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)
678 unsigned int lookup_flags = LOOKUP_FOLLOW;
680 error = user_path_at(dfd, filename, lookup_flags, &path);
682 error = chmod_common(&path, mode);
684 if (retry_estale(error, lookup_flags)) {
685 lookup_flags |= LOOKUP_REVAL;
692 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
695 return do_fchmodat(dfd, filename, mode);
698 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
700 return do_fchmodat(AT_FDCWD, filename, mode);
704 * setattr_vfsuid - check and set ia_fsuid attribute
705 * @kuid: new inode owner
707 * Check whether @kuid is valid and if so generate and set vfsuid_t in
710 * Return: true if @kuid is valid, false if not.
712 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
714 if (!uid_valid(kuid))
716 attr->ia_valid |= ATTR_UID;
717 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
722 * setattr_vfsgid - check and set ia_fsgid attribute
723 * @kgid: new inode owner
725 * Check whether @kgid is valid and if so generate and set vfsgid_t in
728 * Return: true if @kgid is valid, false if not.
730 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
732 if (!gid_valid(kgid))
734 attr->ia_valid |= ATTR_GID;
735 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
739 int chown_common(const struct path *path, uid_t user, gid_t group)
741 struct mnt_idmap *idmap;
742 struct user_namespace *fs_userns;
743 struct inode *inode = path->dentry->d_inode;
744 struct inode *delegated_inode = NULL;
746 struct iattr newattrs;
750 uid = make_kuid(current_user_ns(), user);
751 gid = make_kgid(current_user_ns(), group);
753 idmap = mnt_idmap(path->mnt);
754 fs_userns = i_user_ns(inode);
757 newattrs.ia_vfsuid = INVALID_VFSUID;
758 newattrs.ia_vfsgid = INVALID_VFSGID;
759 newattrs.ia_valid = ATTR_CTIME;
760 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
762 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
765 if (!S_ISDIR(inode->i_mode))
766 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
767 setattr_should_drop_sgid(idmap, inode);
768 /* Continue to send actual fs values, not the mount values. */
769 error = security_path_chown(
771 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
772 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
774 error = notify_change(idmap, path->dentry, &newattrs,
777 if (delegated_inode) {
778 error = break_deleg_wait(&delegated_inode);
785 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
792 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
795 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
796 if (flag & AT_EMPTY_PATH)
797 lookup_flags |= LOOKUP_EMPTY;
799 error = user_path_at(dfd, filename, lookup_flags, &path);
802 error = mnt_want_write(path.mnt);
805 error = chown_common(&path, user, group);
806 mnt_drop_write(path.mnt);
809 if (retry_estale(error, lookup_flags)) {
810 lookup_flags |= LOOKUP_REVAL;
817 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
818 gid_t, group, int, flag)
820 return do_fchownat(dfd, filename, user, group, flag);
823 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
825 return do_fchownat(AT_FDCWD, filename, user, group, 0);
828 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
830 return do_fchownat(AT_FDCWD, filename, user, group,
831 AT_SYMLINK_NOFOLLOW);
834 int vfs_fchown(struct file *file, uid_t user, gid_t group)
838 error = mnt_want_write_file(file);
842 error = chown_common(&file->f_path, user, group);
843 mnt_drop_write_file(file);
847 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
849 struct fd f = fdget(fd);
853 error = vfs_fchown(f.file, user, group);
859 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
861 return ksys_fchown(fd, user, group);
864 static int do_dentry_open(struct file *f,
866 int (*open)(struct inode *, struct file *))
868 static const struct file_operations empty_fops = {};
871 path_get(&f->f_path);
873 f->f_mapping = inode->i_mapping;
874 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
