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 user_namespace *mnt_userns, struct dentry *dentry,
39 loff_t length, unsigned int time_attrs, struct file *filp)
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(mnt_userns, dentry, &newattrs, NULL);
65 inode_unlock(dentry->d_inode);
69 long vfs_truncate(const struct path *path, loff_t length)
71 struct user_namespace *mnt_userns;
75 inode = path->dentry->d_inode;
77 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
78 if (S_ISDIR(inode->i_mode))
80 if (!S_ISREG(inode->i_mode))
83 error = mnt_want_write(path->mnt);
87 mnt_userns = mnt_user_ns(path->mnt);
88 error = inode_permission(mnt_userns, inode, MAY_WRITE);
90 goto mnt_drop_write_and_out;
94 goto mnt_drop_write_and_out;
96 error = get_write_access(inode);
98 goto mnt_drop_write_and_out;
101 * Make sure that there are no leases. get_write_access() protects
102 * against the truncate racing with a lease-granting setlease().
104 error = break_lease(inode, O_WRONLY);
106 goto put_write_and_out;
108 error = security_path_truncate(path);
110 error = do_truncate(mnt_userns, 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)))
189 sb_start_write(inode->i_sb);
190 error = security_path_truncate(&f.file->f_path);
192 error = do_truncate(file_mnt_user_ns(f.file), dentry, length,
193 ATTR_MTIME | ATTR_CTIME, f.file);
194 sb_end_write(inode->i_sb);
201 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
203 return do_sys_ftruncate(fd, length, 1);
207 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
209 return do_sys_ftruncate(fd, length, 1);
213 /* LFS versions of truncate are only needed on 32 bit machines */
214 #if BITS_PER_LONG == 32
215 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
217 return do_sys_truncate(path, length);
220 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
222 return do_sys_ftruncate(fd, length, 0);
224 #endif /* BITS_PER_LONG == 32 */
227 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
229 struct inode *inode = file_inode(file);
232 if (offset < 0 || len <= 0)
235 /* Return error if mode is not supported */
236 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
239 /* Punch hole and zero range are mutually exclusive */
240 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
241 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
244 /* Punch hole must have keep size set */
245 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
246 !(mode & FALLOC_FL_KEEP_SIZE))
249 /* Collapse range should only be used exclusively. */
250 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
251 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
254 /* Insert range should only be used exclusively. */
255 if ((mode & FALLOC_FL_INSERT_RANGE) &&
256 (mode & ~FALLOC_FL_INSERT_RANGE))
259 /* Unshare range should only be used with allocate mode. */
260 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
261 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
264 if (!(file->f_mode & FMODE_WRITE))
268 * We can only allow pure fallocate on append only files
270 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
273 if (IS_IMMUTABLE(inode))
277 * We cannot allow any fallocate operation on an active swapfile
279 if (IS_SWAPFILE(inode))
283 * Revalidate the write permissions, in case security policy has
284 * changed since the files were opened.
286 ret = security_file_permission(file, MAY_WRITE);
290 if (S_ISFIFO(inode->i_mode))
293 if (S_ISDIR(inode->i_mode))
296 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
299 /* Check for wrap through zero too */
300 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
303 if (!file->f_op->fallocate)
306 file_start_write(file);
307 ret = file->f_op->fallocate(file, mode, offset, len);
310 * Create inotify and fanotify events.
312 * To keep the logic simple always create events if fallocate succeeds.
313 * This implies that events are even created if the file size remains
314 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
317 fsnotify_modify(file);
319 file_end_write(file);
322 EXPORT_SYMBOL_GPL(vfs_fallocate);
324 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
326 struct fd f = fdget(fd);
330 error = vfs_fallocate(f.file, mode, offset, len);
336 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
338 return ksys_fallocate(fd, mode, offset, len);
342 * access() needs to use the real uid/gid, not the effective uid/gid.
343 * We do this by temporarily clearing all FS-related capabilities and
344 * switching the fsuid/fsgid around to the real ones.
