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>
36 #include <linux/filelock.h>
40 int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
41 loff_t length, unsigned int time_attrs, struct file *filp)
44 struct iattr newattrs;
46 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
50 newattrs.ia_size = length;
51 newattrs.ia_valid = ATTR_SIZE | time_attrs;
53 newattrs.ia_file = filp;
54 newattrs.ia_valid |= ATTR_FILE;
57 /* Remove suid, sgid, and file capabilities on truncate too */
58 ret = dentry_needs_remove_privs(idmap, dentry);
62 newattrs.ia_valid |= ret | ATTR_FORCE;
64 inode_lock(dentry->d_inode);
65 /* Note any delegations or leases have already been broken: */
66 ret = notify_change(idmap, dentry, &newattrs, NULL);
67 inode_unlock(dentry->d_inode);
71 long vfs_truncate(const struct path *path, loff_t length)
73 struct mnt_idmap *idmap;
77 inode = path->dentry->d_inode;
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80 if (S_ISDIR(inode->i_mode))
82 if (!S_ISREG(inode->i_mode))
85 error = mnt_want_write(path->mnt);
89 idmap = mnt_idmap(path->mnt);
90 error = inode_permission(idmap, inode, MAY_WRITE);
92 goto mnt_drop_write_and_out;
96 goto mnt_drop_write_and_out;
98 error = get_write_access(inode);
100 goto mnt_drop_write_and_out;
103 * Make sure that there are no leases. get_write_access() protects
104 * against the truncate racing with a lease-granting setlease().
106 error = break_lease(inode, O_WRONLY);
108 goto put_write_and_out;
110 error = security_path_truncate(path);
112 error = do_truncate(idmap, path->dentry, length, 0, NULL);
115 put_write_access(inode);
116 mnt_drop_write_and_out:
117 mnt_drop_write(path->mnt);
121 EXPORT_SYMBOL_GPL(vfs_truncate);
123 long do_sys_truncate(const char __user *pathname, loff_t length)
125 unsigned int lookup_flags = LOOKUP_FOLLOW;
129 if (length < 0) /* sorry, but loff_t says... */
133 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
135 error = vfs_truncate(&path, length);
138 if (retry_estale(error, lookup_flags)) {
139 lookup_flags |= LOOKUP_REVAL;
145 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
147 return do_sys_truncate(path, length);
151 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
153 return do_sys_truncate(path, length);
157 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
160 struct dentry *dentry;
172 /* explicitly opened as large or we are on 64-bit box */
173 if (f.file->f_flags & O_LARGEFILE)
176 dentry = f.file->f_path.dentry;
177 inode = dentry->d_inode;
179 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
183 /* Cannot ftruncate over 2^31 bytes without large file support */
184 if (small && length > MAX_NON_LFS)
188 /* Check IS_APPEND on real upper inode */
189 if (IS_APPEND(file_inode(f.file)))
191 sb_start_write(inode->i_sb);
192 error = security_file_truncate(f.file);
194 error = do_truncate(file_mnt_idmap(f.file), dentry, length,
195 ATTR_MTIME | ATTR_CTIME, f.file);
196 sb_end_write(inode->i_sb);
203 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
205 return do_sys_ftruncate(fd, length, 1);
209 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
211 return do_sys_ftruncate(fd, length, 1);
215 /* LFS versions of truncate are only needed on 32 bit machines */
216 #if BITS_PER_LONG == 32
217 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
219 return do_sys_truncate(path, length);
222 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
224 return do_sys_ftruncate(fd, length, 0);
226 #endif /* BITS_PER_LONG == 32 */
228 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
229 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
230 compat_arg_u64_dual(length))
232 return ksys_truncate(pathname, compat_arg_u64_glue(length));
236 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
237 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
238 compat_arg_u64_dual(length))
240 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
244 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
246 struct inode *inode = file_inode(file);
249 if (offset < 0 || len <= 0)
252 /* Return error if mode is not supported */
253 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
256 /* Punch hole and zero range are mutually exclusive */
257 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
258 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
261 /* Punch hole must have keep size set */
262 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
263 !(mode & FALLOC_FL_KEEP_SIZE))
266 /* Collapse range should only be used exclusively. */
267 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
268 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
271 /* Insert range should only be used exclusively. */
272 if ((mode & FALLOC_FL_INSERT_RANGE) &&
273 (mode & ~FALLOC_FL_INSERT_RANGE))
276 /* Unshare range should only be used with allocate mode. */
277 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
278 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
281 if (!(file->f_mode & FMODE_WRITE))
285 * We can only allow pure fallocate on append only files
287 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
290 if (IS_IMMUTABLE(inode))
294 * We cannot allow any fallocate operation on an active swapfile
296 if (IS_SWAPFILE(inode))
300 * Revalidate the write permissions, in case security policy has
301 * changed since the files were opened.
