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_ftruncate(struct file *file, loff_t length, int small)
160 struct dentry *dentry;
163 /* explicitly opened as large or we are on 64-bit box */
164 if (file->f_flags & O_LARGEFILE)
167 dentry = file->f_path.dentry;
168 inode = dentry->d_inode;
169 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
172 /* Cannot ftruncate over 2^31 bytes without large file support */
173 if (small && length > MAX_NON_LFS)
176 /* Check IS_APPEND on real upper inode */
177 if (IS_APPEND(file_inode(file)))
179 sb_start_write(inode->i_sb);
180 error = security_file_truncate(file);
182 error = do_truncate(file_mnt_idmap(file), dentry, length,
183 ATTR_MTIME | ATTR_CTIME, file);
184 sb_end_write(inode->i_sb);
189 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
200 error = do_ftruncate(f.file, length, small);
206 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
208 return do_sys_ftruncate(fd, length, 1);
212 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
214 return do_sys_ftruncate(fd, length, 1);
218 /* LFS versions of truncate are only needed on 32 bit machines */
219 #if BITS_PER_LONG == 32
220 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
222 return do_sys_truncate(path, length);
225 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
227 return do_sys_ftruncate(fd, length, 0);
229 #endif /* BITS_PER_LONG == 32 */
231 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
232 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
233 compat_arg_u64_dual(length))
235 return ksys_truncate(pathname, compat_arg_u64_glue(length));
239 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
240 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
241 compat_arg_u64_dual(length))
243 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
247 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
249 struct inode *inode = file_inode(file);
252 if (offset < 0 || len <= 0)
255 /* Return error if mode is not supported */
256 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
259 /* Punch hole and zero range are mutually exclusive */
260 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
261 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
264 /* Punch hole must have keep size set */
265 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
266 !(mode & FALLOC_FL_KEEP_SIZE))
269 /* Collapse range should only be used exclusively. */
270 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
271 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
274 /* Insert range should only be used exclusively. */
275 if ((mode & FALLOC_FL_INSERT_RANGE) &&
276 (mode & ~FALLOC_FL_INSERT_RANGE))
279 /* Unshare range should only be used with allocate mode. */
280 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
281 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
284 if (!(file->f_mode & FMODE_WRITE))
288 * We can only allow pure fallocate on append only files
290 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
293 if (IS_IMMUTABLE(inode))
297 * We cannot allow any fallocate operation on an active swapfile
299 if (IS_SWAPFILE(inode))
303 * Revalidate the write permissions, in case security policy has
304 * changed since the files were opened.
306 ret = security_file_permission(file, MAY_WRITE);
310 ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
314 if (S_ISFIFO(inode->i_mode))
317 if (S_ISDIR(inode->i_mode))
320 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
323 /* Check for wrap through zero too */
324 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
327 if (!file->f_op->fallocate)
330 file_start_write(file);
331 ret = file->f_op->fallocate(file, mode, offset, len);
334 * Create inotify and fanotify events.
336 * To keep the logic simple always create events if fallocate succeeds.
337 * This implies that events are even created if the file size remains
338 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
341 fsnotify_modify(file);
343 file_end_write(file);
346 EXPORT_SYMBOL_GPL(vfs_fallocate);
348 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
350 struct fd f = fdget(fd);
354 error = vfs_fallocate(f.file, mode, offset, len);
360 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
362 return ksys_fallocate(fd, mode, offset, len);
365 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
366 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
367 compat_arg_u64_dual(len))
369 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
370 compat_arg_u64_glue(len));
375 * access() needs to use the real uid/gid, not the effective uid/gid.
376 * We do this by temporarily clearing all FS-related capabilities and
377 * switching the fsuid/fsgid around to the real ones.
379 * Creating new credentials is expensive, so we try to skip doing it,
380 * which we can if the result would match what we already got.