875 f->f_sb_err = file_sample_sb_err(f);
877 if (unlikely(f->f_flags & O_PATH)) {
878 f->f_mode = FMODE_PATH | FMODE_OPENED;
879 f->f_op = &empty_fops;
883 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
884 i_readcount_inc(inode);
885 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
886 error = get_write_access(inode);
889 error = __mnt_want_write(f->f_path.mnt);
890 if (unlikely(error)) {
891 put_write_access(inode);
894 f->f_mode |= FMODE_WRITER;
897 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
898 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
899 f->f_mode |= FMODE_ATOMIC_POS;
901 f->f_op = fops_get(inode->i_fop);
902 if (WARN_ON(!f->f_op)) {
907 error = security_file_open(f);
911 error = break_lease(file_inode(f), f->f_flags);
915 /* normally all 3 are set; ->open() can clear them if needed */
916 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
918 open = f->f_op->open;
920 error = open(inode, f);
924 f->f_mode |= FMODE_OPENED;
925 if ((f->f_mode & FMODE_READ) &&
926 likely(f->f_op->read || f->f_op->read_iter))
927 f->f_mode |= FMODE_CAN_READ;
928 if ((f->f_mode & FMODE_WRITE) &&
929 likely(f->f_op->write || f->f_op->write_iter))
930 f->f_mode |= FMODE_CAN_WRITE;
931 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
932 f->f_mode &= ~FMODE_LSEEK;
933 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
934 f->f_mode |= FMODE_CAN_ODIRECT;
936 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
937 f->f_iocb_flags = iocb_flags(f);
939 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
941 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
945 * XXX: Huge page cache doesn't support writing yet. Drop all page
946 * cache for this file before processing writes.
948 if (f->f_mode & FMODE_WRITE) {
950 * Paired with smp_mb() in collapse_file() to ensure nr_thps
951 * is up to date and the update to i_writecount by
952 * get_write_access() is visible. Ensures subsequent insertion
953 * of THPs into the page cache will fail.
956 if (filemap_nr_thps(inode->i_mapping)) {
957 struct address_space *mapping = inode->i_mapping;
959 filemap_invalidate_lock(inode->i_mapping);
961 * unmap_mapping_range just need to be called once
962 * here, because the private pages is not need to be
963 * unmapped mapping (e.g. data segment of dynamic
964 * shared libraries here).
966 unmap_mapping_range(mapping, 0, 0, 0);
967 truncate_inode_pages(mapping, 0);
968 filemap_invalidate_unlock(inode->i_mapping);
975 if (WARN_ON_ONCE(error > 0))
980 path_put(&f->f_path);
981 f->f_path.mnt = NULL;
982 f->f_path.dentry = NULL;
988 * finish_open - finish opening a file
989 * @file: file pointer
990 * @dentry: pointer to dentry
991 * @open: open callback
992 * @opened: state of open
994 * This can be used to finish opening a file passed to i_op->atomic_open().
996 * If the open callback is set to NULL, then the standard f_op->open()
997 * filesystem callback is substituted.
999 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1000 * the return value of d_splice_alias(), then the caller needs to perform dput()
1001 * on it after finish_open().
1003 * Returns zero on success or -errno if the open failed.
1005 int finish_open(struct file *file, struct dentry *dentry,
1006 int (*open)(struct inode *, struct file *))
1008 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1010 file->f_path.dentry = dentry;
1011 return do_dentry_open(file, d_backing_inode(dentry), open);
1013 EXPORT_SYMBOL(finish_open);
1016 * finish_no_open - finish ->atomic_open() without opening the file
1018 * @file: file pointer
1019 * @dentry: dentry or NULL (as returned from ->lookup())
1021 * This can be used to set the result of a successful lookup in ->atomic_open().
1023 * NB: unlike finish_open() this function does consume the dentry reference and
1024 * the caller need not dput() it.
1026 * Returns "0" which must be the return value of ->atomic_open() after having
1027 * called this function.