303 ret = security_file_permission(file, MAY_WRITE);
307 if (S_ISFIFO(inode->i_mode))
310 if (S_ISDIR(inode->i_mode))
313 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
316 /* Check for wrap through zero too */
317 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
320 if (!file->f_op->fallocate)
323 file_start_write(file);
324 ret = file->f_op->fallocate(file, mode, offset, len);
327 * Create inotify and fanotify events.
329 * To keep the logic simple always create events if fallocate succeeds.
330 * This implies that events are even created if the file size remains
331 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
334 fsnotify_modify(file);
336 file_end_write(file);
339 EXPORT_SYMBOL_GPL(vfs_fallocate);
341 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
343 struct fd f = fdget(fd);
347 error = vfs_fallocate(f.file, mode, offset, len);
353 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
355 return ksys_fallocate(fd, mode, offset, len);
358 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
359 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
360 compat_arg_u64_dual(len))
362 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
363 compat_arg_u64_glue(len));
368 * access() needs to use the real uid/gid, not the effective uid/gid.
369 * We do this by temporarily clearing all FS-related capabilities and
370 * switching the fsuid/fsgid around to the real ones.
372 * Creating new credentials is expensive, so we try to skip doing it,
373 * which we can if the result would match what we already got.
375 static bool access_need_override_creds(int flags)
377 const struct cred *cred;
379 if (flags & AT_EACCESS)
382 cred = current_cred();
383 if (!uid_eq(cred->fsuid, cred->uid) ||
384 !gid_eq(cred->fsgid, cred->gid))
387 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
388 kuid_t root_uid = make_kuid(cred->user_ns, 0);
389 if (!uid_eq(cred->uid, root_uid)) {
390 if (!cap_isclear(cred->cap_effective))
393 if (!cap_isidentical(cred->cap_effective,
394 cred->cap_permitted))
402 static const struct cred *access_override_creds(void)
404 const struct cred *old_cred;
405 struct cred *override_cred;
407 override_cred = prepare_creds();
412 * XXX access_need_override_creds performs checks in hopes of skipping
413 * this work. Make sure it stays in sync if making any changes in this
417 override_cred->fsuid = override_cred->uid;
418 override_cred->fsgid = override_cred->gid;
420 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
421 /* Clear the capabilities if we switch to a non-root user */
422 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
423 if (!uid_eq(override_cred->uid, root_uid))
424 cap_clear(override_cred->cap_effective);
426 override_cred->cap_effective =
427 override_cred->cap_permitted;
431 * The new set of credentials can *only* be used in
432 * task-synchronous circumstances, and does not need
433 * RCU freeing, unless somebody then takes a separate
436 * NOTE! This is _only_ true because this credential
437 * is used purely for override_creds() that installs
438 * it as the subjective cred. Other threads will be
439 * accessing ->real_cred, not the subjective cred.
441 * If somebody _does_ make a copy of this (using the
442 * 'get_current_cred()' function), that will clear the
443 * non_rcu field, because now that other user may be
444 * expecting RCU freeing. But normal thread-synchronous
445 * cred accesses will keep things non-RCY.
