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>
35 #include <linux/mnt_idmapping.h>
39 int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry,
40 loff_t length, unsigned int time_attrs, struct file *filp)
43 struct iattr newattrs;
45 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
49 newattrs.ia_size = length;
50 newattrs.ia_valid = ATTR_SIZE | time_attrs;
52 newattrs.ia_file = filp;
53 newattrs.ia_valid |= ATTR_FILE;
56 /* Remove suid, sgid, and file capabilities on truncate too */
57 ret = dentry_needs_remove_privs(dentry);
61 newattrs.ia_valid |= ret | ATTR_FORCE;
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);
70 long vfs_truncate(const struct path *path, loff_t length)
72 struct user_namespace *mnt_userns;
76 inode = path->dentry->d_inode;
78 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
79 if (S_ISDIR(inode->i_mode))
81 if (!S_ISREG(inode->i_mode))
84 error = mnt_want_write(path->mnt);
88 mnt_userns = mnt_user_ns(path->mnt);
89 error = inode_permission(mnt_userns, inode, MAY_WRITE);
91 goto mnt_drop_write_and_out;
95 goto mnt_drop_write_and_out;
97 error = get_write_access(inode);
99 goto mnt_drop_write_and_out;
102 * Make sure that there are no leases. get_write_access() protects
103 * against the truncate racing with a lease-granting setlease().
105 error = break_lease(inode, O_WRONLY);
107 goto put_write_and_out;
109 error = security_path_truncate(path);
111 error = do_truncate(mnt_userns, path->dentry, length, 0, NULL);
114 put_write_access(inode);
115 mnt_drop_write_and_out:
116 mnt_drop_write(path->mnt);
120 EXPORT_SYMBOL_GPL(vfs_truncate);
122 long do_sys_truncate(const char __user *pathname, loff_t length)
124 unsigned int lookup_flags = LOOKUP_FOLLOW;
128 if (length < 0) /* sorry, but loff_t says... */
132 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
134 error = vfs_truncate(&path, length);
137 if (retry_estale(error, lookup_flags)) {
138 lookup_flags |= LOOKUP_REVAL;
144 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
146 return do_sys_truncate(path, length);
150 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
152 return do_sys_truncate(path, length);
156 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
159 struct dentry *dentry;
171 /* explicitly opened as large or we are on 64-bit box */
172 if (f.file->f_flags & O_LARGEFILE)
175 dentry = f.file->f_path.dentry;
176 inode = dentry->d_inode;
178 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
182 /* Cannot ftruncate over 2^31 bytes without large file support */
183 if (small && length > MAX_NON_LFS)
187 /* Check IS_APPEND on real upper inode */
188 if (IS_APPEND(file_inode(f.file)))
190 sb_start_write(inode->i_sb);
191 error = security_path_truncate(&f.file->f_path);
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);
202 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
204 return do_sys_ftruncate(fd, length, 1);
208 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
210 return do_sys_ftruncate(fd, length, 1);
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)
218 return do_sys_truncate(path, length);
221 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
223 return do_sys_ftruncate(fd, length, 0);
225 #endif /* BITS_PER_LONG == 32 */
228 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
230 struct inode *inode = file_inode(file);
233 if (offset < 0 || len <= 0)
236 /* Return error if mode is not supported */
237 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
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))
245 /* Punch hole must have keep size set */
246 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
247 !(mode & FALLOC_FL_KEEP_SIZE))
250 /* Collapse range should only be used exclusively. */
251 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
252 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
255 /* Insert range should only be used exclusively. */
256 if ((mode & FALLOC_FL_INSERT_RANGE) &&
257 (mode & ~FALLOC_FL_INSERT_RANGE))
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)))
265 if (!(file->f_mode & FMODE_WRITE))
269 * We can only allow pure fallocate on append only files
271 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
274 if (IS_IMMUTABLE(inode))
278 * We cannot allow any fallocate operation on an active swapfile
280 if (IS_SWAPFILE(inode))
284 * Revalidate the write permissions, in case security policy has
285 * changed since the files were opened.
287 ret = security_file_permission(file, MAY_WRITE);
291 if (S_ISFIFO(inode->i_mode))
294 if (S_ISDIR(inode->i_mode))
297 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
300 /* Check for wrap through zero too */
301 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
304 if (!file->f_op->fallocate)
307 file_start_write(file);
308 ret = file->f_op->fallocate(file, mode, offset, len);
311 * Create inotify and fanotify events.