382 static bool access_need_override_creds(int flags)
384 const struct cred *cred;
386 if (flags & AT_EACCESS)
389 cred = current_cred();
390 if (!uid_eq(cred->fsuid, cred->uid) ||
391 !gid_eq(cred->fsgid, cred->gid))
394 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
395 kuid_t root_uid = make_kuid(cred->user_ns, 0);
396 if (!uid_eq(cred->uid, root_uid)) {
397 if (!cap_isclear(cred->cap_effective))
400 if (!cap_isidentical(cred->cap_effective,
401 cred->cap_permitted))
409 static const struct cred *access_override_creds(void)
411 const struct cred *old_cred;
412 struct cred *override_cred;
414 override_cred = prepare_creds();
419 * XXX access_need_override_creds performs checks in hopes of skipping
420 * this work. Make sure it stays in sync if making any changes in this
424 override_cred->fsuid = override_cred->uid;
425 override_cred->fsgid = override_cred->gid;
427 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
428 /* Clear the capabilities if we switch to a non-root user */
429 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
430 if (!uid_eq(override_cred->uid, root_uid))
431 cap_clear(override_cred->cap_effective);
433 override_cred->cap_effective =
434 override_cred->cap_permitted;
438 * The new set of credentials can *only* be used in
439 * task-synchronous circumstances, and does not need
440 * RCU freeing, unless somebody then takes a separate
443 * NOTE! This is _only_ true because this credential
444 * is used purely for override_creds() that installs
445 * it as the subjective cred. Other threads will be
446 * accessing ->real_cred, not the subjective cred.
448 * If somebody _does_ make a copy of this (using the
449 * 'get_current_cred()' function), that will clear the
450 * non_rcu field, because now that other user may be
451 * expecting RCU freeing. But normal thread-synchronous
452 * cred accesses will keep things non-racy to avoid RCU
455 override_cred->non_rcu = 1;
457 old_cred = override_creds(override_cred);
459 /* override_cred() gets its own ref */
460 put_cred(override_cred);
465 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
470 unsigned int lookup_flags = LOOKUP_FOLLOW;
471 const struct cred *old_cred = NULL;
473 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
476 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
479 if (flags & AT_SYMLINK_NOFOLLOW)
480 lookup_flags &= ~LOOKUP_FOLLOW;
481 if (flags & AT_EMPTY_PATH)
482 lookup_flags |= LOOKUP_EMPTY;
484 if (access_need_override_creds(flags)) {
485 old_cred = access_override_creds();
491 res = user_path_at(dfd, filename, lookup_flags, &path);
495 inode = d_backing_inode(path.dentry);
497 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
499 * MAY_EXEC on regular files is denied if the fs is mounted
500 * with the "noexec" flag.
503 if (path_noexec(&path))
504 goto out_path_release;
507 res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
508 /* SuS v2 requires we report a read only fs too */
509 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
510 goto out_path_release;
512 * This is a rare case where using __mnt_is_readonly()
513 * is OK without a mnt_want/drop_write() pair. Since
514 * no actual write to the fs is performed here, we do
515 * not need to telegraph to that to anyone.
517 * By doing this, we accept that this access is
518 * inherently racy and know that the fs may change
519 * state before we even see this result.
521 if (__mnt_is_readonly(path.mnt))
526 if (retry_estale(res, lookup_flags)) {
527 lookup_flags |= LOOKUP_REVAL;
532 revert_creds(old_cred);
537 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
539 return do_faccessat(dfd, filename, mode, 0);
542 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
545 return do_faccessat(dfd, filename, mode, flags);
548 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
550 return do_faccessat(AT_FDCWD, filename, mode, 0);
553 SYSCALL_DEFINE1(chdir, const char __user *, filename)
557 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
559 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
563 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
567 set_fs_pwd(current->fs, &path);
571 if (retry_estale(error, lookup_flags)) {
572 lookup_flags |= LOOKUP_REVAL;
579 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
581 struct fd f = fdget_raw(fd);
589 if (!d_can_lookup(f.file->f_path.dentry))
592 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
594 set_fs_pwd(current->fs, &f.file->f_path);
601 SYSCALL_DEFINE1(chroot, const char __user *, filename)
605 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
607 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
611 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
616 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
618 error = security_path_chroot(&path);
622 set_fs_root(current->fs, &path);
626 if (retry_estale(error, lookup_flags)) {
627 lookup_flags |= LOOKUP_REVAL;
634 int chmod_common(const struct path *path, umode_t mode)
636 struct inode *inode = path->dentry->d_inode;
637 struct inode *delegated_inode = NULL;
638 struct iattr newattrs;
641 error = mnt_want_write(path->mnt);
646 error = security_path_chmod(path, mode);
649 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
650 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
651 error = notify_change(mnt_idmap(path->mnt), path->dentry,
652 &newattrs, &delegated_inode);
655 if (delegated_inode) {
656 error = break_deleg_wait(&delegated_inode);
660 mnt_drop_write(path->mnt);
664 int vfs_fchmod(struct file *file, umode_t mode)
667 return chmod_common(&file->f_path, mode);
670 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
672 struct fd f = fdget(fd);
676 err = vfs_fchmod(f.file, mode);
682 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
687 unsigned int lookup_flags;
689 if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
692 lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
693 if (flags & AT_EMPTY_PATH)
694 lookup_flags |= LOOKUP_EMPTY;
697 error = user_path_at(dfd, filename, lookup_flags, &path);
699 error = chmod_common(&path, mode);
701 if (retry_estale(error, lookup_flags)) {
702 lookup_flags |= LOOKUP_REVAL;
709 SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
710 umode_t, mode, unsigned int, flags)
712 return do_fchmodat(dfd, filename, mode, flags);
715 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
718 return do_fchmodat(dfd, filename, mode, 0);
721 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
723 return do_fchmodat(AT_FDCWD, filename, mode, 0);
727 * Check whether @kuid is valid and if so generate and set vfsuid_t in
730 * Return: true if @kuid is valid, false if not.
