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 ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
311 if (S_ISFIFO(inode->i_mode))
314 if (S_ISDIR(inode->i_mode))
317 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
320 /* Check for wrap through zero too */
321 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
324 if (!file->f_op->fallocate)
327 file_start_write(file);
328 ret = file->f_op->fallocate(file, mode, offset, len);
331 * Create inotify and fanotify events.
333 * To keep the logic simple always create events if fallocate succeeds.
334 * This implies that events are even created if the file size remains
335 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
338 fsnotify_modify(file);
340 file_end_write(file);
343 EXPORT_SYMBOL_GPL(vfs_fallocate);
345 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
347 struct fd f = fdget(fd);
351 error = vfs_fallocate(f.file, mode, offset, len);
357 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
359 return ksys_fallocate(fd, mode, offset, len);
362 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
363 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
364 compat_arg_u64_dual(len))
366 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
367 compat_arg_u64_glue(len));
372 * access() needs to use the real uid/gid, not the effective uid/gid.
373 * We do this by temporarily clearing all FS-related capabilities and
374 * switching the fsuid/fsgid around to the real ones.
376 * Creating new credentials is expensive, so we try to skip doing it,
377 * which we can if the result would match what we already got.
379 static bool access_need_override_creds(int flags)
381 const struct cred *cred;
383 if (flags & AT_EACCESS)
386 cred = current_cred();
387 if (!uid_eq(cred->fsuid, cred->uid) ||
388 !gid_eq(cred->fsgid, cred->gid))
391 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
392 kuid_t root_uid = make_kuid(cred->user_ns, 0);
393 if (!uid_eq(cred->uid, root_uid)) {
394 if (!cap_isclear(cred->cap_effective))
397 if (!cap_isidentical(cred->cap_effective,
398 cred->cap_permitted))
406 static const struct cred *access_override_creds(void)
408 const struct cred *old_cred;
409 struct cred *override_cred;
411 override_cred = prepare_creds();
416 * XXX access_need_override_creds performs checks in hopes of skipping
417 * this work. Make sure it stays in sync if making any changes in this
421 override_cred->fsuid = override_cred->uid;
422 override_cred->fsgid = override_cred->gid;
424 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
425 /* Clear the capabilities if we switch to a non-root user */
426 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
427 if (!uid_eq(override_cred->uid, root_uid))
428 cap_clear(override_cred->cap_effective);
430 override_cred->cap_effective =
431 override_cred->cap_permitted;
435 * The new set of credentials can *only* be used in
436 * task-synchronous circumstances, and does not need
437 * RCU freeing, unless somebody then takes a separate
440 * NOTE! This is _only_ true because this credential
441 * is used purely for override_creds() that installs
442 * it as the subjective cred. Other threads will be
443 * accessing ->real_cred, not the subjective cred.
445 * If somebody _does_ make a copy of this (using the
446 * 'get_current_cred()' function), that will clear the
447 * non_rcu field, because now that other user may be
448 * expecting RCU freeing. But normal thread-synchronous
449 * cred accesses will keep things non-racy to avoid RCU
452 override_cred->non_rcu = 1;
454 old_cred = override_creds(override_cred);
456 /* override_cred() gets its own ref */
457 put_cred(override_cred);
462 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
467 unsigned int lookup_flags = LOOKUP_FOLLOW;
468 const struct cred *old_cred = NULL;
470 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
473 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
476 if (flags & AT_SYMLINK_NOFOLLOW)
477 lookup_flags &= ~LOOKUP_FOLLOW;
478 if (flags & AT_EMPTY_PATH)
479 lookup_flags |= LOOKUP_EMPTY;
481 if (access_need_override_creds(flags)) {
482 old_cred = access_override_creds();
488 res = user_path_at(dfd, filename, lookup_flags, &path);
492 inode = d_backing_inode(path.dentry);
494 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
496 * MAY_EXEC on regular files is denied if the fs is mounted
497 * with the "noexec" flag.