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.
318 fsnotify_modify(file);
320 file_end_write(file);
323 EXPORT_SYMBOL_GPL(vfs_fallocate);
325 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
327 struct fd f = fdget(fd);
331 error = vfs_fallocate(f.file, mode, offset, len);
337 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
339 return ksys_fallocate(fd, mode, offset, len);
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.
347 static const struct cred *access_override_creds(void)
349 const struct cred *old_cred;
350 struct cred *override_cred;
352 override_cred = prepare_creds();
356 override_cred->fsuid = override_cred->uid;
357 override_cred->fsgid = override_cred->gid;
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);
365 override_cred->cap_effective =
366 override_cred->cap_permitted;
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
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.
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.
386 override_cred->non_rcu = 1;
388 old_cred = override_creds(override_cred);
390 /* override_cred() gets its own ref */
391 put_cred(override_cred);
396 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
401 unsigned int lookup_flags = LOOKUP_FOLLOW;
402 const struct cred *old_cred = NULL;
404 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
407 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
410 if (flags & AT_SYMLINK_NOFOLLOW)
411 lookup_flags &= ~LOOKUP_FOLLOW;
412 if (flags & AT_EMPTY_PATH)
413 lookup_flags |= LOOKUP_EMPTY;
415 if (!(flags & AT_EACCESS)) {
416 old_cred = access_override_creds();
422 res = user_path_at(dfd, filename, lookup_flags, &path);
426 inode = d_backing_inode(path.dentry);
428 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
430 * MAY_EXEC on regular files is denied if the fs is mounted
431 * with the "noexec" flag.
434 if (path_noexec(&path))
435 goto out_path_release;
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;
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.
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.
452 if (__mnt_is_readonly(path.mnt))
457 if (retry_estale(res, lookup_flags)) {
458 lookup_flags |= LOOKUP_REVAL;
463 revert_creds(old_cred);
468 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
470 return do_faccessat(dfd, filename, mode, 0);
473 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
476 return do_faccessat(dfd, filename, mode, flags);
479 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
481 return do_faccessat(AT_FDCWD, filename, mode, 0);
484 SYSCALL_DEFINE1(chdir, const char __user *, filename)
488 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
490 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
494 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
498 set_fs_pwd(current->fs, &path);
502 if (retry_estale(error, lookup_flags)) {
503 lookup_flags |= LOOKUP_REVAL;
510 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
512 struct fd f = fdget_raw(fd);
520 if (!d_can_lookup(f.file->f_path.dentry))
523 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
525 set_fs_pwd(current->fs, &f.file->f_path);
532 SYSCALL_DEFINE1(chroot, const char __user *, filename)
536 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
538 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
542 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
547 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
549 error = security_path_chroot(&path);
553 set_fs_root(current->fs, &path);
557 if (retry_estale(error, lookup_flags)) {
558 lookup_flags |= LOOKUP_REVAL;
565 int chmod_common(const struct path *path, umode_t mode)
567 struct inode *inode = path->dentry->d_inode;
568 struct inode *delegated_inode = NULL;
569 struct iattr newattrs;
572 error = mnt_want_write(path->mnt);
577 error = security_path_chmod(path, mode);
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);
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 static 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 user_namespace *mnt_userns, *fs_userns;
645 struct inode *inode = path->dentry->d_inode;
646 struct inode *delegated_inode = NULL;
648 struct iattr newattrs;
652 uid = make_kuid(current_user_ns(), user);
653 gid = make_kgid(current_user_ns(), group);
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);
661 newattrs.ia_valid = ATTR_CTIME;
662 if (user != (uid_t) -1) {
665 newattrs.ia_valid |= ATTR_UID;
666 newattrs.ia_uid = uid;
668 if (group != (gid_t) -1) {
671 newattrs.ia_valid |= ATTR_GID;
672 newattrs.ia_gid = gid;
674 if (!S_ISDIR(inode->i_mode))
676 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
678 error = security_path_chown(path, uid, gid);
680 error = notify_change(mnt_userns, path->dentry, &newattrs,
683 if (delegated_inode) {
684 error = break_deleg_wait(&delegated_inode);
