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(mnt_userns, 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 */
227 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
228 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
229 compat_arg_u64_dual(length))
231 return ksys_truncate(pathname, compat_arg_u64_glue(length));
235 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
236 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
237 compat_arg_u64_dual(length))
239 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
243 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
245 struct inode *inode = file_inode(file);
248 if (offset < 0 || len <= 0)
251 /* Return error if mode is not supported */
252 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
255 /* Punch hole and zero range are mutually exclusive */
256 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
257 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
260 /* Punch hole must have keep size set */
261 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
262 !(mode & FALLOC_FL_KEEP_SIZE))
265 /* Collapse range should only be used exclusively. */
266 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
267 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
270 /* Insert range should only be used exclusively. */
271 if ((mode & FALLOC_FL_INSERT_RANGE) &&
272 (mode & ~FALLOC_FL_INSERT_RANGE))
275 /* Unshare range should only be used with allocate mode. */
276 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
277 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
280 if (!(file->f_mode & FMODE_WRITE))
284 * We can only allow pure fallocate on append only files
286 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
289 if (IS_IMMUTABLE(inode))
293 * We cannot allow any fallocate operation on an active swapfile
295 if (IS_SWAPFILE(inode))
299 * Revalidate the write permissions, in case security policy has
300 * changed since the files were opened.
302 ret = security_file_permission(file, MAY_WRITE);
306 if (S_ISFIFO(inode->i_mode))
309 if (S_ISDIR(inode->i_mode))
312 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
315 /* Check for wrap through zero too */
316 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
319 if (!file->f_op->fallocate)
322 file_start_write(file);
323 ret = file->f_op->fallocate(file, mode, offset, len);
326 * Create inotify and fanotify events.
328 * To keep the logic simple always create events if fallocate succeeds.
329 * This implies that events are even created if the file size remains
330 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
333 fsnotify_modify(file);
335 file_end_write(file);
338 EXPORT_SYMBOL_GPL(vfs_fallocate);
340 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
342 struct fd f = fdget(fd);
346 error = vfs_fallocate(f.file, mode, offset, len);
352 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
354 return ksys_fallocate(fd, mode, offset, len);
357 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
358 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
359 compat_arg_u64_dual(len))
361 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
362 compat_arg_u64_glue(len));
367 * access() needs to use the real uid/gid, not the effective uid/gid.
368 * We do this by temporarily clearing all FS-related capabilities and
369 * switching the fsuid/fsgid around to the real ones.
371 static const struct cred *access_override_creds(void)
373 const struct cred *old_cred;
374 struct cred *override_cred;
376 override_cred = prepare_creds();
380 override_cred->fsuid = override_cred->uid;
381 override_cred->fsgid = override_cred->gid;
383 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
384 /* Clear the capabilities if we switch to a non-root user */
385 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
386 if (!uid_eq(override_cred->uid, root_uid))
387 cap_clear(override_cred->cap_effective);
389 override_cred->cap_effective =
390 override_cred->cap_permitted;
394 * The new set of credentials can *only* be used in
395 * task-synchronous circumstances, and does not need
396 * RCU freeing, unless somebody then takes a separate
399 * NOTE! This is _only_ true because this credential
400 * is used purely for override_creds() that installs
401 * it as the subjective cred. Other threads will be
402 * accessing ->real_cred, not the subjective cred.
404 * If somebody _does_ make a copy of this (using the
405 * 'get_current_cred()' function), that will clear the
406 * non_rcu field, because now that other user may be
407 * expecting RCU freeing. But normal thread-synchronous
408 * cred accesses will keep things non-RCY.