500 if (path_noexec(&path))
501 goto out_path_release;
504 res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
505 /* SuS v2 requires we report a read only fs too */
506 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
507 goto out_path_release;
509 * This is a rare case where using __mnt_is_readonly()
510 * is OK without a mnt_want/drop_write() pair. Since
511 * no actual write to the fs is performed here, we do
512 * not need to telegraph to that to anyone.
514 * By doing this, we accept that this access is
515 * inherently racy and know that the fs may change
516 * state before we even see this result.
518 if (__mnt_is_readonly(path.mnt))
523 if (retry_estale(res, lookup_flags)) {
524 lookup_flags |= LOOKUP_REVAL;
529 revert_creds(old_cred);
534 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
536 return do_faccessat(dfd, filename, mode, 0);
539 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
542 return do_faccessat(dfd, filename, mode, flags);
545 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
547 return do_faccessat(AT_FDCWD, filename, mode, 0);
550 SYSCALL_DEFINE1(chdir, const char __user *, filename)
554 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
556 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
560 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
564 set_fs_pwd(current->fs, &path);
568 if (retry_estale(error, lookup_flags)) {
569 lookup_flags |= LOOKUP_REVAL;
576 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
578 struct fd f = fdget_raw(fd);
586 if (!d_can_lookup(f.file->f_path.dentry))
589 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
591 set_fs_pwd(current->fs, &f.file->f_path);
598 SYSCALL_DEFINE1(chroot, const char __user *, filename)
602 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
604 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
608 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
613 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
615 error = security_path_chroot(&path);
619 set_fs_root(current->fs, &path);
623 if (retry_estale(error, lookup_flags)) {
624 lookup_flags |= LOOKUP_REVAL;
631 int chmod_common(const struct path *path, umode_t mode)
633 struct inode *inode = path->dentry->d_inode;
634 struct inode *delegated_inode = NULL;
635 struct iattr newattrs;
638 error = mnt_want_write(path->mnt);
643 error = security_path_chmod(path, mode);
646 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
647 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
648 error = notify_change(mnt_idmap(path->mnt), path->dentry,
649 &newattrs, &delegated_inode);
652 if (delegated_inode) {
653 error = break_deleg_wait(&delegated_inode);
657 mnt_drop_write(path->mnt);
661 int vfs_fchmod(struct file *file, umode_t mode)
664 return chmod_common(&file->f_path, mode);
667 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
669 struct fd f = fdget(fd);
673 err = vfs_fchmod(f.file, mode);
679 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
684 unsigned int lookup_flags;
686 if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
689 lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
690 if (flags & AT_EMPTY_PATH)
691 lookup_flags |= LOOKUP_EMPTY;
694 error = user_path_at(dfd, filename, lookup_flags, &path);
696 error = chmod_common(&path, mode);
698 if (retry_estale(error, lookup_flags)) {
699 lookup_flags |= LOOKUP_REVAL;
706 SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
707 umode_t, mode, unsigned int, flags)
709 return do_fchmodat(dfd, filename, mode, flags);
712 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
715 return do_fchmodat(dfd, filename, mode, 0);
718 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
720 return do_fchmodat(AT_FDCWD, filename, mode, 0);
724 * Check whether @kuid is valid and if so generate and set vfsuid_t in
727 * Return: true if @kuid is valid, false if not.
729 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
731 if (!uid_valid(kuid))
733 attr->ia_valid |= ATTR_UID;
734 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
739 * Check whether @kgid is valid and if so generate and set vfsgid_t in
742 * Return: true if @kgid is valid, false if not.