691 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
698 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
701 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
702 if (flag & AT_EMPTY_PATH)
703 lookup_flags |= LOOKUP_EMPTY;
705 error = user_path_at(dfd, filename, lookup_flags, &path);
708 error = mnt_want_write(path.mnt);
711 error = chown_common(&path, user, group);
712 mnt_drop_write(path.mnt);
715 if (retry_estale(error, lookup_flags)) {
716 lookup_flags |= LOOKUP_REVAL;
723 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
724 gid_t, group, int, flag)
726 return do_fchownat(dfd, filename, user, group, flag);
729 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
731 return do_fchownat(AT_FDCWD, filename, user, group, 0);
734 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
736 return do_fchownat(AT_FDCWD, filename, user, group,
737 AT_SYMLINK_NOFOLLOW);
740 int vfs_fchown(struct file *file, uid_t user, gid_t group)
744 error = mnt_want_write_file(file);
748 error = chown_common(&file->f_path, user, group);
749 mnt_drop_write_file(file);
753 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
755 struct fd f = fdget(fd);
759 error = vfs_fchown(f.file, user, group);
765 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
767 return ksys_fchown(fd, user, group);
770 static int do_dentry_open(struct file *f,
772 int (*open)(struct inode *, struct file *))
774 static const struct file_operations empty_fops = {};
777 path_get(&f->f_path);
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);
783 if (unlikely(f->f_flags & O_PATH)) {
784 f->f_mode = FMODE_PATH | FMODE_OPENED;
785 f->f_op = &empty_fops;
789 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
790 error = get_write_access(inode);
793 error = __mnt_want_write(f->f_path.mnt);
794 if (unlikely(error)) {
795 put_write_access(inode);
798 f->f_mode |= FMODE_WRITER;
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;
805 f->f_op = fops_get(inode->i_fop);
806 if (WARN_ON(!f->f_op)) {
811 error = security_file_open(f);
815 error = break_lease(locks_inode(f), f->f_flags);
819 /* normally all 3 are set; ->open() can clear them if needed */
820 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
822 open = f->f_op->open;
824 error = open(inode, f);
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 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
838 f->f_mode |= FMODE_CAN_ODIRECT;
840 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
842 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
844 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
848 * XXX: Huge page cache doesn't support writing yet. Drop all page
849 * cache for this file before processing writes.
851 if (f->f_mode & FMODE_WRITE) {
853 * Paired with smp_mb() in collapse_file() to ensure nr_thps
854 * is up to date and the update to i_writecount by
855 * get_write_access() is visible. Ensures subsequent insertion
856 * of THPs into the page cache will fail.
859 if (filemap_nr_thps(inode->i_mapping)) {
860 struct address_space *mapping = inode->i_mapping;
862 filemap_invalidate_lock(inode->i_mapping);
864 * unmap_mapping_range just need to be called once
865 * here, because the private pages is not need to be
866 * unmapped mapping (e.g. data segment of dynamic
867 * shared libraries here).
869 unmap_mapping_range(mapping, 0, 0, 0);
870 truncate_inode_pages(mapping, 0);
871 filemap_invalidate_unlock(inode->i_mapping);
878 if (WARN_ON_ONCE(error > 0))
881 if (f->f_mode & FMODE_WRITER) {
882 put_write_access(inode);
883 __mnt_drop_write(f->f_path.mnt);
886 path_put(&f->f_path);
887 f->f_path.mnt = NULL;
888 f->f_path.dentry = NULL;
894 * finish_open - finish opening a file
895 * @file: file pointer
896 * @dentry: pointer to dentry
897 * @open: open callback
898 * @opened: state of open
900 * This can be used to finish opening a file passed to i_op->atomic_open().
902 * If the open callback is set to NULL, then the standard f_op->open()
903 * filesystem callback is substituted.
905 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
906 * the return value of d_splice_alias(), then the caller needs to perform dput()
907 * on it after finish_open().
909 * Returns zero on success or -errno if the open failed.
911 int finish_open(struct file *file, struct dentry *dentry,
912 int (*open)(struct inode *, struct file *))
914 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
916 file->f_path.dentry = dentry;
917 return do_dentry_open(file, d_backing_inode(dentry), open);
919 EXPORT_SYMBOL(finish_open);
922 * finish_no_open - finish ->atomic_open() without opening the file
924 * @file: file pointer
925 * @dentry: dentry or NULL (as returned from ->lookup())
927 * This can be used to set the result of a successful lookup in ->atomic_open().