346 static const struct cred *access_override_creds(void)
348 const struct cred *old_cred;
349 struct cred *override_cred;
351 override_cred = prepare_creds();
355 override_cred->fsuid = override_cred->uid;
356 override_cred->fsgid = override_cred->gid;
358 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
359 /* Clear the capabilities if we switch to a non-root user */
360 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
361 if (!uid_eq(override_cred->uid, root_uid))
362 cap_clear(override_cred->cap_effective);
364 override_cred->cap_effective =
365 override_cred->cap_permitted;
369 * The new set of credentials can *only* be used in
370 * task-synchronous circumstances, and does not need
371 * RCU freeing, unless somebody then takes a separate
374 * NOTE! This is _only_ true because this credential
375 * is used purely for override_creds() that installs
376 * it as the subjective cred. Other threads will be
377 * accessing ->real_cred, not the subjective cred.
379 * If somebody _does_ make a copy of this (using the
380 * 'get_current_cred()' function), that will clear the
381 * non_rcu field, because now that other user may be
382 * expecting RCU freeing. But normal thread-synchronous
383 * cred accesses will keep things non-RCY.
385 override_cred->non_rcu = 1;
387 old_cred = override_creds(override_cred);
389 /* override_cred() gets its own ref */
390 put_cred(override_cred);
395 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
400 unsigned int lookup_flags = LOOKUP_FOLLOW;
401 const struct cred *old_cred = NULL;
403 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
406 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
409 if (flags & AT_SYMLINK_NOFOLLOW)
410 lookup_flags &= ~LOOKUP_FOLLOW;
411 if (flags & AT_EMPTY_PATH)
412 lookup_flags |= LOOKUP_EMPTY;
414 if (!(flags & AT_EACCESS)) {
415 old_cred = access_override_creds();
421 res = user_path_at(dfd, filename, lookup_flags, &path);
425 inode = d_backing_inode(path.dentry);
427 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
429 * MAY_EXEC on regular files is denied if the fs is mounted
430 * with the "noexec" flag.
433 if (path_noexec(&path))
434 goto out_path_release;
437 res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS);
438 /* SuS v2 requires we report a read only fs too */
439 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
440 goto out_path_release;
442 * This is a rare case where using __mnt_is_readonly()
443 * is OK without a mnt_want/drop_write() pair. Since
444 * no actual write to the fs is performed here, we do
445 * not need to telegraph to that to anyone.
447 * By doing this, we accept that this access is
448 * inherently racy and know that the fs may change
449 * state before we even see this result.
451 if (__mnt_is_readonly(path.mnt))
456 if (retry_estale(res, lookup_flags)) {
457 lookup_flags |= LOOKUP_REVAL;
462 revert_creds(old_cred);
467 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
469 return do_faccessat(dfd, filename, mode, 0);
472 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
475 return do_faccessat(dfd, filename, mode, flags);
478 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
480 return do_faccessat(AT_FDCWD, filename, mode, 0);
483 SYSCALL_DEFINE1(chdir, const char __user *, filename)
487 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
489 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
493 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
497 set_fs_pwd(current->fs, &path);
501 if (retry_estale(error, lookup_flags)) {
502 lookup_flags |= LOOKUP_REVAL;
509 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
511 struct fd f = fdget_raw(fd);
519 if (!d_can_lookup(f.file->f_path.dentry))
522 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
524 set_fs_pwd(current->fs, &f.file->f_path);
531 SYSCALL_DEFINE1(chroot, const char __user *, filename)
535 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
537 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
541 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
546 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
548 error = security_path_chroot(&path);
552 set_fs_root(current->fs, &path);
556 if (retry_estale(error, lookup_flags)) {
557 lookup_flags |= LOOKUP_REVAL;
564 int chmod_common(const struct path *path, umode_t mode)
566 struct inode *inode = path->dentry->d_inode;
567 struct inode *delegated_inode = NULL;
568 struct iattr newattrs;
571 error = mnt_want_write(path->mnt);
576 error = security_path_chmod(path, mode);
579 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
580 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
581 error = notify_change(mnt_user_ns(path->mnt), path->dentry,
582 &newattrs, &delegated_inode);
585 if (delegated_inode) {
586 error = break_deleg_wait(&delegated_inode);
590 mnt_drop_write(path->mnt);
594 int vfs_fchmod(struct file *file, umode_t mode)
597 return chmod_common(&file->f_path, mode);
600 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