744 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
746 if (!gid_valid(kgid))
748 attr->ia_valid |= ATTR_GID;
749 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
753 int chown_common(const struct path *path, uid_t user, gid_t group)
755 struct mnt_idmap *idmap;
756 struct user_namespace *fs_userns;
757 struct inode *inode = path->dentry->d_inode;
758 struct inode *delegated_inode = NULL;
760 struct iattr newattrs;
764 uid = make_kuid(current_user_ns(), user);
765 gid = make_kgid(current_user_ns(), group);
767 idmap = mnt_idmap(path->mnt);
768 fs_userns = i_user_ns(inode);
771 newattrs.ia_vfsuid = INVALID_VFSUID;
772 newattrs.ia_vfsgid = INVALID_VFSGID;
773 newattrs.ia_valid = ATTR_CTIME;
774 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
776 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
779 if (!S_ISDIR(inode->i_mode))
780 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
781 setattr_should_drop_sgid(idmap, inode);
782 /* Continue to send actual fs values, not the mount values. */
783 error = security_path_chown(
785 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
786 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
788 error = notify_change(idmap, path->dentry, &newattrs,
791 if (delegated_inode) {
792 error = break_deleg_wait(&delegated_inode);
799 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
806 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
809 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
810 if (flag & AT_EMPTY_PATH)
811 lookup_flags |= LOOKUP_EMPTY;
813 error = user_path_at(dfd, filename, lookup_flags, &path);
816 error = mnt_want_write(path.mnt);
819 error = chown_common(&path, user, group);
820 mnt_drop_write(path.mnt);
823 if (retry_estale(error, lookup_flags)) {
824 lookup_flags |= LOOKUP_REVAL;
831 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
832 gid_t, group, int, flag)
834 return do_fchownat(dfd, filename, user, group, flag);
837 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
839 return do_fchownat(AT_FDCWD, filename, user, group, 0);
842 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
844 return do_fchownat(AT_FDCWD, filename, user, group,
845 AT_SYMLINK_NOFOLLOW);
848 int vfs_fchown(struct file *file, uid_t user, gid_t group)
852 error = mnt_want_write_file(file);
856 error = chown_common(&file->f_path, user, group);
857 mnt_drop_write_file(file);
861 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
863 struct fd f = fdget(fd);
867 error = vfs_fchown(f.file, user, group);
873 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
875 return ksys_fchown(fd, user, group);
878 static inline int file_get_write_access(struct file *f)
882 error = get_write_access(f->f_inode);
885 error = mnt_get_write_access(f->f_path.mnt);
888 if (unlikely(f->f_mode & FMODE_BACKING)) {
889 error = mnt_get_write_access(backing_file_user_path(f)->mnt);
896 mnt_put_write_access(f->f_path.mnt);
898 put_write_access(f->f_inode);
902 static int do_dentry_open(struct file *f,
904 int (*open)(struct inode *, struct file *))
906 static const struct file_operations empty_fops = {};
909 path_get(&f->f_path);
911 f->f_mapping = inode->i_mapping;
912 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
913 f->f_sb_err = file_sample_sb_err(f);
915 if (unlikely(f->f_flags & O_PATH)) {
916 f->f_mode = FMODE_PATH | FMODE_OPENED;
917 f->f_op = &empty_fops;
921 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
922 i_readcount_inc(inode);
923 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
924 error = file_get_write_access(f);
927 f->f_mode |= FMODE_WRITER;
930 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
931 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
932 f->f_mode |= FMODE_ATOMIC_POS;
934 f->f_op = fops_get(inode->i_fop);
935 if (WARN_ON(!f->f_op)) {
940 error = security_file_open(f);
944 error = break_lease(file_inode(f), f->f_flags);
948 /* normally all 3 are set; ->open() can clear them if needed */
949 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
951 open = f->f_op->open;
953 error = open(inode, f);
957 f->f_mode |= FMODE_OPENED;
958 if ((f->f_mode & FMODE_READ) &&
959 likely(f->f_op->read || f->f_op->read_iter))
960 f->f_mode |= FMODE_CAN_READ;
961 if ((f->f_mode & FMODE_WRITE) &&
962 likely(f->f_op->write || f->f_op->write_iter))
963 f->f_mode |= FMODE_CAN_WRITE;
964 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
965 f->f_mode &= ~FMODE_LSEEK;
966 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
967 f->f_mode |= FMODE_CAN_ODIRECT;
969 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
970 f->f_iocb_flags = iocb_flags(f);
972 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
974 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
978 * XXX: Huge page cache doesn't support writing yet. Drop all page
979 * cache for this file before processing writes.