447 override_cred->non_rcu = 1;
449 old_cred = override_creds(override_cred);
451 /* override_cred() gets its own ref */
452 put_cred(override_cred);
457 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
462 unsigned int lookup_flags = LOOKUP_FOLLOW;
463 const struct cred *old_cred = NULL;
465 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
468 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
471 if (flags & AT_SYMLINK_NOFOLLOW)
472 lookup_flags &= ~LOOKUP_FOLLOW;
473 if (flags & AT_EMPTY_PATH)
474 lookup_flags |= LOOKUP_EMPTY;
476 if (access_need_override_creds(flags)) {
477 old_cred = access_override_creds();
483 res = user_path_at(dfd, filename, lookup_flags, &path);
487 inode = d_backing_inode(path.dentry);
489 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
491 * MAY_EXEC on regular files is denied if the fs is mounted
492 * with the "noexec" flag.
495 if (path_noexec(&path))
496 goto out_path_release;
499 res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
500 /* SuS v2 requires we report a read only fs too */
501 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
502 goto out_path_release;
504 * This is a rare case where using __mnt_is_readonly()
505 * is OK without a mnt_want/drop_write() pair. Since
506 * no actual write to the fs is performed here, we do
507 * not need to telegraph to that to anyone.
509 * By doing this, we accept that this access is
510 * inherently racy and know that the fs may change
511 * state before we even see this result.
513 if (__mnt_is_readonly(path.mnt))
518 if (retry_estale(res, lookup_flags)) {
519 lookup_flags |= LOOKUP_REVAL;
524 revert_creds(old_cred);
529 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
531 return do_faccessat(dfd, filename, mode, 0);
534 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
537 return do_faccessat(dfd, filename, mode, flags);
540 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
542 return do_faccessat(AT_FDCWD, filename, mode, 0);
545 SYSCALL_DEFINE1(chdir, const char __user *, filename)
549 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
551 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
555 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
559 set_fs_pwd(current->fs, &path);
563 if (retry_estale(error, lookup_flags)) {
564 lookup_flags |= LOOKUP_REVAL;
571 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
573 struct fd f = fdget_raw(fd);
581 if (!d_can_lookup(f.file->f_path.dentry))
584 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
586 set_fs_pwd(current->fs, &f.file->f_path);
593 SYSCALL_DEFINE1(chroot, const char __user *, filename)
597 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
599 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
603 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
608 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
610 error = security_path_chroot(&path);
614 set_fs_root(current->fs, &path);
618 if (retry_estale(error, lookup_flags)) {
619 lookup_flags |= LOOKUP_REVAL;
626 int chmod_common(const struct path *path, umode_t mode)
628 struct inode *inode = path->dentry->d_inode;
629 struct inode *delegated_inode = NULL;
630 struct iattr newattrs;
633 error = mnt_want_write(path->mnt);
638 error = security_path_chmod(path, mode);
641 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
642 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
643 error = notify_change(mnt_idmap(path->mnt), path->dentry,
644 &newattrs, &delegated_inode);
647 if (delegated_inode) {
648 error = break_deleg_wait(&delegated_inode);
652 mnt_drop_write(path->mnt);
656 int vfs_fchmod(struct file *file, umode_t mode)
659 return chmod_common(&file->f_path, mode);
662 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
664 struct fd f = fdget(fd);
668 err = vfs_fchmod(f.file, mode);
674 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
679 unsigned int lookup_flags;
681 if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
684 lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
685 if (flags & AT_EMPTY_PATH)
686 lookup_flags |= LOOKUP_EMPTY;
689 error = user_path_at(dfd, filename, lookup_flags, &path);
691 error = chmod_common(&path, mode);
693 if (retry_estale(error, lookup_flags)) {
694 lookup_flags |= LOOKUP_REVAL;
701 SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
702 umode_t, mode, unsigned int, flags)
704 return do_fchmodat(dfd, filename, mode, flags);
707 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
710 return do_fchmodat(dfd, filename, mode, 0);
713 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
715 return do_fchmodat(AT_FDCWD, filename, mode, 0);
719 * Check whether @kuid is valid and if so generate and set vfsuid_t in
722 * Return: true if @kuid is valid, false if not.
724 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
726 if (!uid_valid(kuid))
728 attr->ia_valid |= ATTR_UID;
729 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
734 * Check whether @kgid is valid and if so generate and set vfsgid_t in
737 * Return: true if @kgid is valid, false if not.