732 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
734 if (!uid_valid(kuid))
736 attr->ia_valid |= ATTR_UID;
737 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
742 * Check whether @kgid is valid and if so generate and set vfsgid_t in
745 * Return: true if @kgid is valid, false if not.
747 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
749 if (!gid_valid(kgid))
751 attr->ia_valid |= ATTR_GID;
752 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
756 int chown_common(const struct path *path, uid_t user, gid_t group)
758 struct mnt_idmap *idmap;
759 struct user_namespace *fs_userns;
760 struct inode *inode = path->dentry->d_inode;
761 struct inode *delegated_inode = NULL;
763 struct iattr newattrs;
767 uid = make_kuid(current_user_ns(), user);
768 gid = make_kgid(current_user_ns(), group);
770 idmap = mnt_idmap(path->mnt);
771 fs_userns = i_user_ns(inode);
774 newattrs.ia_vfsuid = INVALID_VFSUID;
775 newattrs.ia_vfsgid = INVALID_VFSGID;
776 newattrs.ia_valid = ATTR_CTIME;
777 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
779 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
782 if (!S_ISDIR(inode->i_mode))
783 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
784 setattr_should_drop_sgid(idmap, inode);
785 /* Continue to send actual fs values, not the mount values. */
786 error = security_path_chown(
788 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
789 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
791 error = notify_change(idmap, path->dentry, &newattrs,
794 if (delegated_inode) {
795 error = break_deleg_wait(&delegated_inode);
802 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
809 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
812 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
813 if (flag & AT_EMPTY_PATH)
814 lookup_flags |= LOOKUP_EMPTY;
816 error = user_path_at(dfd, filename, lookup_flags, &path);
819 error = mnt_want_write(path.mnt);
822 error = chown_common(&path, user, group);
823 mnt_drop_write(path.mnt);
826 if (retry_estale(error, lookup_flags)) {
827 lookup_flags |= LOOKUP_REVAL;
834 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
835 gid_t, group, int, flag)
837 return do_fchownat(dfd, filename, user, group, flag);
840 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
842 return do_fchownat(AT_FDCWD, filename, user, group, 0);
845 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
847 return do_fchownat(AT_FDCWD, filename, user, group,
848 AT_SYMLINK_NOFOLLOW);
851 int vfs_fchown(struct file *file, uid_t user, gid_t group)
855 error = mnt_want_write_file(file);
859 error = chown_common(&file->f_path, user, group);
860 mnt_drop_write_file(file);
864 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
866 struct fd f = fdget(fd);
870 error = vfs_fchown(f.file, user, group);
876 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
878 return ksys_fchown(fd, user, group);
881 static inline int file_get_write_access(struct file *f)
885 error = get_write_access(f->f_inode);
888 error = mnt_get_write_access(f->f_path.mnt);
891 if (unlikely(f->f_mode & FMODE_BACKING)) {
892 error = mnt_get_write_access(backing_file_user_path(f)->mnt);
899 mnt_put_write_access(f->f_path.mnt);
901 put_write_access(f->f_inode);
905 static int do_dentry_open(struct file *f,
907 int (*open)(struct inode *, struct file *))
909 static const struct file_operations empty_fops = {};
912 path_get(&f->f_path);
914 f->f_mapping = inode->i_mapping;
915 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
916 f->f_sb_err = file_sample_sb_err(f);
918 if (unlikely(f->f_flags & O_PATH)) {
919 f->f_mode = FMODE_PATH | FMODE_OPENED;
920 f->f_op = &empty_fops;
924 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