929 * NB: unlike finish_open() this function does consume the dentry reference and
930 * the caller need not dput() it.
932 * Returns "0" which must be the return value of ->atomic_open() after having
933 * called this function.
935 int finish_no_open(struct file *file, struct dentry *dentry)
937 file->f_path.dentry = dentry;
940 EXPORT_SYMBOL(finish_no_open);
942 char *file_path(struct file *filp, char *buf, int buflen)
944 return d_path(&filp->f_path, buf, buflen);
946 EXPORT_SYMBOL(file_path);
949 * vfs_open - open the file at the given path
950 * @path: path to open
951 * @file: newly allocated file with f_flag initialized
952 * @cred: credentials to use
954 int vfs_open(const struct path *path, struct file *file)
956 file->f_path = *path;
957 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
960 struct file *dentry_open(const struct path *path, int flags,
961 const struct cred *cred)
966 validate_creds(cred);
968 /* We must always pass in a valid mount pointer. */
971 f = alloc_empty_file(flags, cred);
973 error = vfs_open(path, f);
981 EXPORT_SYMBOL(dentry_open);
984 * dentry_create - Create and open a file
985 * @path: path to create
987 * @mode: mode bits for new file
988 * @cred: credentials to use
990 * Caller must hold the parent directory's lock, and have prepared
991 * a negative dentry, placed in @path->dentry, for the new file.
993 * Caller sets @path->mnt to the vfsmount of the filesystem where
994 * the new file is to be created. The parent directory and the
995 * negative dentry must reside on the same filesystem instance.
997 * On success, returns a "struct file *". Otherwise a ERR_PTR
1000 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1001 const struct cred *cred)
1006 validate_creds(cred);
1007 f = alloc_empty_file(flags, cred);
1011 error = vfs_create(mnt_user_ns(path->mnt),
1012 d_inode(path->dentry->d_parent),
1013 path->dentry, mode, true);
1015 error = vfs_open(path, f);
1017 if (unlikely(error)) {
1019 return ERR_PTR(error);
1023 EXPORT_SYMBOL(dentry_create);
1025 struct file *open_with_fake_path(const struct path *path, int flags,
1026 struct inode *inode, const struct cred *cred)
1028 struct file *f = alloc_empty_file_noaccount(flags, cred);
1033 error = do_dentry_open(f, inode, NULL);
1041 EXPORT_SYMBOL(open_with_fake_path);
1043 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1044 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1046 inline struct open_how build_open_how(int flags, umode_t mode)
1048 struct open_how how = {
1049 .flags = flags & VALID_OPEN_FLAGS,
1050 .mode = mode & S_IALLUGO,
1053 /* O_PATH beats everything else. */
1054 if (how.flags & O_PATH)
1055 how.flags &= O_PATH_FLAGS;
1056 /* Modes should only be set for create-like flags. */
1057 if (!WILL_CREATE(how.flags))
1062 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1064 u64 flags = how->flags;
1065 u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1066 int lookup_flags = 0;
1067 int acc_mode = ACC_MODE(flags);
1069 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1070 "struct open_flags doesn't yet handle flags > 32 bits");
1073 * Strip flags that either shouldn't be set by userspace like
1074 * FMODE_NONOTIFY or that aren't relevant in determining struct
1075 * open_flags like O_CLOEXEC.
1080 * Older syscalls implicitly clear all of the invalid flags or argument
1081 * values before calling build_open_flags(), but openat2(2) checks all
1084 if (flags & ~VALID_OPEN_FLAGS)
1086 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1089 /* Scoping flags are mutually exclusive. */
1090 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1093 /* Deal with the mode. */
1094 if (WILL_CREATE(flags)) {
1095 if (how->mode & ~S_IALLUGO)
1097 op->mode = how->mode | S_IFREG;
1105 * In order to ensure programs get explicit errors when trying to use
1106 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1107 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1108 * have to require userspace to explicitly set it.
1110 if (flags & __O_TMPFILE) {
1111 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1113 if (!(acc_mode & MAY_WRITE))
1116 if (flags & O_PATH) {
1117 /* O_PATH only permits certain other flags to be set. */
1118 if (flags & ~O_PATH_FLAGS)
1124 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1125 * check for O_DSYNC if the need any syncing at all we enforce it's
1126 * always set instead of having to deal with possibly weird behaviour
1127 * for malicious applications setting only __O_SYNC.