410 override_cred->non_rcu = 1;
412 old_cred = override_creds(override_cred);
414 /* override_cred() gets its own ref */
415 put_cred(override_cred);
420 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
425 unsigned int lookup_flags = LOOKUP_FOLLOW;
426 const struct cred *old_cred = NULL;
428 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
431 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
434 if (flags & AT_SYMLINK_NOFOLLOW)
435 lookup_flags &= ~LOOKUP_FOLLOW;
436 if (flags & AT_EMPTY_PATH)
437 lookup_flags |= LOOKUP_EMPTY;
439 if (!(flags & AT_EACCESS)) {
440 old_cred = access_override_creds();
446 res = user_path_at(dfd, filename, lookup_flags, &path);
450 inode = d_backing_inode(path.dentry);
452 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
454 * MAY_EXEC on regular files is denied if the fs is mounted
455 * with the "noexec" flag.
458 if (path_noexec(&path))
459 goto out_path_release;
462 res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS);
463 /* SuS v2 requires we report a read only fs too */
464 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
465 goto out_path_release;
467 * This is a rare case where using __mnt_is_readonly()
468 * is OK without a mnt_want/drop_write() pair. Since
469 * no actual write to the fs is performed here, we do
470 * not need to telegraph to that to anyone.
472 * By doing this, we accept that this access is
473 * inherently racy and know that the fs may change
474 * state before we even see this result.
476 if (__mnt_is_readonly(path.mnt))
481 if (retry_estale(res, lookup_flags)) {
482 lookup_flags |= LOOKUP_REVAL;
487 revert_creds(old_cred);
492 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
494 return do_faccessat(dfd, filename, mode, 0);
497 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
500 return do_faccessat(dfd, filename, mode, flags);
503 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
505 return do_faccessat(AT_FDCWD, filename, mode, 0);
508 SYSCALL_DEFINE1(chdir, const char __user *, filename)
512 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
514 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
518 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
522 set_fs_pwd(current->fs, &path);
526 if (retry_estale(error, lookup_flags)) {
527 lookup_flags |= LOOKUP_REVAL;
534 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
536 struct fd f = fdget_raw(fd);
544 if (!d_can_lookup(f.file->f_path.dentry))
547 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
549 set_fs_pwd(current->fs, &f.file->f_path);
556 SYSCALL_DEFINE1(chroot, const char __user *, filename)
560 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
562 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
566 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
571 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
573 error = security_path_chroot(&path);
577 set_fs_root(current->fs, &path);
581 if (retry_estale(error, lookup_flags)) {
582 lookup_flags |= LOOKUP_REVAL;
589 int chmod_common(const struct path *path, umode_t mode)
591 struct inode *inode = path->dentry->d_inode;
592 struct inode *delegated_inode = NULL;
593 struct iattr newattrs;
596 error = mnt_want_write(path->mnt);
601 error = security_path_chmod(path, mode);
604 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
605 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
606 error = notify_change(mnt_user_ns(path->mnt), path->dentry,
607 &newattrs, &delegated_inode);
610 if (delegated_inode) {
611 error = break_deleg_wait(&delegated_inode);
615 mnt_drop_write(path->mnt);
619 int vfs_fchmod(struct file *file, umode_t mode)
622 return chmod_common(&file->f_path, mode);
625 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
627 struct fd f = fdget(fd);
631 err = vfs_fchmod(f.file, mode);
637 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
641 unsigned int lookup_flags = LOOKUP_FOLLOW;
643 error = user_path_at(dfd, filename, lookup_flags, &path);
645 error = chmod_common(&path, mode);
647 if (retry_estale(error, lookup_flags)) {
648 lookup_flags |= LOOKUP_REVAL;
655 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
658 return do_fchmodat(dfd, filename, mode);
661 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
663 return do_fchmodat(AT_FDCWD, filename, mode);
667 * setattr_vfsuid - check and set ia_fsuid attribute
668 * @kuid: new inode owner
670 * Check whether @kuid is valid and if so generate and set vfsuid_t in
673 * Return: true if @kuid is valid, false if not.
