR_CTIME;
656 if (user != (uid_t) -1) {
659 newattrs.ia_valid |= ATTR_UID;
660 newattrs.ia_uid = uid;
662 if (group != (gid_t) -1) {
665 newattrs.ia_valid |= ATTR_GID;
666 newattrs.ia_gid = gid;
668 if (!S_ISDIR(inode->i_mode))
670 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
672 error = security_path_chown(path, uid, gid);
674 error = notify_change(path->dentry, &newattrs, &delegated_inode);
676 if (delegated_inode) {
677 error = break_deleg_wait(&delegated_inode);
684 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
691 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
694 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
695 if (flag & AT_EMPTY_PATH)
696 lookup_flags |= LOOKUP_EMPTY;
698 error = user_path_at(dfd, filename, lookup_flags, &path);
701 error = mnt_want_write(path.mnt);
704 error = chown_common(&path, user, group);
705 mnt_drop_write(path.mnt);
708 if (retry_estale(error, lookup_flags)) {
709 lookup_flags |= LOOKUP_REVAL;
716 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
717 gid_t, group, int, flag)
719 return do_fchownat(dfd, filename, user, group, flag);
722 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
724 return do_fchownat(AT_FDCWD, filename, user, group, 0);
727 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
729 return do_fchownat(AT_FDCWD, filename, user, group,
730 AT_SYMLINK_NOFOLLOW);
733 int vfs_fchown(struct file *file, uid_t user, gid_t group)
737 error = mnt_want_write_file(file);
741 error = chown_common(&file->f_path, user, group);
742 mnt_drop_write_file(file);
746 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
748 struct fd f = fdget(fd);
752 error = vfs_fchown(f.file, user, group);
758 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
760 return ksys_fchown(fd, user, group);
763 static int do_dentry_open(struct file *f,
765 int (*open)(struct inode *, struct file *))
767 static const struct file_operations empty_fops = {};
770 path_get(&f->f_path);
772 f->f_mapping = inode->i_mapping;
773 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
774 f->f_sb_err = file_sample_sb_err(f);
776 if (unlikely(f->f_flags & O_PATH)) {
777 f->f_mode = FMODE_PATH | FMODE_OPENED;
778 f->f_op = &empty_fops;
782 /* Any file opened for execve()/uselib() has to be a regular file. */
783 if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
788 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
789 error = get_write_access(inode);
792 error = __mnt_want_write(f->f_path.mnt);
793 if (unlikely(error)) {
794 put_write_access(inode);
797 f->f_mode |= FMODE_WRITER;
800 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
801 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
802 f->f_mode |= FMODE_ATOMIC_POS;
804 f->f_op = fops_get(inode->i_fop);
805 if (WARN_ON(!f->f_op)) {
810 error = security_file_open(f);
814 error = break_lease(locks_inode(f), f->f_flags);
818 /* normally all 3 are set; ->open() can clear them if needed */
819 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
821 open = f->f_op->open;
823 error = open(inode, f);
827 f->f_mode |= FMODE_OPENED;
828 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
829 i_readcount_inc(inode);
830 if ((f->f_mode & FMODE_READ) &&
831 likely(f->f_op->read || f->f_op->read_iter))
832 f->f_mode |= FMODE_CAN_READ;
833 if ((f->f_mode & FMODE_WRITE) &&
834 likely(f->f_op->write || f->f_op->write_iter))
835 f->f_mode |= FMODE_CAN_WRITE;
837 f->f_write_hint = WRITE_LIFE_NOT_SET;
838 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
840 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
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)
849 * XXX: Huge page cache doesn't support writing yet. Drop all page
850 * cache for this file before processing writes.