602 struct fd f = fdget(fd);
606 err = vfs_fchmod(f.file, mode);
612 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
616 unsigned int lookup_flags = LOOKUP_FOLLOW;
618 error = user_path_at(dfd, filename, lookup_flags, &path);
620 error = chmod_common(&path, mode);
622 if (retry_estale(error, lookup_flags)) {
623 lookup_flags |= LOOKUP_REVAL;
630 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
633 return do_fchmodat(dfd, filename, mode);
636 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
638 return do_fchmodat(AT_FDCWD, filename, mode);
641 int chown_common(const struct path *path, uid_t user, gid_t group)
643 struct user_namespace *mnt_userns;
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);
654 mnt_userns = mnt_user_ns(path->mnt);
655 uid = kuid_from_mnt(mnt_userns, uid);
656 gid = kgid_from_mnt(mnt_userns, gid);
659 newattrs.ia_valid = ATTR_CTIME;
660 if (user != (uid_t) -1) {
663 newattrs.ia_valid |= ATTR_UID;
664 newattrs.ia_uid = uid;
666 if (group != (gid_t) -1) {
669 newattrs.ia_valid |= ATTR_GID;
670 newattrs.ia_gid = gid;
672 if (!S_ISDIR(inode->i_mode))
674 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
676 error = security_path_chown(path, uid, gid);
678 error = notify_change(mnt_userns, path->dentry, &newattrs,
681 if (delegated_inode) {
682 error = break_deleg_wait(&delegated_inode);
689 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
696 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
699 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
700 if (flag & AT_EMPTY_PATH)
701 lookup_flags |= LOOKUP_EMPTY;
703 error = user_path_at(dfd, filename, lookup_flags, &path);
706 error = mnt_want_write(path.mnt);
709 error = chown_common(&path, user, group);
710 mnt_drop_write(path.mnt);
713 if (retry_estale(error, lookup_flags)) {
714 lookup_flags |= LOOKUP_REVAL;
721 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
722 gid_t, group, int, flag)
724 return do_fchownat(dfd, filename, user, group, flag);
727 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
729 return do_fchownat(AT_FDCWD, filename, user, group, 0);
732 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
734 return do_fchownat(AT_FDCWD, filename, user, group,
735 AT_SYMLINK_NOFOLLOW);
738 int vfs_fchown(struct file *file, uid_t user, gid_t group)
742 error = mnt_want_write_file(file);
746 error = chown_common(&file->f_path, user, group);
747 mnt_drop_write_file(file);
751 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
753 struct fd f = fdget(fd);
757 error = vfs_fchown(f.file, user, group);
763 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
765 return ksys_fchown(fd, user, group);
768 static int do_dentry_open(struct file *f,
770 int (*open)(struct inode *, struct file *))
772 static const struct file_operations empty_fops = {};
775 path_get(&f->f_path);
777 f->f_mapping = inode->i_mapping;
778 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
779 f->f_sb_err = file_sample_sb_err(f);
781 if (unlikely(f->f_flags & O_PATH)) {
782 f->f_mode = FMODE_PATH | FMODE_OPENED;
783 f->f_op = &empty_fops;
787 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
788 error = get_write_access(inode);
791 error = __mnt_want_write(f->f_path.mnt);
792 if (unlikely(error)) {
793 put_write_access(inode);
796 f->f_mode |= FMODE_WRITER;
799 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
800 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
801 f->f_mode |= FMODE_ATOMIC_POS;
803 f->f_op = fops_get(inode->i_fop);
804 if (WARN_ON(!f->f_op)) {
809 error = security_file_open(f);
813 error = break_lease(locks_inode(f), f->f_flags);
817 /* normally all 3 are set; ->open() can clear them if needed */
818 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
820 open = f->f_op->open;
822 error = open(inode, f);
826 f->f_mode |= FMODE_OPENED;
827 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
828 i_readcount_inc(inode);
829 if ((f->f_mode & FMODE_READ) &&
830 likely(f->f_op->read || f->f_op->read_iter))
831 f->f_mode |= FMODE_CAN_READ;
832 if ((f->f_mode & FMODE_WRITE) &&
833 likely(f->f_op->write || f->f_op->write_iter))
834 f->f_mode |= FMODE_CAN_WRITE;
836 f->f_write_hint = WRITE_LIFE_NOT_SET;
837 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
839 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
841 /* NB: we're sure to have correct a_ops only after f_op->open */
842 if (f->f_flags & O_DIRECT) {
843 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
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 truncate_pagecache(inode, 0);
866 if (WARN_ON_ONCE(error > 0))
869 if (f->f_mode & FMODE_WRITER) {
870 put_write_access(inode);
871 __mnt_drop_write(f->f_path.mnt);
874 path_put(&f->f_path);
875 f->f_path.mnt = NULL;
876 f->f_path.dentry = NULL;
882 * finish_open - finish opening a file
883 * @file: file pointer
884 * @dentry: pointer to dentry
885 * @open: open callback
886 * @opened: state of open
888 * This can be used to finish opening a file passed to i_op->atomic_open().