739 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
741 if (!gid_valid(kgid))
743 attr->ia_valid |= ATTR_GID;
744 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
748 int chown_common(const struct path *path, uid_t user, gid_t group)
750 struct mnt_idmap *idmap;
751 struct user_namespace *fs_userns;
752 struct inode *inode = path->dentry->d_inode;
753 struct inode *delegated_inode = NULL;
755 struct iattr newattrs;
759 uid = make_kuid(current_user_ns(), user);
760 gid = make_kgid(current_user_ns(), group);
762 idmap = mnt_idmap(path->mnt);
763 fs_userns = i_user_ns(inode);
766 newattrs.ia_vfsuid = INVALID_VFSUID;
767 newattrs.ia_vfsgid = INVALID_VFSGID;
768 newattrs.ia_valid = ATTR_CTIME;
769 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
771 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
774 if (!S_ISDIR(inode->i_mode))
775 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
776 setattr_should_drop_sgid(idmap, inode);
777 /* Continue to send actual fs values, not the mount values. */
778 error = security_path_chown(
780 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
781 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
783 error = notify_change(idmap, path->dentry, &newattrs,
786 if (delegated_inode) {
787 error = break_deleg_wait(&delegated_inode);
794 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
801 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
804 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
805 if (flag & AT_EMPTY_PATH)
806 lookup_flags |= LOOKUP_EMPTY;
808 error = user_path_at(dfd, filename, lookup_flags, &path);
811 error = mnt_want_write(path.mnt);
814 error = chown_common(&path, user, group);
815 mnt_drop_write(path.mnt);
818 if (retry_estale(error, lookup_flags)) {
819 lookup_flags |= LOOKUP_REVAL;
826 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
827 gid_t, group, int, flag)
829 return do_fchownat(dfd, filename, user, group, flag);
832 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
834 return do_fchownat(AT_FDCWD, filename, user, group, 0);
837 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
839 return do_fchownat(AT_FDCWD, filename, user, group,
840 AT_SYMLINK_NOFOLLOW);
843 int vfs_fchown(struct file *file, uid_t user, gid_t group)
847 error = mnt_want_write_file(file);
851 error = chown_common(&file->f_path, user, group);
852 mnt_drop_write_file(file);
856 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
858 struct fd f = fdget(fd);
862 error = vfs_fchown(f.file, user, group);
868 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
870 return ksys_fchown(fd, user, group);
873 static int do_dentry_open(struct file *f,
875 int (*open)(struct inode *, struct file *))
877 static const struct file_operations empty_fops = {};
880 path_get(&f->f_path);
882 f->f_mapping = inode->i_mapping;
883 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
884 f->f_sb_err = file_sample_sb_err(f);
886 if (unlikely(f->f_flags & O_PATH)) {
887 f->f_mode = FMODE_PATH | FMODE_OPENED;
888 f->f_op = &empty_fops;
892 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
893 i_readcount_inc(inode);
894 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
895 error = get_write_access(inode);
898 error = __mnt_want_write(f->f_path.mnt);
899 if (unlikely(error)) {
900 put_write_access(inode);
903 f->f_mode |= FMODE_WRITER;
906 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
907 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
908 f->f_mode |= FMODE_ATOMIC_POS;
910 f->f_op = fops_get(inode->i_fop);
911 if (WARN_ON(!f->f_op)) {
916 error = security_file_open(f);
920 error = break_lease(file_inode(f), f->f_flags);
924 /* normally all 3 are set; ->open() can clear them if needed */
925 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
927 open = f->f_op->open;
929 error = open(inode, f);
933 f->f_mode |= FMODE_OPENED;
934 if ((f->f_mode & FMODE_READ) &&
935 likely(f->f_op->read || f->f_op->read_iter))
936 f->f_mode |= FMODE_CAN_READ;
937 if ((f->f_mode & FMODE_WRITE) &&
938 likely(f->f_op->write || f->f_op->write_iter))
939 f->f_mode |= FMODE_CAN_WRITE;
940 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
941 f->f_mode &= ~FMODE_LSEEK;
942 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
943 f->f_mode |= FMODE_CAN_ODIRECT;
945 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
946 f->f_iocb_flags = iocb_flags(f);
948 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
950 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
954 * XXX: Huge page cache doesn't support writing yet. Drop all page
955 * cache for this file before processing writes.