1129 if (flags & __O_SYNC)
1132 op->open_flag = flags;
1134 /* O_TRUNC implies we need access checks for write permissions */
1135 if (flags & O_TRUNC)
1136 acc_mode |= MAY_WRITE;
1138 /* Allow the LSM permission hook to distinguish append
1139 access from general write access. */
1140 if (flags & O_APPEND)
1141 acc_mode |= MAY_APPEND;
1143 op->acc_mode = acc_mode;
1145 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1147 if (flags & O_CREAT) {
1148 op->intent |= LOOKUP_CREATE;
1149 if (flags & O_EXCL) {
1150 op->intent |= LOOKUP_EXCL;
1151 flags |= O_NOFOLLOW;
1155 if (flags & O_DIRECTORY)
1156 lookup_flags |= LOOKUP_DIRECTORY;
1157 if (!(flags & O_NOFOLLOW))
1158 lookup_flags |= LOOKUP_FOLLOW;
1160 if (how->resolve & RESOLVE_NO_XDEV)
1161 lookup_flags |= LOOKUP_NO_XDEV;
1162 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1163 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1164 if (how->resolve & RESOLVE_NO_SYMLINKS)
1165 lookup_flags |= LOOKUP_NO_SYMLINKS;
1166 if (how->resolve & RESOLVE_BENEATH)
1167 lookup_flags |= LOOKUP_BENEATH;
1168 if (how->resolve & RESOLVE_IN_ROOT)
1169 lookup_flags |= LOOKUP_IN_ROOT;
1170 if (how->resolve & RESOLVE_CACHED) {
1171 /* Don't bother even trying for create/truncate/tmpfile open */
1172 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1174 lookup_flags |= LOOKUP_CACHED;
1177 op->lookup_flags = lookup_flags;
1182 * file_open_name - open file and return file pointer
1184 * @name: struct filename containing path to open
1185 * @flags: open flags as per the open(2) second argument
1186 * @mode: mode for the new file if O_CREAT is set, else ignored
1188 * This is the helper to open a file from kernelspace if you really
1189 * have to. But in generally you should not do this, so please move
1190 * along, nothing to see here..
1192 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1194 struct open_flags op;
1195 struct open_how how = build_open_how(flags, mode);
1196 int err = build_open_flags(&how, &op);
1198 return ERR_PTR(err);
1199 return do_filp_open(AT_FDCWD, name, &op);
1203 * filp_open - open file and return file pointer
1205 * @filename: path to open
1206 * @flags: open flags as per the open(2) second argument
1207 * @mode: mode for the new file if O_CREAT is set, else ignored
1209 * This is the helper to open a file from kernelspace if you really
1210 * have to. But in generally you should not do this, so please move
1211 * along, nothing to see here..
1213 struct file *filp_open(const char *filename, int flags, umode_t mode)
1215 struct filename *name = getname_kernel(filename);
1216 struct file *file = ERR_CAST(name);
1218 if (!IS_ERR(name)) {
1219 file = file_open_name(name, flags, mode);
1224 EXPORT_SYMBOL(filp_open);
1226 struct file *file_open_root(const struct path *root,
1227 const char *filename, int flags, umode_t mode)
1229 struct open_flags op;
1230 struct open_how how = build_open_how(flags, mode);
1231 int err = build_open_flags(&how, &op);
1233 return ERR_PTR(err);
1234 return do_file_open_root(root, filename, &op);
1236 EXPORT_SYMBOL(file_open_root);
1238 static long do_sys_openat2(int dfd, const char __user *filename,
1239 struct open_how *how)
1241 struct open_flags op;
1242 int fd = build_open_flags(how, &op);
1243 struct filename *tmp;
1248 tmp = getname(filename);
1250 return PTR_ERR(tmp);
1252 fd = get_unused_fd_flags(how->flags);
1254 struct file *f = do_filp_open(dfd, tmp, &op);
1267 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1269 struct open_how how = build_open_how(flags, mode);
1270 return do_sys_openat2(dfd, filename, &how);
1274 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1276 if (force_o_largefile())
1277 flags |= O_LARGEFILE;
1278 return do_sys_open(AT_FDCWD, filename, flags, mode);
1281 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1284 if (force_o_largefile())
1285 flags |= O_LARGEFILE;
1286 return do_sys_open(dfd, filename, flags, mode);
1289 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1290 struct open_how __user *, how, size_t, usize)
1293 struct open_how tmp;
1295 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1296 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1298 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1301 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1305 audit_openat2_how(&tmp);
1307 /* O_LARGEFILE is only allowed for non-O_PATH. */
1308 if (!(tmp.flags & O_PATH) && force_o_largefile())
1309 tmp.flags |= O_LARGEFILE;
1311 return do_sys_openat2(dfd, filename, &tmp);
1314 #ifdef CONFIG_COMPAT
1316 * Exactly like sys_open(), except that it doesn't set the
1319 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1321 return do_sys_open(AT_FDCWD, filename, flags, mode);
1325 * Exactly like sys_openat(), except that it doesn't set the
1328 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1330 return do_sys_open(dfd, filename, flags, mode);
1337 * For backward compatibility? Maybe this should be moved
1338 * into arch/i386 instead?