981 if (f->f_mode & FMODE_WRITE) {
983 * Paired with smp_mb() in collapse_file() to ensure nr_thps
984 * is up to date and the update to i_writecount by
985 * get_write_access() is visible. Ensures subsequent insertion
986 * of THPs into the page cache will fail.
989 if (filemap_nr_thps(inode->i_mapping)) {
990 struct address_space *mapping = inode->i_mapping;
992 filemap_invalidate_lock(inode->i_mapping);
994 * unmap_mapping_range just need to be called once
995 * here, because the private pages is not need to be
996 * unmapped mapping (e.g. data segment of dynamic
997 * shared libraries here).
999 unmap_mapping_range(mapping, 0, 0, 0);
1000 truncate_inode_pages(mapping, 0);
1001 filemap_invalidate_unlock(inode->i_mapping);
1006 * Once we return a file with FMODE_OPENED, __fput() will call
1007 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
1013 if (WARN_ON_ONCE(error > 0))
1018 path_put(&f->f_path);
1019 f->f_path.mnt = NULL;
1020 f->f_path.dentry = NULL;
1026 * finish_open - finish opening a file
1027 * @file: file pointer
1028 * @dentry: pointer to dentry
1029 * @open: open callback
1031 * This can be used to finish opening a file passed to i_op->atomic_open().
1033 * If the open callback is set to NULL, then the standard f_op->open()
1034 * filesystem callback is substituted.
1036 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1037 * the return value of d_splice_alias(), then the caller needs to perform dput()
1038 * on it after finish_open().
1040 * Returns zero on success or -errno if the open failed.
1042 int finish_open(struct file *file, struct dentry *dentry,
1043 int (*open)(struct inode *, struct file *))
1045 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1047 file->f_path.dentry = dentry;
1048 return do_dentry_open(file, d_backing_inode(dentry), open);
1050 EXPORT_SYMBOL(finish_open);
1053 * finish_no_open - finish ->atomic_open() without opening the file
1055 * @file: file pointer
1056 * @dentry: dentry or NULL (as returned from ->lookup())
1058 * This can be used to set the result of a successful lookup in ->atomic_open().
1060 * NB: unlike finish_open() this function does consume the dentry reference and
1061 * the caller need not dput() it.
1063 * Returns "0" which must be the return value of ->atomic_open() after having
1064 * called this function.
1066 int finish_no_open(struct file *file, struct dentry *dentry)
1068 file->f_path.dentry = dentry;
1071 EXPORT_SYMBOL(finish_no_open);
1073 char *file_path(struct file *filp, char *buf, int buflen)
1075 return d_path(&filp->f_path, buf, buflen);
1077 EXPORT_SYMBOL(file_path);
1080 * vfs_open - open the file at the given path
1081 * @path: path to open
1082 * @file: newly allocated file with f_flag initialized
1084 int vfs_open(const struct path *path, struct file *file)
1086 file->f_path = *path;
1087 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1090 struct file *dentry_open(const struct path *path, int flags,
1091 const struct cred *cred)
1096 /* We must always pass in a valid mount pointer. */
1099 f = alloc_empty_file(flags, cred);
1101 error = vfs_open(path, f);
1109 EXPORT_SYMBOL(dentry_open);
1112 * dentry_create - Create and open a file
1113 * @path: path to create
1115 * @mode: mode bits for new file
1116 * @cred: credentials to use
1118 * Caller must hold the parent directory's lock, and have prepared
1119 * a negative dentry, placed in @path->dentry, for the new file.
1121 * Caller sets @path->mnt to the vfsmount of the filesystem where
1122 * the new file is to be created. The parent directory and the
1123 * negative dentry must reside on the same filesystem instance.