1029 int finish_no_open(struct file *file, struct dentry *dentry)
1031 file->f_path.dentry = dentry;
1034 EXPORT_SYMBOL(finish_no_open);
1036 char *file_path(struct file *filp, char *buf, int buflen)
1038 return d_path(&filp->f_path, buf, buflen);
1040 EXPORT_SYMBOL(file_path);
1043 * vfs_open - open the file at the given path
1044 * @path: path to open
1045 * @file: newly allocated file with f_flag initialized
1046 * @cred: credentials to use
1048 int vfs_open(const struct path *path, struct file *file)
1050 file->f_path = *path;
1051 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1054 struct file *dentry_open(const struct path *path, int flags,
1055 const struct cred *cred)
1060 validate_creds(cred);
1062 /* We must always pass in a valid mount pointer. */
1065 f = alloc_empty_file(flags, cred);
1067 error = vfs_open(path, f);
1075 EXPORT_SYMBOL(dentry_open);
1078 * dentry_create - Create and open a file
1079 * @path: path to create
1081 * @mode: mode bits for new file
1082 * @cred: credentials to use
1084 * Caller must hold the parent directory's lock, and have prepared
1085 * a negative dentry, placed in @path->dentry, for the new file.
1087 * Caller sets @path->mnt to the vfsmount of the filesystem where
1088 * the new file is to be created. The parent directory and the
1089 * negative dentry must reside on the same filesystem instance.
1091 * On success, returns a "struct file *". Otherwise a ERR_PTR
1094 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1095 const struct cred *cred)
1100 validate_creds(cred);
1101 f = alloc_empty_file(flags, cred);
1105 error = vfs_create(mnt_idmap(path->mnt),
1106 d_inode(path->dentry->d_parent),
1107 path->dentry, mode, true);
1109 error = vfs_open(path, f);
1111 if (unlikely(error)) {
1113 return ERR_PTR(error);
1117 EXPORT_SYMBOL(dentry_create);
1119 struct file *open_with_fake_path(const struct path *path, int flags,
1120 struct inode *inode, const struct cred *cred)
1122 struct file *f = alloc_empty_file_noaccount(flags, cred);
1127 error = do_dentry_open(f, inode, NULL);
1135 EXPORT_SYMBOL(open_with_fake_path);
1137 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1138 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1140 inline struct open_how build_open_how(int flags, umode_t mode)
1142 struct open_how how = {
1143 .flags = flags & VALID_OPEN_FLAGS,
1144 .mode = mode & S_IALLUGO,
1147 /* O_PATH beats everything else. */
1148 if (how.flags & O_PATH)
1149 how.flags &= O_PATH_FLAGS;
1150 /* Modes should only be set for create-like flags. */
1151 if (!WILL_CREATE(how.flags))
1156 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1158 u64 flags = how->flags;
1159 u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1160 int lookup_flags = 0;
1161 int acc_mode = ACC_MODE(flags);
1163 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1164 "struct open_flags doesn't yet handle flags > 32 bits");
1167 * Strip flags that either shouldn't be set by userspace like
1168 * FMODE_NONOTIFY or that aren't relevant in determining struct
1169 * open_flags like O_CLOEXEC.
1174 * Older syscalls implicitly clear all of the invalid flags or argument
1175 * values before calling build_open_flags(), but openat2(2) checks all
1178 if (flags & ~VALID_OPEN_FLAGS)
1180 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1183 /* Scoping flags are mutually exclusive. */
1184 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1187 /* Deal with the mode. */
1188 if (WILL_CREATE(flags)) {
1189 if (how->mode & ~S_IALLUGO)
1191 op->mode = how->mode | S_IFREG;
1199 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1200 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1201 * O_TMPFILE below which requires O_DIRECTORY being raised.
1203 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1206 /* Now handle the creative implementation of O_TMPFILE. */
1207 if (flags & __O_TMPFILE) {
1209 * In order to ensure programs get explicit errors when trying
1210 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1211 * is raised alongside __O_TMPFILE.
1213 if (!(flags & O_DIRECTORY))
1215 if (!(acc_mode & MAY_WRITE))
1218 if (flags & O_PATH) {
1219 /* O_PATH only permits certain other flags to be set. */
1220 if (flags & ~O_PATH_FLAGS)
1226 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1227 * check for O_DSYNC if the need any syncing at all we enforce it's
1228 * always set instead of having to deal with possibly weird behaviour
1229 * for malicious applications setting only __O_SYNC.