852 if ((f->f_mode & FMODE_WRITE) && filemap_nr_thps(inode->i_mapping))
853 truncate_pagecache(inode, 0);
858 if (WARN_ON_ONCE(error > 0))
861 if (f->f_mode & FMODE_WRITER) {
862 put_write_access(inode);
863 __mnt_drop_write(f->f_path.mnt);
866 path_put(&f->f_path);
867 f->f_path.mnt = NULL;
868 f->f_path.dentry = NULL;
874 * finish_open - finish opening a file
875 * @file: file pointer
876 * @dentry: pointer to dentry
877 * @open: open callback
878 * @opened: state of open
880 * This can be used to finish opening a file passed to i_op->atomic_open().
882 * If the open callback is set to NULL, then the standard f_op->open()
883 * filesystem callback is substituted.
885 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
886 * the return value of d_splice_alias(), then the caller needs to perform dput()
887 * on it after finish_open().
889 * Returns zero on success or -errno if the open failed.
891 int finish_open(struct file *file, struct dentry *dentry,
892 int (*open)(struct inode *, struct file *))
894 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
896 file->f_path.dentry = dentry;
897 return do_dentry_open(file, d_backing_inode(dentry), open);
899 EXPORT_SYMBOL(finish_open);
902 * finish_no_open - finish ->atomic_open() without opening the file
904 * @file: file pointer
905 * @dentry: dentry or NULL (as returned from ->lookup())
907 * This can be used to set the result of a successful lookup in ->atomic_open().
909 * NB: unlike finish_open() this function does consume the dentry reference and
910 * the caller need not dput() it.
912 * Returns "0" which must be the return value of ->atomic_open() after having
913 * called this function.
915 int finish_no_open(struct file *file, struct dentry *dentry)
917 file->f_path.dentry = dentry;
920 EXPORT_SYMBOL(finish_no_open);
922 char *file_path(struct file *filp, char *buf, int buflen)
924 return d_path(&filp->f_path, buf, buflen);
926 EXPORT_SYMBOL(file_path);
929 * vfs_open - open the file at the given path
930 * @path: path to open
931 * @file: newly allocated file with f_flag initialized
932 * @cred: credentials to use
934 int vfs_open(const struct path *path, struct file *file)
936 file->f_path = *path;
937 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
940 struct file *dentry_open(const struct path *path, int flags,
941 const struct cred *cred)
946 validate_creds(cred);
948 /* We must always pass in a valid mount pointer. */
951 f = alloc_empty_file(flags, cred);
953 error = vfs_open(path, f);
961 EXPORT_SYMBOL(dentry_open);
963 struct file *open_with_fake_path(const struct path *path, int flags,
964 struct inode *inode, const struct cred *cred)
966 struct file *f = alloc_empty_file_noaccount(flags, cred);
971 error = do_dentry_open(f, inode, NULL);
979 EXPORT_SYMBOL(open_with_fake_path);
981 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
982 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
984 inline struct open_how build_open_how(int flags, umode_t mode)
986 struct open_how how = {
987 .flags = flags & VALID_OPEN_FLAGS,
988 .mode = mode & S_IALLUGO,
991 /* O_PATH beats everything else. */
992 if (how.flags & O_PATH)
993 how.flags &= O_PATH_FLAGS;
994 /* Modes should only be set for create-like flags. */
995 if (!WILL_CREATE(how.flags))
1000 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1002 int flags = how->flags;
1003 int lookup_flags = 0;
1004 int acc_mode = ACC_MODE(flags);
1006 /* Must never be set by userspace */
1007 flags &= ~(FMODE_NONOTIFY | O_CLOEXEC);
1010 * Older syscalls implicitly clear all of the invalid flags or argument
1011 * values before calling build_open_flags(), but openat2(2) checks all
1014 if (flags & ~VALID_OPEN_FLAGS)
1016 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1019 /* Deal with the mode. */
1020 if (WILL_CREATE(flags)) {
1021 if (how->mode & ~S_IALLUGO)
1023 op->mode = how->mode | S_IFREG;
1031 * In order to ensure programs get explicit errors when trying to use
1032 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1033 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1034 * have to require userspace to explicitly set it.