957 if (f->f_mode & FMODE_WRITE) {
959 * Paired with smp_mb() in collapse_file() to ensure nr_thps
960 * is up to date and the update to i_writecount by
961 * get_write_access() is visible. Ensures subsequent insertion
962 * of THPs into the page cache will fail.
965 if (filemap_nr_thps(inode->i_mapping)) {
966 struct address_space *mapping = inode->i_mapping;
968 filemap_invalidate_lock(inode->i_mapping);
970 * unmap_mapping_range just need to be called once
971 * here, because the private pages is not need to be
972 * unmapped mapping (e.g. data segment of dynamic
973 * shared libraries here).
975 unmap_mapping_range(mapping, 0, 0, 0);
976 truncate_inode_pages(mapping, 0);
977 filemap_invalidate_unlock(inode->i_mapping);
982 * Once we return a file with FMODE_OPENED, __fput() will call
983 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
989 if (WARN_ON_ONCE(error > 0))
994 path_put(&f->f_path);
995 f->f_path.mnt = NULL;
996 f->f_path.dentry = NULL;
1002 * finish_open - finish opening a file
1003 * @file: file pointer
1004 * @dentry: pointer to dentry
1005 * @open: open callback
1007 * This can be used to finish opening a file passed to i_op->atomic_open().
1009 * If the open callback is set to NULL, then the standard f_op->open()
1010 * filesystem callback is substituted.
1012 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1013 * the return value of d_splice_alias(), then the caller needs to perform dput()
1014 * on it after finish_open().
1016 * Returns zero on success or -errno if the open failed.
1018 int finish_open(struct file *file, struct dentry *dentry,
1019 int (*open)(struct inode *, struct file *))
1021 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1023 file->f_path.dentry = dentry;
1024 return do_dentry_open(file, d_backing_inode(dentry), open);
1026 EXPORT_SYMBOL(finish_open);
1029 * finish_no_open - finish ->atomic_open() without opening the file
1031 * @file: file pointer
1032 * @dentry: dentry or NULL (as returned from ->lookup())
1034 * This can be used to set the result of a successful lookup in ->atomic_open().
1036 * NB: unlike finish_open() this function does consume the dentry reference and
1037 * the caller need not dput() it.
1039 * Returns "0" which must be the return value of ->atomic_open() after having
1040 * called this function.
1042 int finish_no_open(struct file *file, struct dentry *dentry)
1044 file->f_path.dentry = dentry;
1047 EXPORT_SYMBOL(finish_no_open);
1049 char *file_path(struct file *filp, char *buf, int buflen)
1051 return d_path(&filp->f_path, buf, buflen);
1053 EXPORT_SYMBOL(file_path);
1056 * vfs_open - open the file at the given path
1057 * @path: path to open
1058 * @file: newly allocated file with f_flag initialized
1060 int vfs_open(const struct path *path, struct file *file)
1062 file->f_path = *path;
1063 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1066 struct file *dentry_open(const struct path *path, int flags,
1067 const struct cred *cred)
1072 /* We must always pass in a valid mount pointer. */
1075 f = alloc_empty_file(flags, cred);
1077 error = vfs_open(path, f);
1085 EXPORT_SYMBOL(dentry_open);
1088 * dentry_create - Create and open a file
1089 * @path: path to create
1091 * @mode: mode bits for new file
1092 * @cred: credentials to use
1094 * Caller must hold the parent directory's lock, and have prepared
1095 * a negative dentry, placed in @path->dentry, for the new file.
1097 * Caller sets @path->mnt to the vfsmount of the filesystem where
1098 * the new file is to be created. The parent directory and the
1099 * negative dentry must reside on the same filesystem instance.
1101 * On success, returns a "struct file *". Otherwise a ERR_PTR
1104 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1105 const struct cred *cred)
1110 f = alloc_empty_file(flags, cred);
1114 error = vfs_create(mnt_idmap(path->mnt),
1115 d_inode(path->dentry->d_parent),
1116 path->dentry, mode, true);
1118 error = vfs_open(path, f);
1120 if (unlikely(error)) {
1122 return ERR_PTR(error);
1126 EXPORT_SYMBOL(dentry_create);
1129 * kernel_file_open - open a file for kernel internal use
1130 * @path: path of the file to open
1131 * @flags: open flags
1133 * @cred: credentials for open
1135 * Open a file for use by in-kernel consumers. The file is not accounted
1136 * against nr_files and must not be installed into the file descriptor
1139 * Return: Opened file on success, an error pointer on failure.