1340 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1342 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1344 if (force_o_largefile())
1345 flags |= O_LARGEFILE;
1346 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1351 * "id" is the POSIX thread ID. We use the
1352 * files pointer for this..
1354 int filp_close(struct file *filp, fl_owner_t id)
1358 if (!file_count(filp)) {
1359 printk(KERN_ERR "VFS: Close: file count is 0\n");
1363 if (filp->f_op->flush)
1364 retval = filp->f_op->flush(filp, id);
1366 if (likely(!(filp->f_mode & FMODE_PATH))) {
1367 dnotify_flush(filp, id);
1368 locks_remove_posix(filp, id);
1374 EXPORT_SYMBOL(filp_close);
1377 * Careful here! We test whether the file pointer is NULL before
1378 * releasing the fd. This ensures that one clone task can't release
1379 * an fd while another clone is opening it.
1381 SYSCALL_DEFINE1(close, unsigned int, fd)
1383 int retval = close_fd(fd);
1385 /* can't restart close syscall because file table entry was cleared */
1386 if (unlikely(retval == -ERESTARTSYS ||
1387 retval == -ERESTARTNOINTR ||
1388 retval == -ERESTARTNOHAND ||
1389 retval == -ERESTART_RESTARTBLOCK))
1396 * close_range() - Close all file descriptors in a given range.
1398 * @fd: starting file descriptor to close
1399 * @max_fd: last file descriptor to close
1400 * @flags: reserved for future extensions
1402 * This closes a range of file descriptors. All file descriptors
1403 * from @fd up to and including @max_fd are closed.
1404 * Currently, errors to close a given file descriptor are ignored.
1406 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1407 unsigned int, flags)
1409 return __close_range(fd, max_fd, flags);
1413 * This routine simulates a hangup on the tty, to arrange that users
1414 * are given clean terminals at login time.
1416 SYSCALL_DEFINE0(vhangup)
1418 if (capable(CAP_SYS_TTY_CONFIG)) {
1426 * Called when an inode is about to be open.
1427 * We use this to disallow opening large files on 32bit systems if
1428 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1429 * on this flag in sys_open.
1431 int generic_file_open(struct inode * inode, struct file * filp)
1433 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1438 EXPORT_SYMBOL(generic_file_open);
1441 * This is used by subsystems that don't want seekable
1442 * file descriptors. The function is not supposed to ever fail, the only
1443 * reason it returns an 'int' and not 'void' is so that it can be plugged
1444 * directly into file_operations structure.
1446 int nonseekable_open(struct inode *inode, struct file *filp)
1448 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1452 EXPORT_SYMBOL(nonseekable_open);
1455 * stream_open is used by subsystems that want stream-like file descriptors.
1456 * Such file descriptors are not seekable and don't have notion of position
1457 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1458 * Contrary to file descriptors of other regular files, .read() and .write()
1459 * can run simultaneously.
1461 * stream_open never fails and is marked to return int so that it could be
1462 * directly used as file_operations.open .
1464 int stream_open(struct inode *inode, struct file *filp)
1466 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1467 filp->f_mode |= FMODE_STREAM;
1471 EXPORT_SYMBOL(stream_open);