675 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
677 if (!uid_valid(kuid))
679 attr->ia_valid |= ATTR_UID;
680 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
685 * setattr_vfsgid - check and set ia_fsgid attribute
686 * @kgid: new inode owner
688 * Check whether @kgid is valid and if so generate and set vfsgid_t in
691 * Return: true if @kgid is valid, false if not.
693 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
695 if (!gid_valid(kgid))
697 attr->ia_valid |= ATTR_GID;
698 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
702 int chown_common(const struct path *path, uid_t user, gid_t group)
704 struct user_namespace *mnt_userns, *fs_userns;
705 struct inode *inode = path->dentry->d_inode;
706 struct inode *delegated_inode = NULL;
708 struct iattr newattrs;
712 uid = make_kuid(current_user_ns(), user);
713 gid = make_kgid(current_user_ns(), group);
715 mnt_userns = mnt_user_ns(path->mnt);
716 fs_userns = i_user_ns(inode);
719 newattrs.ia_vfsuid = INVALID_VFSUID;
720 newattrs.ia_vfsgid = INVALID_VFSGID;
721 newattrs.ia_valid = ATTR_CTIME;
722 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
724 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
727 if (!S_ISDIR(inode->i_mode))
728 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
729 setattr_should_drop_sgid(mnt_userns, inode);
730 /* Continue to send actual fs values, not the mount values. */
731 error = security_path_chown(
733 from_vfsuid(mnt_userns, fs_userns, newattrs.ia_vfsuid),
734 from_vfsgid(mnt_userns, fs_userns, newattrs.ia_vfsgid));
736 error = notify_change(mnt_userns, path->dentry, &newattrs,
739 if (delegated_inode) {
740 error = break_deleg_wait(&delegated_inode);
747 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
754 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
757 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
758 if (flag & AT_EMPTY_PATH)
759 lookup_flags |= LOOKUP_EMPTY;
761 error = user_path_at(dfd, filename, lookup_flags, &path);
764 error = mnt_want_write(path.mnt);
767 error = chown_common(&path, user, group);
768 mnt_drop_write(path.mnt);
771 if (retry_estale(error, lookup_flags)) {
772 lookup_flags |= LOOKUP_REVAL;
779 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
780 gid_t, group, int, flag)
782 return do_fchownat(dfd, filename, user, group, flag);
785 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
787 return do_fchownat(AT_FDCWD, filename, user, group, 0);
790 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
792 return do_fchownat(AT_FDCWD, filename, user, group,
793 AT_SYMLINK_NOFOLLOW);
796 int vfs_fchown(struct file *file, uid_t user, gid_t group)
800 error = mnt_want_write_file(file);
804 error = chown_common(&file->f_path, user, group);
805 mnt_drop_write_file(file);
809 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
811 struct fd f = fdget(fd);
815 error = vfs_fchown(f.file, user, group);
821 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
823 return ksys_fchown(fd, user, group);
826 static int do_dentry_open(struct file *f,
828 int (*open)(struct inode *, struct file *))
830 static const struct file_operations empty_fops = {};
833 path_get(&f->f_path);
835 f->f_mapping = inode->i_mapping;
836 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
837 f->f_sb_err = file_sample_sb_err(f);
839 if (unlikely(f->f_flags & O_PATH)) {
840 f->f_mode = FMODE_PATH | FMODE_OPENED;
841 f->f_op = &empty_fops;
845 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
846 i_readcount_inc(inode);
847 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
848 error = get_write_access(inode);
851 error = __mnt_want_write(f->f_path.mnt);
852 if (unlikely(error)) {
853 put_write_access(inode);
856 f->f_mode |= FMODE_WRITER;
859 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
860 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
861 f->f_mode |= FMODE_ATOMIC_POS;
863 f->f_op = fops_get(inode->i_fop);
864 if (WARN_ON(!f->f_op)) {
869 error = security_file_open(f);
873 error = break_lease(locks_inode(f), f->f_flags);
877 /* normally all 3 are set; ->open() can clear them if needed */
878 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
880 open = f->f_op->open;
882 error = open(inode, f);
886 f->f_mode |= FMODE_OPENED;
887 if ((f->f_mode & FMODE_READ) &&
888 likely(f->f_op->read || f->f_op->read_iter))
889 f->f_mode |= FMODE_CAN_READ;
890 if ((f->f_mode & FMODE_WRITE) &&
891 likely(f->f_op->write || f->f_op->write_iter))
892 f->f_mode |= FMODE_CAN_WRITE;
893 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
894 f->f_mode &= ~FMODE_LSEEK;
895 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
896 f->f_mode |= FMODE_CAN_ODIRECT;
898 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
899 f->f_iocb_flags = iocb_flags(f);
901 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
903 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
907 * XXX: Huge page cache doesn't support writing yet. Drop all page
908 * cache for this file before processing writes.