1141 struct file *kernel_file_open(const struct path *path, int flags,
1142 struct inode *inode, const struct cred *cred)
1147 f = alloc_empty_file_noaccount(flags, cred);
1152 error = do_dentry_open(f, inode, NULL);
1159 EXPORT_SYMBOL_GPL(kernel_file_open);
1162 * backing_file_open - open a backing file for kernel internal use
1163 * @path: path of the file to open
1164 * @flags: open flags
1165 * @real_path: path of the backing file
1166 * @cred: credentials for open
1168 * Open a backing file for a stackable filesystem (e.g., overlayfs).
1169 * @path may be on the stackable filesystem and backing inode on the
1170 * underlying filesystem. In this case, we want to be able to return
1171 * the @real_path of the backing inode. This is done by embedding the
1172 * returned file into a container structure that also stores the path of
1173 * the backing inode on the underlying filesystem, which can be
1174 * retrieved using backing_file_real_path().
1176 struct file *backing_file_open(const struct path *path, int flags,
1177 const struct path *real_path,
1178 const struct cred *cred)
1183 f = alloc_empty_backing_file(flags, cred);
1188 path_get(real_path);
1189 *backing_file_real_path(f) = *real_path;
1190 error = do_dentry_open(f, d_inode(real_path->dentry), NULL);
1198 EXPORT_SYMBOL_GPL(backing_file_open);
1200 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1201 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1203 inline struct open_how build_open_how(int flags, umode_t mode)
1205 struct open_how how = {
1206 .flags = flags & VALID_OPEN_FLAGS,
1207 .mode = mode & S_IALLUGO,
1210 /* O_PATH beats everything else. */
1211 if (how.flags & O_PATH)
1212 how.flags &= O_PATH_FLAGS;
1213 /* Modes should only be set for create-like flags. */
1214 if (!WILL_CREATE(how.flags))
1219 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1221 u64 flags = how->flags;
1222 u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1223 int lookup_flags = 0;
1224 int acc_mode = ACC_MODE(flags);
1226 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1227 "struct open_flags doesn't yet handle flags > 32 bits");
1230 * Strip flags that either shouldn't be set by userspace like
1231 * FMODE_NONOTIFY or that aren't relevant in determining struct
1232 * open_flags like O_CLOEXEC.
1237 * Older syscalls implicitly clear all of the invalid flags or argument
1238 * values before calling build_open_flags(), but openat2(2) checks all
1241 if (flags & ~VALID_OPEN_FLAGS)
1243 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1246 /* Scoping flags are mutually exclusive. */
1247 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1250 /* Deal with the mode. */
1251 if (WILL_CREATE(flags)) {
1252 if (how->mode & ~S_IALLUGO)
1254 op->mode = how->mode | S_IFREG;
1262 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1263 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1264 * O_TMPFILE below which requires O_DIRECTORY being raised.
1266 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1269 /* Now handle the creative implementation of O_TMPFILE. */
1270 if (flags & __O_TMPFILE) {
1272 * In order to ensure programs get explicit errors when trying
1273 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1274 * is raised alongside __O_TMPFILE.
1276 if (!(flags & O_DIRECTORY))
1278 if (!(acc_mode & MAY_WRITE))
1281 if (flags & O_PATH) {
1282 /* O_PATH only permits certain other flags to be set. */
1283 if (flags & ~O_PATH_FLAGS)
1289 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1290 * check for O_DSYNC if the need any syncing at all we enforce it's
1291 * always set instead of having to deal with possibly weird behaviour
1292 * for malicious applications setting only __O_SYNC.