890 * If the open callback is set to NULL, then the standard f_op->open()
891 * filesystem callback is substituted.
893 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
894 * the return value of d_splice_alias(), then the caller needs to perform dput()
895 * on it after finish_open().
897 * Returns zero on success or -errno if the open failed.
899 int finish_open(struct file *file, struct dentry *dentry,
900 int (*open)(struct inode *, struct file *))
902 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
904 file->f_path.dentry = dentry;
905 return do_dentry_open(file, d_backing_inode(dentry), open);
907 EXPORT_SYMBOL(finish_open);
910 * finish_no_open - finish ->atomic_open() without opening the file
912 * @file: file pointer
913 * @dentry: dentry or NULL (as returned from ->lookup())
915 * This can be used to set the result of a successful lookup in ->atomic_open().
917 * NB: unlike finish_open() this function does consume the dentry reference and
918 * the caller need not dput() it.
920 * Returns "0" which must be the return value of ->atomic_open() after having
921 * called this function.
923 int finish_no_open(struct file *file, struct dentry *dentry)
925 file->f_path.dentry = dentry;
928 EXPORT_SYMBOL(finish_no_open);
930 char *file_path(struct file *filp, char *buf, int buflen)
932 return d_path(&filp->f_path, buf, buflen);
934 EXPORT_SYMBOL(file_path);
937 * vfs_open - open the file at the given path
938 * @path: path to open
939 * @file: newly allocated file with f_flag initialized
940 * @cred: credentials to use
942 int vfs_open(const struct path *path, struct file *file)
944 file->f_path = *path;
945 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
948 struct file *dentry_open(const struct path *path, int flags,
949 const struct cred *cred)
954 validate_creds(cred);
956 /* We must always pass in a valid mount pointer. */
959 f = alloc_empty_file(flags, cred);
961 error = vfs_open(path, f);
969 EXPORT_SYMBOL(dentry_open);
971 struct file *open_with_fake_path(const struct path *path, int flags,
972 struct inode *inode, const struct cred *cred)
974 struct file *f = alloc_empty_file_noaccount(flags, cred);
979 error = do_dentry_open(f, inode, NULL);
987 EXPORT_SYMBOL(open_with_fake_path);
989 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
990 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
992 inline struct open_how build_open_how(int flags, umode_t mode)
994 struct open_how how = {
995 .flags = flags & VALID_OPEN_FLAGS,
996 .mode = mode & S_IALLUGO,
999 /* O_PATH beats everything else. */
1000 if (how.flags & O_PATH)
1001 how.flags &= O_PATH_FLAGS;
1002 /* Modes should only be set for create-like flags. */
1003 if (!WILL_CREATE(how.flags))
1008 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1010 u64 flags = how->flags;
1011 u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1012 int lookup_flags = 0;
1013 int acc_mode = ACC_MODE(flags);
1015 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1016 "struct open_flags doesn't yet handle flags > 32 bits");
1019 * Strip flags that either shouldn't be set by userspace like
1020 * FMODE_NONOTIFY or that aren't relevant in determining struct
1021 * open_flags like O_CLOEXEC.
1026 * Older syscalls implicitly clear all of the invalid flags or argument
1027 * values before calling build_open_flags(), but openat2(2) checks all
1030 if (flags & ~VALID_OPEN_FLAGS)
1032 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1035 /* Scoping flags are mutually exclusive. */
1036 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1039 /* Deal with the mode. */
1040 if (WILL_CREATE(flags)) {
1041 if (how->mode & ~S_IALLUGO)
1043 op->mode = how->mode | S_IFREG;
1051 * In order to ensure programs get explicit errors when trying to use
1052 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1053 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1054 * have to require userspace to explicitly set it.