1125 * On success, returns a "struct file *". Otherwise a ERR_PTR
1128 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1129 const struct cred *cred)
1134 f = alloc_empty_file(flags, cred);
1138 error = vfs_create(mnt_idmap(path->mnt),
1139 d_inode(path->dentry->d_parent),
1140 path->dentry, mode, true);
1142 error = vfs_open(path, f);
1144 if (unlikely(error)) {
1146 return ERR_PTR(error);
1150 EXPORT_SYMBOL(dentry_create);
1153 * kernel_file_open - open a file for kernel internal use
1154 * @path: path of the file to open
1155 * @flags: open flags
1157 * @cred: credentials for open
1159 * Open a file for use by in-kernel consumers. The file is not accounted
1160 * against nr_files and must not be installed into the file descriptor
1163 * Return: Opened file on success, an error pointer on failure.
1165 struct file *kernel_file_open(const struct path *path, int flags,
1166 struct inode *inode, const struct cred *cred)
1171 f = alloc_empty_file_noaccount(flags, cred);
1176 error = do_dentry_open(f, inode, NULL);
1183 EXPORT_SYMBOL_GPL(kernel_file_open);
1185 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1186 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1188 inline struct open_how build_open_how(int flags, umode_t mode)
1190 struct open_how how = {
1191 .flags = flags & VALID_OPEN_FLAGS,
1192 .mode = mode & S_IALLUGO,
1195 /* O_PATH beats everything else. */
1196 if (how.flags & O_PATH)
1197 how.flags &= O_PATH_FLAGS;
1198 /* Modes should only be set for create-like flags. */
1199 if (!WILL_CREATE(how.flags))
1204 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1206 u64 flags = how->flags;
1207 u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1208 int lookup_flags = 0;
1209 int acc_mode = ACC_MODE(flags);
1211 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1212 "struct open_flags doesn't yet handle flags > 32 bits");
1215 * Strip flags that either shouldn't be set by userspace like
1216 * FMODE_NONOTIFY or that aren't relevant in determining struct
1217 * open_flags like O_CLOEXEC.
1222 * Older syscalls implicitly clear all of the invalid flags or argument
1223 * values before calling build_open_flags(), but openat2(2) checks all
1226 if (flags & ~VALID_OPEN_FLAGS)
1228 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1231 /* Scoping flags are mutually exclusive. */
1232 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1235 /* Deal with the mode. */
1236 if (WILL_CREATE(flags)) {
1237 if (how->mode & ~S_IALLUGO)
1239 op->mode = how->mode | S_IFREG;
1247 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1248 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1249 * O_TMPFILE below which requires O_DIRECTORY being raised.
1251 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1254 /* Now handle the creative implementation of O_TMPFILE. */
1255 if (flags & __O_TMPFILE) {
1257 * In order to ensure programs get explicit errors when trying
1258 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1259 * is raised alongside __O_TMPFILE.
1261 if (!(flags & O_DIRECTORY))
1263 if (!(acc_mode & MAY_WRITE))
1266 if (flags & O_PATH) {
1267 /* O_PATH only permits certain other flags to be set. */
1268 if (flags & ~O_PATH_FLAGS)
1274 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1275 * check for O_DSYNC if the need any syncing at all we enforce it's
1276 * always set instead of having to deal with possibly weird behaviour
1277 * for malicious applications setting only __O_SYNC.