910 if (f->f_mode & FMODE_WRITE) {
912 * Paired with smp_mb() in collapse_file() to ensure nr_thps
913 * is up to date and the update to i_writecount by
914 * get_write_access() is visible. Ensures subsequent insertion
915 * of THPs into the page cache will fail.
918 if (filemap_nr_thps(inode->i_mapping)) {
919 struct address_space *mapping = inode->i_mapping;
921 filemap_invalidate_lock(inode->i_mapping);
923 * unmap_mapping_range just need to be called once
924 * here, because the private pages is not need to be
925 * unmapped mapping (e.g. data segment of dynamic
926 * shared libraries here).
928 unmap_mapping_range(mapping, 0, 0, 0);
929 truncate_inode_pages(mapping, 0);
930 filemap_invalidate_unlock(inode->i_mapping);
937 if (WARN_ON_ONCE(error > 0))
942 path_put(&f->f_path);
943 f->f_path.mnt = NULL;
944 f->f_path.dentry = NULL;
950 * finish_open - finish opening a file
951 * @file: file pointer
952 * @dentry: pointer to dentry
953 * @open: open callback
954 * @opened: state of open
956 * This can be used to finish opening a file passed to i_op->atomic_open().
958 * If the open callback is set to NULL, then the standard f_op->open()
959 * filesystem callback is substituted.
961 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
962 * the return value of d_splice_alias(), then the caller needs to perform dput()
963 * on it after finish_open().
965 * Returns zero on success or -errno if the open failed.
967 int finish_open(struct file *file, struct dentry *dentry,
968 int (*open)(struct inode *, struct file *))
970 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
972 file->f_path.dentry = dentry;
973 return do_dentry_open(file, d_backing_inode(dentry), open);
975 EXPORT_SYMBOL(finish_open);
978 * finish_no_open - finish ->atomic_open() without opening the file
980 * @file: file pointer
981 * @dentry: dentry or NULL (as returned from ->lookup())
983 * This can be used to set the result of a successful lookup in ->atomic_open().
985 * NB: unlike finish_open() this function does consume the dentry reference and
986 * the caller need not dput() it.
988 * Returns "0" which must be the return value of ->atomic_open() after having
989 * called this function.
991 int finish_no_open(struct file *file, struct dentry *dentry)
993 file->f_path.dentry = dentry;
996 EXPORT_SYMBOL(finish_no_open);
998 char *file_path(struct file *filp, char *buf, int buflen)
1000 return d_path(&filp->f_path, buf, buflen);
1002 EXPORT_SYMBOL(file_path);
1005 * vfs_open - open the file at the given path
1006 * @path: path to open
1007 * @file: newly allocated file with f_flag initialized
1008 * @cred: credentials to use
1010 int vfs_open(const struct path *path, struct file *file)
1012 file->f_path = *path;
1013 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1016 struct file *dentry_open(const struct path *path, int flags,
1017 const struct cred *cred)
1022 validate_creds(cred);
1024 /* We must always pass in a valid mount pointer. */
1027 f = alloc_empty_file(flags, cred);
1029 error = vfs_open(path, f);
1037 EXPORT_SYMBOL(dentry_open);
1040 * dentry_create - Create and open a file
1041 * @path: path to create
1043 * @mode: mode bits for new file
1044 * @cred: credentials to use
1046 * Caller must hold the parent directory's lock, and have prepared
1047 * a negative dentry, placed in @path->dentry, for the new file.