925 i_readcount_inc(inode);
926 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
927 error = file_get_write_access(f);
930 f->f_mode |= FMODE_WRITER;
933 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
934 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
935 f->f_mode |= FMODE_ATOMIC_POS;
937 f->f_op = fops_get(inode->i_fop);
938 if (WARN_ON(!f->f_op)) {
943 error = security_file_open(f);
947 error = break_lease(file_inode(f), f->f_flags);
951 /* normally all 3 are set; ->open() can clear them if needed */
952 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
954 open = f->f_op->open;
956 error = open(inode, f);
960 f->f_mode |= FMODE_OPENED;
961 if ((f->f_mode & FMODE_READ) &&
962 likely(f->f_op->read || f->f_op->read_iter))
963 f->f_mode |= FMODE_CAN_READ;
964 if ((f->f_mode & FMODE_WRITE) &&
965 likely(f->f_op->write || f->f_op->write_iter))
966 f->f_mode |= FMODE_CAN_WRITE;
967 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
968 f->f_mode &= ~FMODE_LSEEK;
969 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
970 f->f_mode |= FMODE_CAN_ODIRECT;
972 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
973 f->f_iocb_flags = iocb_flags(f);
975 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
977 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
981 * XXX: Huge page cache doesn't support writing yet. Drop all page
982 * cache for this file before processing writes.
984 if (f->f_mode & FMODE_WRITE) {
986 * Paired with smp_mb() in collapse_file() to ensure nr_thps
987 * is up to date and the update to i_writecount by
988 * get_write_access() is visible. Ensures subsequent insertion
989 * of THPs into the page cache will fail.
992 if (filemap_nr_thps(inode->i_mapping)) {
993 struct address_space *mapping = inode->i_mapping;
995 filemap_invalidate_lock(inode->i_mapping);
997 * unmap_mapping_range just need to be called once
998 * here, because the private pages is not need to be
999 * unmapped mapping (e.g. data segment of dynamic
1000 * shared libraries here).
1002 unmap_mapping_range(mapping, 0, 0, 0);
1003 truncate_inode_pages(mapping, 0);
1004 filemap_invalidate_unlock(inode->i_mapping);
1009 * Once we return a file with FMODE_OPENED, __fput() will call
1010 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
1016 if (WARN_ON_ONCE(error > 0))
1021 path_put(&f->f_path);
1022 f->f_path.mnt = NULL;
1023 f->f_path.dentry = NULL;
1029 * finish_open - finish opening a file
1030 * @file: file pointer
1031 * @dentry: pointer to dentry
1032 * @open: open callback
1034 * This can be used to finish opening a file passed to i_op->atomic_open().
1036 * If the open callback is set to NULL, then the standard f_op->open()
1037 * filesystem callback is substituted.
1039 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1040 * the return value of d_splice_alias(), then the caller needs to perform dput()
1041 * on it after finish_open().
1043 * Returns zero on success or -errno if the open failed.
1045 int finish_open(struct file *file, struct dentry *dentry,
1046 int (*open)(struct inode *, struct file *))
1048 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1050 file->f_path.dentry = dentry;
1051 return do_dentry_open(file, d_backing_inode(dentry), open);
1053 EXPORT_SYMBOL(finish_open);
1056 * finish_no_open - finish ->atomic_open() without opening the file
1058 * @file: file pointer
1059 * @dentry: dentry or NULL (as returned from ->lookup())
1061 * This can be used to set the result of a successful lookup in ->atomic_open().
1063 * NB: unlike finish_open() this function does consume the dentry reference and
1064 * the caller need not dput() it.
1066 * Returns "0" which must be the return value of ->atomic_open() after having
1067 * called this function.