1279 if (flags & __O_SYNC)
1282 op->open_flag = flags;
1284 /* O_TRUNC implies we need access checks for write permissions */
1285 if (flags & O_TRUNC)
1286 acc_mode |= MAY_WRITE;
1288 /* Allow the LSM permission hook to distinguish append
1289 access from general write access. */
1290 if (flags & O_APPEND)
1291 acc_mode |= MAY_APPEND;
1293 op->acc_mode = acc_mode;
1295 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1297 if (flags & O_CREAT) {
1298 op->intent |= LOOKUP_CREATE;
1299 if (flags & O_EXCL) {
1300 op->intent |= LOOKUP_EXCL;
1301 flags |= O_NOFOLLOW;
1305 if (flags & O_DIRECTORY)
1306 lookup_flags |= LOOKUP_DIRECTORY;
1307 if (!(flags & O_NOFOLLOW))
1308 lookup_flags |= LOOKUP_FOLLOW;
1310 if (how->resolve & RESOLVE_NO_XDEV)
1311 lookup_flags |= LOOKUP_NO_XDEV;
1312 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1313 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1314 if (how->resolve & RESOLVE_NO_SYMLINKS)
1315 lookup_flags |= LOOKUP_NO_SYMLINKS;
1316 if (how->resolve & RESOLVE_BENEATH)
1317 lookup_flags |= LOOKUP_BENEATH;
1318 if (how->resolve & RESOLVE_IN_ROOT)
1319 lookup_flags |= LOOKUP_IN_ROOT;
1320 if (how->resolve & RESOLVE_CACHED) {
1321 /* Don't bother even trying for create/truncate/tmpfile open */
1322 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1324 lookup_flags |= LOOKUP_CACHED;
1327 op->lookup_flags = lookup_flags;
1332 * file_open_name - open file and return file pointer
1334 * @name: struct filename containing path to open
1335 * @flags: open flags as per the open(2) second argument
1336 * @mode: mode for the new file if O_CREAT is set, else ignored
1338 * This is the helper to open a file from kernelspace if you really
1339 * have to. But in generally you should not do this, so please move
1340 * along, nothing to see here..
1342 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1344 struct open_flags op;
1345 struct open_how how = build_open_how(flags, mode);
1346 int err = build_open_flags(&how, &op);
1348 return ERR_PTR(err);
1349 return do_filp_open(AT_FDCWD, name, &op);
1353 * filp_open - open file and return file pointer
1355 * @filename: path to open
1356 * @flags: open flags as per the open(2) second argument
1357 * @mode: mode for the new file if O_CREAT is set, else ignored
1359 * This is the helper to open a file from kernelspace if you really
1360 * have to. But in generally you should not do this, so please move
1361 * along, nothing to see here..
1363 struct file *filp_open(const char *filename, int flags, umode_t mode)
1365 struct filename *name = getname_kernel(filename);
1366 struct file *file = ERR_CAST(name);
1368 if (!IS_ERR(name)) {
1369 file = file_open_name(name, flags, mode);
1374 EXPORT_SYMBOL(filp_open);
1376 struct file *file_open_root(const struct path *root,
1377 const char *filename, int flags, umode_t mode)
1379 struct open_flags op;
1380 struct open_how how = build_open_how(flags, mode);
1381 int err = build_open_flags(&how, &op);
1383 return ERR_PTR(err);
1384 return do_file_open_root(root, filename, &op);
1386 EXPORT_SYMBOL(file_open_root);
1388 static long do_sys_openat2(int dfd, const char __user *filename,
1389 struct open_how *how)
1391 struct open_flags op;
1392 int fd = build_open_flags(how, &op);
1393 struct filename *tmp;
1398 tmp = getname(filename);
1400 return PTR_ERR(tmp);
1402 fd = get_unused_fd_flags(how->flags);
1404 struct file *f = do_filp_open(dfd, tmp, &op);
1416 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1418 struct open_how how = build_open_how(flags, mode);
1419 return do_sys_openat2(dfd, filename, &how);
1423 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1425 if (force_o_largefile())
1426 flags |= O_LARGEFILE;
1427 return do_sys_open(AT_FDCWD, filename, flags, mode);
1430 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1433 if (force_o_largefile())
1434 flags |= O_LARGEFILE;
1435 return do_sys_open(dfd, filename, flags, mode);
1438 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1439 struct open_how __user *, how, size_t, usize)
1442 struct open_how tmp;
1444 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1445 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1447 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1450 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1454 audit_openat2_how(&tmp);
1456 /* O_LARGEFILE is only allowed for non-O_PATH. */
1457 if (!(tmp.flags & O_PATH) && force_o_largefile())
1458 tmp.flags |= O_LARGEFILE;
1460 return do_sys_openat2(dfd, filename, &tmp);
1463 #ifdef CONFIG_COMPAT
1465 * Exactly like sys_open(), except that it doesn't set the
1468 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1470 return do_sys_open(AT_FDCWD, filename, flags, mode);
1474 * Exactly like sys_openat(), except that it doesn't set the
1477 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1479 return do_sys_open(dfd, filename, flags, mode);
1486 * For backward compatibility? Maybe this should be moved
1487 * into arch/i386 instead?