1049 * Caller sets @path->mnt to the vfsmount of the filesystem where
1050 * the new file is to be created. The parent directory and the
1051 * negative dentry must reside on the same filesystem instance.
1053 * On success, returns a "struct file *". Otherwise a ERR_PTR
1056 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1057 const struct cred *cred)
1062 validate_creds(cred);
1063 f = alloc_empty_file(flags, cred);
1067 error = vfs_create(mnt_user_ns(path->mnt),
1068 d_inode(path->dentry->d_parent),
1069 path->dentry, mode, true);
1071 error = vfs_open(path, f);
1073 if (unlikely(error)) {
1075 return ERR_PTR(error);
1079 EXPORT_SYMBOL(dentry_create);
1081 struct file *open_with_fake_path(const struct path *path, int flags,
1082 struct inode *inode, const struct cred *cred)
1084 struct file *f = alloc_empty_file_noaccount(flags, cred);
1089 error = do_dentry_open(f, inode, NULL);
1097 EXPORT_SYMBOL(open_with_fake_path);
1099 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1100 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1102 inline struct open_how build_open_how(int flags, umode_t mode)
1104 struct open_how how = {
1105 .flags = flags & VALID_OPEN_FLAGS,
1106 .mode = mode & S_IALLUGO,
1109 /* O_PATH beats everything else. */
1110 if (how.flags & O_PATH)
1111 how.flags &= O_PATH_FLAGS;
1112 /* Modes should only be set for create-like flags. */
1113 if (!WILL_CREATE(how.flags))
1118 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1120 u64 flags = how->flags;
1121 u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1122 int lookup_flags = 0;
1123 int acc_mode = ACC_MODE(flags);
1125 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1126 "struct open_flags doesn't yet handle flags > 32 bits");
1129 * Strip flags that either shouldn't be set by userspace like
1130 * FMODE_NONOTIFY or that aren't relevant in determining struct
1131 * open_flags like O_CLOEXEC.
1136 * Older syscalls implicitly clear all of the invalid flags or argument
1137 * values before calling build_open_flags(), but openat2(2) checks all
1140 if (flags & ~VALID_OPEN_FLAGS)
1142 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1145 /* Scoping flags are mutually exclusive. */
1146 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1149 /* Deal with the mode. */
1150 if (WILL_CREATE(flags)) {
1151 if (how->mode & ~S_IALLUGO)
1153 op->mode = how->mode | S_IFREG;
1161 * In order to ensure programs get explicit errors when trying to use
1162 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1163 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1164 * have to require userspace to explicitly set it.
1166 if (flags & __O_TMPFILE) {
1167 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1169 if (!(acc_mode & MAY_WRITE))
1172 if (flags & O_PATH) {
1173 /* O_PATH only permits certain other flags to be set. */
1174 if (flags & ~O_PATH_FLAGS)
1180 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1181 * check for O_DSYNC if the need any syncing at all we enforce it's
1182 * always set instead of having to deal with possibly weird behaviour
1183 * for malicious applications setting only __O_SYNC.