1231 if (flags & __O_SYNC)
1234 op->open_flag = flags;
1236 /* O_TRUNC implies we need access checks for write permissions */
1237 if (flags & O_TRUNC)
1238 acc_mode |= MAY_WRITE;
1240 /* Allow the LSM permission hook to distinguish append
1241 access from general write access. */
1242 if (flags & O_APPEND)
1243 acc_mode |= MAY_APPEND;
1245 op->acc_mode = acc_mode;
1247 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1249 if (flags & O_CREAT) {
1250 op->intent |= LOOKUP_CREATE;
1251 if (flags & O_EXCL) {
1252 op->intent |= LOOKUP_EXCL;
1253 flags |= O_NOFOLLOW;
1257 if (flags & O_DIRECTORY)
1258 lookup_flags |= LOOKUP_DIRECTORY;
1259 if (!(flags & O_NOFOLLOW))
1260 lookup_flags |= LOOKUP_FOLLOW;
1262 if (how->resolve & RESOLVE_NO_XDEV)
1263 lookup_flags |= LOOKUP_NO_XDEV;
1264 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1265 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1266 if (how->resolve & RESOLVE_NO_SYMLINKS)
1267 lookup_flags |= LOOKUP_NO_SYMLINKS;
1268 if (how->resolve & RESOLVE_BENEATH)
1269 lookup_flags |= LOOKUP_BENEATH;
1270 if (how->resolve & RESOLVE_IN_ROOT)
1271 lookup_flags |= LOOKUP_IN_ROOT;
1272 if (how->resolve & RESOLVE_CACHED) {
1273 /* Don't bother even trying for create/truncate/tmpfile open */
1274 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1276 lookup_flags |= LOOKUP_CACHED;
1279 op->lookup_flags = lookup_flags;
1284 * file_open_name - open file and return file pointer
1286 * @name: struct filename containing path to open
1287 * @flags: open flags as per the open(2) second argument
1288 * @mode: mode for the new file if O_CREAT is set, else ignored
1290 * This is the helper to open a file from kernelspace if you really
1291 * have to. But in generally you should not do this, so please move
1292 * along, nothing to see here..
1294 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1296 struct open_flags op;
1297 struct open_how how = build_open_how(flags, mode);
1298 int err = build_open_flags(&how, &op);
1300 return ERR_PTR(err);
1301 return do_filp_open(AT_FDCWD, name, &op);
1305 * filp_open - open file and return file pointer
1307 * @filename: path to open
1308 * @flags: open flags as per the open(2) second argument
1309 * @mode: mode for the new file if O_CREAT is set, else ignored
1311 * This is the helper to open a file from kernelspace if you really
1312 * have to. But in generally you should not do this, so please move
1313 * along, nothing to see here..
1315 struct file *filp_open(const char *filename, int flags, umode_t mode)
1317 struct filename *name = getname_kernel(filename);
1318 struct file *file = ERR_CAST(name);
1320 if (!IS_ERR(name)) {
1321 file = file_open_name(name, flags, mode);
1326 EXPORT_SYMBOL(filp_open);
1328 struct file *file_open_root(const struct path *root,
1329 const char *filename, int flags, umode_t mode)
1331 struct open_flags op;
1332 struct open_how how = build_open_how(flags, mode);
1333 int err = build_open_flags(&how, &op);
1335 return ERR_PTR(err);
1336 return do_file_open_root(root, filename, &op);
1338 EXPORT_SYMBOL(file_open_root);
1340 static long do_sys_openat2(int dfd, const char __user *filename,
1341 struct open_how *how)
1343 struct open_flags op;
1344 int fd = build_open_flags(how, &op);
1345 struct filename *tmp;
1350 tmp = getname(filename);
1352 return PTR_ERR(tmp);
1354 fd = get_unused_fd_flags(how->flags);
1356 struct file *f = do_filp_open(dfd, tmp, &op);
1369 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1371 struct open_how how = build_open_how(flags, mode);
1372 return do_sys_openat2(dfd, filename, &how);
1376 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1378 if (force_o_largefile())
1379 flags |= O_LARGEFILE;
1380 return do_sys_open(AT_FDCWD, filename, flags, mode);
1383 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1386 if (force_o_largefile())
1387 flags |= O_LARGEFILE;
1388 return do_sys_open(dfd, filename, flags, mode);
1391 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1392 struct open_how __user *, how, size_t, usize)
1395 struct open_how tmp;
1397 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1398 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1400 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1403 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1407 audit_openat2_how(&tmp);
1409 /* O_LARGEFILE is only allowed for non-O_PATH. */
1410 if (!(tmp.flags & O_PATH) && force_o_largefile())
1411 tmp.flags |= O_LARGEFILE;
1413 return do_sys_openat2(dfd, filename, &tmp);
1416 #ifdef CONFIG_COMPAT
1418 * Exactly like sys_open(), except that it doesn't set the
1421 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1423 return do_sys_open(AT_FDCWD, filename, flags, mode);
1427 * Exactly like sys_openat(), except that it doesn't set the
1430 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1432 return do_sys_open(dfd, filename, flags, mode);
1439 * For backward compatibility? Maybe this should be moved
1440 * into arch/i386 instead?