1069 int finish_no_open(struct file *file, struct dentry *dentry)
1071 file->f_path.dentry = dentry;
1074 EXPORT_SYMBOL(finish_no_open);
1076 char *file_path(struct file *filp, char *buf, int buflen)
1078 return d_path(&filp->f_path, buf, buflen);
1080 EXPORT_SYMBOL(file_path);
1083 * vfs_open - open the file at the given path
1084 * @path: path to open
1085 * @file: newly allocated file with f_flag initialized
1087 int vfs_open(const struct path *path, struct file *file)
1089 file->f_path = *path;
1090 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1093 struct file *dentry_open(const struct path *path, int flags,
1094 const struct cred *cred)
1099 /* We must always pass in a valid mount pointer. */
1102 f = alloc_empty_file(flags, cred);
1104 error = vfs_open(path, f);
1112 EXPORT_SYMBOL(dentry_open);
1115 * dentry_create - Create and open a file
1116 * @path: path to create
1118 * @mode: mode bits for new file
1119 * @cred: credentials to use
1121 * Caller must hold the parent directory's lock, and have prepared
1122 * a negative dentry, placed in @path->dentry, for the new file.
1124 * Caller sets @path->mnt to the vfsmount of the filesystem where
1125 * the new file is to be created. The parent directory and the
1126 * negative dentry must reside on the same filesystem instance.
1128 * On success, returns a "struct file *". Otherwise a ERR_PTR
1131 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1132 const struct cred *cred)
1137 f = alloc_empty_file(flags, cred);
1141 error = vfs_create(mnt_idmap(path->mnt),
1142 d_inode(path->dentry->d_parent),
1143 path->dentry, mode, true);
1145 error = vfs_open(path, f);
1147 if (unlikely(error)) {
1149 return ERR_PTR(error);
1153 EXPORT_SYMBOL(dentry_create);
1156 * kernel_file_open - open a file for kernel internal use
1157 * @path: path of the file to open
1158 * @flags: open flags
1160 * @cred: credentials for open
1162 * Open a file for use by in-kernel consumers. The file is not accounted
1163 * against nr_files and must not be installed into the file descriptor
1166 * Return: Opened file on success, an error pointer on failure.
1168 struct file *kernel_file_open(const struct path *path, int flags,
1169 struct inode *inode, const struct cred *cred)
1174 f = alloc_empty_file_noaccount(flags, cred);
1179 error = do_dentry_open(f, inode, NULL);
1186 EXPORT_SYMBOL_GPL(kernel_file_open);
1188 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1189 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1191 inline struct open_how build_open_how(int flags, umode_t mode)
1193 struct open_how how = {
1194 .flags = flags & VALID_OPEN_FLAGS,
1195 .mode = mode & S_IALLUGO,
1198 /* O_PATH beats everything else. */
1199 if (how.flags & O_PATH)
1200 how.flags &= O_PATH_FLAGS;
1201 /* Modes should only be set for create-like flags. */
1202 if (!WILL_CREATE(how.flags))
1207 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1209 u64 flags = how->flags;
1210 u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1211 int lookup_flags = 0;
1212 int acc_mode = ACC_MODE(flags);
1214 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1215 "struct open_flags doesn't yet handle flags > 32 bits");
1218 * Strip flags that either shouldn't be set by userspace like
1219 * FMODE_NONOTIFY or that aren't relevant in determining struct
1220 * open_flags like O_CLOEXEC.
1225 * Older syscalls implicitly clear all of the invalid flags or argument
1226 * values before calling build_open_flags(), but openat2(2) checks all
1229 if (flags & ~VALID_OPEN_FLAGS)
1231 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1234 /* Scoping flags are mutually exclusive. */
1235 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1238 /* Deal with the mode. */
1239 if (WILL_CREATE(flags)) {
1240 if (how->mode & ~S_IALLUGO)
1242 op->mode = how->mode | S_IFREG;
1250 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1251 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1252 * O_TMPFILE below which requires O_DIRECTORY being raised.
1254 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1257 /* Now handle the creative implementation of O_TMPFILE. */
1258 if (flags & __O_TMPFILE) {
1260 * In order to ensure programs get explicit errors when trying
1261 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1262 * is raised alongside __O_TMPFILE.
1264 if (!(flags & O_DIRECTORY))
1266 if (!(acc_mode & MAY_WRITE))
1269 if (flags & O_PATH) {
1270 /* O_PATH only permits certain other flags to be set. */
1271 if (flags & ~O_PATH_FLAGS)
1277 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1278 * check for O_DSYNC if the need any syncing at all we enforce it's
1279 * always set instead of having to deal with possibly weird behaviour
1280 * for malicious applications setting only __O_SYNC.