1185 if (flags & __O_SYNC)
1188 op->open_flag = flags;
1190 /* O_TRUNC implies we need access checks for write permissions */
1191 if (flags & O_TRUNC)
1192 acc_mode |= MAY_WRITE;
1194 /* Allow the LSM permission hook to distinguish append
1195 access from general write access. */
1196 if (flags & O_APPEND)
1197 acc_mode |= MAY_APPEND;
1199 op->acc_mode = acc_mode;
1201 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1203 if (flags & O_CREAT) {
1204 op->intent |= LOOKUP_CREATE;
1205 if (flags & O_EXCL) {
1206 op->intent |= LOOKUP_EXCL;
1207 flags |= O_NOFOLLOW;
1211 if (flags & O_DIRECTORY)
1212 lookup_flags |= LOOKUP_DIRECTORY;
1213 if (!(flags & O_NOFOLLOW))
1214 lookup_flags |= LOOKUP_FOLLOW;
1216 if (how->resolve & RESOLVE_NO_XDEV)
1217 lookup_flags |= LOOKUP_NO_XDEV;
1218 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1219 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1220 if (how->resolve & RESOLVE_NO_SYMLINKS)
1221 lookup_flags |= LOOKUP_NO_SYMLINKS;
1222 if (how->resolve & RESOLVE_BENEATH)
1223 lookup_flags |= LOOKUP_BENEATH;
1224 if (how->resolve & RESOLVE_IN_ROOT)
1225 lookup_flags |= LOOKUP_IN_ROOT;
1226 if (how->resolve & RESOLVE_CACHED) {
1227 /* Don't bother even trying for create/truncate/tmpfile open */
1228 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1230 lookup_flags |= LOOKUP_CACHED;
1233 op->lookup_flags = lookup_flags;
1238 * file_open_name - open file and return file pointer
1240 * @name: struct filename containing path to open
1241 * @flags: open flags as per the open(2) second argument
1242 * @mode: mode for the new file if O_CREAT is set, else ignored
1244 * This is the helper to open a file from kernelspace if you really
1245 * have to. But in generally you should not do this, so please move
1246 * along, nothing to see here..
1248 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1250 struct open_flags op;
1251 struct open_how how = build_open_how(flags, mode);
1252 int err = build_open_flags(&how, &op);
1254 return ERR_PTR(err);
1255 return do_filp_open(AT_FDCWD, name, &op);
1259 * filp_open - open file and return file pointer
1261 * @filename: path to open
1262 * @flags: open flags as per the open(2) second argument
1263 * @mode: mode for the new file if O_CREAT is set, else ignored
1265 * This is the helper to open a file from kernelspace if you really
1266 * have to. But in generally you should not do this, so please move
1267 * along, nothing to see here..
1269 struct file *filp_open(const char *filename, int flags, umode_t mode)
1271 struct filename *name = getname_kernel(filename);
1272 struct file *file = ERR_CAST(name);
1274 if (!IS_ERR(name)) {
1275 file = file_open_name(name, flags, mode);
1280 EXPORT_SYMBOL(filp_open);
1282 struct file *file_open_root(const struct path *root,
1283 const char *filename, int flags, umode_t mode)
1285 struct open_flags op;
1286 struct open_how how = build_open_how(flags, mode);
1287 int err = build_open_flags(&how, &op);
1289 return ERR_PTR(err);
1290 return do_file_open_root(root, filename, &op);
1292 EXPORT_SYMBOL(file_open_root);
1294 static long do_sys_openat2(int dfd, const char __user *filename,
1295 struct open_how *how)
1297 struct open_flags op;
1298 int fd = build_open_flags(how, &op);
1299 struct filename *tmp;
1304 tmp = getname(filename);
1306 return PTR_ERR(tmp);
1308 fd = get_unused_fd_flags(how->flags);
1310 struct file *f = do_filp_open(dfd, tmp, &op);
1323 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1325 struct open_how how = build_open_how(flags, mode);
1326 return do_sys_openat2(dfd, filename, &how);
1330 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1332 if (force_o_largefile())
1333 flags |= O_LARGEFILE;
1334 return do_sys_open(AT_FDCWD, filename, flags, mode);
1337 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1340 if (force_o_largefile())
1341 flags |= O_LARGEFILE;
1342 return do_sys_open(dfd, filename, flags, mode);
1345 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1346 struct open_how __user *, how, size_t, usize)
1349 struct open_how tmp;
1351 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1352 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1354 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1357 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1361 audit_openat2_how(&tmp);
1363 /* O_LARGEFILE is only allowed for non-O_PATH. */
1364 if (!(tmp.flags & O_PATH) && force_o_largefile())
1365 tmp.flags |= O_LARGEFILE;
1367 return do_sys_openat2(dfd, filename, &tmp);
1370 #ifdef CONFIG_COMPAT
1372 * Exactly like sys_open(), except that it doesn't set the
1375 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1377 return do_sys_open(AT_FDCWD, filename, flags, mode);
1381 * Exactly like sys_openat(), except that it doesn't set the
1384 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1386 return do_sys_open(dfd, filename, flags, mode);
1393 * For backward compatibility? Maybe this should be moved
1394 * into arch/i386 instead?