1489 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1491 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1493 if (force_o_largefile())
1494 flags |= O_LARGEFILE;
1495 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1500 * "id" is the POSIX thread ID. We use the
1501 * files pointer for this..
1503 static int filp_flush(struct file *filp, fl_owner_t id)
1507 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1508 "VFS: Close: file count is 0 (f_op=%ps)",
1513 if (filp->f_op->flush)
1514 retval = filp->f_op->flush(filp, id);
1516 if (likely(!(filp->f_mode & FMODE_PATH))) {
1517 dnotify_flush(filp, id);
1518 locks_remove_posix(filp, id);
1523 int filp_close(struct file *filp, fl_owner_t id)
1527 retval = filp_flush(filp, id);
1532 EXPORT_SYMBOL(filp_close);
1535 * Careful here! We test whether the file pointer is NULL before
1536 * releasing the fd. This ensures that one clone task can't release
1537 * an fd while another clone is opening it.
1539 SYSCALL_DEFINE1(close, unsigned int, fd)
1544 file = file_close_fd(fd);
1548 retval = filp_flush(file, current->files);
1551 * We're returning to user space. Don't bother
1552 * with any delayed fput() cases.
1556 /* can't restart close syscall because file table entry was cleared */
1557 if (unlikely(retval == -ERESTARTSYS ||
1558 retval == -ERESTARTNOINTR ||
1559 retval == -ERESTARTNOHAND ||
1560 retval == -ERESTART_RESTARTBLOCK))
1567 * sys_close_range() - Close all file descriptors in a given range.
1569 * @fd: starting file descriptor to close
1570 * @max_fd: last file descriptor to close
1571 * @flags: reserved for future extensions
1573 * This closes a range of file descriptors. All file descriptors
1574 * from @fd up to and including @max_fd are closed.
1575 * Currently, errors to close a given file descriptor are ignored.
1577 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1578 unsigned int, flags)
1580 return __close_range(fd, max_fd, flags);
1584 * This routine simulates a hangup on the tty, to arrange that users
1585 * are given clean terminals at login time.
1587 SYSCALL_DEFINE0(vhangup)
1589 if (capable(CAP_SYS_TTY_CONFIG)) {
1597 * Called when an inode is about to be open.
1598 * We use this to disallow opening large files on 32bit systems if
1599 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1600 * on this flag in sys_open.
1602 int generic_file_open(struct inode * inode, struct file * filp)
1604 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1609 EXPORT_SYMBOL(generic_file_open);
1612 * This is used by subsystems that don't want seekable
1613 * file descriptors. The function is not supposed to ever fail, the only
1614 * reason it returns an 'int' and not 'void' is so that it can be plugged
1615 * directly into file_operations structure.
1617 int nonseekable_open(struct inode *inode, struct file *filp)
1619 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1623 EXPORT_SYMBOL(nonseekable_open);
1626 * stream_open is used by subsystems that want stream-like file descriptors.
1627 * Such file descriptors are not seekable and don't have notion of position
1628 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1629 * Contrary to file descriptors of other regular files, .read() and .write()
1630 * can run simultaneously.
1632 * stream_open never fails and is marked to return int so that it could be
1633 * directly used as file_operations.open .
1635 int stream_open(struct inode *inode, struct file *filp)
1637 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1638 filp->f_mode |= FMODE_STREAM;
1642 EXPORT_SYMBOL(stream_open);