1294 if (flags & __O_SYNC)
1297 op->open_flag = flags;
1299 /* O_TRUNC implies we need access checks for write permissions */
1300 if (flags & O_TRUNC)
1301 acc_mode |= MAY_WRITE;
1303 /* Allow the LSM permission hook to distinguish append
1304 access from general write access. */
1305 if (flags & O_APPEND)
1306 acc_mode |= MAY_APPEND;
1308 op->acc_mode = acc_mode;
1310 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1312 if (flags & O_CREAT) {
1313 op->intent |= LOOKUP_CREATE;
1314 if (flags & O_EXCL) {
1315 op->intent |= LOOKUP_EXCL;
1316 flags |= O_NOFOLLOW;
1320 if (flags & O_DIRECTORY)
1321 lookup_flags |= LOOKUP_DIRECTORY;
1322 if (!(flags & O_NOFOLLOW))
1323 lookup_flags |= LOOKUP_FOLLOW;
1325 if (how->resolve & RESOLVE_NO_XDEV)
1326 lookup_flags |= LOOKUP_NO_XDEV;
1327 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1328 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1329 if (how->resolve & RESOLVE_NO_SYMLINKS)
1330 lookup_flags |= LOOKUP_NO_SYMLINKS;
1331 if (how->resolve & RESOLVE_BENEATH)
1332 lookup_flags |= LOOKUP_BENEATH;
1333 if (how->resolve & RESOLVE_IN_ROOT)
1334 lookup_flags |= LOOKUP_IN_ROOT;
1335 if (how->resolve & RESOLVE_CACHED) {
1336 /* Don't bother even trying for create/truncate/tmpfile open */
1337 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1339 lookup_flags |= LOOKUP_CACHED;
1342 op->lookup_flags = lookup_flags;
1347 * file_open_name - open file and return file pointer
1349 * @name: struct filename containing path to open
1350 * @flags: open flags as per the open(2) second argument
1351 * @mode: mode for the new file if O_CREAT is set, else ignored
1353 * This is the helper to open a file from kernelspace if you really
1354 * have to. But in generally you should not do this, so please move
1355 * along, nothing to see here..
1357 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1359 struct open_flags op;
1360 struct open_how how = build_open_how(flags, mode);
1361 int err = build_open_flags(&how, &op);
1363 return ERR_PTR(err);
1364 return do_filp_open(AT_FDCWD, name, &op);
1368 * filp_open - open file and return file pointer
1370 * @filename: path to open
1371 * @flags: open flags as per the open(2) second argument
1372 * @mode: mode for the new file if O_CREAT is set, else ignored
1374 * This is the helper to open a file from kernelspace if you really
1375 * have to. But in generally you should not do this, so please move
1376 * along, nothing to see here..
1378 struct file *filp_open(const char *filename, int flags, umode_t mode)
1380 struct filename *name = getname_kernel(filename);
1381 struct file *file = ERR_CAST(name);
1383 if (!IS_ERR(name)) {
1384 file = file_open_name(name, flags, mode);
1389 EXPORT_SYMBOL(filp_open);
1391 struct file *file_open_root(const struct path *root,
1392 const char *filename, int flags, umode_t mode)
1394 struct open_flags op;
1395 struct open_how how = build_open_how(flags, mode);
1396 int err = build_open_flags(&how, &op);
1398 return ERR_PTR(err);
1399 return do_file_open_root(root, filename, &op);
1401 EXPORT_SYMBOL(file_open_root);
1403 static long do_sys_openat2(int dfd, const char __user *filename,
1404 struct open_how *how)
1406 struct open_flags op;
1407 int fd = build_open_flags(how, &op);
1408 struct filename *tmp;
1413 tmp = getname(filename);
1415 return PTR_ERR(tmp);
1417 fd = get_unused_fd_flags(how->flags);
1419 struct file *f = do_filp_open(dfd, tmp, &op);
1431 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1433 struct open_how how = build_open_how(flags, mode);
1434 return do_sys_openat2(dfd, filename, &how);
1438 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1440 if (force_o_largefile())
1441 flags |= O_LARGEFILE;
1442 return do_sys_open(AT_FDCWD, filename, flags, mode);
1445 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1448 if (force_o_largefile())
1449 flags |= O_LARGEFILE;
1450 return do_sys_open(dfd, filename, flags, mode);
1453 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1454 struct open_how __user *, how, size_t, usize)
1457 struct open_how tmp;
1459 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1460 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1462 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1465 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1469 audit_openat2_how(&tmp);
1471 /* O_LARGEFILE is only allowed for non-O_PATH. */
1472 if (!(tmp.flags & O_PATH) && force_o_largefile())
1473 tmp.flags |= O_LARGEFILE;
1475 return do_sys_openat2(dfd, filename, &tmp);
1478 #ifdef CONFIG_COMPAT
1480 * Exactly like sys_open(), except that it doesn't set the
1483 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1485 return do_sys_open(AT_FDCWD, filename, flags, mode);
1489 * Exactly like sys_openat(), except that it doesn't set the
1492 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1494 return do_sys_open(dfd, filename, flags, mode);
1501 * For backward compatibility? Maybe this should be moved
1502 * into arch/i386 instead?