1036 if (flags & __O_TMPFILE) {
1037 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1039 if (!(acc_mode & MAY_WRITE))
1042 if (flags & O_PATH) {
1043 /* O_PATH only permits certain other flags to be set. */
1044 if (flags & ~O_PATH_FLAGS)
1050 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1051 * check for O_DSYNC if the need any syncing at all we enforce it's
1052 * always set instead of having to deal with possibly weird behaviour
1053 * for malicious applications setting only __O_SYNC.
1055 if (flags & __O_SYNC)
1058 op->open_flag = flags;
1060 /* O_TRUNC implies we need access checks for write permissions */
1061 if (flags & O_TRUNC)
1062 acc_mode |= MAY_WRITE;
1064 /* Allow the LSM permission hook to distinguish append
1065 access from general write access. */
1066 if (flags & O_APPEND)
1067 acc_mode |= MAY_APPEND;
1069 op->acc_mode = acc_mode;
1071 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1073 if (flags & O_CREAT) {
1074 op->intent |= LOOKUP_CREATE;
1075 if (flags & O_EXCL) {
1076 op->intent |= LOOKUP_EXCL;
1077 flags |= O_NOFOLLOW;
1081 if (flags & O_DIRECTORY)
1082 lookup_flags |= LOOKUP_DIRECTORY;
1083 if (!(flags & O_NOFOLLOW))
1084 lookup_flags |= LOOKUP_FOLLOW;
1086 if (how->resolve & RESOLVE_NO_XDEV)
1087 lookup_flags |= LOOKUP_NO_XDEV;
1088 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1089 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1090 if (how->resolve & RESOLVE_NO_SYMLINKS)
1091 lookup_flags |= LOOKUP_NO_SYMLINKS;
1092 if (how->resolve & RESOLVE_BENEATH)
1093 lookup_flags |= LOOKUP_BENEATH;
1094 if (how->resolve & RESOLVE_IN_ROOT)
1095 lookup_flags |= LOOKUP_IN_ROOT;
1097 op->lookup_flags = lookup_flags;
1102 * file_open_name - open file and return file pointer
1104 * @name: struct filename containing path to open
1105 * @flags: open flags as per the open(2) second argument
1106 * @mode: mode for the new file if O_CREAT is set, else ignored
1108 * This is the helper to open a file from kernelspace if you really
1109 * have to. But in generally you should not do this, so please move
1110 * along, nothing to see here..
1112 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1114 struct open_flags op;
1115 struct open_how how = build_open_how(flags, mode);
1116 int err = build_open_flags(&how, &op);
1118 return ERR_PTR(err);
1119 return do_filp_open(AT_FDCWD, name, &op);
1123 * filp_open - open file and return file pointer
1125 * @filename: path to open
1126 * @flags: open flags as per the open(2) second argument
1127 * @mode: mode for the new file if O_CREAT is set, else ignored
1129 * This is the helper to open a file from kernelspace if you really
1130 * have to. But in generally you should not do this, so please move
1131 * along, nothing to see here..