1282 if (flags & __O_SYNC)
1285 op->open_flag = flags;
1287 /* O_TRUNC implies we need access checks for write permissions */
1288 if (flags & O_TRUNC)
1289 acc_mode |= MAY_WRITE;
1291 /* Allow the LSM permission hook to distinguish append
1292 access from general write access. */
1293 if (flags & O_APPEND)
1294 acc_mode |= MAY_APPEND;
1296 op->acc_mode = acc_mode;
1298 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1300 if (flags & O_CREAT) {
1301 op->intent |= LOOKUP_CREATE;
1302 if (flags & O_EXCL) {
1303 op->intent |= LOOKUP_EXCL;
1304 flags |= O_NOFOLLOW;
1308 if (flags & O_DIRECTORY)
1309 lookup_flags |= LOOKUP_DIRECTORY;
1310 if (!(flags & O_NOFOLLOW))
1311 lookup_flags |= LOOKUP_FOLLOW;
1313 if (how->resolve & RESOLVE_NO_XDEV)
1314 lookup_flags |= LOOKUP_NO_XDEV;
1315 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1316 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1317 if (how->resolve & RESOLVE_NO_SYMLINKS)
1318 lookup_flags |= LOOKUP_NO_SYMLINKS;
1319 if (how->resolve & RESOLVE_BENEATH)
1320 lookup_flags |= LOOKUP_BENEATH;
1321 if (how->resolve & RESOLVE_IN_ROOT)
1322 lookup_flags |= LOOKUP_IN_ROOT;
1323 if (how->resolve & RESOLVE_CACHED) {
1324 /* Don't bother even trying for create/truncate/tmpfile open */
1325 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1327 lookup_flags |= LOOKUP_CACHED;
1330 op->lookup_flags = lookup_flags;
1335 * file_open_name - open file and return file pointer
1337 * @name: struct filename containing path to open
1338 * @flags: open flags as per the open(2) second argument
1339 * @mode: mode for the new file if O_CREAT is set, else ignored
1341 * This is the helper to open a file from kernelspace if you really
1342 * have to. But in generally you should not do this, so please move
1343 * along, nothing to see here..
1345 struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1347 struct open_flags op;
1348 struct open_how how = build_open_how(flags, mode);
1349 int err = build_open_flags(&how, &op);
1351 return ERR_PTR(err);
1352 return do_filp_open(AT_FDCWD, name, &op);
1356 * filp_open - open file and return file pointer
1358 * @filename: path to open
1359 * @flags: open flags as per the open(2) second argument
1360 * @mode: mode for the new file if O_CREAT is set, else ignored
1362 * This is the helper to open a file from kernelspace if you really
1363 * have to. But in generally you should not do this, so please move
1364 * along, nothing to see here..
1366 struct file *filp_open(const char *filename, int flags, umode_t mode)
1368 struct filename *name = getname_kernel(filename);
1369 struct file *file = ERR_CAST(name);
1371 if (!IS_ERR(name)) {
1372 file = file_open_name(name, flags, mode);
1377 EXPORT_SYMBOL(filp_open);
1379 struct file *file_open_root(const struct path *root,
1380 const char *filename, int flags, umode_t mode)
1382 struct open_flags op;
1383 struct open_how how = build_open_how(flags, mode);
1384 int err = build_open_flags(&how, &op);
1386 return ERR_PTR(err);
1387 return do_file_open_root(root, filename, &op);
1389 EXPORT_SYMBOL(file_open_root);
1391 static long do_sys_openat2(int dfd, const char __user *filename,
1392 struct open_how *how)
1394 struct open_flags op;
1395 int fd = build_open_flags(how, &op);
1396 struct filename *tmp;
1401 tmp = getname(filename);
1403 return PTR_ERR(tmp);
1405 fd = get_unused_fd_flags(how->flags);
1407 struct file *f = do_filp_open(dfd, tmp, &op);
1419 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1421 struct open_how how = build_open_how(flags, mode);
1422 return do_sys_openat2(dfd, filename, &how);
1426 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1428 if (force_o_largefile())
1429 flags |= O_LARGEFILE;
1430 return do_sys_open(AT_FDCWD, filename, flags, mode);
1433 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1436 if (force_o_largefile())
1437 flags |= O_LARGEFILE;
1438 return do_sys_open(dfd, filename, flags, mode);
1441 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1442 struct open_how __user *, how, size_t, usize)
1445 struct open_how tmp;
1447 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1448 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1450 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1453 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1457 audit_openat2_how(&tmp);
1459 /* O_LARGEFILE is only allowed for non-O_PATH. */
1460 if (!(tmp.flags & O_PATH) && force_o_largefile())
1461 tmp.flags |= O_LARGEFILE;
1463 return do_sys_openat2(dfd, filename, &tmp);
1466 #ifdef CONFIG_COMPAT
1468 * Exactly like sys_open(), except that it doesn't set the
1471 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1473 return do_sys_open(AT_FDCWD, filename, flags, mode);
1477 * Exactly like sys_openat(), except that it doesn't set the
1480 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1482 return do_sys_open(dfd, filename, flags, mode);
1489 * For backward compatibility? Maybe this should be moved
1490 * into arch/i386 instead?