1056 if (flags & __O_TMPFILE) {
1057 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1059 if (!(acc_mode & MAY_WRITE))
1062 if (flags & O_PATH) {
1063 /* O_PATH only permits certain other flags to be set. */
1064 if (flags & ~O_PATH_FLAGS)
1070 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1071 * check for O_DSYNC if the need any syncing at all we enforce it's
1072 * always set instead of having to deal with possibly weird behaviour
1073 * for malicious applications setting only __O_SYNC.
1075 if (flags & __O_SYNC)
1078 op->open_flag = flags;
1080 /* O_TRUNC implies we need access checks for write permissions */
1081 if (flags & O_TRUNC)
1082 acc_mode |= MAY_WRITE;
1084 /* Allow the LSM permission hook to distinguish append
1085 access from general write access. */
1086 if (flags & O_APPEND)
1087 acc_mode |= MAY_APPEND;
1089 op->acc_mode = acc_mode;
1091 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1093 if (flags & O_CREAT) {
1094 op->intent |= LOOKUP_CREATE;
1095 if (flags & O_EXCL) {
1096 op->intent |= LOOKUP_EXCL;
1097 flags |= O_NOFOLLOW;
1101 if (flags & O_DIRECTORY)
1102 lookup_flags |= LOOKUP_DIRECTORY;
1103 if (!(flags & O_NOFOLLOW))
1104 lookup_flags |= LOOKUP_FOLLOW;
1106 if (how->resolve & RESOLVE_NO_XDEV)
1107 lookup_flags |= LOOKUP_NO_XDEV;
1108 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1109 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1110 if (how->resolve & RESOLVE_NO_SYMLINKS)
1111 lookup_flags |= LOOKUP_NO_SYMLINKS;
1112 if (how->resolve & RESOLVE_BENEATH)
1113 lookup_flags |= LOOKUP_BENEATH;
1114 if (how->resolve & RESOLVE_IN_ROOT)
1115 lookup_flags |= LOOKUP_IN_ROOT;
1116 if (how->resolve & RESOLVE_CACHED) {
1117 /* Don't bother even trying for create/truncate/tmpfile open */
1118 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1120 lookup_flags |= LOOKUP_CACHED;
1123 op->lookup_flags = lookup_flags;
1128 * file_open_name - open file and return file pointer
1130 * @name: struct filename containing path to open
1131 * @flags: open flags as per the open(2) second argument
1132 * @mode: mode for the new file if O_CREAT is set, else ignored
1134 * This is the helper to open a file from kernelspace if you really
1135 * have to. But in generally you should not do this, so please move
1136 * along, nothing to see here..
1138 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1140 struct open_flags op;
1141 struct open_how how = build_open_how(flags, mode);
1142 int err = build_open_flags(&how, &op);
1144 return ERR_PTR(err);
1145 return do_filp_open(AT_FDCWD, name, &op);
1149 * filp_open - open file and return file pointer
1151 * @filename: path to open
1152 * @flags: open flags as per the open(2) second argument
1153 * @mode: mode for the new file if O_CREAT is set, else ignored
1155 * This is the helper to open a file from kernelspace if you really
1156 * have to. But in generally you should not do this, so please move
1157 * along, nothing to see here..
1159 struct file *filp_open(const char *filename, int flags, umode_t mode)
1161 struct filename *name = getname_kernel(filename);
1162 struct file *file = ERR_CAST(name);
1164 if (!IS_ERR(name)) {
1165 file = file_open_name(name, flags, mode);
1170 EXPORT_SYMBOL(filp_open);
1172 struct file *file_open_root(const struct path *root,
1173 const char *filename, int flags, umode_t mode)
1175 struct open_flags op;
1176 struct open_how how = build_open_how(flags, mode);
1177 int err = build_open_flags(&how, &op);
1179 return ERR_PTR(err);
1180 return do_file_open_root(root, filename, &op);
1182 EXPORT_SYMBOL(file_open_root);
1184 static long do_sys_openat2(int dfd, const char __user *filename,
1185 struct open_how *how)
1187 struct open_flags op;
1188 int fd = build_open_flags(how, &op);
1189 struct filename *tmp;
1194 tmp = getname(filename);
1196 return PTR_ERR(tmp);
1198 fd = get_unused_fd_flags(how->flags);
1200 struct file *f = do_filp_open(dfd, tmp, &op);
1213 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1215 struct open_how how = build_open_how(flags, mode);
1216 return do_sys_openat2(dfd, filename, &how);
1220 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1222 if (force_o_largefile())
1223 flags |= O_LARGEFILE;
1224 return do_sys_open(AT_FDCWD, filename, flags, mode);
1227 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1230 if (force_o_largefile())
1231 flags |= O_LARGEFILE;
1232 return do_sys_open(dfd, filename, flags, mode);
1235 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1236 struct open_how __user *, how, size_t, usize)
1239 struct open_how tmp;
1241 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1242 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1244 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1247 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1251 /* O_LARGEFILE is only allowed for non-O_PATH. */
1252 if (!(tmp.flags & O_PATH) && force_o_largefile())
1253 tmp.flags |= O_LARGEFILE;
1255 return do_sys_openat2(dfd, filename, &tmp);
1258 #ifdef CONFIG_COMPAT
1260 * Exactly like sys_open(), except that it doesn't set the
1263 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1265 return do_sys_open(AT_FDCWD, filename, flags, mode);
1269 * Exactly like sys_openat(), except that it doesn't set the
1272 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1274 return do_sys_open(dfd, filename, flags, mode);
1281 * For backward compatibility? Maybe this should be moved
1282 * into arch/i386 instead?