1442 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1444 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1446 if (force_o_largefile())
1447 flags |= O_LARGEFILE;
1448 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1453 * "id" is the POSIX thread ID. We use the
1454 * files pointer for this..
1456 int filp_close(struct file *filp, fl_owner_t id)
1460 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1461 "VFS: Close: file count is 0 (f_op=%ps)",
1466 if (filp->f_op->flush)
1467 retval = filp->f_op->flush(filp, id);
1469 if (likely(!(filp->f_mode & FMODE_PATH))) {
1470 dnotify_flush(filp, id);
1471 locks_remove_posix(filp, id);
1477 EXPORT_SYMBOL(filp_close);
1480 * Careful here! We test whether the file pointer is NULL before
1481 * releasing the fd. This ensures that one clone task can't release
1482 * an fd while another clone is opening it.
1484 SYSCALL_DEFINE1(close, unsigned int, fd)
1486 int retval = close_fd(fd);
1488 /* can't restart close syscall because file table entry was cleared */
1489 if (unlikely(retval == -ERESTARTSYS ||
1490 retval == -ERESTARTNOINTR ||
1491 retval == -ERESTARTNOHAND ||
1492 retval == -ERESTART_RESTARTBLOCK))
1499 * close_range() - Close all file descriptors in a given range.
1501 * @fd: starting file descriptor to close
1502 * @max_fd: last file descriptor to close
1503 * @flags: reserved for future extensions
1505 * This closes a range of file descriptors. All file descriptors
1506 * from @fd up to and including @max_fd are closed.
1507 * Currently, errors to close a given file descriptor are ignored.
1509 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1510 unsigned int, flags)
1512 return __close_range(fd, max_fd, flags);
1516 * This routine simulates a hangup on the tty, to arrange that users
1517 * are given clean terminals at login time.
1519 SYSCALL_DEFINE0(vhangup)
1521 if (capable(CAP_SYS_TTY_CONFIG)) {
1529 * Called when an inode is about to be open.
1530 * We use this to disallow opening large files on 32bit systems if
1531 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1532 * on this flag in sys_open.
1534 int generic_file_open(struct inode * inode, struct file * filp)
1536 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1541 EXPORT_SYMBOL(generic_file_open);
1544 * This is used by subsystems that don't want seekable
1545 * file descriptors. The function is not supposed to ever fail, the only
1546 * reason it returns an 'int' and not 'void' is so that it can be plugged
1547 * directly into file_operations structure.
1549 int nonseekable_open(struct inode *inode, struct file *filp)
1551 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1555 EXPORT_SYMBOL(nonseekable_open);
1558 * stream_open is used by subsystems that want stream-like file descriptors.
1559 * Such file descriptors are not seekable and don't have notion of position
1560 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1561 * Contrary to file descriptors of other regular files, .read() and .write()
1562 * can run simultaneously.
1564 * stream_open never fails and is marked to return int so that it could be
1565 * directly used as file_operations.open .
1567 int stream_open(struct inode *inode, struct file *filp)
1569 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1570 filp->f_mode |= FMODE_STREAM;
1574 EXPORT_SYMBOL(stream_open);