1492 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1494 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1496 if (force_o_largefile())
1497 flags |= O_LARGEFILE;
1498 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1503 * "id" is the POSIX thread ID. We use the
1504 * files pointer for this..
1506 static int filp_flush(struct file *filp, fl_owner_t id)
1510 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1511 "VFS: Close: file count is 0 (f_op=%ps)",
1516 if (filp->f_op->flush)
1517 retval = filp->f_op->flush(filp, id);
1519 if (likely(!(filp->f_mode & FMODE_PATH))) {
1520 dnotify_flush(filp, id);
1521 locks_remove_posix(filp, id);
1526 int filp_close(struct file *filp, fl_owner_t id)
1530 retval = filp_flush(filp, id);
1535 EXPORT_SYMBOL(filp_close);
1538 * Careful here! We test whether the file pointer is NULL before
1539 * releasing the fd. This ensures that one clone task can't release
1540 * an fd while another clone is opening it.
1542 SYSCALL_DEFINE1(close, unsigned int, fd)
1547 file = file_close_fd(fd);
1551 retval = filp_flush(file, current->files);
1554 * We're returning to user space. Don't bother
1555 * with any delayed fput() cases.
1559 /* can't restart close syscall because file table entry was cleared */
1560 if (unlikely(retval == -ERESTARTSYS ||
1561 retval == -ERESTARTNOINTR ||
1562 retval == -ERESTARTNOHAND ||
1563 retval == -ERESTART_RESTARTBLOCK))
1570 * sys_close_range() - Close all file descriptors in a given range.
1572 * @fd: starting file descriptor to close
1573 * @max_fd: last file descriptor to close
1574 * @flags: reserved for future extensions
1576 * This closes a range of file descriptors. All file descriptors
1577 * from @fd up to and including @max_fd are closed.
1578 * Currently, errors to close a given file descriptor are ignored.
1580 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1581 unsigned int, flags)
1583 return __close_range(fd, max_fd, flags);
1587 * This routine simulates a hangup on the tty, to arrange that users
1588 * are given clean terminals at login time.
1590 SYSCALL_DEFINE0(vhangup)
1592 if (capable(CAP_SYS_TTY_CONFIG)) {
1600 * Called when an inode is about to be open.
1601 * We use this to disallow opening large files on 32bit systems if
1602 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1603 * on this flag in sys_open.
1605 int generic_file_open(struct inode * inode, struct file * filp)
1607 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1612 EXPORT_SYMBOL(generic_file_open);
1615 * This is used by subsystems that don't want seekable
1616 * file descriptors. The function is not supposed to ever fail, the only
1617 * reason it returns an 'int' and not 'void' is so that it can be plugged
1618 * directly into file_operations structure.
1620 int nonseekable_open(struct inode *inode, struct file *filp)
1622 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1626 EXPORT_SYMBOL(nonseekable_open);
1629 * stream_open is used by subsystems that want stream-like file descriptors.
1630 * Such file descriptors are not seekable and don't have notion of position
1631 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1632 * Contrary to file descriptors of other regular files, .read() and .write()
1633 * can run simultaneously.
1635 * stream_open never fails and is marked to return int so that it could be
1636 * directly used as file_operations.open .
1638 int stream_open(struct inode *inode, struct file *filp)
1640 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1641 filp->f_mode |= FMODE_STREAM;
1645 EXPORT_SYMBOL(stream_open);