1284 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1286 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1288 if (force_o_largefile())
1289 flags |= O_LARGEFILE;
1290 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1295 * "id" is the POSIX thread ID. We use the
1296 * files pointer for this..
1298 int filp_close(struct file *filp, fl_owner_t id)
1302 if (!file_count(filp)) {
1303 printk(KERN_ERR "VFS: Close: file count is 0\n");
1307 if (filp->f_op->flush)
1308 retval = filp->f_op->flush(filp, id);
1310 if (likely(!(filp->f_mode & FMODE_PATH))) {
1311 dnotify_flush(filp, id);
1312 locks_remove_posix(filp, id);
1318 EXPORT_SYMBOL(filp_close);
1321 * Careful here! We test whether the file pointer is NULL before
1322 * releasing the fd. This ensures that one clone task can't release
1323 * an fd while another clone is opening it.
1325 SYSCALL_DEFINE1(close, unsigned int, fd)
1327 int retval = close_fd(fd);
1329 /* can't restart close syscall because file table entry was cleared */
1330 if (unlikely(retval == -ERESTARTSYS ||
1331 retval == -ERESTARTNOINTR ||
1332 retval == -ERESTARTNOHAND ||
1333 retval == -ERESTART_RESTARTBLOCK))
1340 * close_range() - Close all file descriptors in a given range.
1342 * @fd: starting file descriptor to close
1343 * @max_fd: last file descriptor to close
1344 * @flags: reserved for future extensions
1346 * This closes a range of file descriptors. All file descriptors
1347 * from @fd up to and including @max_fd are closed.
1348 * Currently, errors to close a given file descriptor are ignored.
1350 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1351 unsigned int, flags)
1353 return __close_range(fd, max_fd, flags);
1357 * This routine simulates a hangup on the tty, to arrange that users
1358 * are given clean terminals at login time.
1360 SYSCALL_DEFINE0(vhangup)
1362 if (capable(CAP_SYS_TTY_CONFIG)) {
1370 * Called when an inode is about to be open.
1371 * We use this to disallow opening large files on 32bit systems if
1372 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1373 * on this flag in sys_open.
1375 int generic_file_open(struct inode * inode, struct file * filp)
1377 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1382 EXPORT_SYMBOL(generic_file_open);
1385 * This is used by subsystems that don't want seekable
1386 * file descriptors. The function is not supposed to ever fail, the only
1387 * reason it returns an 'int' and not 'void' is so that it can be plugged
1388 * directly into file_operations structure.
1390 int nonseekable_open(struct inode *inode, struct file *filp)
1392 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1396 EXPORT_SYMBOL(nonseekable_open);
1399 * stream_open is used by subsystems that want stream-like file descriptors.
1400 * Such file descriptors are not seekable and don't have notion of position
1401 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1402 * Contrary to file descriptors of other regular files, .read() and .write()
1403 * can run simultaneously.
1405 * stream_open never fails and is marked to return int so that it could be
1406 * directly used as file_operations.open .
1408 int stream_open(struct inode *inode, struct file *filp)
1410 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1411 filp->f_mode |= FMODE_STREAM;
1415 EXPORT_SYMBOL(stream_open);