1133 struct file *filp_open(const char *filename, int flags, umode_t mode)
1135 struct filename *name = getname_kernel(filename);
1136 struct file *file = ERR_CAST(name);
1138 if (!IS_ERR(name)) {
1139 file = file_open_name(name, flags, mode);
1144 EXPORT_SYMBOL(filp_open);
1146 struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1147 const char *filename, int flags, umode_t mode)
1149 struct open_flags op;
1150 struct open_how how = build_open_how(flags, mode);
1151 int err = build_open_flags(&how, &op);
1153 return ERR_PTR(err);
1154 return do_file_open_root(dentry, mnt, filename, &op);
1156 EXPORT_SYMBOL(file_open_root);
1158 static long do_sys_openat2(int dfd, const char __user *filename,
1159 struct open_how *how)
1161 struct open_flags op;
1162 int fd = build_open_flags(how, &op);
1163 struct filename *tmp;
1168 tmp = getname(filename);
1170 return PTR_ERR(tmp);
1172 fd = get_unused_fd_flags(how->flags);
1174 struct file *f = do_filp_open(dfd, tmp, &op);
1187 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1189 struct open_how how = build_open_how(flags, mode);
1190 return do_sys_openat2(dfd, filename, &how);
1194 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1196 if (force_o_largefile())
1197 flags |= O_LARGEFILE;
1198 return do_sys_open(AT_FDCWD, filename, flags, mode);
1201 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1204 if (force_o_largefile())
1205 flags |= O_LARGEFILE;
1206 return do_sys_open(dfd, filename, flags, mode);
1209 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1210 struct open_how __user *, how, size_t, usize)
1213 struct open_how tmp;
1215 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1216 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1218 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1221 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1225 /* O_LARGEFILE is only allowed for non-O_PATH. */
1226 if (!(tmp.flags & O_PATH) && force_o_largefile())
1227 tmp.flags |= O_LARGEFILE;
1229 return do_sys_openat2(dfd, filename, &tmp);
1232 #ifdef CONFIG_COMPAT
1234 * Exactly like sys_open(), except that it doesn't set the
1237 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1239 return do_sys_open(AT_FDCWD, filename, flags, mode);
1243 * Exactly like sys_openat(), except that it doesn't set the
1246 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1248 return do_sys_open(dfd, filename, flags, mode);
1255 * For backward compatibility? Maybe this should be moved
1256 * into arch/i386 instead?
1258 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1260 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1262 if (force_o_largefile())
1263 flags |= O_LARGEFILE;
1264 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1269 * "id" is the POSIX thread ID. We use the
1270 * files pointer for this..
1272 int filp_close(struct file *filp, fl_owner_t id)
1276 if (!file_count(filp)) {
1277 printk(KERN_ERR "VFS: Close: file count is 0\n");
1281 if (filp->f_op->flush)
1282 retval = filp->f_op->flush(filp, id);
1284 if (likely(!(filp->f_mode & FMODE_PATH))) {
1285 dnotify_flush(filp, id);
1286 locks_remove_posix(filp, id);
1292 EXPORT_SYMBOL(filp_close);
1295 * Careful here! We test whether the file pointer is NULL before
1296 * releasing the fd. This ensures that one clone task can't release
1297 * an fd while another clone is opening it.
1299 SYSCALL_DEFINE1(close, unsigned int, fd)
1301 int retval = __close_fd(current->files, fd);
1303 /* can't restart close syscall because file table entry was cleared */
1304 if (unlikely(retval == -ERESTARTSYS ||
1305 retval == -ERESTARTNOINTR ||
1306 retval == -ERESTARTNOHAND ||
1307 retval == -ERESTART_RESTARTBLOCK))
1314 * This routine simulates a hangup on the tty, to arrange that users
1315 * are given clean terminals at login time.
1317 SYSCALL_DEFINE0(vhangup)
1319 if (capable(CAP_SYS_TTY_CONFIG)) {
1327 * Called when an inode is about to be open.
1328 * We use this to disallow opening large files on 32bit systems if
1329 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1330 * on this flag in sys_open.
1332 int generic_file_open(struct inode * inode, struct file * filp)
1334 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1339 EXPORT_SYMBOL(generic_file_open);
1342 * This is used by subsystems that don't want seekable
1343 * file descriptors. The function is not supposed to ever fail, the only
1344 * reason it returns an 'int' and not 'void' is so that it can be plugged
1345 * directly into file_operations structure.
1347 int nonseekable_open(struct inode *inode, struct file *filp)
1349 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1353 EXPORT_SYMBOL(nonseekable_open);
1356 * stream_open is used by subsystems that want stream-like file descriptors.
1357 * Such file descriptors are not seekable and don't have notion of position
1358 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1359 * Contrary to file descriptors of other regular files, .read() and .write()
1360 * can run simultaneously.
1362 * stream_open never fails and is marked to return int so that it could be
1363 * directly used as file_operations.open .
1365 int stream_open(struct inode *inode, struct file *filp)
1367 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1368 filp->f_mode |= FMODE_STREAM;
1372 EXPORT_SYMBOL(stream_open);