1396 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1398 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1400 if (force_o_largefile())
1401 flags |= O_LARGEFILE;
1402 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1407 * "id" is the POSIX thread ID. We use the
1408 * files pointer for this..
1410 int filp_close(struct file *filp, fl_owner_t id)
1414 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1415 "VFS: Close: file count is 0 (f_op=%ps)",
1420 if (filp->f_op->flush)
1421 retval = filp->f_op->flush(filp, id);
1423 if (likely(!(filp->f_mode & FMODE_PATH))) {
1424 dnotify_flush(filp, id);
1425 locks_remove_posix(filp, id);
1431 EXPORT_SYMBOL(filp_close);
1434 * Careful here! We test whether the file pointer is NULL before
1435 * releasing the fd. This ensures that one clone task can't release
1436 * an fd while another clone is opening it.
1438 SYSCALL_DEFINE1(close, unsigned int, fd)
1440 int retval = close_fd(fd);
1442 /* can't restart close syscall because file table entry was cleared */
1443 if (unlikely(retval == -ERESTARTSYS ||
1444 retval == -ERESTARTNOINTR ||
1445 retval == -ERESTARTNOHAND ||
1446 retval == -ERESTART_RESTARTBLOCK))
1453 * close_range() - Close all file descriptors in a given range.
1455 * @fd: starting file descriptor to close
1456 * @max_fd: last file descriptor to close
1457 * @flags: reserved for future extensions
1459 * This closes a range of file descriptors. All file descriptors
1460 * from @fd up to and including @max_fd are closed.
1461 * Currently, errors to close a given file descriptor are ignored.
1463 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1464 unsigned int, flags)
1466 return __close_range(fd, max_fd, flags);
1470 * This routine simulates a hangup on the tty, to arrange that users
1471 * are given clean terminals at login time.
1473 SYSCALL_DEFINE0(vhangup)
1475 if (capable(CAP_SYS_TTY_CONFIG)) {
1483 * Called when an inode is about to be open.
1484 * We use this to disallow opening large files on 32bit systems if
1485 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1486 * on this flag in sys_open.
1488 int generic_file_open(struct inode * inode, struct file * filp)
1490 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1495 EXPORT_SYMBOL(generic_file_open);
1498 * This is used by subsystems that don't want seekable
1499 * file descriptors. The function is not supposed to ever fail, the only
1500 * reason it returns an 'int' and not 'void' is so that it can be plugged
1501 * directly into file_operations structure.
1503 int nonseekable_open(struct inode *inode, struct file *filp)
1505 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1509 EXPORT_SYMBOL(nonseekable_open);
1512 * stream_open is used by subsystems that want stream-like file descriptors.
1513 * Such file descriptors are not seekable and don't have notion of position
1514 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1515 * Contrary to file descriptors of other regular files, .read() and .write()
1516 * can run simultaneously.
1518 * stream_open never fails and is marked to return int so that it could be
1519 * directly used as file_operations.open .
1521 int stream_open(struct inode *inode, struct file *filp)
1523 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1524 filp->f_mode |= FMODE_STREAM;
1528 EXPORT_SYMBOL(stream_open);