1504 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1506 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1508 if (force_o_largefile())
1509 flags |= O_LARGEFILE;
1510 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1515 * "id" is the POSIX thread ID. We use the
1516 * files pointer for this..
1518 static int filp_flush(struct file *filp, fl_owner_t id)
1522 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1523 "VFS: Close: file count is 0 (f_op=%ps)",
1528 if (filp->f_op->flush)
1529 retval = filp->f_op->flush(filp, id);
1531 if (likely(!(filp->f_mode & FMODE_PATH))) {
1532 dnotify_flush(filp, id);
1533 locks_remove_posix(filp, id);
1538 int filp_close(struct file *filp, fl_owner_t id)
1542 retval = filp_flush(filp, id);
1547 EXPORT_SYMBOL(filp_close);
1550 * Careful here! We test whether the file pointer is NULL before
1551 * releasing the fd. This ensures that one clone task can't release
1552 * an fd while another clone is opening it.
1554 SYSCALL_DEFINE1(close, unsigned int, fd)
1559 file = close_fd_get_file(fd);
1563 retval = filp_flush(file, current->files);
1566 * We're returning to user space. Don't bother
1567 * with any delayed fput() cases.
1571 /* can't restart close syscall because file table entry was cleared */
1572 if (unlikely(retval == -ERESTARTSYS ||
1573 retval == -ERESTARTNOINTR ||
1574 retval == -ERESTARTNOHAND ||
1575 retval == -ERESTART_RESTARTBLOCK))
1582 * sys_close_range() - Close all file descriptors in a given range.
1584 * @fd: starting file descriptor to close
1585 * @max_fd: last file descriptor to close
1586 * @flags: reserved for future extensions
1588 * This closes a range of file descriptors. All file descriptors
1589 * from @fd up to and including @max_fd are closed.
1590 * Currently, errors to close a given file descriptor are ignored.
1592 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1593 unsigned int, flags)
1595 return __close_range(fd, max_fd, flags);
1599 * This routine simulates a hangup on the tty, to arrange that users
1600 * are given clean terminals at login time.
1602 SYSCALL_DEFINE0(vhangup)
1604 if (capable(CAP_SYS_TTY_CONFIG)) {
1612 * Called when an inode is about to be open.
1613 * We use this to disallow opening large files on 32bit systems if
1614 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1615 * on this flag in sys_open.
1617 int generic_file_open(struct inode * inode, struct file * filp)
1619 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1624 EXPORT_SYMBOL(generic_file_open);
1627 * This is used by subsystems that don't want seekable
1628 * file descriptors. The function is not supposed to ever fail, the only
1629 * reason it returns an 'int' and not 'void' is so that it can be plugged
1630 * directly into file_operations structure.
1632 int nonseekable_open(struct inode *inode, struct file *filp)
1634 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1638 EXPORT_SYMBOL(nonseekable_open);
1641 * stream_open is used by subsystems that want stream-like file descriptors.
1642 * Such file descriptors are not seekable and don't have notion of position
1643 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1644 * Contrary to file descriptors of other regular files, .read() and .write()
1645 * can run simultaneously.
1647 * stream_open never fails and is marked to return int so that it could be
1648 * directly used as file_operations.open .
1650 int stream_open(struct inode *inode, struct file *filp)
1652 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1653 filp->f_mode |= FMODE_STREAM;
1657 EXPORT_SYMBOL(stream_open);